An Abstract of the Thesis of Greg Gaston for the Degree of Doctor of Philosophy in Physical Geography presented on 13 July 1993. Title: A Procedure for Estimating the Carbon Budget of Terrestrial Ecosystems in the Former Soviet Union Using Geographic Information System Analysis and Classification of AVHRR Derived Global Vegetation Index. Abstract approved: Philip L. Jackson The purpose of this research was to develop methods for using geographic information system (GIS) analysis and unsupervised classification of Global Vegetation Index (GVI) images to improve carbon budget estimates of the former Soviet Union (FSU). A GIS was used to locate the geographic distribution and estimate the area of agricultural lands in the FSU. Climatic factors were used to isolate agricultural lands suitable for no-till management. The effects of various tillage practices on the carbon cycle were estimated. Four year average, monthly maximum value GVI composites were clustered on the basis of timing, magnitude, and duration of "greenness" as recorded by the normalized difference of vegetation index (NDVI). The resulting 42 image classes were identified by a variety of quantitative and qualitative methods including expert assessment by scientists with extensive field experience in the FSU and analysis of temporal NDVI response curves. The image classes appear to accurately delineate vegetation/landcover regions. The percent agricultural land, percent forest cover and percent natural nonforest was estimated for each image class. Equivalent ecosystems described in the Bazilevich carbon data base were used to calculate the carbon cycle parameters forest and natural non-forest ecosystems. Carbon estimates for agricultural lands were calculated from production data. Comparisons between areal estimates of land cover from this research and reported statistics are very good. For example, forested land was estimated at 876 million hectares (MHa) forest statistical data from the FSU estimate "commercial" forests to be 814 MHa. Total agricultural lands are estimated in research to be 211 MHa, the same as reported by the USDA and very close to the 217 MHa reported by statistics from the FSU. Results from carbon budget analysis include: - Complete conversion of all climatically suitable lands to no-till management would sequester over 3 gigatons (Gt) of carbon in the agricultural soils of the FSU. - Phytomass and net primary productivity (NPP) estimates for forest and non-forest ecosystems compare well with carbon budget estimates that used maps to identify carbon quantifiable regions. A Procedure for Estimating the Carbon Budget of Terrestrial Ecosystems in the Former Soviet Union by Geographic Information System Analysis and Classification of AVHRR Derived Global Vegetation Index. by Greg G. Gaston A THESIS submitted to Oregon State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy Completed 13 July 1993 Commencement June 1994 APPROVED: Philip L. Jackson, Geosciences, in charge of major Chairman, Geosciences Department Dean of Graduate School Date thesis is Presented: 13 July 1993 Typed by the Author; Greg Gaston Acknowledgement I have had the opportunity of working for and with some of the finest and most supportive people in the world. I would like to thank my advisor Dr. Phil Jackson for his constant support and encouragement. Dr. Chuck Rosenfeld for (among other things) letting me fly backseat on air photo missions. Dr. Keith Muckleston for his expertise and support concerning the (now former) Soviet Union. Dr. Ted Vinson for bringing me on as a part of his team, and for his incredible ability and perseverance in navigating the minefields of paperwork necessary to obtain funding. His dedication, honesty and hard work have been an inspiration. I'd like to thank my Russian colleagues, especially, Dr. Tatayana Kolchugina, Dr. Marina Botch, and Dr. Kira Kobak, for opening the eyes of a cold warrior to the warmth and friendliness of the Russian people and for their invaluable assistance in completing this project. I wish to thank my co-authors who edited my drafts, translated from Russian, and provided invaluable insight into this research effort. I wish to thank my Daddy, Dr. Leonard Gaston who has been constantly supportive and encouraging in a multitude of ways. My Mother, Brother and Sister who also have been unfailing in their encouragement. I'd like to thank my dear wife who hung tough through the whole graduate school thing. Through it all she held down a job, kept me from going off the deep end, and gave birth to the two most beautiful sons a man could ask for. And finally I thank the Lord Jesus Christ, who started us down this road and has never turned away and I know that he will always be there. I also acknowledge the funding provided by cooperative Agreement (CR820239) between the U.S. Environmental Protection Agency (EPA) - Environmental Research Laboratory, Corvallis, Oregon and Oregon State University. Jeffrey J. Lee is the Project Officer for the project entitled "Carbon Cycling in Terrestrial Ecosystems of the Former Soviet Union." The work presented is a component of the U.S. EPA Global Climate Research Program, Global Mitigation and Adaptation program, Robert K. Dixon, Program Leader. The papers presented in this thesis have not been subjected to the EPA's review and, therefore, do not necessarily reflect the views of the EPA, and no official endorsement should be inferred. TABLE OF CONTENTS Introduction Estimating Changes in Carbon from Agricultural Management Remote Sensing and GIS in Carbon Budget Research Carbon Budget Estimates for the FSU Based on Classified Image Beyond Global Warming Combined References Potential Effect of No-Till Management On Carbon in the Agricultural Soils of the Former Soviet Union. Abstract Introduction Current Estimate of Soil Carbon in Agricultural Soils Purpose Continued Conventional Tillage Technical Management Options and Soil Carbon Defining Climatic Limits Potential Effect of Conservation Tillage on Soil Carbon Benefits and Problems Associated with No-Till Management Summary and Conclusions References Identification of Carbon Quantifiable Regions in the Former Soviet Union Using Unsupervised Classification of AVHRR Global Vegetation Index. Abstract Introduction An Appropriate Remote Sensing System Data and Methods Research Background GVI Data Used to Identify Carbon Quantifiable Regions Problems With Weekly GVI Composites Clustering and Classification Interpreting the Image Quantitative Tools Qualitative Tools Basic Geographic Patterns in the FSU The Image Classes Agricultural Land Discrepancies in Classification Summary and Conclusions Acknowledgements References TABLE OF CONTENTS (contd.) The Use of Global Vegetation Index (GVI) to Estimate Phytomass and Net Primary Productivity in Terrestrial Ecosystems of the Former Soviet Union. 96 Abstract 97 Introduction 99 A Carbon Budget Satellite Remote Sensing Systems 100 102 Approaches for the Use of AVHRR Data 106 Image-Map of the FSU Linking Phytomass and NPP Data to Image Classes 108 111 Estimates of Area Evaluating Phytomass and NPP Estimates 113 Phytomass and NPP 115 Summary and Conclusions Acknowledgement 117 118 References 127 Appendix 114 132 LIST OF FIGURES Figure Page Agricultural Soils of FSU Suitable for No-Till Management 41 2. Soil Organic Carbon Versus Depth 42 3. Classified Image of the FSU 83 1. 4. Temporal NDVI Response Curves for Class 11 and 16 84 5. Vegetation Map (Sochava,1960): Amur Region 85 6. Classified Image: Amur Region (Inset: Elevation) 86 7. Classified Image: South Central Asia (Inset: Elevation) 87 8. Comparison of Soil/Vegetation Map (Ryabchikov,1988) and Classified Image in Yamal Region 88 Classified Image: Kamchatka (Inset: Elevation) 89 9. 10. Classified Image: Kamchatka (Inset: Vegetation Map) 90 11. Comparison of the Signatures of Sample Image Classes 122 12. Area of Biomes Identified by Image Classes and Thematic Maps 123 13. Comparison of Phytomass Densities 124 14. Comparison of NPP Densities 125 15. Image Based NPP Estimates Compared to Fung et al. (1987) 126 LIST OF TABLES Table Page 1. Carbon Content Of Agricultural Soils of the FSU 43 2. Carbon Content of Agricultural Soils Climatically Suitable for No-Till Management 44 3. Soil Carbon Content Under No-Till Management 45 4. Percent Distribution of Soil Carbon 46 5. Maps Used as an Aid in Image Interpretation 76 6. A Brief Description of Image Classes 77 7. Arable Lands Located in Image Classes 79 8. Descriptions of Figures (Figure Captions) 80 9. Landcover, Phytomass and NPP Densities for Image Classes 119 10. Total Phytomass and NPP for Image Classes and Landcover Type 121 A Procedure for Estimating the Carbon Budget of Terrestrial Ecosystems in the Former Soviet Union by Geographic Information System Analysis and Classification of AVHRR Derived Global Vegetation Index. Introduction Natural Processes in the oceans and in terrestrial ecosystems, together with human activities, have caused a measurable increase in the atmospheric concentration of CO2. In 1988 the atmosphere contained 748 gigatons (Gt) of carbon, the largest amount during the last 160,000 years (Post et al., 1990). On average, the CO2 concentrations in the atmosphere are increasing by 3.0 Gt/yr (Keeling, 1983; Tans et al., 1990) The long-term ecological consequences of change in the chemical composition of the atmosphere are not fully understood; however, a warmer global climate is possible (PPIGW, 1992). If CO. concentrations were to double, the earth's temperatures could rise between 1 and 5° C (Schneider, 1990). Global warming may further disrupt the equilibrium of the natural carbon cycle by accelerating the rates of plant respiration (Keeling et al., 1989) and decay of organic matter (Dixon and Turner, 1991). It may be possible to manage terrestrial ecosystems to help offset increased amounts of atmospheric CO. (PPIGW, 1992). Before any global strategy for environmental management aimed at carbon mitigation can be formulated, accurate assessments of the current carbon budget at a national level must be made. The Former Soviet Union (FSU) was the largest country in the world, occupying one-sixth of the land surface of the earth. An understanding of the 2 pools and fluxes of biogenic carbon of the FSU is essential to the development of a global strategy aimed at mitigating the potential negative impacts of climate change. The three papers presented in this dissertation are focused on the use of a geographic information system (GIS) and satellite remote sensing data to improve carbon budget estimates for the FSU. The first paper concentrates on the changes in the carbon cycle brought about by conversion of land from natural ecosystems to agricultural production. GIS techniques were used to identify and measure agricultural lands climatically suited to no-till management. The effect of changing agricultural management on the carbon budget of agricultural soils was estimated. The second paper presents the methodology developed for using remote sensing information to accurately identify homogeneous regions with similar vegetation and land cover. These regions formed the basis for new estimates of carbon storage and flux in the terrestrial ecosystems of the FSU. The linkage of carbon data from vegetation data bases and calculated values for agricultural lands to image classes are presented in the third paper. A comparison of these carbon budget estimates with previous carbon budget budget estimates for the FSU is also presented in the third paper. Estimating Changes in Carbon from Agricultural Management A significant part of the increase in atmospheric carbon dioxide results from land use change. The reservoir of carbon stored in soils and vegetation has been estimated at 2000 Gt (Post et al., 1990). The amount of carbon stored in soil is greater than the combined carbon storage in vegetation and the atmosphere. Conversion of lands to agricultural production results in a sharp decrease in carbon stored in soil (Haas et al., 1957; Hobbs and Brown, 1965; Mann, 1986). Houghton et al. (1983) state that a significant percentage of previously stored soil carbon is released to the atmosphere upon conversion to agriculture. Further, they note that conversion of land to agricultural production worldwide is estimated to have released 180 Gt of carbon into the atmosphere. The first paper in this dissertation volume presents the results of the use of geographic information system (GIS) analysis to estimate the spatial distribution of arable lands in the FSU. Climatic factors (i.e growing degree days and moisture balance) were used to predict lands climatically suitable to no-till management. No-till management has been suggested as a way to increase the amount of atmospheric carbon sequestered in agricultural soils. Using the carbon distribution through a soil profile and equations predicting the change in soil carbon storage from no-till management an estimate was made for the changes in carbon emission/sequestration in agricultural soils resulting from changes in management. The results of this research indicate, that while additional atmospheric carbon could be sequestered in agricultural soils of the FSU by a conversion to no-till management, the potential change in carbon equilibrium may not be of major significance for global carbon cycle management. This research supported the methodology for manipulating, with the GIS, spatial data from mapped sources which were then linked to carbon data from field measurements. Remote Sensing and GIS in Carbon Budget Research The carbon cycle consists of a combination of pools and fluxes. The pools are carbon stored in soil and vegetation, including living vegetation (i.e. phytomass) and plant detritus (i.e. mortmass) . The effluxes are carbon emissions resulting from plant respiration and decomposition of organic matter. Influxes of carbon from the atmosphere into terrestrial ecosystems are represented by formation of new organic matter in soil and vegetation. The annual accumulation of carbon in vegetation after subtracting the carbon returned to the atmosphere through respiration is known as Net Primary Productivity (NPP). Carbon cycle parameters have been quantified by soil scientists, agricultural and forest scientists, ecologists and botanists for several decades. Carbon cycle parameters may be expressed in terms of carbon content (for pools) or rate (for influxes or effluxes). One approach for the estimation of a carbon budget is to extrapolate a limited number of actual field measurements to characterize the carbon cycle parameters for a specific ecoregion. An ecoregion has been defined as a region similar in soil, vegetation, and land cover characteristics (Omernik, 1987). This approach to carbon budget estimates assumes homogenous carbon quantifiable ecoregions regions can be identified based on the distribution of soil, and/or vegetation and land cover. If the soil,vegetation and land cover characteristics which comprise an ecosystem are accurately portrayed on the thematic maps which are used to isolate ecoregions, then the carbon budget for an ecoregion can be established simply by multiplying the area of the ecoregion (in hectares) by the contents content(s) and flux(es). The carbon contents and fluxes for all the ecoregions may be summed to arrive at the carbon budget for a larger region, or nation. Kolchugina and Vinson (1993) (and Vinson and Kolchugina, 1993) used this approach to create a framework to assess pools and fluxes of biogenic carbon in the FSU. Initially, maps of soil/vegetation complexes (Ryabchikov, 1988), wetlands (Isachenko, 1988), and arable lands (Cherdantsev,1961) were used to isolate ecoregions, and data bases which contain natural carbon cycle parameters were compiled. The areal coverage of the ecoregions was integrated with the carbon content and flux data bases to establish the carbon budget within the ecoregion. There were, however, concerns about the information content of the maps used to identify the carbon quantifiable ecoregions. Each map, at best "...is a snapshot of the situation seen through the particular filter of a given surveyor, in a given discipline at a certain moment in time"(Burrough, 1987). Cartographic data are generally the result of observations taken over a very short time, thus maps are static sources of information. There are problems with scale changes between maps, significant differences in the level of detail between maps, and changes in classification schemes between maps and across political boundaries. The placement of boundary lines between areas that grade into each other over long distances is an arbitrary decision of the individual cartographer. Additionally, the size and inaccessibility of much of the land area of the FSU necessitates cartographic generalization, introducing interpolation errors. There is wide agreement that satellite data can be used to overcome the limitations of thematic maps, providing timely, consistent and reliable information for the entire globe on the areal extent of, and conditions in, terrestrial ecosystems (Norwin and Greegor, 1983, Badwhar, 1986, Janssen et al., 1990, Justice et al.1985, Hall et al., 1991). However, satellite sensor systems record only reflected radiation from the surface. Surface conditions represent a complex mixture of vegetation, exposed soil or rock, water and shadows. The species composition, structure, density and levels of photosynthetic activity of both the dominant vegetation and understory vegetation are all a significant part of the signal recorded by satellite remote sensing systems (Spanner et el., 1990; Knipling, 1970). Other conditions such as the "brightness" or color of the exposed soil (rock) and the amount and density of shadows present are all integrated into a satellite observation of terrestrial ecosystems (Huete et al., 1985). Not only is the data recorded by a satellite sensor a result of complex surface interactions but many important ecosystem processes cannot be directly measured. Soil processes, litter accumulation, and mortmass accumulation for example are not part of the signal received by satellite remote sensing systems. In order to successfully use remote sensing information to assist in carbon budget estimates it was necessary to: 1. Identify the most appropriate sensor system and data set. 2. Develop a methodology that makes remote sensing information useful in carbon budget estimates. The sensor system selected was the Advanced Very High Resolution Radiometer (AVHRR). The AVHRR is a part of NOAA's meteorological satellite program which is designed to provide daily coverage of atmospheric conditions over the entire globe. The orbit and resolution of this sensor system while designed for meteorological data collection, has been successfully used to monitor conditions in terrestrial ecosystems (Tucker et al., 1983, Hayes, 1985, Townshend et al., 1987, and Fung et al.,1987). Specific characteristics of the spectral and spatial resolution of this sensor system and the reasons for selecting this sensor and data set are presented in the second paper in this volume. The data chosen for this research effort were monthly average Global Vegetation Index (GVI) composites from a four year period (1985-1988). Two basic approaches can be taken to the use of remote sensing data from satellite sensors to estimate carbon budget dynamics: 1. Direct quantitative calculation of ecosystem characteristics using remote sensing data as a primary indicator, or 2. Using the unique signal created by various ecosystem characteristics to identify homogenous vegetation/land cover regions. These regions should be an accurate representation of surface conditions which can be linked with plot level field studies to extrapolate carbon budget estimates. While various methods of direct assessment show some success (for example Tucker et al., 1981 and 1983. Goward et al. 1985 and 1987). There are enough questions that are not well answered about the relationships of terrestrial ecosystem processes and the signals recorded by satellite remote sensing systems (Badwhar et al., 1986 and Runyon et al., 1991 for example) that the second approach was taken for this research. The second paper also describes the methodology of using unsupervised classification of monthly GVI composites to create an image that delineates regions of similar vegetation and landcover in the FSU. The selection process of the most appropriate image and the methods used to identify each image class (the conversion from image class to "informational" class Swain and Davis, 1978) are an important part of the research results presented in this paper. Carbon Budget Estimates for the FSU Based on the Classified Image Each of the image classes have been described in terms of vegetation cover, agricultural land use and geographic location. Appendix one of this volume details the identification of each of the 42 image classes that characterize the vegetation and land cover of the FSU. We have assumed that each of the image classes represent homogenous vegetation and land cover conditions. Further, we have assumed that these homogenous areas can be described in terms of carbon content by linking them with a carbon data base. The carbon data base of N.I. Bazilevich (1986) was used to assign carbon values to the forest and natural non-forest ecosystems contained in each of the image classes. The Bazilevich data base is comprised of carbon accumulation data for 1500 vegetation complexes in the FSU. This data base is a comprehensive source of carbon information for all the vegetation formations in the FSU. The descriptions of the image classes and descriptions of vegetation communities in the Bazilevich data base were often slightly different. Expert assessment considering the geographic extent and regional characteristics in vegetation communities was necessary to assign the most appropriate carbon values to each image class. Accurate assignment of carbon values to each image class using of the data provided by the Bazilevich data base required the estimation of the individual components that make up the vegetation and land cover of each image class. For example, image class 35 is characterized as "forest meadow steppe, with significant agricultural land use". The forested portions have a very different carbon storage and flux from the meadow steppe vegetation, both of which are very different from the carbon cycle found in agricultural land. The percent forest cover for forests in the FSU averages 65% (Vorobyov, 1985). We have assumed that even the most dense of forests identified in an image class will contain non-forested patches that will make up at least 20% of the area identified as forest. When available, published forest cover was used to assign the percent forest cover to image classes identified as forest. For example, forest steppe, represented by image classes 32 and 35 average 15% forest cover according to Vorobyov (1985). In the absence of published data an expert assessment of forest density was made for each image class. This assessment considered the geographic location of image classes and the density of "greenness" as recorded by NDVI. The forest cover calculated for the image classes was 876 million hectares (MHa). This compares very well with forest statistical data (Alimov, 1989) which estimates "commercial forests" at 814 MHa. The non-forest areas in each image class were either considered agricultural lands or natural non-forest vegetation. If the natural non-forest vegetation types found in each image class were not specifically described in the Bazilevich data 10 base, carbon values were assigned that reflect the natural non-forest vegetation most likely to be found in each geographic location. For example, image class 11, Sparse Northern Taiga Larch was assigned a percent forest cover of 50%. As there is no arable land in this image class the non-forest vegetation was assigned phytomass and NPP values from "typical tundra" which most likely represent the character of the non forested vegetation in this region. The influence of agricultural land use on image classes was estimated using the methods outlined in the first paper in this dissertation volume. Again, good overall correlation was seen between total agricultural lands reported in agricultural statistics (USDA, 1990; and Sokolovskiy et al. 1989) and the total area of agricultural lands calculated in this research. and Carbon values for agricultural lands were calculated using production data and "harvest index" equations from Sharp et. al. (1976). The results presented in this final paper confirm the validity of the methods used to create the image classes. Comparisons of area calculated from image classes an available statistical data show very good correspondence. While it is not possible to compare carbon cycle estimates to "reality", a good correlation with other carbon budget estimates for the FSU was found. The other estimates of carbon budget of the FSU were made using very different sources of information to identify carbon quantifiable regions. Beyond Global Warming Most of the scientific literature in the last decade warns that global warming will result from increased levels of atmospheric carbon dioxide. Warming of global climate is projected to bring about an number of disasters including drastic sea level rise that will flood coastal populations and the virtual destruction of agriculture in the most productive parts of the world. In many cases it appears that some sort of social agenda or scientific fashion are driving the conclusions published by some scientists, for example, Houghton and Woodwell (1989) call for "radical" social and economic changes to "halt any further [climate] change". The scientific community is not, however, monolithic in this view. Schlesinger and Jiang (1991) suggest that much more study is needed before calling for massive social and economic change to possibly avert an uncertain potential climate change. Reifsnyder (1989) holds Aspect all models that call for CO2 induced warming over the entire globe. Recent data from Kahl et al. (1993) suggest in fact that rather than warming in the arctic regions, there has been a cooling trend over the past 40 years in the atmosphere over the arctic regions. Regardless of the effect (if any) on the global climate from increased levels of atmospheric CO, the research presented in this volume is useful and valid in a number of ways. At this time, no other research has used remote sensing data to identify landcover regions across the entire FSU. This research further validates the process of land cover mapping using coarse resolution remote sensing data. There are a number of opportunities to expand on these initial results. As higher resolution remote sensing data becomes available (or 12 cost effective) the image classes developed using GVI in this research could be used as a basis for further classification of regions and regional characteristics. Another significant possibilitiy, is the integration of these results into ecological process models that can be used to predict the effect of management activities on natural and managed ecosystems. The image/map produced by this research provides a baseline of vegetation and landcover conditions in the FSU that can be used to detect and assess future change in terrestrial conditions. Dramatic social and economic changes brought about by the fall of the Soviet empire make it very important to understand and inventory current ecosystem conditions. The results of this research help provide a solid base of geographic information which is essential for intelligent decisions regarding resource development. I Combined References Alimov, Y.P., I.V. Golovikhin, L.B. Zdanevich and I.V. Yunov (Eds.), 1989. Dynamics of Forests Under Forest Management Organization Regarding the Main Forest Forming Species in 1986-1988, U.S.S.R. State Forestry Committee, 159 pp. Amemiya M., 1977, Conservation Tillage in the Western Corn Belt, Jrnl. of Soil and Water Conservation, Jan.- Feb. 1977 pp.29-36. Anderberg, M.R., 1973, Cluster analysis for Applications. Academic Press, New York, New York, 345 pp. Arshad M.A., M. Schnitizer, D.A. Angers and J.A. Ripmeester, 1990, Effects of Till vs, No-Till on the Quality of Soil Organic Matter, Soil Biological Biochemistry, Vol. 22 No. 5 pp.595-599. Badhwar, G.D., R.B. MacDonald, and N.C. Mehta. 1986. Satellite-derived leaf area index and vegetation maps as input to global carbon cycle models - a hierarchical approach. International J. of Remote Sensing Vol.7, No.2, pp.265-281. Barnwell, T.O., R.B. Jackson, IV, E.T. Elliott, I.C. Burke, C.V. Cole, K. Paustaian, E.A. Paul, A.S. Donigian, A.S. Patwardhan, A. Rowell and K. Weinrich, 1992, An Approach to Assessment of Management Impacts on Agricultural Carbon, Journal of Water, Air & Soil Pollution (in press). Bazilevich, N.I. 1986. Biological productivity of soil-vegetation formations in the U.S.S.R. Bulletin of Academy of Sciences of the U.S.S.R. Geographical Series 2 pp.49-66. Blevins R.L., M.S. Smith, G.W. Thomas and W.W. Frye, 1983, Influence of Conservation Tillage on Soil Properties, Jrnl. of Soil and Water Conservation, May-June 1983 pp.301-305. Bohn H.L., 1982, Estimates of Organic Carbon in World Soils: II, Journal of Soil Science Society of Arizona, No. 4 pp.118-119. Brown L.R. and E.C. Wolf, 1984, Soil Erosion: The Quiet Crisis in the World Economy, World Watch, Paper 60, Sept. 1984 Bryant, J., 1978, Applications of Clustering in Multi-Image Data Analysis. Report #18 Earth observation division, NASA/Johnson Space Center, Houston, Texas, NASA Contract 9-4689-85. Burough, P.A., 1987, Principles of Geographical Information Systems for Land Resource Assessment. Claredon Press, Oxford, 193 pp. Carter M.R. and D.A. Rennie, 1982, Changes in Soil Quality Under Zero Tillage Farming Systems: Distribution of Microbial Biomass and Mineralizable C and N Potentials, Canadian Jrnl. of Soil Science, Vol. 62 pp.587-597. Cherdantsev, G.N., 1961, Map: Arable Land in the U.S.S.R. in 1954. From A Geography of the U.S.S.R. - Background to a Planned Economy by J.P. Cole, F.C. German, 290 pp. Chistobayev A.I., 1990, A State Farm in the Depths of the Non-Chernozem Zone, Soviet Geography, 1990 pp.151-159. Dick W.A., 1983, Organic Carbon, Nitrogen and Phosphorus Concentrations and pH in Soil Profiles as Affected by Tillage Intensity, Soil Science Society of America Jrnl., Vol. 47 pp.102-107. Dixon, R.K. and D.P. Turner. 1991. The global carbon cycle and climate change: responses and feedbacks from below-ground systems. Environmental Pollution 73 pp.245-262. Fenster C.R., 1977, Conservation Tillage in the Northern Plains, Jrnl. of Soil and Water Conservation, Jan.- Feb. 1977 pp. 37-42. Fung, I.Y., C.J. Tucker, and K.C.Prentice, 1987, Application of Advanced Very High Resolution Radiometer Vegetation Index to Study Atmosphere-Biosphere Exchange of CO, Journal of Geophysical Research, Vol. 92, No. D3, pp. 2999-3015 Gaston, G.G., T.P. Kolchugina, and T.S. Vinson, 1993, Potential Effect of No-Till Management on Carbon in the Agricultural Soils of the Former Soviet Union. Agriculture Ecosystems and Environment, (in press) Gebhardt M.R., T.C. Daniel, E.E. Schweizer and R.R. Almaras, 1985, Conservation Tillage, Science, Vol. 230 pp.625-630. Gervin, J.C., A.G. Kerber, R.G. Witt, Y.C. Lu, and R. Sekhon, 1985, Comparison of Level One Land Cover Classification Accuracy for MSS and AVHRR Data. International. J. of Remote Sensing, Vol. 6, No. 1, pp. 1271-1318 Glazovskaya M.A., 1972, Soils of the World, Vol. I, Moscow University Publishers, Moscow, 214 pp. Goward, S.N., B. Markham, D.C. Dye, W. Dulaney, and J. Yang, 1991, Derivation of Quantitative NDVI Measurements From AVHRR Observations. Remote Sensing of Env., Vol. 35, pp. 257-277 Goward, S.N., C.J. Tucker, and D.G. Dye, 1985, North American Vegetation Patterns Observed With the NOA4-7 Advanced Very High Resolution Radiometer. Vegetatio, Vol. 64, pp. 3-14 Goward, S.N., D. Dye, A. Kerber, and V. Kalb, 1987, Comparison of North and South American Biomass From AVHRR Observations. Geocarto International, Vol. 1, 1987. pp. 27-39. Haas H.J., C.E.Evans, E.F. Miles, 1957, Nitrogen and Carbon Changes in Great Plains Soils as Influenced by Cropping and Soil Treatments, U.S. Dept. of Agriculture Technical Bulletin 1164. Hall, F.G., D.B. Botkin, D.E. Strebel, K.D. Woods, and S.J. Goetz, 1991, Large Scale Patterns of Forest Succession as Determined by Remote Sensing. Ecology, Vol. 72, No. 2, pp. 628-690 Harmon, V. and M. Shapiro, 1992, Grass Tutorial: Image Processing. U.S. Army Construction Engineering Res. Lab., CERL, Champaign, Illinois. Hayes, L., 1985, The Current Use of TIROS-N Series Meteorological Satellites for Land Cover Studies. International. J. of Remote Sensing, Vol. 6, No. 1, pp. 35-40 Hinkle M.K., 1983, Problems With Conservation Tillage, Jrnl. of Soil and Water Conservation, May-June 1983 pp.201-206. Hobbs J.A. and P.L. Brown, 1965, Effects of Cropping and Management on Nitrogen and Organic Carbon Contents of a Western Kansas Soil, Kansas Agricultural Experiment Station Technical Bulletin 144. Holben, B.N., 1986, Characteristics of Maximum Value Composite Images From Multi-Temporal AVHRR Data. International. J. of Remote Sensing, Vol. 7, No. 11, pp. 1395-1434 Houghton R.A., J.E. Hobbie,J.M. Melillo, B. Moore, B.J. Peterson, G.R. Shaver and G.M. Woodwell, 1983, Changes in Carbon Content of Terrestrial Biota and Soils Between 1860 and 1980: A Net Release of CO2 to the Atmosphere, Ecological Monographs Vol. 53 No. 3 pp.235-262. Houghton, R.A. and G.M. Woodwell, 1989, Global Climate Change. Scientific American, Vol. 260 No. 4 pp.36-44 Huete A.R., R.D. Jackson and D.F. Post. 1985. Spectral Response of a plant canopy with different soil backgrounds. Remote Sensing of Env. Vol 17, pp. 37-53. IPCC, 1992,1992 Supplemental Report of AFOS Working Group, IPCC Secreteriate, Geneva, Switzerland (in press). Isachenko, A.G., ed. 1988. Landscape Map of the U.S.S.R. Institute of Geography, Leningrad State University. Janssen, L.F., M.N. Jaarsma, and E.T.M. Vanderlinden, 1990, Integrating Topographic Data With Remote Sensing for Land Cover Classification. Photogrammetric Engineering and Remote Sensing, Vol. 65, No. 11, pp. 1503-1506 1 Jenkinson D.S., 1991, The Rothamsted Long-Term Experiments: Are They Still of Use?, Agronomy Journal, Vol. 83 pp.2-10. Justice, C.O., J.R.H. Townshend, B.N. Holben, and C.J. Tucker, 1985, Analysis of the Phenology of Global Vegetation Using Meteorological Satellite Data. International J. of Remote Sensing, Vol. 6, No. 8, pp. 1271-1318 Kahl, J.D., DJ. Charlevoix, N.A. Zaitseva, R.C. Schnell, and M.C. Serrez, 1993, Absence of Greenhouse Warming Over the Arctic Ocean in the Past 40 Years. Nature Vol. 361 pp. 335-337 Keeling, C.D., R.B. Bacastow, A.F. Carter, S.C. Piper, T.P. Whorf, M. Heimann, W.G. Mook, and H. Roeloffzen. 1989. A three-dimensional model of atmospheric CO2 transport based on observed winds: 1. Analysis of observational data. Geophysical Monographs, Vol. 55, pp. 165 - 236. Keeling, C.D., 1983. The Global Carbon Cycle: What we know and could know from atmospheric, biospheric and oceanic observations. Proceedings of Carbon Dioxide Research Conference: Carbon Dioxide, Science and Concencesus, Berkley Springs, West Virginia, DOE Conf. - 820970, U.S. Dept of Energy, Washington, D.C. Kern J.S. and M.G. Johnson, 1991, Impact of Conservation Tillage Use on Soil and Atmospheric Carbon in the Contiguous United States, EPA/600/3-91/056, Env. Res. Lab, Corvallis OR. Kidwell, M., 1990, Global Vegetation Index Users Guide, U.S. Dept. of Commerce, NOAA, NESDES, NCDC, SDSD, Washington, D.C. Kimes, D.S., B.L. Markham, C.J. Tucker, and J.E. McMurtrey. 1981. Temporal relationships between between spectral response and agronomic variables of a corn canopy. Remote Sensing of Env. No. 11, pp. 410-411. Knipling, E.B., 1970. Physical and Phisiological basis for the reflectance of visible and near-infrared radiation from vegetation. Remote Sensing of Env., No.1 pp.155-159. Kobak K.I. and N. Yu. Kondrashova, 1986, Distribution of Organic Carbon in Soils of the Globe, Trudy GGI, Vol. 320 pp.61-76. Kobak K.I., 1988, Biotical Compounds of Carbon Cycle, Hydrometeoizdat, Leningrad, 248 pp. Kolchugina, T.P. and T.S. Vinson, 1993, Equilibrium Analysis of Carbon Pools and Fluxes of Forest Biomes in the Soviet Union. Canadian Journal of Forest Research, Vol. 23, pp. 81-88 Kononova M.M., 1963, Soil Organic Matter, Moscow, Nauka, 314 pp. Loveland, T., J.W. Merchant, D.O. Ohlen, and J.F. Brown, 1991. Development of a Land-Cover Characteristics Data Base for the Conterminous United States, Photogrammetric Engineering and Remote Sensing, Vol. 57, No. 11, pp. 1453-1463 Lydolph, P., 1970, Geography of the USSR. Wiley and Sons, New York, New York. 667 pp. Mann L.K., 1986, Changes in Soil Carbon Storage After Cultivation, Soil Science, Vol. 142 No. 5 pp.279-288. Mann L.K., 1985, A Regional Comparison of Carbon in Cultivated and Uncultivated Alfisols and Mollisols in the Continental United States, Geoderma, Vol. 36 pp.241-253. Medvedev Z.A., 1987, Soviet Agriculture, Norton and Co., New York, 464 pp. Milligan, G.W. and M.C. Cooper, 1985, An Examination of Procedures For Determining the Number of Clusters in a Data Set. Psychometrika, Vol. 50, No. 2, pp. 159-179 Nalivkin, D.V., 1973, Geology of the USSR. University of Toronto Press, Toronto, 855 pp NOAA-EPA, Global EcoSystems Data Base Version 1.0: Users guide, Documentation and Digital Data on C.D. Rom, USDOGNOAA, NCDC, Boulder, Colorado Norwin, J. and D.H. Greegor, 1983, Vegetation Classification Based on AVHRR Satellite Imagery. Remote Sensing of Env., Vol. 13, pp. 69-87 Omernik, J.M., 1987. Ecoregions of the conterminous United States. AAG Annals. Vol. 77, No.1 pp. 118-125. Panel on Policy Implications of Greenhouse Warming (PPIGW). 1992. Policy Implications of Greenhouse Warming - Mitigation, Adaptation, and the Science Base, Committee on Science, Engineering, and Public Policy, Natl. Academy of Sciences, Natl. Academy of Engineering, Inst. of Medicine,National Academy Press, Washington, D.C., Parker, W.H., 1983, The World's Landscapes: The Soviet Union, Longman Publishing Co.. London, England, 207 pp. Phillips R.E., R.L. Blevins, G.W. Thomas, W.W. Frye and S.H. Thomas, 1980, No-Tillage Agriculture, Science, Vol. 208 pp.1108-1113. Post, W.M., W.R. Emanuel, P.I. Zinke, and A.G. Stagenberger. 1982. Soil carbon pools and world life zones. Nature 298(5870) pp. 156-159. Post W.M., T.-H. Peng, W.R. Emanuel, A.W. King, V.H. Dale and D.L. DeAngelis, 1990, The Global Carbon Cycle, American Scientist, Vol.78 pp.310-326. Priputina I.V., 1989, Lowering of the Humus Content of Chernozem Soils of the Russian Plain as a Result of Human Action, Soviet Geography, 1989 pp.759-762. Reifsnyder,W.E., 1989, A Tale of Ten Fallacies: The Skeptical Enquirer's View of the Carbon Dioxide/Climate Controversy. Agricultural and Forest Meterology, No. 47 pp. 349-371 Runyon, J., R.H. Waring, R.W. McCrieght. 1991. Assessing the impact of climatic stress on forest production. in Proceedings of the International Workshop on Carbon Cycling in Boreal Forest and Sub-Arctic Ecosystems. T.S. Vinson and T.K. Kolchugina eds. Corvallis, Oregon. Ryabchikov, A.M. (ed.), 1988, Map: Geographical Belts and Zonal Types of Landscapes in the World. School of Geography, Moscow State University, Moscow, Russia SAS/STAT 1989, User's Guide, Version 6, Fourth Ed. Vol. 1. Cary, North Carolina: SAS Institute Inc., 1989, 943 pp. Schlesinger, M.E., and X. Jiang, 1991, Revised Projection of Greenhouse Warming. Nature, Vol.350 No.6315 pp.219-221. Schlesinger, W.H., 1977, Carbon Balance in Terrestrial Detritus, Ann. Rev. Ecolog. Syst., No. 8 pp.51-81. Schneider, S.H. 1990. Global Warming. Vintage Books, New York.. Schneider,S.H., 1989, The Greenhouse Effect: Science and Policy, Science, No.243 pp.771-781. Sochava, V.B. (ed.), 1960, Map: Vegetation of the USSR. Ministry of the Interior, Moscow, Russia Sokolovskiy, V.G. ed., 1989. Report on the State of the Envorinment in the USSR:1988. State Committee for the Protection of Nature. Moscow, 184 pp. Spanner, M.A., L.L. Pierce, D.L. Peterson, and S.W. Running, 1990, Remote Sensing of Temperate Coniferous Forest Leaf Area Index: The Influence of Canopy Closure, Understory Vegetation and Background Reflectance. International J. of Remote Sensing, Vol. 11, No. 1, pp. 95-111 Swain, P.H. and S.M. Davis, Ed., 1978, Remote Sensing: The Quantitative Approach. McGraw Hill, New York, New York. 385 pp. Symons L.S., 1972, Russian Agriculture: A Geographic Survey, J.Wiley and Sons, New York, 348 pp. Tans,P.P., I.Y. Fung, and T.Takahash. 1990. Observational constraints of the global atmospheric CO2 budget. Science No. 247 pp. 1431-1438 Touchton J.T., R.R. Sharp and D.W. Reeves, 1989, Tillage Systems for Double Cropped Wheat and Soybeans, Applied Agricultural Research, Vol. 4 No. 4 pp. 264-269. Townshend, J.R.G. and C.J. Tucker, 1984, Objective Assessment of AVHRR Data for Land Cover Mapping, International J. of Remote Sensing. Vol. 5, No. 2, pp. 497-504 Townshend, J.R.G. and C.O. Justice, 1986, Analysis of the Dynamics of African Vegetation Using the Normalized Difference Vegetation Index. International J. of Remote Sensing, Vol. 7, No. 11, pp. 1435-1445 Townshend, J.R.G., C.O. Justice, and V. Kalb, 1987, Characteristics And Classification of South American Land Cover Types Using Satellite Data. International J. of Remote Sensing, Vol. 8, No. 8, pp. 1189-1207 Tucker, C.J., B.N. Holben, J.H. Eglin, and J.E. McMurtrey. 1981. Remote Sensing Total Dry Matter Accumulation in Winter Wheat. Remote Sensing of Env. No.11 pp. 171-189. Tucker, C.J., C. Van Praet, E. Boerwinkle, and A. Gaston, 1983, Satellite Remote Sensing of Total Dry Matter Production in the Senegalese Sahel. Remote Sensing of Env., Vol. 13, pp. 461-474 Tucker, C.J. and P.J. Sellars, 1986, Satellite Remote Sensing of Primary Production. International J. of Remote Sensing, Vol. 6, No. 1, pp. 1395-1415 Tucker, C.J., J.R.G. Townshend, T.E. Goff, 1985, African Land Cover Classification Using Satellite Data. Science. Vol. 223, No.4685, pp. 369-375 U.S. Army Civil Engineering Research Lab, 1992, GRASS, Geographical Resource Analysis Support System, Program Documentation. US Army CERL, Champaign, Illinois U.S. Dept. of Agriculture, 1990, USSR Trade and Agriculture Report: Situation and Outlook. USDA, RS-90-1, May 1990 Unger P.W., 1991, Organic Matter, Nutrient and pH Distribution in No- and Conventional-Tillage Semi-Arid Soils, Agronomy Journal, Vol. 83 pp.186-189. Unger P.W., 1968, Soil Organic Matter and Nitrogen Changes During 24 Years of Dryland Wheat and Cropping Practices, Soil Science Society of America Proceedings, Vol. 32 pp.427-429. USSR Atlas of Agriculture, 1960, Moscow, 318 pp. Vinson, T.S., T.K. Kolchugina, Pools and Fluxes of Biogenic Carbon in the Former Soviet Union. Journal of Water, Air and Soil Pollution. (accepted for publication). Vinson, T.S., T.P. Kolchugina, R.K. Dixon, P.M. Bradley and G.G. Gaston, 1992, Proceedings of the IPCC AFOS Workshop, University of Joensuu, Joensuu, Finland, May 11-15, 1992. Vorobyov, G.I., ed. 1985. Forest Encyclopedia. Moscow. Sovetskaya Ecyclopedia Press, POTENTIAL EFFECT OF NO-TILL MANAGEMENT ON CARBON IN THE AGRICULTURAL SOILS OF THE FORMER SOVIET UNION by Greg G. Gaston Graduate Research Assistant Tatyana Kolchugina Visiting Research Associate and Ted S. Vinson Professor Oregon State University Corvallis, OR 97331 Submitted to Agriculture, Ecosystems & Environment September 21, 1992 Potential Effect of No-Till Management on Carbon in the Agricultural Soils of the Former Soviet Union Abstract Agricultural soils act as both a source and a sink for atmospheric carbon. Since the onset of cultivation, the 211.5 million hectares of agricultural soils in the former Soviet Union (FSU) have lost 10.2 gigatons of carbon. No-till management represents a promising option to increase the amount of carbon sequestered in the agricultural soil of the FSU. No-till management reduces erosion and sequesters additional carbon in the soil by lowering the soil temperature and raising soil moisture. To determine the carbon sequested under no-till management, a data base containing precultivation estimates of soil carbon for the seven major classes of soil found in the agricultural areas of the FSU was used to establish an equilibrium carbon content for each soil. Other published data provided a method to quantify the change in soil carbon brought about by converting to no-till management. Soils suitable for no-till management were analyzed and estimates of changes in carbon storage were made. No-till management is not suitable in areas where crop production is limited by cold, wet soils. Based on the results of a geographic information system analysis using maps of climatic factors and soil characteristics, 181 million hectares in the FSU were identified as climatically suitable for no-till management (almost 86% of all agricultural land). Complete conversion of all climatically suitable land to notill management would sequester 3.3 gigatons of carbon. This represents a 10% 25 increase in carbon in the agricultural soils of the FSU. This estimated accumulation of carbon is associated with a new soil carbon equilibrium condition. Accumulation of carbon in the soil produced by a conversion from conventional to no-till management is expected to take at least 10 years. The carbon accumulation produced by conversion to no-till management is not a continuing process; once a new no-till equilibrium condition has been reached, additional quantities of atmospheric carbon will not be sequestered in agricultural soils through continued no-till management. Key Words: Agricultural Soils, No-Till Management, Carbon, Carbon Sequestration, Former Soviet Union Introduction The reservoir of carbon stored in soils and vegetation has been estimated at 2000 gigatons (Post et al., 1990). The amount of carbon stored in soil is greater than the combined carbon storage in vegetation and the atmosphere. Conversion of lands to agricultural production results in a sharp decrease in carbon stored in soil (Haas et al., 1957; Hobbs and Brown, 1965; Mann, 1986). Houghton et al. (1983) state that a significant percentage of previously stored soil carbon is released to the atmosphere upon conversion to agriculture. Further, they note that conversion of land to agricultural production worldwide is estimated to have released 180 gigatons of carbon into the atmosphere. In 1988 the cultivated area of the former Soviet Union (FSU) was 211.5 million hectares (USDA, 1990). This huge area is almost twice the cultivated area of the United States (120 min ha.) and over four times the area of Canadian agricultural lands (46 min ha.). Clearly, agroecosystems in the FSU play a significant role in the global carbon cycle (Schneider 1989). A number of management strategies have great potential for conserving carbon in agricultural soils. By reducing soil erosion and in some cases stabilizing the amounts of carbon input into the soil, management practices such as ridge tillage, contour cropping, stubble mulching, and shelter belts all serve to reduce the loss of carbon from agricultural areas (IPCC, 1992; Barnwell, 1992). It has been estimated that 50% of the carbon transported from agricultural soils by erosion eventually ends up in the atmosphere (Kern and Johnson, 1991). However, when considering addtional storage of atmospheric carbon in agricultural soils, fewer management options are available. Growing cover crops for green manure, irrigation of dry lands, addition of carbon from outside sources (municipal sewage sludge, for example), and conservation tillage have all been suggested as options for sequestering atmospheric carbon in agricultural soils. Kern and Johnson (1991) indicate that of the possible conservation tillage options "... only no-till significantly results in increased [carbon] sequestration." Purpose The purpose of the work reported herein is to assess the potential effect of no- till management on carbon in the agricultural soils of the FSU. The scope of work includes: (1) an assessment of the carbon contents currently found in the agricultural soils of the FSU, (2) a consideration of climatological factors that would limit the use of no-till management, (3) a spatial analysis of the effect of limiting climate factors on agriculturally productive areas, and (4) an estimate of changes in soil carbon brought about by the introduction of no-till management to suitable agricultural areas. The findings of this research have implications for policy development. Given the policy development potential, a discussion of the problems associated with implementing no-till management in the FSU is presented. Current Estimates of Soil Carbon in Agricultural Soils To investigate the potential effects of various management strategies on carbon sequestration in agricultural soils, current carbon contents in the soils must be established. Current carbon contents can be established knowing the area of agricultural lands in the FSU, identifying the soil types within the agricultural lands, and assessing the effect of long-term cultivation on the soil carbon in the agricultural lands. Agricultural lands of the FSU were located using a map of arable lands produced by Cherdantsev (1961) (Figure 1). The location and extent of agricultural lands described and mapped in other available sources (Symons, 1972; Medvedev, 1987; Lydolph, 1970) appeared to correlate well with the first three classes of the arable lands map (class 1, < 60% arable; class 2, 30-60% arable; class 3, >30% arable). Using an average percentage for each class (i.e. 80%, 45%, and 15%) the total area of agricultural lands in the FSU was calculated to be 210 million hectares. This compares very favorably with the 211.5 million hectares reported in USDA statistics (USDA, 1990). The areal extent of various soil types within the limits of arable land were established using a soil/vegetation association map produced by Moscow State University (Ryabchikov, 1988). The arable lands and soil/vegetation maps were converted to a computer compatible format by manual digitization. A spatial analysis of the maps was performed with a geographic information system (GIS) and the area of agricultural land and corresponding soil types were identified. The results of this effort are presented in Table 1. Kobak (1988) compiled a data base of soil carbon parameters from 70 Soviet and international sources (e.g., Kononova, 1963; Schlesinger, 1977; Bohn, 1978; Post et al., 1982; Kobak and Kondrashova, 1986). Average soil carbon cycle parameters were presented for over 40 soil types in the FSU. The soil types in Kobak's data base were related to the soil/vegetation complexes defined on the soil/vegetation association map. Based on Kobak's data, carbon contents were assigned to each soil type. The carbon contents were assumed to be representative of an initial or precultivation condition. These data are presented in Table 1 (column 3). The initial or precultivation carbon pools are the product of the area for a given soil type and the soil carbon content. The results are presented in Table 1 (column 4). Conversion of soils to agricultural production releases carbon to the atmosphere. The rate of carbon release is relatively high after the initial cultivation of virgin soil, but the rate decreases with time (Houghton et al., 1983; Haas et al., 1957). The greatest rates of change in carbon storage occur in the first 20 years of cultivation. Mann (1986) used published data from 625 paired soil samples to develop regression equations that allow one to predict current carbon contents of cultivated soil as a function of the initial (precultivation) carbon contents. He found most soils lost at least 20% of their organic carbon after cultivation. Further, the loss of soil carbon content was strongly dependent on the length of time a soil had been cultivated and the initial carbon content. For soils under cultivation for 70- 100 years the current organic carbon present in the soil is given by the following equation: SOC, = (SOC; * 0.76) + 0.88 (in t/ha) (1) (R = 0.77, n = 129) in which, SOC, = final or current soil organic carbon SOC, = initial carbon content of the soil. Equation 1 was applied to each initial soil carbon content and an estimate was made of current soil organic carbon in the agricultural soils of the FSU. These estimates are presented in Table 1 (columns 5 and 6). The current estimate of soil carbon in agricultural soils of the FSU is 32.4 gigatons. The productive chernozem soils account for more than 60% of this total. On average there has been a 24% decline in soil carbon and a total loss of 10.2 gigatons of carbon since the onset of cultivation. Continued Conventional Tillage It is generally assumed that over time cultivated soils reach a new equilibrium at a lower level of organic carbon (Hobbs and Brown, 1965; Haas et al., 1957; Unger, 1968; Mann, 1985). The time required to reach equilibrium and the carbon content under the new equilibrium condition depends on the soil type, climatic conditions and agricultural management practices (i.e., crop rotations, tillage, fertilizer usage, etc.). Long-term investigations of soil properties at the Rothamsted (England) Experiment Station indicate soil organic carbon has long term stability. Jenkinson (1991) studied the soil carbon content of unfertilized test plots that have been cultivated for over 150 years and observed that organic carbon appears to 3 reach equilibrium after approximately 45 years. However, most researchers report that soils cultivated less than 100 years tend to show a continued slow loss of soil carbon to the atmosphere (Hobbs and Brown, 1965; Haas, 1957; Unger, 1968). It is difficult to predict the slow rates of soil carbon loss for soils approaching a new equilibrium condition. Since 1913 there has been no significant expansion of agricultural lands. Considering this long period of cultivation it is reasonable to assume, notwithstanding ongoing erosion, that an equilibrium condition has been reached. An exception to this assumption is the 33 million hectares (approximately 15% of all agricultural land) of the so called "virgin lands", which were first converted to agricultural use in the mid 1950's (Medvedev, 1987). At this time it is not possible to quantify the condition of soil carbon in the "virgin lands". However, these soils should continue to lose carbon to the atmosphere up to the beginning of the next century. Technical Management Options and Soil Carbon The type of agricultural management employed can significantly effect the equilibrium level of organic carbon in soil. Conventional tillage practices encourage the oxidation of soil organic carbon by mixing crop residues into the soil, adding air into soil macropores, and exposing the soil to direct solar radiation. Large amounts of available food in an oxygen rich environment which is heated by solar radiation produce higher metabolic rates in soil microbial communities, thus increasing the efflux of carbon from the soil. Erosion rates for conventional tillage can be quite high. The soil is usually completely inverted, the structure broken down through secondary tillage operations, and then left totally exposed to the action of wind and water for extended periods of time. One approach suggested for reducing the rates of carbon release from agricultural soils is conservation tillage. Conservation tillage management practice leaves most or all of the crop residues on the soil surface. Conservation tillage is generally defined as tillage practices that leave more than 30% of the crop residue on the soil surface. No-till management leaves virtually 100% of the organic residues undisturbed on the surface of the soil. No-till is the most extreme form of conservation tillage where all residues are left on the surface and the soil is disturbed only the minimum amount necessary to seed the next crop. A number of benefits accrue from this management strategy including lower rates of carbon oxidation and erosion. The cover of crop residue left on the surface of the soil acts as a mulch, reducing soil temperatures and increasing soil moisture. Cool moist soils have been shown to have significantly slower rates of carbon loss and in some cases act as a long term sink for atmospheric carbon. Conservation tillage reduces the frequency and magnitude of mechanical disturbance. This, in turn, reduces the creation of air filled soil macropores and slows the rate of carbon oxidation. Further, less mixing of crop residue into the soil, reduces the rate of biological and chemical interactions (Dick, 1983). No-till management has been shown to significantly reduce the rates of soil erosion (Phillips et al., 1980). Reduction in the rate of erosion tends to preserve the upper carbon rich layers of the soil. It has been estimated that up to 50% of the 33 soil organic carbon removed by erosion is later oxidized and enters the atmosphere (Kern and Johnson 1991). Defining the Climatic Limits The existence of climatic limits for no-till management in the northern areas of the United States have been reported by a number of authors (Phillips et al., 1980; Fenster, 1977; Gebhardt et al., 1985). In general terms, those areas where crop production is limited by cold wet soils and short growing seasons are considered unsuitable for no-till management. It is reasonable to assume, based on the latitudinal extent of most of the agricultural areas of the FSU (the most southern extremes of agricultural land in the FSU are roughly equivalent to the northern Great Plains of the United States), that a significant portion of the agricultural land in the FSU will not be climatically suitable for no-till management. The agricultural areas of the FSU are characterized by a strong continental climate with long cold winters and short warm summers. In significant portions of the FSU agricultural production is limited by short growing seasons and high soil moisture that restricts tillage operations in the spring (Medvedev, 1987). In the northwestern parts of the FSU "the intense winter cold ... (is) less important than the shortness of the summers and the lateness of spring..." (Symons, 1972). It is reasonable to assume that areas where crop production is limited by low soil temperatures, low rates of evapotranspiration, relatively high winter precipitation, and a short growing season will not be suited for no-till management, an agricultural management practice that produces cooler, moister soil. A standard unit of measurement for estimating the heat balance of agricultural lands is the growing degree-day. Data from Symons (1972) concerning the climatic production limits for various crops in the FSU suggest that areas of less than 1600 growing degree-days (above 10°C or 50°F) would be too cool for no-till manage- ment. Maps with climatic factors for the FSU are contained in the USSR Atlas of Agriculture (1960). Based on these maps, areas with less than 1600 growing degree-days were removed from further consideration. Areas having a significant moisture surplus were assumed to be located in regions with a precipitation to potential evapotranspiration ratio greater than 1.33. These areas were assumed to be generally unsuited to no-till management and were also removed. Generally, the areas eliminated by these assumptions were characterized in descriptive terms as poorly drained, and limited in agricultural production by cold wet soils and short growing seasons (Symons, 1972; Medvedev, 1987). The areas of each soil type that were included within the general agro-climatic limits suitable for no-till management are summarized in Table 2 (column 3). The results presented indicate the amount of agricultural land suitable for no-till management is 181 million hectares. This represents a reduction of 29 million hectares (or almost 14% of the total agricultural lands) when the climate limits previously discussed were applied. Figure 3 shows the location and extent of arable lands (>30% arable) and the portions of those lands suitable for no till management in the PSi T Potential Effect of Conservation Tillage on Soil Carbon An increase in soil organic carbon has been reported in agricultural soils where conservation tillage has been practiced (Unger, 1968 and 1991; Arshad et al, 1990; Blevins et al., 1983; Dick, 1983), with few exceptions (e.g., Carter and Rennie, 1982). Further, a significant change in the distribution of organic carbon in the soil profile occurs with conservation tillage. Reduced mixing of crop residues into the soil column produces a high concentration of organic carbon at or very near the soil surface. Kern and Johnson (1991) developed regression equations predicting increases in organic carbon resulting from no-till management in agricultural soils of the United States. Their equations were based on published research results from 15 different studies of no-till management. Their work resulted in two general equations for predicting the response of soil carbon to the no-till management option. Research results concerning the effects of conservation tillage short of complete no-till (such as ridge tillage and mulch tillage) on soil carbon were inconclusive. Therefore, in the present study the effect of conservation tillage on soil carbon has been restricted to the no-till management option. In the uppermost eight centimeters of the soil column the response of soil carbon to no-till is predicted by the following equation (Kern and Johnson, 1991): SOC, = (1.283 * SOCC) + .0510 (in t/ha) (2) (R = 0.75, n = 15) in which, SOC, = the current soil organic carbon (re. equation 1) SOC , = the soil carbon equilibrium condition for the no-till management option. In the 8-30 centimeter portion of the soil column the rate of carbon accumulation is less under no-till management because the tillage operations no longer transport organic matter from the surface into the lower portions of the soil profile. The equation for this portion of the soil column is (Kern and Johnson, 1991): SOCK, = (1.16 * SOCC) - 0.018 (in t/ha) (3) (R = 0.89, n = 34) Below 30 centimeters, no significant differences in the rates of carbon accumulation or loss were observed between no-till management and conventional tillage practices. To use equations 2 and 3 it was necessary to have carbon profiles for each soil type and distribute the carbon contents through each profile. Organic carbon profiles for soil types in the FSU reported by Glazovskaya (1972) were used to calculate the percent distribution of carbon in each soil profile. These profiles and percent distribution of carbon with depth are presented in Figure 2 and Table 4, respectively. The distributed organic carbon contents and the current soil carbon equilibrium (SOCC from Table 1 (column 4) for each soil type were used with equations 2 and 3 to predict potential carbon sequestration for each agricultural soil type in the FSU. The changes in soil carbon for each soil type are presented in Table 3 (columns 46). The carbon pool for a no-till management scenario is the product of the climatically suitable area of each soil type, the appropriate percent distribution of carbon in the soil profile, and the coefficients in equations 2 and 3. Comparison of the estimated current carbon pool for climatically suitable lands now under conventional cultivation (Table 2 (column 7)) and the no-till carbon pool (Table 3 (column 8)) suggests that a complete conversion of all climatically suitable land to a no-till management practice would sequester 3.3 gigatons of carbon (Table 3 (column 9)). The overall increase in carbon would be slightly more than 10% of the total carbon presently stored in agricultural soils. This estimated accumulation of carbon in the soil represents the new equilibrium for carbon in agricultural soils under no-till management. Sequestration of additional atmospheric carbon in the soil produced by a conversion from conventional to no-till is expected to take at least 10 years (Kern and Johnson, 1991). Accumulation of carbon in the soil produced by this change in management strategy is not a continuing process; once a new carbon equilibrium has been reached under no-till management no additional atmospheric carbon will be sequestered by continued no-till management (IPCC, 1992). Benefits and Problems Associated with No-Till Management No-till management represents a useful agricultural management option that when properly applied produces a number of benefits. Aside from the long-term benefits of carbon sequestration, conservation tillage has been proven to reduce soil erosion. The amount of soil that is saved by implementing conservation tillage was estimated to be from 30 to 90% (Amemiya, 1977; Phillips et al., 1980; Gebhardt, 1985). While accurate data for soil erosion are not available for the FSU considerable anecdotal evidence indicates that soil erosion may be an even greater problem in the FSU than it is in the United States (Brown and Wolf, 1984; Medvedev, 1987). For example, the rich chemozems which produce 80% of the grain crops in the FSU have been especially hard hit by soil erosion and the loss of organic matter (carbon) (Priputina, 1989). Additionally, crop residues reduce wind speed at the soil surface and act as a sponge and buffer for incoming precipitation. The mulch of crop residues also increases the rate at which water infiltrates the soil (Fenster, 1977). The introduction of no-till management has been shown to increase the aggregate stability and raise the percentage of organic matter in the soil (Blevins et al., 1983). These increases have a high positive correlation with increased root growth and production in agricultural crops. The increase in organic matter also results in higher storage of carbon in the soil. In the United States 6.48 million hectares (6% of agricultural lands) are currently under no-till management (Kern and Johnson, 1991) with the percentage increasing each year. Agricultural lands managed with the no-till option are projected to reach 45% of all crop lands in the United States by 2000 (Phillips et al., 1980). Projecting the same percentages for no-till management in the FSU, the amount of carbon sequestered in agricultural lands would be .18 and 1.35 gigatons of carbon at 6 and 45%, respectively. A number of additional factors must be considered when forecasting changes in the agricultural sector of the FSU. Even though conversion to no-till management requires lower capital investment than do other management practices that have been proposed to sequester carbon in agro-ecosystems, no-till management requires different or specially modified agricultural equipment. In the United States the cost of new equipment often prohibits the introduction of conservation tillage. It is unlikely that individual farmers (there are few in the FSU) or individual collective farms will be able to finance the cost of new equipment, at least in the next decade. It is also unlikely that the governments of the republics in the FSU will be able to spend their foreign currency on equipment that may produce little short-term increase in yield. A number of authors (Philips et al., 1980; Kern and Johnson, 1991; Hinkle, 1983) suggest that no-till management is more complex than conventional tillage with respect to the control of weeds, insects and other pests. For example, crop residues on the surface prohibit mechanical removal of weeds. Annual plowing does not bury the previous season's weed seeds nor can mechanical cultivators be used to remove weeds between the developing crops. Increased (or complete) reliance on chemical weed control my result from the adoption of no-till management. The production and timely delivery of an increased number of agricultural chemicals is probably beyond the capability of the current agricultural infrastructure in the FSU (Chistobayev, 1990). Summary and Conclusions One-hundred and eighty-one million hectares of agricultural land in the FSU are climatically suitable for no-till management. Complete conversion of all climatically suitable land to no-till management would sequester 3.3 gigatons of carbon. This represents a 10% increase in carbon in the agricultural soils of the FSU. Accumulation of carbon in the soil produced by a conversion from conventional to no-till is expected to take at least 10 years. All successful agricultural operations require energetic, involved and educated operators who have a wide variety of skills and often have the ability to recognize the long-term benefit of a particular management practice. Up to the present time, these traits have not been encouraged in the FSU. Medvedev (1987) gave the following analysis of the agricultural population of the FSU: "... in the Soviet Union ... the most active and productive part of the rural population has been destroyed ... Thus while only the most active, able, and devoted farmers remain on the land in the United States and Western Europe, in the Soviet Union it is the least able, the most passive, and the oldest who stay voluntarily." Dramatic changes in infrastructure, education, and social structures will be necessary in the agricultural sector of the FSU before no-till management will be implemented to sequester carbon in the terrestrial biosphere. — - - - - -. - —- ——' - - - - I oGpa - S w (I (IC II — -.-- • -- - L - - --•---.-- stp i:a Russia ç Arable Lands (Greater than 30% A able) Lands Suitable for No-Till Manag anent LNon-Arable Lands (Less than 30%Ira4e) L -: I — Kazaka e Iran / Afgani s ki s an an j islan China ;Kor*a Figure 1. Agricujtgral Soils in the Former Soviet Union Suitableift lip-Till Management IL E ( ° L G 11 U Desert " Q 0 O N +. O C/) 1015 5 1015 2 4 1 4 6--8 N.. 1, Id. 60,;. -C Chestnut; O E N 2 3 1 40 E U 50 Q 60 (1) O 70 L 80 Organic Matter, percent by weight Figure 2. Soil Organic Carbon versus Depth (after Glazovskaya, 1972) Table 1. Carbon Content of Agricultural Soils of the Former Soviet Union Soil Type Total Area Initial or Initial or Cultivated of Precultivation Carbon Conten t Precultivation Carbon Pool (109 t) Equilibrium C arbon Conte nt SOCc (4) (t/ha) (5) Current Carbon Pool for Aagricultural So ils (109 t) (6) Agricultural S oils (106 Ha) SOC1 (t/ha) (1) --T (2) (3) 51.7 120.0 6.2 91.3 4.7 3.9 190.0 .7 144.5 .6 Gray-Forest 44.1 160.0 7.1 121.7 5.4 Chernozem 78.0 333.0 26.0 253.8 19.8 Chestnut 17.5 110.0 1.9 83.7 1.5 Solonetz 6.7 42.0 .3 32.0 .2 Gray-Brown Desert 8.3 45.0 .4 34.3 .3 Podzol Floodplain-Meadow Totals 210 42.6 32.5 I It 10 Table 2. Carbon Content of Agricultural Soils Climatically Suitable for .. il u, i', No-Till Manage ment I Soil Type Total Area Area of Decrease in of Agricultural Soils Climaticall y (106 Ha) Soils (106 Ha) Area Percent Current Equilibrium Carbon Reduction in Area Content SOCK (t/ha) Suitable for Conservatio n Tillage Podzol (2) (3) Suitable for Conservation Tillage (10 t) (106 Ha) (1) Current Carbon Pool for Soils Climatically (4) (5) (6) (7) 51.7 31.4 20.3 39 91.3 2.9 3.9 1.9 2.0 52 144.5 .3 Gary-Forest 44.1 40.1 4.0 10 121.7 4.9 Chernozem 78.0 76.5 1.5 2 253.2 19.4 Chestnut 17.5 16.9 .6 36 83.7 1.4 Solonetz 6.7 6.7 4.0 32.0 .2 Gray-Brown Desert 8.3 7.6 .7 34.3 .3 Floodplain-Meadow Total 210 181 29 .6 8 29.4 13.8 Ii I 10 Table 3. Soil Carbon Content Under No-Till Management . t. Soil Type Area Current (l) of Equilibrium Soils Carbon Climatically Suitable Content (SOCd for (t/ha) Conservation Tillage (3) Carbon Content 0-8 cm (t/ha) (4) Carbon Content 8-30 cm (t/ha) (5) Carbon Content > 30 cm (t/ha) (6) Ii ICI ). "No-Till" Total Increase Equilibrium "No-Till" in Carbon Content (t/ba) (7) Carbon Current Equilibrium Carbon in Carbon Pool (109 t) (9) from "No-Till" Pool (109 t) (8) (ha) Percent Increase Pool Management (10) (2) Podzol 31.4 91.3 35.6 24.7 31.0 105.3 3.3 1.9 144.5 30.3 56.4 57.8 162.2 .3 Gray-Forest 40.1 121.7 12.2 35.3 74.2 130.9 5.3 .4 Chernozem 76.5 253.2 38.0 114.0 101.2 21.5 2.1 ll Chestnut 16.9 83.7 23.4 54.4 5.9 99.1 1.7 .3 21 6.7 32.0 13.8 18.2 0.0 38.9 .3 1 21 7.6 34.3 11.3 23.0 0.0 41.2 .3 Floodplain-Meadow Solonetz Gray-Brown Desert Total 282 181 32.7 11 It P 1 P .4 0 15 0 0 3.3 8.6 0 Table 4. Percent Distribution of Soil Carbon > 30 cm 0-8 cm 8-30 cm Podzol 39 27 34 Acidic Floodplain-Meadow 21 39 40 Gray-Forest 10 29 61 Chemozem 15 45 40 Chestnut 28 65 7 Solonetz 43 57 0 Gray-Brown Desert 33 67 0 Soil Type References Amemiya M., 1977, "Conservation Tillage in the Western Corn Belt," Jrnl. of Soil and Water Conservation, Jan.- Feb. 1977:29-36. Arshad M.A., M. Schnitizer, D.A. Angers and J.A. Ripmeester, 1990, "Effects of Till vs, No-Till on the Quality of Soil Organic Matter," Soil Biological Biochemistry, Vol. 22 No. 5 :595-599. Barnwell, T.O., R.B. Jackson, IV, E.T. Elliott, I.C. Burke, C.V. Cole, K.Paustaian, E.A. Paul, A.S. Donigian, A.S. Patwardhan, A. Rowell and K. Weinrich, 1992, "An Approach to Assessment of Management Impacts on Agricultural Carbon," Journal of Water, Air & Soil Pollution (in press). Blevins R.L., M.S. Smith, G.W. Thomas and W.W. Frye, 1983, "Influence of Conservation Tillage on Soil Properties," Jrnl. of Soil and Water Conservation, May-June 1983 :301-305. Bohn H.L., 1982, "Estimates of Organic Carbon in World Soils: II," Journal of Soil Science Society of Arizona, No. 4:118-119. Brown L.R. and E.C. Wolf, 1984, "Soil Erosion: The Quiet Crisis in the World Economy," World Watch, Paper 60, Sept. 1984 Carter M.R. and D.A. Rennie, 1982, "Changes in Soil Quality Under Zero Tillage Farming Systems: Distribution of Microbial Biomass and Mineralizable C and N Potentials," Canadian Jrnl. of Soil Science, Vol. 62:587-597. Cherdantsev, G.N., 1961, Map: Arable Land in the U.S.S.R. in 1954. From A Geography of the U.S.S.R. - Background to a Planned Economy by J.P. Cole, F.C. German, 290 pp. Chistobayev A.I., 1990, "A State Farm in the Depths of the Non-Chernozem Zone," Soviet Geography, 1990 :151-159. Dick W.A., 1983, "Organic Carbon, Nitrogen and Phosphorus Concentrations and PH in Soil Profiles as Affected by Tillage Intensity," Soil Science Society of America Jrnl., Vol. 47:102-107. Fenster C.R., 1977, "Conservation Tillage in the Northern Plains," Jrnl. of Soil and Water Conservation, Jan.- Feb. 1977: 37-42. Gebhardt M.R., T.C. Daniel, E.E. Schweizer and R.R. Almaras, 1985, "Conservation Tillage," Science, Vol. 230:625-630. Glazovskaya M.A., 1972, Soils of the World, Vol. I, Moscow University Publishers, Moscow, 214 pp. Haas H.J., C.E.Evans, E.F. Miles, 1957, "Nitrogen and Carbon Changes in Great Plains Soils as Influenced by Cropping and Soil Treatments," U.S. Dept. of Agriculture Technical Bulletin #1164. Hinkle M.K., 1983, "Problems With Conservation Tillage," Jrnl. of Soil and Water Conservation, May-June 1983 :201-206. Hobbs J.A. and P.L. Brown, 1965, "Effects of Cropping and Management on Nitrogen and Organic Carbon Contents of a Western Kansas Soil," Kansas Agricultural Experiment Station Technical Bulletin 144. Houghton R.A., J.E. Hobbie,J.M. Melillo, B. Moore, B.J. Peterson, G.R. Shaver and G.M. Woodwell, 1983, "Changes in Carbon Content of Terrestrial Biota and Soils Between 1860 and 1980: A Net Release of CO2 to the Atmosphere," Ecological Monographs Vol. 53 No. 3:235-262. IPCC, 1992, "1992 Supplemental Report of AFOS Working Group," IPCC Secreteriate, Geneva, Switzerland (in press). Jenkinson D.S., 1991, "The Rothamsted Long-Term Experiments: Are They Still of Use?," Agronomy Journal, Vol. 83:2-10. Kern J.S. and M.G. Johnson, 1991, "Impact of Conservation Tillage Use on Soil and Atmospheric Carbon in the Contiguous United States," EPA/600/3-91/056, Env. Res. Lab, Corvallis OR. Kobak K.I., 1988, Biotical Compounds of Carbon Cycle, Hydrometeoizdat, Leningrad, 248 pp. Kobak K.I. and N. Yu. Kondrashova, 1986, "Distribution of Organic Carbon in Soils of the Globe," Trudy GGI, Vol. 320:61-76. Kononova M.M., 1963, Soil Organic Matter, Moscow, Nauka, 314 pp. Lydolph P.E., 1970, Geography of the USSR, 2nd Ed., John Wiley & Sons, New York, 683 pp. Mann L.K., 1985, "A Regional Comparison of Carbon in Cultivated and Uncultivated Alfisols and Mollisols in the Continental United States," Geoderma,Vol. 36:241-253. Mann L.K., 1986, "Changes in Soil Carbon Storage After Cultivation, " Soil Science, Vol. 142 No. 5:279-288. Medvedev Z.A., 1987, Soviet Agriculture, Norton and Co., New York, 464 pp. Phillips R.E., R.L. Blevins, G.W. Thomas, W.W. Frye and S.H. Thomas, 1980, "No Tillage Agriculture," Science, Vol. 208:1108-1113. Post W.M., W.R. Emanuel, P.I. Zinke and A.G. Stagenberger, 1982, "Soil Carbon Pools and World Life Zones," Nature, Vol. 298, No. 5870:156-190. Post W.M., T.-H. Peng, W.R. Emanuel, A.W. King, V.H. Dale and D.L. DeAngelis, 1990, "The Global Carbon Cycle," American Scientist, Vol. 78:310-326. Priputina I.V., 1989, "Lowering of the Humus Content of Chemozem Soils of the Russian Plain as a Result of Human Action," Soviet Geography, 1989 :759-762. Ryabchikov A.M., ed., 1988, Map: Geographical Belts and Zonal Types of Landscapes of the World, School of Geography, Moscow State University. Schlesinger, W.H., 1977, "Carbon Balance in Terrestrial Detritus," Ann. Rev. Ecolog. Syst., No. 8:51-81. 5 Schneider, S.H., 1989, "The Greenhouse Effect: Science and Policy," Science, 243:771-781. Symons L.S., 1972, Russian Agriculture: A Geographic Survey, J.Wiley and Sons, New York, 348 pp. Touchton IT., R.R. Sharp and D.W. Reeves, 1989, "Tillage Systems for Double Cropped Wheat and Soybeans," Applied Agricultural Research, Vol. 4 No. 4: 264-269. Unger P.W., 1968, "Soil Organic Matter and Nitrogen Changes During 24 Years of Dryland Wheat and Cropping Practices," Soil Science Society of America Proceedings, Vol. 32:427-429. Unger P.W., 1991, "Organic Matter, Nutrient and pH Distribution in No- and Conventional-Tillage Semi-Arid Soils," Agronomy Journal, Vol. 83:186-189. U.S. Depart. of Agriculture, 1990, "USSR Trade and Agriculture Report: Situation and Outlook," USDA RS-90-1, May 1990. USSR Atlas of Agriculture, 1960, Moscow, 318 pp. Vinson, T.S., T.P. Kolchugina, R.K. Dixon, P.M. Bradley and G.G. Gaston, 1992, Proceedings of the IPCC AFOS Workshop, University of Joensuu, Joensuu, Finland, May 11-15, 1992. Identification of Carbon Quantifiable Regions in the Former Soviet Union Using Unsupervised Classification of AVHRR Global Vegetation Index Images. Greg G. Gaston Ted S. Vinson 2 Philip L. Jackson ' Tatayana P. Kolchugina 2 1. Department of Geosciences Oregon State University Corvallis, Oregon 2. Civil Engineering Department Oregon State University Corvallis, Oregon Please Address Correspondence to Greg Gaston U.S. EPA Research Lab 200 S.W. 35th Corvallis, Oregon 97333 Abstract Global VegetationIndex (GVI) data from the Advanced Very High Resolution Radiometer (AVHRR) was used to identify macro-scale vegetation/land cover regions in the former Soviet Union (FSU). These regions are a better representation of surface vegetation and land cover than can be obtained from existing thematic maps of the FSU. Image classes were identified through cluster analysis using the ISODATA clustering algorithm and a maximum likelihood classifier. Qualitative analysis of the image variants produced with different input parameters indicated that an image with 42 classes best represented significant details in vegetation and land cover patterns without producing uninterpretable levels of detail that represent artifacts of the clustering algorithm. Initial identification of image classes has been made by considering the weight of evidence provided by quantitative and qualitative analysis of existing maps, analytical tools from class statistics, ancillary data from a variety of sources and expert assessment by Russian scientists with extensive field experience in the FSU. Overall, this method of image classification using GVI data appears to accurately describe regions with similar vegetation and land cover across the FSU. Some questions regarding the identification of wetlands and potential problems with classification in the Russian high arctic are discussed. The products of this research will help improve carbon budget estimates of the FSU by providing accurate delineation and definition of carbon quantifiable regions. Introduction To date, direct assessment of the carbon pools and fluxes (i.e. carbon budget) in terrestrial ecosystems in the former Soviet Union (FSU) has not been performed. The best current estimates of the carbon budget for this vast land area (one sixth of the earth's land surface) which contains the largest expanse of boreal and tundra ecosystems in the world are based exclusively on extrapolation of site specific carbon estimates to the regions displayed on a limited number of thematic maps (Kolchugina and Vinson, 1993; Vinson and Kolchugina, 1993). However, there are serious questions about the information content of these maps. Each map, at best "...is a snapshot of the situation seen through the particular filter of a given surveyor, in a given discipline at a certain moment in time"(Burrough, 1987). Cartographic data are generally the result of observations taken over a very short time, thus maps are static sources of information. There are problems with scale changes between maps, significant differences in the level of detail between maps, and changes in classification schemes between maps and across political boundaries. The placement of boundary lines between areas that grade into each other over long distances is an arbitrary decision of the individual cartographer. Additionally, the size and inaccessibility of much of the land area of the FSU necessitates cartographic generalization, introducing interpolation errors.. Political restrictions on availability and even purposeful inaccuracy of USSR maps are believed to be limited to large scale maps. Continental scale maps, used in this research, are likely to be as accurate as field sampling and cartographic representation allows. There is wide agreement that satellite data can be used to overcome the limitations of thematic maps, providing timely, consistent and reliable information on the areal extent of, and conditions in, terrestrial ecosystems (Norwin and Greegor, 1983, Janssen et al., 1990, Justice et al.1985, Hall et al., 1991). The purpose of this research is therefore to utilize satellite data to identify macro-scale vegetation regions across the FSU. These regions should improve the accuracy of carbon budget estimates. An Appropriate Remote Sensing System The Advanced Very High Resolution Radiometer (AVHRR) appears to be the most useful satellite remote sensing system presently available for continental scale studies of terrestrial ecosystems. The AVHRR is a part of NOAA's meteorological satellite program which is designed to provide daily coverage of atmospheric conditions over the entire globe. The orbit and resolution of this sensor system while designed for meteorological data collection, has been successfully used to monitor conditions in terrestrial ecosystems (Tucker et al., 1983, Hayes, 1985, Townshend et al., 1987, and Fung et al.,1987). A comparison of landcover mapping based on the accuracy of 1.1 km AVHRR data to landcover maps based on 80 m Landsat MSS indicated the accuracy for classifications based on AVHRR data was 71.9% compared to 76.8% for MSS. This accuracy is considered acceptable for most research and given the disparity in spatial resolution, the accuracy of the AVHRR is exceptional (Gervin et al.,1985). For analysis of conditions in terrestrial ecosystems the most commonly used product from the AVHRR is a normalized ratio of channel 1 (0.58-0.68 microns) and channel 2 (0.725-1.1 microns). This normalized ratio of visible and near infra red reflectance is commonly termed the Normalized Difference of Vegetation Index (NDVI). The equation used to calculate NDVI is as follows (Kidwell, 1990): CH.2 (Near IR) - CH.1 (Visible) NDVI = --------------------------------CH.2 (Near IR) + CH.1 (Visible) (1) Normally active photosynthetic vegetation absorbs the visible wavelengths (AVHRR Ch.1) (esp. red) and reflects near infra-red radiation (AVHRR Ch.2). Higher photosynthetic activity results in increased red absorption and even higher near infra-red reflection. NDVI values range from -1 to +1, with most vegetation clustering around +.6 NDVI; higher values are characteristic of higher photosyn- thetic activity. This index has been related to Primary Productivity (Tucker et al.,1983), Leaf Area Index (LAI)(Spanner et a1.,1990), Percent Vegetative Cover (Townshend et a1.,1987) and Biomass and Absorbed Photosynthetically Active Radiation (PAR or IPAR)(Goward et al.,1990). There appears to be a strong correlation between NDVI and the vegetation in terrestrial ecosystems. Most research using AVHRR data concentrates on this index to evaluate vegetation attributes in the analysis of terrestrial ecosystems. The highest spatial resolution for the AVHRR scanner is 1.1 km at the nadir. As the scan widens to acquire data across the entire 112 degrees of the sensor swath width, the spectral and spatial resolution degrades due to increases in atmospheric path length,curvature of the earth, and changes in reflection angles. The wide scan angles and daily repeat of orbits produce a tremendous amount of data. A lack of on-board storage and limited down-link connections require on- board pre-processing and sampling of the data (Goward et al.,1990). The first of the pre-processed products is known as Global Area Coverage (GAC). GAC data are produced on a daily basis by averaging the first 4 pixels in a 5 x 3 pixel array, and assigning this average value to the entire array, producing a nominal 4 x 4 km pixel (Goward et al.,1987). GAC data are further processed by NOAA, by selecting the "greenest" pixel (the highest NDVI value) in a 4 x 4 pixel array. The resulting product, known as the Global Vegetation Index (GVI) reflects the highest NDVI value occurring in a seven day period over a nominal 16 x 16 km pixel (Kidwell,1990). The presence of clouds (on an average day one half the surface of the globe is obscured by clouds), shadows, off nadir 'look' angles, and different reflection ratios all tend to reduce NDVI values. In order to reduce the effects of these factors a temporal compositing routine is used. The maximum NDVI is usually obtained in an unobscured near nadir view. A maximum value composite is constructed by taking the maximum value for a given pixel in a given time frame. Weekly and monthly maximum value composites are regularly produced using a set of georeferenced AVHRR images (Kidwell, 1990). This process appears to remove most of the problems associated with clouds, off nadir views, and other factors that reduce NDVI (Holben,1986). Goward et al. (1990) maintain that errors can be reduced to ± 10% (.1 NDVI unit) using a maximum value compositing process on a monthly time step. The maximum value compositing procedure assumes that the surface features are homogenous over the sampling area. This procedure may overestimate the actual extent of photosynthetically active vegetation. In choosing the "greenest" pixel it is possible that an oasis would be chosen to characterize an otherwise barren desert (Townshend and Justice, 1986). The temporal signature of a heterogenous landscape may also be confused using this procedure. For example, regenerating broadleaf vegetation in a disturbed area in a conifer forest may produce "greenest" pixels that represent these areas of regeneration rather than the bulk of the conifer forest. Further, the temporal compositing procedure which simply assigns the value of a single 4 x 4 km GAC pixel to the entire 16 x 16 km GVI pixel, regardless of time or the spatial location of each pixel, makes locating the exact geographic position of individual pixels impossible. Data and Methods Research Background The use of AVHRR derived products to characterize the location, areal extent, and characteristics of land cover (terrestrial eco-systems) has been reported by a number of authors. Goward et al. (1985), used three week composites of GVI data to identify and characterize vegetation patterns of North America. They selected 3 x 3 pixel plots 58 of "known" surface conditions. The change in NDVI with time for these plots was graphed. The area of the temporal NDVI signature was determined for each test area and this statistic was used to relate each class to measures of biomass productivity. Their research demonstrated that major land cover regions could be identified using seasonal patterns of NDVI change. Justice et al. (1985) using, in part, GVI data concluded that the extent and seasonal dynamics of global vegetation could be successfully mapped on a global scale. A variety of classification schemes were tested by Townshend et al. (1987) in research directed towards classifying vegetation complexes in South America. They concluded that a maximum likelihood classification produced the most satisfactory delineation of South American vegetation complexes, when compared to training sites selected from the UNESCO 1981 vegetation map. Loveland et al. (1991) utilized unsupervised classification of monthly maximum value composites (1 km resolution) to characterize the land cover of the conterminous United States. They reported that large homogenous land cover regions were well identified, citing forest ecosystems as prime examples of successful classification. GVI Data Used To Identify Carbon Quantifiable Regions The goal of this research was to use Global Vegetation Index data to identify macro-scale vegetation/land cover regions across the FSU. The motivation for identifying these regions is an improvement of carbon budget estimates in the FSU. The use of a single year time series could produce bias in regional identification due to variations from climatic norms occurring in a single year. In order to reduce the possible effects of inter-annual variation, monthly GVI composites from 1985 through 1988 were averaged, creating a single set of monthly observations that represented conditions over a 4 year period. The GVI composites used in this research are slightly different in construction from a simple maximum value composite. It is possible that the use of a single maximum value from a single observation may cause further compression of the temporal resolution. For example, the maximum NDVI value for the month of June may occur on the 30th, and the maximum July may occur on the 1st, consequently two "monthly" observations may be separated by a single day. The GVI data used in this research are constructed from weekly maximum value composites for the weeks occurring in a calendar month, the maximum and minimum values are removed and a root mean square (RMS) average is calculated from the remaining weekly values (NOAA-EPA,1992). This RMS average data set reduces the chances of temporal compression possible with the use of single observations while preserving most of the advantages found in a true maximum value compositing scheme. At high latitudes, with low winter sun angles, anomalously high NDVI values are commonly found in AVHRR images. This phenomenon (known as the "terminator effect") is apparently caused by the different transmission rates of red and near infra-red radiation (Justice et al.,1985). Examination of the GVI composites of the FSU in the winter months displayed extremely low (usually 0) NDVI values, with the exception of the high arctic which appeared to be vigor- ously photosynthetic throughout December, January and February. Of course, the actual levels of vegetative growth are negligible between October and March-April over most of the FSU. Consequently, the NDVI data used in the present study were restricted to composites from March through October. Problems with Weekly GVI Composites The problems previously discussed (e.g. the oasis effect and lack of spatial certainty) are inherent to the sampling scheme used to create GVI data. It is also recognized that utilizing a monthly time step for maximum value composites may result in the loss of temporal detail. Examination of GVI data sets at a weekly time step (1988 only) suggests that in the extremely continental climates of the FSU the rapid northward advance of the "green wave" in the spring takes place at a rate much faster than can be captured in a monthly time step . However, the use of weekly composites is not without problems. Our experiments with the 1988 weekly time step data showed what are believed to be significant problems with cloud contamination. Since the Russian high arctic is one of the most cloud covered regions on earth (Parker,1983), a weekly compositing period does not appear to be sufficient time to significantly reduce the problem of cloud cover. Holben (1986) discusses the problem of temporal compositing period, and suggests that the best guide is the quantity of information desired and a priori knowledge of the phenological response curves of the vegetation. He states that longer compositing periods significantly reduce problems with clouds, etc., but that more observations make identification of vegetation complexes based on phenological change easier. A four year average of monthly maximum value composites represents a compromise in data selection, and in the author's opinion are the most appropriate data for initial identification of carbon quantifiable regions across the entire FSU. Clustering and Classification Unsupervised classification is a means of analyzing data that allows the inherent structure of the data to be seen (Anderberg,1973). Unsupervised classification was recommended by Townshend and Justice (1986) as the most effective way to identify vegetation communities using GVI data. This method was used by Loveland et al. (1991) with eight monthly maximum composites from 1990 as a first step in creating a landcover map of the conterminous United States. An unsupervised classification requires two steps: (1). identification of "clusters" within the data (cluster analysis), and (2). a maximum likelihood grouping of the entire data set around these "cluster centers". Combined multi-year data for each of the seven months in the growing season of the FSU were clustered using the ISODATA algorithm imbedded in the GRASS geographic information system (GIS)(U.S Army CERL,1992). ISODATA is the most common clustering algorithm found in image processing systems (Bryant,1978). ISODATA requires a number of input parameters, most significantly the initial number of classes and the class separation distance. These parameters define the level of generalization in the resulting image. The initial number of classes sets the maximum number of classes that can be generated. The class separation distance defines a threshold distance below which pixel clusters will be grouped together. Changes in these parameters produce some very noticeable differences in the resolution and level of detail in the resulting image. Most discussions of cluster analysis suggest that extensive experimentation and expert assessment are the primary means of selecting appropriate input parameters and the most useful resulting image (Swain and Davis,1978, Anderberg,1973, Harmon and Shapiro,1990). A number of quantitative methods have been proposed to determine the number of "natural" clusters present in a data set. Milligan and Cooper (1985) compared 30 methods for estimating the natural number of clusters in a test data set, and identified the Calinski/Harabaz pseudo F statistic as the most reliable. This statistic is supplied by the SAS module 'FASTCLUS' (SAS/STAT,1990). This module was executed with the GVI data for the FSU. The results of this analysis suggest that the "natural" number of clusters present in the combined monthly maximum value composites from March-October for the FSU is only two. The authors suspect this statistic represents a vegetation/no vegetation split, which is much too simplistic for a detailed analysis of the FSU. The most effective method of identifying the appropriate number of clusters (image classes) was based on expert assessment and visual comparison of experimental images. This was an iterative process that involved changing cluster parameters to create a large number of image variants. Examination of the images revealed some stable overall patterns with varying levels of detail in class separation. The fundamental problem became the identification of the image variant that successfully identified important vegetation/landcover complexes without creating an unnecessarily large number of classes as an artifact of processing. 63 Key to the selection of the most useful image variant was the identification of a set of features that should be seen on a detailed vegetation/landcover map of the FSU. This set of features was identified with the aid of maps, ancillary data sources, and expert assessment provided by scientists with extensive field experi- ence in the FSU . Image variants were examined to see if these details were successfully identified while minimizing confusion in the overall patterns seen in the balance of the image. Images produced using a smaller number of initial clusters (i.e. 15) were smoother and more generalized with a high correlation with the classes displayed on a generalized soil/vegetation complex map of the FSU (Ryabchikov, 1988). However, significant detail was lost on this image, with areas of very different characteristics lumped together into a single image class. Images produced with a large number of classes (i.e. 75) were extremely noisy with overall patterns degraded with unacceptable (and uninterpretable) detail. The image variant chosen for this research (42 classes) had sufficient detail to locate the specified features and at the same time preserved the overall patterns described and mapped for the vegetation communities of the FSU. The image map used to classify macro-scale vegetation/land cover patterns of the FSU is shown in Figure 1. There are 42 image classes each reflecting a different timing and magnitude of photosynthetic activity as recorded by NDVI. The image is a simple latitude - longitude projection. Interpreting the Image Quantitative Tools The critical step in utilizing an unsupervised classification scheme is trans- forming the image classes into "informational classes" (Swain and Davis,1978). A number of quantitative tools can be brought to bear on the problem of image interpretation. Computerized GIS tools make it extremely simple to produce coincidence tabulations, showing the percentage of image classes present in each class of a thematic map or maps. In this research, for example, image classes were compared with the best available digitized vegetation map of the FSU (Sochava,1960), giving some indication of vegetation present in each image class. For example, image class 2 which has been identified as mountain tundra/arctic tundra contains the following mapped vegetation classes; northern mountain tundra (43.3%), arctic tundra (12.9%), arctic desert (9.7%), southern mountain tundra (5.8%) and typical tundra (3.2%). The remainder of this image class (25.1%) is represented by very small percentages of a number of mapped vegetation classes. Overall, this image class has a high correlation with the vegetation map. In contrast, class 21, identified as predominantly agriculture contains 49 mapped vegetation classes scattered with no apparent pattern or dominant characteristic. Direct comparison of image classes to thematic map classes is a useful technique but does not provide a complete explanation of class characteristics. Goward et al. (1987, p.29) states that "...comparison of AVHRR observations to 'known' distributions produce varying results based on the map selected". It is possible that vegetation patterns displayed by classified remote sensing data may show significant spatial patterns independent of previous classification schemes. Townshend (1987) asserts that because existing maps of vegetation and land use have been developed using different methods they may not be acceptable standards of reference for AVHRR classifications. Another useful interpretative tool is the construction and comparison of temporal NDVI response curves (TNRC). These can provide a great deal of information about individual classes and the relationships between classes. TNRC's provide useful information for identifying similarities between classes and allow one to separate classes with significantly different reflectance response. It may be possible to develop additional analytical tools based on TNRCs. Goward et al. (1985 and 1987) and Tucker et. al. (1985) suggested the use of the area under the TNRC could be used to classify vegetation types and that this statistic might be related directly to biomass estimates of terrestrial ecosystems (at least in North America). It may also be possible to develop additional statistical tools to identify patterns of class groupings and to quantify differences between classes. Realistically, however, image interpretation is not done solely with quantitative tools. The combined weight of evidence from quantitative tools, information from a wide variety of maps, descriptive geographies, and the assessment provided by experts with field experience must all be used by the interpreter to convert image classes into information classes. Qualitative Tools Table 5 lists the maps that were used to aid visual interpretation of the image classes and in the case of the vegetation map (Sochava 1960) were used in a coincidence tabulation to provide quantitative information for each image class. In addition to the maps presented in Table 1, several descriptive geographies by Russian geographers were extremely useful (e.g. 'Physical Geography of Asiatic Russia' by S.P. Suslov (1961), and 'Natural Regions in the USSR' by L.S. Berg (1950)). A variety of other ancillary data sources were also consulted to provide additional information for image interpretation. Expert assessment was provided primarily by two Russian scientists with extensive field experience in the FSU. These scientists are identified and their credentials are presented in Table 5. Basic Geographic Patterns in the FSU Successful interpretation of the image created through unsupervised classifica- tion of GVI images which represents the land cover of the FSU is facilitated by an understanding of the climatic and physiographic factors that influence vegetation and human settlement patterns. In simple physiographic terms the FSU can be described as a plateau, open to the north and bounded on the south and east by high mountains. The Russian platform and the Siberian Platform are "typically flat, almost planar" (Nalivkin,1973) and the lowlands of western siberia are "flat as a 67 floor" (Lydolph,1970). The Ural mountains, the Putorana mountains and other small ranges in the central portions of the FSU do not provide a significant topographic barrier to climatic events. The high complex mountain ranges of the Kopet Dag, the Pamir, Tien Shan, Altai, Sayans and other ranges across the south and east of the FSU "shield" the center of the country from tropical and marine tropical air masses. This shielding from moisture in the south, coupled with the distance that marine air masses from the Atlantic must travel over land and the frozen surface of most of the arctic ocean, produce moisture deficit conditions in much of the FSU (Parker,1983) Physiographically, the FSU is predominantly a stage on which a strong continental climate with a limited moisture regime at high latitudes shapes the vegetation complexes, producing definite zones of soil and vegetation. Physiographic and climatic factors produce pronounced latitudinal zones of vegetation. Along a north to south transect the vegetation transitions from arctic tundra through shrubby tundra to northern taiga forest. This north to south transition is marked by vegetation communities with increasing biomass. Further south, moisture deficits produce a broadleaf forest transition, steppe, semi-desert, and desert, an order corresponding to vegetation communities with decreasing biomass. An additional shift in vegetation communities generally occurs across a west-east transect, from European species to predominantly Asian species. The Image Classes The patterns of the image classes reflect the expected patterns in vegetation that result from climatic factors (re. Figure 3). Latitudinal banding in the image classes reflects the effects of extreme cold and moisture deficits. Image class differences reflect species change from west to east. From north to south there is an increase in biomass as related to NDVI "greenness" with an increase in the percentage of woody vegetation which is represented in image classes 2, (3-5), 4, 6, 8, 10. These classes represent arctic (high mountain) tundra (image class 2) transitioning through shrubby tundra (image class 6, 8, 10) to the forest/tundra (image class 12) boundary where the sparse forests of the northern taiga begin to dominate the vegetation communities. A similar pattern may be observed in the elevational gradient of the Burranga mountains, the Putorana mountains and the mountain ranges of eastern siberia. In the south, the effects of moisture deficits on vegetation communities are reflected in image classes 13 (desert) through 14 (28), 17, 26 (23), 35 (32) which represent increasing vegetation grading from semi-desert steppe (image class 14 and 28) into forested meadows (image classes 35 and 32). The Northern taiga forest of the FSU are identified by classes 11, 15, 16, 18, 20, and 23. The dwarf and shrubby forests of the pacific coast are identified by image classes 9, 19, and 22. Middle taiga forests are found in image classes 24, 25, 29, 30, and 31. Southern taiga forests are represented by image classes 33, 36, and 37. Mixed conifer and broadleaf forests are identified by image classes 38, 39, and 40. Broadleaf forests 69 with a smaller percentage of conifers and agricultural lands are identified by image classes 41 and 42. Table 6 presents a summary description of all image classes. A complete description of the identification process for all 42 image classes is beyond the scope of this paper. However, some examples of the identification process are provided for clarity. In the forest regions of the FSU a number of classes have been generated. There is some question concerning the exact species composition of several classes but most maps and ancillary data sources describe the central siberian plateau as an unbroken, unmixed larch forest. This expanse of larch is differentiated only by the density of larch trees and in some cases the composition of the understory. Image classes 11 and 16 appear as divisions in this unbroken siberian larch forest. Examination of the TNRC for each class reveals that these image classes are virtually identical in temporal patterns with the only differences occurring in the magnitude of the NDVI peak (Figure 4). A large body of evidence, including that provided by comparing TNRCs, aided the identification of these classes as larch forest of varying density. In the Amur river region class 42 has been identified as a high biomass mixed broadleaf forest. Detailed maps of the vegetation communities in this area (Suslov, 1961; Sochava 1960) are very similar in their location and description of broadleaf and mixed broadleaf "Ussurian" taiga vegetation (re. Figure 5). A comparison between the mapped distribution of broadleaf and mixed "Ussuri" taiga forests and the distribution of class 42 show a very high correlation (re. Figure 6). 70 In the mountainous region of south central Asia, image classes reflect very clear delineations of vegetation communities (re. Figure 7). Vegetation in this region is strongly influenced by elevation and the presence of water for irrigation. The distribution of irrigated agriculture is very well identified by image class 21, concentrated along the Amu-Darya and Syr-Darya rivers and indicating the course of the Hi river flowing into Lake Balkash. Class 21 also locates very accurately the Fergana valley, a fertile agricultural valley that has been under irrigation for several centuries. Detailed inspection of this region helps illustrate the accuracy of the methods used in this research. The Aral sea, Lake Balkash, and Lake Issul-kul all appear as image class 1. An extremely small area of class 1 (two or three pixels east of Lake Balkash) locate and identify Lake Ala-kul, one of the brackish remanents of a much larger Lake Balkash that once filled this synclinal valley (Suslov,1971). The semi-desert transitional zones are represented in the region by both class 14 and class 28. Suslov (1971, p.569) describes strong differences in the "origin, composition and ecology... (of the) predominate vegetation" in these semi desert steppes. The image classes show clear differences that appear to correspond to the differences described by Suslov. Comparison of the TNRCs for each class reveals the similarity of these classes in timing, development and magnitude of vegetation growth confirming the identification of these classes as a complex that represent semi-desert steppe. 7 Agricultural Land Locating the areas where agriculture is a significant factor to the character of the landscape is an important consideration in interpreting the image classes. The maps available to aid in interpreting the image classes are "potential natural" vegetation maps that do not reflect actual agricultural land use. Maps that accurately show current agricultural land use were not available at this time. Currently, information on the location and extent of agricultural lands is provided by the arable lands map of Cherdantsev (1961). The first three categories of this map (greater than 60% arable, 30-60% arable, and less than 30% arable) appear to accurately identify agricultural lands in the FSU. Assigning appropriate factors (eg. 80%, 45%, 15%) for each of these three classes, Gaston et al. (1993) calculated the area of agricultural lands in the FSU at 210 million hectares. This estimate compares very favorably with USDA estimates of 211 million hectares (USDA, 1990). A similar method was used to estimate the area of arable lands in each image class. Image classes 41, 38 and 35 appear to correlate well with the arable lands mapped by Cherdantsev (1961). A coincidence tabulation between image classes and arable lands indicated that 50% (approximately 100 million hectares) are located in these three image classes. The estimated area and percent coverage of agricultural land in each image class are presented in Table 7. Agricultural land uses are a significant factor in a limited number of image classes. The percentage of arable land in each image class (re. Table 7) were considered when defining and describing each image class. 7 Discrepancies in Classification A number of examples where the classification procedures appear to work very well have been described. It is also possible to offer examples of discrepancies between the spatial patterns in the image classes and the conditions described and mapped in other sources. For example, the tundra of the Tamir peninsula is divided on almost all vegetation maps into three nearly equal divisions. The dividing lines for the tundra divisions closely follow lines of latitude. The image classes of the Tamir do not exactly follow these mapped patterns (re. Figure 8). There is a more complex north to south distribution of tundra as displayed in image classes 2, 3 and 5. The image classes appear to accurately reflect the GVI data and are not an artifact of clustering. The reason for this disagreement is unknown. Perhaps the four year average of monthly maximum value composites used in this study does not completely remove the influence of clouds or extreme off nadir look angles in the remote sensing data. It is possible that the thematic maps tend to over generalize the patterns of vegetation distribution. In order to successfully answer these questions additional information should be provided by higher resolution sensor data (such as Landsat Thematic Mapper) or by field reconnaissance. It appears that wetlands are not well identified in the image. For example the west coast of the Kamchatka peninsula is known to be covered with coastal marshes and "blanket bogs" (Sochava 1960). While the image classes in this region appear to clearly identify vegetation especially as influenced by elevational gradients (r.e Figure 9). The image displays only a narrow band of classes 7 and 21 along the western shore of the Kamchatka peninsula. The image classes are much more narrow than the bogs mapped by Sochava (1960)(r.e. Figure 10) or described by Botch (personal communication, 1992). The extensive wetlands of the western siberian lowlands are also not well identified. Under representation of wetland areas may be a product of the "oasis effect" inherent in GVI sampling. A large island of trees in a wetland matrix could represent the maximum NDVI which is used to characterize the entire GVI pixel. Summary and Conclusions The research presented provides an example of the use of satellite data to identify and characterize vegetation/land cover complexes over one sixth of the earth's land surface. The image described in this paper represents a first attempt to utilize satellite data to identify carbon quantifiable regions across the FSU. This effort should be viewed as the beginning of a process that improves the identification and characterization of vegetation/land cover classes over the FSU. The use of the ISODATA clustering algorithm and a maximum likelihood classification of monthly maximum value GVI composites has produced a clear image of the FSU in which terrestrial vegetation/land cover complexes are clearly separated. The separation of image classes is based on "greenness" and not on species composition. The timing, magnitude, and duration of photosynthetic activity as reflected by NDVI are the critical factors that influence class separation. There often appears to be a high correlation between images classes and species composition (eg. class 42 in the Amur river region). However, some of the divisions described by different image classes across the boreal forest may not 74 reflect species changes. The differences in image classes appear to be a result of a number of factors including: species composition of all vegetation present in a pixel, the density of vegetation, and the latitudinal position of the image pixel. It is not possible at this time to fully isolate and explain the individual influence of all these factors when characterizing the image classes. Initial identification of class characteristics was possible through the combined use of qualitative and quantitative tools. Previous research helps confirm the conclusions from this research that while quantitative comparisons with existing maps are useful, cartographic data does not fully describe the image classes generated from satellite data. The conversion of image classes to "information" classes must remain the result of qualitative analysis. It may be possible to use the image classes generated in this research to create a stratification for sampling higher resolution sensor data and provide structure for ground reconnaissance. Higher resolution data (both temporal and spatial) would provide support for classification refinement and definition. Acknowledgement The work presented herein was funded by the U.S. Environmental Protection Agency (EPA)- Environmental Research Laboratory Corvallis, Oregon, under Cooperative Agreement CR820239 to Oregon State University. Jeffery J. Lee is the Project Officer for the project entitled "Carbon Cycling in Terrestrial Ecosystems of the Former Soviet Union". The work presented is a component of the U.S. EPA Global Climate Research Program, Global Mitigation and Adaption Program, Robert K. Dixon, Program Leader. This paper has not been subjected to the EPA's review and, therefore, does not necessarily reflect the views of the U.S. EPA, and no official endorsement should be inferred. Table 5 Maps used as an aid in image interpretation 'Vegetation of the USSR', 1960, Sochava V.B.(ed.), Ministry of Interior, Moscow 'Forest Tree Species of the USSR', 1960, Kruchinin A.P.(ed.), Ministry of Interior, Moscow 'Landscape Units of the USSR', 1988, Isachenko A.G. Inst.of Geog. Leningrad State University 'Geographic Belts and Zonal Types of Landscapes of the World' 1988, Ryabchikov A.M., School of Geography, Moscow State Univ. 'Arable Lands of the USSR in 1954', 1961, Cherdantsev G.N., In: A Geography of the U.S.S.R.- Background to a Planned Economy ed. J.P. Cole and F.C. German 'Permafrost Regions in the USSR', 1985, Ministry of Interior, Moscow Scientific Experts Dr. Marina Botch Dr. Botch is a professor and senior research scientist at the Komarov Botanical Institute in St. Petersburg, Russia. She is the Chairperson of the Russian committee on peatlands which is the highest appointment related to peatlands in Russia. She has extensive field experience and is the author of over 300 papers. Dr. Kira Kobak Dr Kobak is a professor and senior research scientist at the State Hydrologic Institute in St. Petersburg, Russia. She has extensive experience in plant physiology, soil processes, carbon cycling and climatic influences on ecological systems. Table 6 A Brief Description of Image Classes Image Class 1 Description Water / No Vegetation 2 High Mountain Tundra (Arctic Tundra) 3 Arctic Tundra / Mountain Tundra 4 Mountain Tundra 5 "Typical" Tundra 6 Shrubby Tundra of Siberian Mountains 7 Bunch Grass Steppe 8 "Typical" / Shrubby Tundra 9 Dwarf Shrub Pine / Alder (Riparian Grass and Willows of Ob river) 10 Shrubby Tundra 11 Sparse Northern Taiga Larch 12 Forest Tundra 13 Desert 14 Semi-Desert (Artemisia and Bunch Grass) 15 Scattered Larch on Mountain Slopes 16 Northern Taiga Larch 17 Artemisia and Bunch Grass: Transition from Semi-Desert to Grassy Steppe 18 Northern Taiga Spruce / Larch Forest 19 Mountain Shrubs of Siberia; Pinus pumila, Alnus spp. 20 Northern Taiga Spruce Forest 21 Irrigated Agriculture (Some northern steppe east of Ural Mountains and Kamchatka) 22 Mixed Dwarf Forest of Pacific Coast; Pinus pumila, Betula ermani 23 Scattered Forest (probably Pine/Aspen in the south; Spruce/Birch in the north) 24 Middle Taiga Forest; Larch / Pine 25 Open Forest: Larch in Vilyuy basin and Birch / Pine / Aspen in central Siberia 26 Grassy Steppe 27 Pinus pumila / Betula ermani in Kamchatka Table 6: (contd.) 28 Semi-Desert Steppe in Kopet-Dag and Pamir Mountains. ("Iranian in character") 29 Middle Taiga Pine and Spruce 30 Middle Taiga Spruce / Fir / Pine Forest 31 Conifer Forest on Mountain Slopes 32 Forest Meadow Steppe: (Boggy Larch Forest in Vilyuy and Lena basins) 33 Coniferous Forest of Russian Plain (Pine / Spruce) 34 Grassy Steppe on Mountain Slopes of Central Asia; Agricultural land use is significant part of land cover matrix 35 Forest Meadow Steppe ; steppe portions are in agricultural crops 36 Larch / Pine (spruce) Forests on Mountain Slopes of Southern Siberia 37 Southern Taiga Forest of Southern Russia (predominantly mixed conifer) 38 Mixed Conifer / Broadleaf Forest (spruce/oak) (West of Ural Mountains) agricultural crop patterns significant part land cover matrix 39 Mixed Conifer / Broadleaf Forest (aspen-birch/spruce-pine) East of Urals) 40 Mixed Forest of Southern Siberia (birch and spruce-larch) 41 Broadleaf (mixed) forest of Russia, Beyloruss, and the Ukraine agricultural crop patterns significant part of land cover matrix 42 Highly Productive mixed forest (the species mix changes with location) Table 7 Arable Lands located in Image Classes Image Class Area of Arable Land (million ha) Percentage 35 43.7 64% 38 29.5 43% 41 27.7 42% 26 14.25 36% 37 13.3 27% 34 10.2 44% 33 10.0 14% 32 7.8 13% 23 5.5 11% in Arable Land total 162 million ha. *Image classes 35, 38, and 41 account for over 50% of all arable lands estimated for the FSU Table 8 Descriptions of Figures Figure 3. Unsupervised classification of four year average Global Vegetation Index (GVI) images for the seven months of the growing season (March-October). The image has 42 classes each representing areas with similar timing, duration and peak "greenness", as recorded by NDVI values. The image is displayed in a simple latitude-longitude grid array. Figure 4. Temporal NDVI response curves (TNRC) of class 11 and 16. These classes represent Larch forests of northern and middle siberia. These classes appear to be separated only by the peak greenness. Classes of Larch forest are discriminated only by density of the forest. Figures 5 and 6. Vegetation map (Sochava 1960) of the Amur/Primorski region. Class 55 is Ussurian Taiga, Class 261 is Broadleaf forest, and Class 221 is Mixed Broadleaf Forest. These mapped vegetation classes correspond very well to image classes 39 and 42. Further examination of Figure 4 demonstrates a good correspondence of image classes to elevation. Table 8 (contd.) Figure 7. Image details in south central asia. The distribution of image classes appears to correspond well to elevation (inset) and to vegetation communities described in ancillary data sources. For example class 28 along the slopes of the Kopet Dag represent semi-desert steppe. The timing and magnitude of greenness for this class is different (earlier and higher) than for other semi-desert classes (class 14). This differentiation between areas of semi-desert steppe is well described by Suslov (1961). The snow covered peaks of the Pamir mountains are clearly located by image classes 1 (no vegetation), 2 and 4 (high mountain tundra). In this region class 21 appears to clearly locate irrigated agriculture along the Amu Darya and Syr Darya rivers. The Fergana valley, the irrigated slopes near Tashkent and the Ili river that flows into Lake Balkash are also identified as irrigated agriculture in image class 21. Figure 8. Comparison of soil/vegetation association map (Ryabchikov 1988) to image class distribution in the vicinity of the Yamal Peninsula. The nearly equal divisions in vegetation distribution displayed on the map have been confirmed by expert assessment. Image classes 3 and 5 are not distributed equally along latitude lines. Examination of all GVI images used in the classification display a persistent pattern of values that reflect the distribution of image classes 3 and 5. The distribution of image classes may reflect persistent cloud cover or extremely steep look angles inherent in GVI data for this region. Figure 9 and 10. Image class distribution in Kamchatka. The distribution of image classes clearly reflect the effects of elevation on the distribution of vegetation. Overall, the image classes correspond to mapped vegetation classes. However, class 271 on the vegetation map is "blanket bogs" or wetlands along the western shore of the Kamchatka peninsula. The greater extent of wetlands on the vegetation map has been confirmed by expert assessment. Only a narrow band of image classes 7 and 21 appears to reflect the distribution of bogs on the west coast of the Kamchatka peninsula. The under representation of wetlands may be a result of the sampling inherent in the creation of GVI. uotun laTAOS iauuo3 JO 08EWi pa JissEID II Figure 3 Cl 7 —t -Th OZL 09 Oti OZ class 11 class 16 0 April May June July August September October Figure 4 09 0 Temporal NDVI Response Curves for Class 11 and 16 OOL T --------' -------- Figure 5 Vegetaton Map (Sochava, 1960): Amur Region z JJ 85 Figure 6 Classified Image: Amur Region Inset: Elevation / 86 LSO I A n Y WP'erTana Valley .N f Y 7 iq Pa,ur Mts. ! C': N_ I. Figure 8 and Classified Image in Yamal Region as Comparison of Soil/Vegetation Complex Map (Ryabichikov, 1988) tic Unsupervised Classification 4 (Otto lPerr +C Figure 9 Inset: Elevation Classified Image: Kamchatka 7_' S 'a I• is4J- ft 'p. 9 S I— - S SS - N UI". 'p I• I Figure 10 Classified Image: Kamchatka Inset: Vegetation (Sochava, 1960) 19: 9 • • 9 References Anderberg, M.R., 1973, Cluster analysis for Applications. Academic Press, New York, New York, 345 pp. Badhwar, G.D., R.B. Mac Donald, and N.C. Mehta, 1986, Satellite-Derived Leaf Area Index and Vegetation Maps as Input to Global Carbon Cycle Models: A Hierarchical Approach. International. J. of Remote Sensing, Vol. 7, No. 2, pp. 265-281 Bryant, J., 1978, Applications of Clustering in Multi-Image Data Analysis. Report #18 Earth observation division, NASA/Johnson Space Center, Houston, Texas, NASA Contract 9-4689-85. Burough, P.A., 1987, Principles of Geographical Information Systems for Land Resource Assessment. Claredon Press, Oxford, 193 pp. Cherdantsev, G.N., 1961, Map: Arable Land in the USSR in 1954, In: A Geography of the USSR - Background to a Planned Economy. Ed., J.P. Cole and F.C. German, 290 pp. Fung, I.Y., C.J. Tucker, and K.C.Prentice, 1987, Application of Advanced Very High Resolution Radiometer Vegetation Index to Study Atmosphere-Biosphere Exchange of CO2. Journal of Geophysical Research, Vol. 92, No. D3, pp.2999-3015 Gaston, G.G., T.P. Kolchugina, and T.S. Vinson, 1993, Potential Effect of No-Till Management on Carbon in the Agricultural Soils of the Former Soviet Union. Agriculture Ecosystems and Environment, (in press) Gervin, J.C., A.G. Kerber, R.G. Witt, Y.C. Lu, and R. Sekhon, 1985, Comparison of Level One Land Cover Classification Accuracy for MSS and AVHRR Data. International. J. of Remote Sensing, Vol. 6, No. 1, pp. 1271-1318 Goward, S.N., B. Markham, D.C. Dye, W. Dulaney, and J. Yang, 1991, Derivation of Quantitative NDVI Measurements From AVHRR Observations. Remote Sensing of Env., Vol. 35, pp. 257-277 Goward, S.N., D. Dye, A. Kerber, and V. Kalb, 1987, Comparison of North and South American Biomass From AVHRR Observations. Geocarto International, Vol. 1, 1987. pp. 27-39. Goward, S.N., C.J. Tucker, and D.G. Dye, 1985, North American Vegetation Patterns Observed With the NOA4-7 Advanced Very High Resolution Radiometer. Vegetatio, Vol. 64, pp. 3-14 Hall, F.G., D.B. Botkin, D.E. Strebel, K.D. Woods, and S.J. Goetz, 1991, Large Scale Patterns of Forest Succession as Determined by Remote Sensing. Ecology, Vol. 72, No. 2, pp. 628-690 Harmon, V. and M. Shapiro, 1992, Grass Tutorial: Image Processing. U.S. Army Construction Engineering Res. Lab., CERL, Champaign, Illinois. Hayes, L., 1985, The Current Use of TIROS-N Series Meteorological Satellites for Land Cover Studies. International. J. of Remote Sensing, Vol. 6, No. 1, pp. 35-40 Holben, B.N., 1986, Characteristics of Maximum Value Composite Images From Multi-Temporal AVHRR Data. International. J. of Remote Sensing, Vol. 7, No. 11, pp. 1395-1434 Janssen, L.F., M.N. Jaarsma, and E.T.M. Vanderlinden, 1990, Integrating Topographic Data With Remote Sensing for Land Cover Classification. Photogrammetric Engineering and Remote Sensing, Vol. 65, No. 11, pp. 1503-1506 Justice, C.O., J.R.H. Townshend, B.N. Holben, and C.J. Tucker, 1985, Analysis of the Phenology of Global Vegetation Using Meteorological Satellite Data. International J. of Remote Sensing, Vol. 6, No. 8, pp. 1271-1318 Kidwell, M., 1990, Global Vegetation Index Users Guide, U.S. Dept. of Commerce, NOAA, NESDES, NCDC, SDSD, Washington, D.C. Kolchugina, T.P. and T.S. Vinson, 1993, Equilibrium Analysis of Carbon Pools and Fluxes of Forest Biomes in the Soviet Union. Canadian Journal of Forest Research, Vol. 23, pp. 81-88 Loveland, T., J.W. Merchant, D.O. Ohlen, and J.F. Brown, 1991. Development of a Land-Cover Characteristics Data Base for the Conterminous United States, Photogrammetric Engineering and Remote Sensing, Vol. 57, No. 11, pp. 1453-1463 Lydolph, P., 1970, Geography of the USSR. Wiley and Sons, New York, New York. 667 pp. Milligan, G.W. and M.C. Cooper, 1985, An Examination of Procedures For Determining the Number of Clusters in a Data Set. Psychometrika, Vol. 50, No.2, pp. 159-179 Nalivkin, D.V., 1973, Geology of the USSR. University of Toronto 855 pp Press, Toronto, NOAA-EPA, Global EcoSystems Data Base Version 1.0: Users guide, Documentation and Digital Data on C.D. Rom, USDOC/NOAA, NCDC, Boulder, Colorado Norwin, J. and D.H. Greegor, 1983, Vegetation Classification Based on AVHRR Satellite Imagery. Remote Sensing of Env., Vol. 13, pp. 69-87 Parker, W.H., 1983, The World's Landscapes: The Soviet Union, Longman Publishing Co.. London, England, 207 pp. Ryabchikov, A.M. (ed.), 1988, Map: Geographical Belts and Zonal Types of Landscapes in the World. School of Geography, Moscow State University, Moscow, Russia SAS/STAT 1989, User's Guide, Version 6, Fourth Ed. Vol. 1. Cary, North Carolina: SAS Institute Inc., 1989, 943 pp. Sochava, V.B. (ed.), 1960, Map: Vegetation of the USSR. Ministry of the Interior, Moscow, Russia Spanner, M.A., L.L. Pierce, D.L. Peterson, and S.W. Running, 1990, Remote Sensing of Temperate Coniferous Forest Leaf Area Index: The Influence of Canopy Closure, Understory Vegetation and Background Reflectance. International J. of Remote Sensing, Vol. 11, No. 1, pp. 95-111 Swain, P.H. and S.M. Davis, Ed., 1978, Remote Sensing: The Quantitative Approach. McGraw Hill, New York, New York. 385 pp. Townshend, J.R.G., C.O. Justice, and V. Kalb, 1987, Characteristics And Classification of South American Land Cover Types Using Satellite Data. International J. of Remote Sensing, Vol. 8, No. 8, pp. 1189-1207 Townshend, J.R.G. and C.O. Justice, 1986, Analysis of the Dynamics of African Vegetation Using the Normalized Difference Vegetation Index. International J. of Remote Sensing, Vol. 7, No. 11, pp. 1435-1445 Townshend, J.R.G. and C.J. Tucker, 1984, Objective Assessment of AVHRR Data for Land Cover Mapping, International J. of Remote Sensing. Vol. 5, No. 2, pp.497-504 Tucker, C.J., J.R.G. Townshend, T.E. Goff, 1985, African Land Cover Classification Using Satellite Data. Science. Vol. 223, No.4685, pp. 369-375 Tucker, C.J. and P.J. Sellars, 1986, Satellite Remote Sensing of Primary Production. International J. of Remote Sensing, Vol. 6, No. 1, pp. 1395-1415 Tucker, C.J., C. Van Praet, E. Boerwinkle, and A. Gaston, 1983, Satellite Remote Sensing of Total Dry Matter Production in the Senegalese Sahel. Remote Sensing of Env., Vol. 13, pp. 461-474 U.S. Army Civil Engineering Research Lab, 1992, GRASS, Geographical Resource Analysis Support System, Program Documentation. US Army CERL, Champaign, Illinois U.S. Dept. of Agriculture, 1990, USSR Trade and Agriculture Report: Situation and Outlook. USDA, RS-90-1, May 1990 Vinson, T.S., T.K. Kolchugina, Pools and Fluxes of Biogenic Carbon in the Former Soviet Union. Journal of Water, Air and Soil Pollution. (accepted for publication). The Use of Global Vegetation Index (GVI) To Estimate Phytomass and Net Primary Productivity in Terresterial Ecosystems of the Former Soviet Union Greg G. Gaston' * Tatayana P.Kolchugina2 1. Geosciences Dept. Oregon State University Corvallis, OR 97331 2. Dept. of Civil Engineering 'Oregon State University Corvallis, OR 97331 * Please Direct Correspondence to: Greg Gaston U.S. EPA Research Lab 200 SW 35th Corvallis, OR 97333 (503)754-4496 (503)754-4338 FAX gregg@russia.cor2.epa.gov (internet) Abstract A limited number of functions in a terrestrial carbon budget can be assessed by satellite remote sensing systems. One approach for estimating phytomass accumulation and net primary productivity (NPP) using Global Vegetation Index (GVI) is to identify relatively homogenous vegetation and landcover regions. Regions with similar vegetation and land cover in the former Soviet Union (FSU) were identified using unsupervised classification of four year average monthly maximum value GVI composites. Using a variety of information and techniques, each of the 42 image classes was identified in terms of vegetation and land cover (i.e., % forest, % agricultural, and % natural non-forest). Accurate delineation and description of most vegetation and land cover complexes in the FSU was possible using classified GVI imagery. This method provided more accurate and precise descriptions for most ecosystems than is available on general thematic maps. There was good comparison between reported area statistics from the FSU and land cover estimates for image classes. The area of forest lands were estimated at 1,330 Mha with the actual area of forest ecosystems estimated to be 875 Mha. Arable lands were estimated to be 211 Mha. Using the classified image with the Bazilevich data base of phytomass and NPP for vegetation in the FSU and calculated estimates of phytomass and NPP in agricultural lands, the total phytomass in the FSU was estimated at 97.1 Gt C, with 86.8 in forest vegetation and 0.6 Gt C in arable lands. The total NPP for the FSU was estimated at 8.6 Gt C/yr, with 3.2 and 4.8 Gt C/yr in forest and natural non-forest ecosystems. The total phytomass of natural ecosystems compared well with estimates which used thematic maps to identify ecosystems. The phytomass estimated for forest ecosystems was higher than another assessment which considered the age-class distribution of forest stands in the FSU. The total NPP of natural ecosystems estimated in this study was 23 % greater than previous estimates which used thematic maps to identify ecosystems. The NPP densities estimated in this study were greater than a published where NDVI data was correlated to NPP parameters. assessment Introduction Natural processes in the oceans and terrestrial ecosystems, together with human activities, have caused a measurable increase in the atmospheric concentration of CO2. In 1988 the atmosphere contained 748 gigatons (Gt) of carbon, the largest amount during the last 160,000 years (Post et al., 1990). On average, the CO2 concentrations in the atmosphere appear to be increasing by 3.0 Gt/yr (Keeling, 1983; Tans et al., 1990) The long-term ecological consequences of change in the chemical composition of the atmosphere are not fully understood; however, a warmer global climate is possible (PPIGW, 1992). If CO2 concentrations were to double, the earth's temperatures could rise between 1 and 5° C (Schneider, 1990). Global warming may disrupt the equilibrium of the natural carbon cycle by accelerating the rates of plant respiration (Keeling et al., 1989) and decay of organic matter (Dixon and Turner, 1991; Raich and Schlesinger, 1992). It may be possible to manage terrestrial ecosystems to help offset increased amounts of atmospheric CO2 (PPIGW, 1992). Before any global strategy for environmental management aimed at mitigation of climate change can be formulated, accurate assessment of the carbon cycle within the national boundaries is necessary (Vinson and Kolchugina, 1993). The quantification of the carbon cycle following an assessment of carbon pools and fluxes is generally referred to as the carbon budget. The Former Soviet Union (FSU) was the largest country in the world, occupying one-sixth of the land surface of the earth. An understanding of the 100 carbon budget of the FSU is essential to the development of a global strategy aimed at mitigation the potential negative impacts of climate change. A Carbon Budget The biogenic carbon cycle consists of a combination of pools and fluxes. The pools are carbon stored in soil and vegetation. Effluxes are carbon emissions resulting from plant respiration and decomposition of organic matter. Influxes of carbon are represented by formation of new organic matter in soil and vegetation. The annual accumulation of organic matter in vegetation after carbon is expended for autotrophic respiration is known as net primary production (NPP). To calculate the carbon budget of any region, a geographic area within which carbon cycle parameters can be quantified must be isolated. The term ecoregion has been applied to the boundaries and areal extent of the geographic area and the term ecosystem has been applied to the soil-vegetation associations within an ecoregion (Vinson and Kolchugina, 1993). Carbon budget parameters may be expressed in terms of carbon content (for pools) or rate per hectare (for fluxes) for a variety of soil-vegetation associations. If the soil, vegetation, and land cover characteristics are accurately portrayed on the thematic maps used to identify ecosystems, then the carbon budget for an ecoregion can be established simply by multiplying the area of the individual ecosystems by the associated carbon contents. The carbon contents and fluxes for all ecoregions may be summed to arrive at the carbon budget for a larger region, or nation. Kolchugina and Vinson (1991) and Vinson and Kolchugina (1993) used this approach to assess the carbon budget of the FSU (including Russia, Ukraine, 101 Byelorussia, Kazakhstan, and the Baltic States). The southern republics (Turkmenia, Tadjikistan, Kyrgizia, and Uzbekistan), and the Trans-Caucasian republics (Armenia, Azerbaijan, and Georgia) were not included. Maps containing information on distribution of soil-vegetation associations within the FSU (Ryabchikov, 1988), wetlands (Isachenko, 1988), and arable lands (Cherdantsev, 1961) were used to isolate ecoregions. The areal coverage of the ecoregions was integrated with the carbon content and flux data for major vegetation and soil types in the FSU (Bazilevich, 1986; Kobak, 1988) to establish the carbon budget within each ecoregion. The key map in this study was the "Geographical Belts and Zonal Types of Landscapes of the World" (Ryabchikov, 1988). This global map represents a synthesis of soil, vegetation, and climatic parameters. The use of this very general map with the detailed Bazilevich data base required equating a range of carbon parameters from the data base to map legend categories. For example, the map legend describes only a single division in the coniferous forests in the FSU, namely "light-crowned coniferous" and "dark-crowned coniferous" forests. No other descriptions are provided. A range of carbon values from the Bazilevich data base were assigned to each map class, representing the range of possible vegetation types and phytomass and NPP densities possible within map legend categories. There was concern, however, about the information content of the maps used to identify the carbon quantifiable ecoregions. Each map, at best "...is a snapshot of the situation seen through the particular filter of a given surveyor, in a given discipline at a certain moment in time" (Burrough, 1987). Cartographic data are generally the result of observations taken over a very short time, thus maps are 102 static sources of information. There are problems with scale changes between maps, significant differences in the level of detail between maps, and changes in classification schemes between maps and across political boundaries. The placement of boundary lines between areas that grade into each other over long distances is an arbitrary decision of the individual cartographer. Additionally, the size and inaccessibility of much of the land area of the FSU necessitates cartographic generalization, introducing interpolation errors. There is wide agreement that satellite data can be used to overcome the limitations of thematic maps, providing timely, consistent and reliable information for the entire globe on the areal extent and conditions of terrestrial ecosystems (Norwin and Greegor, 1983, Badwhar, 1986, Janssen et al., 1990, Justice et al.1985, Hall et al., 1991). Satellite Remote Sensing Systems The abilities of satellite sensor systems to measure ecosystem functions are limited as these systems generally record only reflected radiation from the surface. Surface conditions represent a complex mixture of vegetation, exposed soil or rock, water and shadows. The species composition, structure, density and levels of photosynthetic activity of both the dominant vegetation and understory vegetation are all a significant part of the signal recorded by satellite remote sensing systems (Spanner et al., 1990; Knipling, 1970). Other conditions such as the "brightness" or color of the exposed soil (rock) and the amount and density of shadows present are all integrated into a satellite observation of terrestrial ecosystems (Huete et al., 1985). Not only are the data recorded by a satellite sensor a result of complex surface interactions but many important ecosystem processes cannot be directly measured. Soil processes, litter accumulation, and mortmass accumulation for example are not part of the signal received by satellite remote sensing systems. Before further discission of how remote sensing data from satellite systems can be used to estimate components of the terrestrial carbon cycle it is important to define the system most appropriate for regional, contentional, and global scale investigations of the carbon cycle. The Advanced Very High Resolution Radiometer (AVHRR) appears to be the most useful satellite remote sensing system presently available for continental scale studies of terrestrial ecosystems. The AVHRR is a part of NOAA's meteorological satellite program which is designed to provide daily coverage of atmospheric conditions over the entire globe. The orbit and resolution of this sensor system while designed for meteorological data collection, has been successfully used to monitor conditions in terrestrial ecosystems (Tucker et al., 1983, Hayes, 1985, Townshend et al., 1987, and Fung et al.,1987). A comparison of landcover mapping based on the 1.1 km AVHRR data to landcover maps based on 80 m Landsat MSS indicated the accuracy for classifications based on AVHRR data was 71.9% compared to 76.8% for MSS. This accuracy is considered acceptable for most research and given the disparity in spatial resolution, the accuracy of the AVHRR is exceptional (Gervin et al.,1985). A normalized ratio of channel 1 (0.58-0.68 microns) and channel 2 (0.725- 1.1 microns) is the most commonly used product from the AVHRR for analysis of conditions in terrestrial ecosystems. The normalized ratio of visible and near infra red reflectance is commonly termed the normalized difference of vegetation index (NDVI). The equation used to calculate NDVI is as follows (Kidwell, 1990): CH.2 (near IR) - CH.1 (visible) NDVI = --------------------------------- (1) CH.2 (near IR) + CH.1 (visible) Normally active photosynthetic vegetation absorbs the visible wavelengths (AVHRR Ch.1) (especially red) and reflects near infra-red radiation (AVHRR Ch.2). Higher photosynthetic activity results in increased red absorption and even higher near infra-red reflection. The NDVI values range from -1 to +1, with most vegetation clustering around +0.6 NDVI; higher values are characteristic of higher photosynthetic activity. This index has been related to primary productivity (Tucker et al.,1983), leaf area index (LAI) (Spanner et al.,1990), percent vegetative cover (Townshend et al.,1987) and biomass and absorbed photosynthetically active radiation (PAR or IPAR) (Goward et al.,1990). There appears to be a strong correlation between NDVI and the vegetation in terrestrial ecosystems. Most research using AVHRR data concentrates on the NDVI to evaluate vegetation attributes in the analysis of terrestrial ecosystems. The highest spatial resolution for the AVHRR scanner is 1.1 km at the nadir. As the scan widens to acquire data across the entire 112 degrees of the sensor swath width, the spectral and spatial resolution degrades due to increases in atmospheric path length, curvature of the earth, and changes in reflection angles. The wide scan angles and daily repeat of orbits produce a tremendous amount of data. A lack of on-board storage and limited down-link connections require on-board pre-processing and sampling of the data (Goward et al.,1990). The first of the pre-processed products is known as global area coverage (GAC). The GAC data are produced on a daily basis by averaging the first 4 pixels in a 5 x 3 pixel array, and assigning this average value to the entire array, producing a nominal 4 x 4 km pixel (Goward et al.,1987). The GAC data are further processed by NOAA, by selecting the "greenest" pixel (the highest NDVI value) in a 4 x 4 pixel array. The resulting product, known as the global vegetation index (GVI) reflects the highest NDVI value occurring in a seven day period over a nominal 16 x 16 km pixel (Kidwell,1990). The presence of clouds (on an average day one half the surface of the globe is obscured by clouds), shadows, off nadir `look' angles, and different reflection ratios all tend to reduce NDVI values. In order to reduce the effects of these factors a temporal compositing routine is used. The maximum NDVI is usually obtained in an unobscured near nadir view. A maximum value composite is constructed by taking the maximum value for a given pixel in a given time frame. Weekly and monthly maximum value composites are regularly produced using a set of georeferenced AVHRR images (Kidwell, 1990). This process appears to remove most of the problems associated with clouds, off nadir views, and other factors that reduce NDVI (Holben,1986). According to Goward et al. (1990) errors can be reduced to ± 10% (0.1 NDVI unit) using a maximum value compositing process on a monthly time step. 106 The maximum value compositing procedure assumes that the surface features are homogenous over the sampling area. This procedure may overestimate the actual extent of photosynthetically active vegetation. In choosing the "greenest" pixel it is possible that an oasis would be chosen to characterize an otherwise barren desert (Townshend and Justice, 1986). The temporal signature of a heterogenous landscape may also be confused using this procedure. For example, regenerating broadleaf vegetation in a disturbed area in a conifer forest may produce "greenest" pixels that represent these areas of regeneration rather than the bulk of the conifer forest. Further, the temporal compositing procedure which simply assigns the value of a single 4 x 4 km GAC pixel to the entire 16 x 16 km GVI pixel, regardless of time or the spatial location of each pixel, makes locating the exact geographic position of individual pixels impossible. For identification of vegetation and land cover in the FSU, GVI monthly maximum value composites from 1985 through 1988 were selected as the most appropriate remote sensing data presently available. Approaches for the use of AVHRR data Two basic approaches can be taken to the use of remote sensing data from satellite sensors to estimate carbon budget dynamics: (1) direct quantitative calculation of ecosystem characteristics using remote sensing data as a primary indicator, or (2) use of the unique signal created by various ecosystem characteristics to identify homogenous vegetation/land cover regions. These regions 107 should be an accurate representation of surface conditions which can be linked with plot level field studies to extrapolate carbon budget estimates. A strong correlation between NDVI ("greenness") and biomass has been reported by a number of authors, Tucker et al. (1981,1983), Kimes et al. (1981) and Goward et al (1985, 1987). Many of the studies have focused on homogenous agricultural crops or relatively simple grassland plant canopies (Tucker, 1983). In these environments a linear relationship appears to exist between LAI (calculated as a function of NDVI) and NPP. In more complex ecosystems a relationship exists but its form is less predictable. Goward et al. (1985, 1987) identified the apparent linkage of integrated area under the curve produced by plotting NDVI values as they change with time against biomass estimates for different ecosystems. Fung et al. (1987) reported a linkage between NDVI in terrestrial vegetation and measured changes in the concentration of atmospheric CO2. Runyon et al. (1991) however, reported that NDVI (as a surrogate for intercepted radiation) was not a good predictor of higher biomass levels (greater above-ground NPP). Badwhar et al. (1986) emphasizes that the use of ratios calculated for agricultural crops or range land may not be suitable for estimating the biomass of a deciduous forest canopy. Rather than attempting to isolate factors in the overall remote sensing signal which can be related to ecosystem productivity, a second approach, utilizing all the unique factors that make up the signal to identify complexes of vegetation and landcover has been used by a number of authors. For example, Goward et al. (1985), used three week composites of GVI data to successfully identify and characterize vegetation patterns of North America. lob Justice et al. (1985) using, in part, GVI data concluded that the extent and seasonal dynamics of global vegetation could be successfully mapped on a global scale. A variety of classification schemes were tested by Townshend et al. (1987) in research directed towards classifying vegetation complexes in South America. The research concluded that a maximum likelihood classification produced the most satisfactory delineation of South American vegetation complexes, when compared to the UNESCO 1981 vegetation map. Loveland et al. (1991) used unsupervised classification of monthly maximum value composites (1 km resolution) to characterize the land cover of the conterminous United States. They reported that large homogenous land cover regions were well identified, citing forest ecosystems as prime examples of successful classification. The Image-Map of the FSU Unsupervised classification of monthly GVI composites was used to create an image that delineates regions of similar vegetation and landcover in the FSU. In order to reduce the possible effects of inter-annual variation, monthly GVI composites from 1985 through 1988 were averaged, creating a single set of monthly observations that represented conditions over a four-year period. Unsupervised classification required two steps: (1). identification of "clusters" within the data (cluster analysis) and (2) a maximum likelihood grouping of the entire data set around these "cluster renter-," Clusters were identified on the basis of timing, magnitude, and duration of "greenness" as recorded by NDVI change with time. The complete signature of understory vegetation, geometric structure of vegetation communities, and land cover differences within the instantaneous field of view (IFOV) of the AVHRR sensor, were integrated into the remote sensing observations. While it was nearly impossible to separate the individual influences of each factor, it was possible to identify regions with similar signatures. Figure 11 presents the NDVI response with time for four sample image classes. The shape and magnitude of these curves form the signature of each class and were the basis for identifying clusters in the data. These signatures were very useful in defining the vegetation and land cover matrix that comprise each image class. Using a variety of quantitative and qualitative methods including comparison with existing vegetation maps and consultations with scientists with extensive field experience in the FSU' it was possible to describe each of the 42 spectral/temporal classes on the image in terms of vegetation and land cover. The image map of the FSU created by clustering eight months of four year average GVI data is presented in a simple latitude-longitude format in Figure 3 (page 83). The patterns of the image classes reflect the expected patterns in vegetation that result from climatic factors (re. Figure 3, page 83). Latitudinal banding in the image classes reflects the effects of extreme cold and moisture deficits. Image class differences also reflect species change from west to east. From north to south there 'M.S. Botch, Senior Researcher, Komarov Botanical Institute, St. Petersburg, Russia, and K.I. Kobak, Senior Researcher, State Hydrological Institute, St. Petersburg, Russia is an increase in biomass as related to NDVI "greenness" with an increase in the percentage of woody vegetation which is represented in image classes 2, 3 and 5, 4, 6, 8, 10. These classes represent arctic (high mountain) tundra (image class 2) transitioning through shrubby tundra (image class 6, 8, 10) to the forest/tundra (image class 12) boundary where the sparse forests of the northern taiga begin to dominate the vegetation communities. A similar pattern may be observed in the elevational gradient of the Burranga mountains, the Putorana mountains and other mountain ranges of Eastern Siberia. In the south, the effects of moisture deficits on vegetation communities are reflected in image classes 13 (desert) through 14 and 28, 17, 26 and 23, 35 and 32, which represent increasing vegetation grading from semidesert steppe (image class 14 and 28) into forested meadows (image classes 35 and 32). The northern taiga forests of the FSU are identified by classes 11, 15, 16, 18, 20, and 23. The dwarf and shrubby forests of the Pacific coast are represented by image classes 9, 19, and 22. The middle taiga forests are identified by image classes 24, 25, 29, 30, and 31. The southern taiga forests are represented by image classes 33, 36, and 37. The mixed-conifer and broadleaf forests are identified by image classes 38, 39, and 40. The broadleaf forests with a smaller percentage of conifers and agricultural lands are identified by image classes 41 and 42. Underestimation of wetlands appears to be a problem with these image processing techniques. Areas on the image known to contain extensive wetlands (e.g. the western shore of the Kamchatka Peninsula and the lower Ob basin) are not well identified by image class differences. This underestimation of wetlands is also a problem with thematic maps of the FSU (M.S. Botch, 1992, personal communication). Linking Phytomass and NPP Data to Image Classes Image classes described in terms of vegetation, agricultural land use, and geographic location were correlated to the data base compiled by Bazilevich (1986). The Bazilevich data base contains phytomass and NPP estimates for all major vegetation formations in the FSU. Along with the descriptions of vegetation and land cover provided for each image class, the image class signature curves were a useful tool for assigning phytomass and NPP estimates from the Bazilevich data base. In order to further refine the carbon assignments for each image class using Bazilevich data base it was necessary to estimate some of the individual components that make up the vegetation and land cover of each image class. For example, image class 35 was characterized as "forest-meadow-steppe, with significant agricultural land use". Forest vegetation has a very different phytomass and NPP than meadow-steppe vegetation, both of which are very different from the phytomass and NPP found in agricultural land. The percent forest cover for the taiga forests in the FSU averages 65 % (Kolchugina and Vinson, 1993). It was assumed that even the most dense of forests 112 identified in an image class contained non-forested patches that made up at least 20 percent of the area identified as forest. Using available published data on percent forest cover of forest land in the FSU (Vorobyov, 1985; State Forestry Committee, 1990), areal extent of forest ecosystems was identified in each image class. For example, the forest-steppe ecosystems represented by image classes 32 and 35 (forest-meadow-steppe) average 15 % forest cover according to Vorobyov (1985). In the absence of published data an expert assessment of forest density was made for each image class. This assessment considered the geographic location of image classes and the density of "greenness" as recorded by NDVI and graphed in the signature of each image class. The total estimates of area for each land cover type in all image classes are presented in Table 9. The non-forest areas in each image class were either considered agricultural lands or natural non-forest vegetation. If the natural non-forest vegetation types found in each image class were not specifically described in the Bazilevich data base, carbon values which reflect the natural non-forest vegetation most likely to be found in each geographic location were used. For example, image class 18 (i.e., northern taiga spruce/larch) was assigned a percent forest cover of 80 % (re. Table 1). Phytomass and NPP values for forest meadows which most likely represent the character of non-forested vegetation in this region were used. Agricultural statistics provide data which may be used to estimate phytomass and NPP of agricultural lands. Most agricultural statistics are expressed in terms of total yield by crop, or yield per unit area (t/ha). Equations that convert crop yield to total phytomass and NPP were calculated by Sharp et al. (1976). These equations were used together with production data reported in USDA agricultural 113 statistics for the FSU (USDA, 1990). Both phytomass and NPP were calculated for each crop reported (phytomass is equivalent to NPP in the case of annual crops). The coarse resolution of the remote sensing data and the classification limits imposed by the diversity of the data set prevented the separate identification of individual cropping patterns. Therefore, all production data were combined and a single weighted average NPP calculated for all agricultural lands in the FSU. This single value (2.85 t C/ha) was used to estimate the phytomass and NPP for agricultural lands in the FSU. Estimates of Area Information on the location and extent of agricultural lands was provided by the arable lands map of Cherdantsev (1961). When compared to all available descriptions of agricultural land in the FSU, the first three categories of this map (greater than 60 % arable, 30-60 % arable, and less than 30 % arable) appeared to accurately identify agricultural lands in the FSU. Assigning appropriate factors (e.g., 80 %, 45 %, 15 %) for each of these three classes it was possible to calculate the area of agricultural lands in the FSU at 211 Mha. This estimate was identical to USDA estimates of agricultural land in the FSU (USDA, 1990). Sokolovskiy et al. (1989) report the agricultural lands in the FSU to be 217 Mha, which compares favorably with the 211 Mha estimated in this study. It must be acknowledged that this excellent comparison in total area may be the result of a number of errors that, in the final calculation, cancel each other out. The estimates for forest and natural non-forest landcover (re. Table 9) correspond well to published data for each land cover type in the statistical reports from the FSU. The forest statistical data base for the FSU (Alimov et al., 1989) reports the total area of forest land in the FSU (area of forest biomes as defined in Kolchugina and Vinson, 1993a) at 1,254.2 Mha (36.6% of the total land area in the FSU). The total area of image classes described as forested with no allowance made for forest density (percent forest cover) is 1,330 Mha (39 % of the total land area in the FSU). Forest statistical data of the FSU report the area of commercial forests at 811 to 814 Mha. This compares favorably with the estimates of 875 Mha of forest ecosystems calculated for the image classes using the percentages of actual forest cover provided by an expert assessment (re. Table 9). The higher estimate of forest ecosystems from image classification, may result from: 1. Exclusion of forests with no commercial value from the forest statistical data of the FSU, while image class descriptions included all types of forest cover. For example, in the vicinity of the Taimir peninsula a significant area of forest-tundra was identified by image class descriptions. However, the forest statistical data for the Taimir oblast (an administrative division), report no commercial forests in this area, or 2. The "oasis effect" inherent in GVI sampling. It is possible that isolated patches of forest with high NDVI values incorrectly characterize entire GVI pixels producing an overestimation of forest ecosystems. Kolchugina and Vinson (1993a) estimated area of forest ecosystems in the FSU at approximately 800 Mha. Evaluating Phytomass and NPP Estimates In order to compare results based on image classification to the estimates made by Kolchugina and Vinson (1991) and Vinson and Kolchugina (1993), it was necessary to (1) "mask out" areas not evaluated by Kolchugina and Vinson (1991) and Vinson and Kolchugina (1993), and (2) group image classes into the same general biomes used in their analysis. The term biome is applied to the complex of ecosystems within a climatic region. Kolchugina and Vinson (1991) and Vinson and Kolchugina (1993) identified nine biomes in the FSU: polar desert, tundra, forest-tundra/sparse taiga, taiga, mixed-deciduous forest, forest-steppe, desert- semidesert, and mountainous subtropical woodlands. Figure 12 presents a comparison of area by biome. Overall, the comparisons between area estimates based on thematic maps and the image based estimates are good. The most significant differences between the image based classification and the estimates made by Kolchugina and Vinson (1991) and Vinson and Kolchugina (1993) resulted from the differences in the biome description. For example image class 11 described as "sparse northern taiga larch" forest was assigned to the taiga biome, while in the previous assessment this vegetation type was assigned to the forest- tundra/sparse taiga biome. Phytomass and NPP Table 10 presents the phytomass and NPP estimates based on image classification. The total phytomass in the FSU was estimated at 97.1 Gt C, with 86.8 Gt C in forest vegetation, and 0.6 Gt C in crops (re. Table 10). Thus the total phytomass of natural ecosystems was estimated at 96.5 Gt C, which compared favorably with the estimate made by Vinson and Kolchugina (1993) (91.0 Gt C). Estimated forest phytomass compared well with the estimate of Kolchugina and Vinson (1993b) (87.7 Gt Q. The image based estimate of forest phytomass was 116 twice as large as an estimate made for forests considering age class distribution of forest stands in the FSU (Kolchugina and Vinson, 1993c). Total NPP was estimated at 8.6 Gt C/yr (re. Table 10), with 3.2, 4.8, and 0.6 Gt C/yr NPP in forest, natural non-forest, and arable ecosystems, respectively. The NPP of non-arable land was 8.0 Gt C/yr, which was 23 % greater than estimated by Vinson and Kolchugina (1993). Forest NPP was 34 % greater than the estimate made for forest ecosystems (2.1 Gt C/yr) which considered age-class distribution of forest stands (Kolchugina and Vinson, 1993c). The phytomass and NPP densities as depicted in Figures 13 and 14 reflect generally good agreement between the present assessment an earlier estimates, with the exception of desert-semidesert biome. Vinson and Kolchugina assigned higher values of phytomass and NPP to the desert-semidesert biome reflecting a variation of vegetation types from low productive scattered herb cover to the woody shrub formations with well developed root structure and high values of phytomass and NPP. Present assessment considered relatively low values of GVI of the desert- semidesert classes when correlating them to Bazilevich data base. In addition, irrigated agriculture is a significant feature in this region. Vinson and Kolchugina (1993) did not separate this ecosystem from their estimate of desert-semidesert biome. Inclusion of irrigated agriculture into the desert-semidesert biome produced higher estimate of NPP made for the desert-semidesert biome (re. Figure 14). NPP estimates were also compared to estimates based on Fung et al. (1987) (re. Figure 15). In almost every case the image based estimates were greater than those based on Fung et al. (1987). This disagreement reflects the overall differences in NPP parameters of the Bazilevich data base and NPP data of Fung et 117 al. (1987). The Bazilevich data base tends to emphasize mature (established) forest ecosystems with higher phytomass and NPP than developing forests. It is possible that Fung et al (1987) used NPP data which underestimated the production of below-ground and woody parts of plants. The exception to this pattern, again, was in the desert-semidesert biome where image based estimates were lower than those of Fung et al. (1987). This difference may be a result of generalization of the desert and semi-desert ecoregion by Fung et al. (1987). By including isolated ecosystems with high annual NPP, a higher average NPP value for the entire class may be generated. It also may be possible that in relying on the image classes to accurately identify ecosystems and using the temporal NDVI response curves to help select representative phytomass and NPP values from the Bazilevich data base, there was a tendency to select values that underestimate the phytomass and NPP for these ecosystems. Summary and Conclusion Accurate delineation and description of most vegetation and land cover complexes in the FSU was possible using classified GVI imagery. This method provided more accurate and precise descriptions for most ecosystems than is available on general thematic maps. Good comparison was found between reported area statistics from the FSU and land cover estimates for image classes, The area of forest biomes was estimated at 1,330 Mha, the area of forest ecosystems was estimated at 875 Mha. Arable land was reported by the USDA (1990) at 211 Mha, and Sokolovskiy et al. (1989) reported the agricultural lands in the FSU to be 217 Mha, which compares favorably with the 211 Mha estimated in this study. Using the Bazilevich data base for natural ecosystems and a weighted average calculated from yield data for agricultural systems, image class based estimates for the total phytomass in the FSU were 97.1 Gt C, with 86.8 and 0.6 Gt C in forest vegetation and arable land, respectively. Total NPP was estimated at 8.6 Gt C/yr, with 3.2, 4.8, and 0.6 GtC/yr of forest, natural non-forest, and arable ecosystems, respectively. The total phytomass of natural ecosystems compared well with the published estimates based on the use of thematic maps to identify ecosystems. The phytomass estimated for forest ecosystems was significantly higher than a previous assessment which considered the age-class distribution of forest stands in the FSU. The total NPP of natural ecosystems was 23 % greater than estimates which used on thematic maps to identify ecoregions. NPP densities estimated for nine biomes in the FSU in this study were greater than an assessment where NDVI data was correlated to NPP parameters. The differences in estimated NPP may be a result of differences in the data base used to estimate carbon storage and flux. Acknowledgment The work presented was funded by the U.S. Environmental Protection Agency (EPA) - Environmental Research Laboratory, Corvallis, Oregon, under a cooperative Agreement CR820239 to Oregon State University. Jeffrey J. Lee is the Project Officer for the project entitled "Carbon Cycling in Terrestrial Ecosystems of the Former Soviet Union." This paper has not been subjected to the EPA's review and, therefore, does not necessarily reflect the views of the EPA, and no official endorsement should be inferred. Table 9 Landcover, Phytomass and NPP Densities for Image Classes Phvtomass (1/ha) NPP (t/ha/vr) NonDescription Class Ih Non- Area (Mha) 3 Arctic tundra % Fores % Arabic Non-forest Forest Arabic forest Forest Arabic forest 25.74106399 0.00°/ 0 00% 0 00 2.8! O.OC 100.00°/ 0.0( 0 OC 0.00 31.10372839 0.00°/ 0.00% 100.00% 0 00 2.8! O.80 00( 2 85 0.01 35.63446643 0.00% 0 00% 100 00% O.OC 2.8! 2.95 0.00 2.85 0.7! 4 Mountain tundra 29.39552146 0.00% 1.70% 98.30% O.OC 2.85 4 8C 0.00 2.85 1.1! 46.82851701 0.00% 0.00% 100 00% O.OC 2 85 11.2C 0.00 2 85 1.3! 46.40196614 0.00°/ 0.00% 100.00% 0 OC 2.85 11.50 0.00 2 85 2.65 42.6740631 0.00°/ 11.25% 88 75% 0.00 2.85 3 75 0.00 2.85 2.90 5 "Typical"/shrubby tundra 37.84658304 0.00% 000% 100.00% 0 OC 2.85 120C 0.00 2.85 2.70 9 Dwarf shrub (NE coast) 27.71765332 0.00% 0.00% 100.00% 0 OC 2.85 490( 0.00 2.85 1.45 1C Shrubby tundra 44.58128004 0.00% 0.00% 100.00% 2.85 11.9! 0.00 2.8! 2.70 11 Northern taiga larch (sparse) 80.59291861 50.00% 0.000% 00( 5000% 306( 2.85 11.20 1.75 2.8! 1.50 12 Forest-tundra 48.89610815 20.00% 0.00% 80.000% 30.7! 2.85 11.20 23C 2.85 1.50 13 Desert 109.7763894 0.00% 0.680% 99.320% 00( 2 85 1.00 0.00 2.85 0.50 14 Semidesert steppe 71.60893556 0.00% 0.89% 99.11% 0.0( 2.85 4 90 0.00 2.8! 3.30 15 Larch on mountain slopes 66.67558662 50.00% 2.90% 47.10% 31.0( 2 85 6.70 18( 2.85 7.30 16 Northern taiga larch 78.98773814 80.00% 0.00% 20 00% 32.45 1 Water 2 Mountain tundra 5 "Typical'/ tundra 6 Shrubby mountainous tundra 7 Solonchack 6.70 1.8! 2.85 7.30 0.00% 8.26% 91.74% 0 00 2.8! 7.30 00( 2 8! 6.85 l8 Northern taiga spruce/larch 82.79395443 8000% 1.96% 18040/4 61.90 2 8! 19 Mountain shrub (Asia) 22.41326032 0.00% 0000/4 100.00% 0.00 20 Northern taiga spruce 46.41969123 80.00% 0000/4 20.00% 52.64188225 0.00% 8.000% 22 Dwarf forest (Pacific Coast) 55.09691881 70.00% 23 Scattered forest 57.10990325 24 Middle taiga larch/pine 25 21 Irrigated agriculturbr t I it 6 7C 24( 2.85 7.30 2.8! 40.50 O.0C 2.8! 2.15 61.50 2 8! 6 7C 2.4( 2.85 7.30 92.00% 32.50 2 8! 7.45 2.85 2.8! 2.85 0.00% 30.00% 42.45 2 8! 6 70 3.4! 2 85 73C 50.00% 17.46% 32.54% 65.25 2.8! 7.45 3 55 2.85 6.55 52.2949575E 80 00% 0.00% 20.00% 70.45 2 8! 6.70 2 05 2.85 7.3( 75.39311802 50.00% 5.20% 44.808/4 33.5! 2.85 6 70 12! 2.85 7.3( 119 2 85 68.7003375 11 Artem. and grass steppe I Table 9 (contd.) Landcover, Phytomass and NPP Densities for Image Classes Phytomaa3 Class 11 Description 26 Grassy -steppe 2111 Kamchatka forests 28 Semidesert steppe 29 Middle taiga pine/sprucc 30 M:JJI_ wuuic Taiga sprucevnr/pme (t/ha) NPP (t/ha/vr) Non- Non Area (Mha) % Fores I % Arable Non-forest Forest Arable forest Forest Arable foxes 1 39.72402402 0.00% 36000/1 64 00% 151.30 2 85 6 95 5.50 2 8! 7.4. 23.89714841 70.00% 2 07% 27.93°/, 1181( 2 85 6.70 3.85 2.85 7.3( 21.3125764E 0.00% 7.10% 92.90°/ 103.4( 2.85 8.10 2.9( 2 85 8.9( 67.31948441 65 001/4 6.20% 28.800/, 95.30 2.85 6.70 2.60 2.8.) 74 27422755 73( 70.00% 2 20% 27.806/, 128 7C 2.85 6.70 3.1( 2.85 7.3( 51.1174943E 70.006% 3.05% 26.96% 64 9C 2.85 6.70 2.6( 2.85 58.13327201 7.3( 15.00% 13 00% 72.00% 118 7C 2.85 8.10 4.15 2.85 74.00376915 65.004% 23.12878312 10.00% 8.90 14 00% 21.00% 118 25 2.85 7.45 3.15 44.00°,1 46.000% 144.2C 2.85 7.45 7.40 68.67734205 2.85 10.4( 2.85 10.4( 15 008/c 64.008/4 21 000/, 151.3C 2.8! 8.10 7.00 72.4816053! 2.85 80.004% 2.580/1 37 Southern taiga mixed forest (Russia) 17.42% 128 OC 2.85 7.45 41C 48.72437352 2.85 10.4( 50.0061, 38 Mixed conifer/b roadleaf z7.00% 23 00% 99.45 2 85 8 10 3.05 69.3674436E 2 85 56.90% 43.000/1 39 Mixed conifer/broadleaf (Asia) 0 10% 160 85 2 85 7.45 7.2C 56.50604175 2.85 10.4( 70 00% 40 Mixed forest (southern taiga) 9 25% 20 75% 160 85 2 85 7.45 7 20 48 52276405 2 85 10 4C 80 00°/ 2.45% 17.55% 210.50 2 85 81C 4.20 66.29212437 2 85 8.9C 5800°4 4200% 000°/ 2160( 2 85 8 10 6 50 28.21865235 2 85 8.9C 80.000% 736% 12.64% 235.5( 2 85 8 10 12.45 2 85 8 9C 31 Middle taiga conifer on moanrains 37 Forest-meadow-s teppe 33 Southern taiga pine/spruce 34 Grassy-steppe (mountain) 35 Forest meadow steppe (agriculture) 3E Southern taiga conifer forest on mountains 41 Broadleaf forest (agriculture) 42 Mixed forest (very productive) i 8.9( 8.9( Table 10 Total Phytomass and NPP for Image Classes and Landcover Type Area (Mha) Class # 1 Forest Arable Total Phytomass (Mt C) Non-Forest Arable Forest Non-forest Total NPP (Mt C) Forest Arable Non-forest 0.00 0.00 0.00 25.74 0.OC 0.00 0.00 0.0C 0.00 0.00 0.00 31.10 0.00 0.00 24.8E 0.00 0.00 0.31 0.00 26.72 33.23 0.00 0.00 35.63 0.00 0.00 105.12 0.00 0.00 0.5C 28.90 0.00 1.42 138.70 0.00 1.42 0.00 0.00 46.83 0.00 0.00 524.48 0.00 0.00 63.22 0.00 0.00 46.40 0.00 0.00 533.62 0.00 0.00 122.91 0.00 4.8C 37.87 0.00 13.68 142.02 0.00 13.68 109.83 8 0.00 0.00 37.85 0.00 0.00 454.16 0.00 0.06 102.19 9 0.00 0.00 27.72 0.00 0.00 1358.17 0.00 0.00 40.19 10 0.00 0.00 44.58 0.00 0.00 532.75 0.00 0.00 120.37 11 40.30 0.00 40.30 1233.01 0.00 451.32 70.52 0.00 60.44 12 9.78 0.00 39.12 300.71 0.00 438.11 22.49 0.00 58.68 13 0.0C 0.75 109.03 0.00 2.13 109.03 0.00 2.13 54.51 14 0.00 0.64 70.97 0.00 1.82 347.76 0.00 1.82 234.21 229.25 4 6 33.34 1.93 31.40 1033.4 7 5.51 210.41 60.01 5.51 16 63.1S 0.00 15.80 2050.52 0.00 105.84 116.90 0.00 115.32 17 0.00 5.67 63.03 0.00 16.17 460.09 0.00 16.11 431.73 18 66.24 1.62 14.94 4099.96 4.62 100.01 158.96 4.62 109.03 15 0.00 0.00 22.41 0.00 0.00 907.74 0.00 0.00 48.19 2C 37.14 0.00 9.28 2283.85 0.00 62.20 21 0.00 4.21 48.43 0.00 12.00 360.81 0.00 22 38.51 0.00 16.53 1637.20 0.00 110.74 28.42 138.4! 23 28.55 9.91 24 41.84 25 37.70 26 Iq, 15 0.00 67.77 12.00 138.03 133.0E 0.00 120.64 101.37 28.42 121.72 89.1 1863.21 0.00 10.4E 2947.34 0.00 70.0E 85.76 3.92 33.7E 1264.72 11.17 226.30 47.12 11.1 246.57 0.00 14.30 25.42 0.00 40.7E 176.65 0.00 40.76 189.40 27 16.73 0.49 6.61 1975.51 1.41 44.72 64.40 1.41 48.72 28 0.00 1.51 19.80 0.00 4.31 160.38 0.00 -4.31 176.21 25 43.76 4.11 19.35 4170.11 11.90 129.90 113.77 11.90 141.53 30 51.95 1.63 20.65 6691.31 4.66 138.34 161.18 4.66 150.73 31 35.78 1.5E 13.78 2322.21 4.44 92.32 93.03 4.44 100.58 32 8.72 7.5E 41.86 1035.06 21.54 339.03 36.19 21.54 372.52 33 48.10 10.36 15.54 5688.11 29.53 115.78 151.52 29.53 161.62 34 2.31 10.1E 10.64 333.52 29.00 79.26 17.1 29.00 110.65 35 10.30 43.95 14.42 1558.63 125.27 116.82 72.11 125.21 128.36 36 57.99 1.81 12.63 7422.12 5.33 94.07 237.74 5.33 131.31 31 24.36 13.16 11.21 2422.82 37.49 90.71 74.30 37.49 99.74 38 39.41 29.83 0.07 6348.76 85.01 0.52 284.18 85.01 0.72 39 39.55 5.23 11.73 6362.30 14.85 87.361 284.79 14.89 121.95 40 38.82 1.19 8.52 8171.23 3.35 68.98 163.04 3.35 75.79 41 38.45 27.84 0.00 8305.08 79.35 0.00 249.92 79.35 42 22.57 2.08 3.51 5316.39 5.92 28.891 281.06 5.921 31.74 875.54 210.93 1142.56 86837.40 9676.671 3169.6E 601.141 4773.05 -- . 18.58 0.00 76.3! 1, E 601.14 --=i =-4 _ D Total 0.00 1 I v9 Ii" f A'A!;a. 119. I.' Class 16 Clas-g 2 zim a iu IN in n 1.M ".r Apd Max MY Ions dro A.pV.. - Sid Dev _ Avcragc - Min 0 I..J u- 1 Ap.a - I.I A.. .. Mn a- Sid Dcv _ _ Avcraee __ M1;,I Iw. 1.1., S.p.... r o... S.p. u . 0o Northern Taiga Larch Mountain Tundra Class 39 3M) Class 14 60 Im -in 0 App A1.r Mar I.ry Ialr A.p..x Lp.nM. Sid I)- _._ Avclagc , MIn Mixed Broadleaf/Coniter''IFlirest Mb.'- Apa 16p -.-.. Max 1- IQ, Aup.. _+ - SW Dcv __ Avcragc -0- Min Semi-Desert Steppe Figure 11. Comparison of the Signatures of Sample Image Classes .Arable; irrigated. land and meadow (237 Mha, I 1%) Tundra (261 Mha, 13%) Desert, semi-desert (126 Mha, 6%) Fairest tundra, shrubland and dwarf forest (153 Mha, 7%) Steppe (123 Mha, 6%) Arable (191 Mha, 10%) Tundra (226 Mha, I1%) Desert, semi-desert (174 Mha, 9%) Forest tundra (340 Mha, I 7%) Stcppe (1-17 Mha, 6%) : as, Forest steppe Forest steppe- (144 Mha, 7%) (11.0 Mha, 5%) Mixed deciduous (196 Mha, 10%) Mixed deciduous (141 Mha, 7%) Taiga (822 Mha, 40%) Total: 2065. Mha Present :Study (Note: Non-arable land inc.ludes forest and non-forest vegetatio+n in class) I Taiga (709 Mha, 35%) Total: 2008 Mha Vinson and Kolchugina (1993) If Figure 12. Area of Biomes Identified by Image Classes and Thematic Maps a 'II 1,0 VIII II .0"0 (1993): Figure 13. Comparison of Phytomass Densities MOR : Al sei>::x::c. . O C O ba saftsuOU ddNI JO uosLmduuoD C rO3 ftl 1'25 1i ain2t3 meadows land/ Arablelimgated Forest-steppe Mixed-deciduous Taiga forest dwarf Forest-tundra/shrubland/ 0-4 LA ao M 9ZT References Alimov, Y.P., I.V. Golovikhin, L.B. Zdanevich and I.V. Yunov (Eds.), 1989. Dynamics of Forests Under Forest Management Organization Regarding the Main Forest Forming Species in 1986-1988, U.S.S.R. State Forestry Committee, 159 pp. Badhwar, G.D., R.B. MacDonald, and N.C. Mehta, 1986. Satellite-derived leaf area index and vegetation maps as input to global carbon cycle models - hierarchical approach. International J. of Remote Sensing Vol. 7, No. 2, pp. 265-281. Bazilevich, N.I., 1986. Biological productivity of soil-vegetation formations in the U.S.S.R. Bulletin of Academy of Sciences of the U.S.S.R. Geographical Series, Vol. 2, pp. 49-66. Burrough, P.A., 1987. Principles of Geographical Information Systems for Land Resources Assessment. Clarendon Press, Oxford, New York, 193 pp. Cherdantsev, G N., 1961. Map: Arable land in the U.S.S.R. in 1954. In: J.P. Cole and F.C. German (ed.) A Geography of the U.S.S.R. - Background to a Planned Economy. 290 pp. Dixon, R.K. and D.P. Turner, 1991. The global carbon cycle and climatechange: responses and feedbacks from below-ground systems. Environmental Pollution Vol. 73, pp. 245-262 Fung, I.Y., C.J. Tucker, and K.C. Prentice, 1987. Application of advanced very high resolution radiometer vegetation index to study atmosphere - biosphere exchange of C02. Journal of Geophysical Research Vol. 92(D3,) pp. 2999 3015. Gervin, J.C., A.G. Kerber, R.G. Witt, Y.C. Lu, and R. Sekhon, 1985. Comparison of Level One Land Cover Classification Accuracy for MSS and AVHRR Data. International. J. of Remote Sensing, Vol. 6, No. 1, pp. 1271-1318 Goward, S.N., B. Markham, D.C. Dye, W. Dulaney, and J. Yang, 1991. Derivation of Quantitative NDVI Measurements From AVHRR Observations. Remote Sensing of Env., Vol. 35, pp. 257-277 Coward, S.N., D. Dye, A. Kerber, and V. Kalb, 1987. Comparison of North and South American Biomass From AVHRR Observations. Geocarto International, Vol. 1, 1987. pp. 27-39. Coward, S.N., C.J. Tucker, and D.G. Dye, 1985, North American Vegetation Patterns Observed With the NOAA-7 Advanced Very High Resolution Radiometer. Vegetatio, Vol. 64, pp. 3-14 Hall, F.G., D.B. Botkin, D.E. Strebel, K.D. Woods, and S.J. Goetz, 1991. Large Scale Patterns of Forest Succession as Determined by Remote Sensing. Ecology, Vol. 72, No. 2, pp. 628-690 Holben, B.N., 1986. Characteristics of Maximum Value Composite Images From Multi-Temporal AVHRR Data. International. J. of Remote Sensing, Vol. 7, No. 11, pp. 1395-1434 Huete A.R., R.D. Jackson and D.F. Post, 1985. Spectral Response of a Plant Canopy With Different Soil Backgrounds. Remote Sensing of Env. Vol 17, pp. 3753. [sachenko, A.G. ed., 1988. Map: Landscape Map of the U.S.S.R. Institute of Geography, Leningrad State University. Janssen, L.F., M.N. Jaarsma, and E.T.M. Vanderlinden, 1990. Integrating Topographic Data With Remote Sensing for Land Cover Classification. Photogrammetric Engineering and Remote Sensing, Vol. 65, No. 11, pp.15031506 Justice, C.O., J.R.H. Townshend, B.N. Holben, and C.J. Tucker, 1985. Analysis of the Phenology of Global Vegetation Using Meteorological Satellite Data. International J. of Remote Sensing, Vol. 6, No. 8, pp. 1271-1318 Keeling, C.D., R.B. Bacastow, A.F. Carter, S.C. Piper, T.P. Whorf, M. Heimann,W.G. Mook, and H. Roeloffzen, 1989. A three-dimensional model of atmospheric CO2 transport based on observed winds: 1. Analysis of observational data. Geophysical Monographs, Vol. 55, pp. 165 - 236. 129 Keeling, C.D., 1983. The Global Carbon Cycle: What we know and could know from atmospheric, biospheric and oceanic observations. Proceedings of Carbon Dioxide Research Conference: Carbon Dioxide, Science and Concencesus, Berkley Springs, West Virginia, DOE Conf. - 820970, U.S. Dept of Energy, Wash. D.C. Kidwell, M., 1990. Global Vegetation Index Users Guide, U.S. Dept. of Commerce, NOAA, NESDES, NCDC, SDSD, Washington, D.C. Kimes, D.S., B.L. Markham, C.J. Tucker, and J.E. McMurtrey. 1981. Temporal Relationships Between Spectral Response and Agronomic Variables of a Corn Canopy. Remote Sensing of Env., No. 11, pp. 410-411. Knipling, E.B., 1970. Physical and Phisiological Basis for the Reflectance of Visible and Near-Infrared Radiation from Vegetation. Remote Sensing of Env., No.1 pp.155-159. Kolchugina, T.P. and T.S. Vinson, 1993a. Carbon Sources and Sinks in the Forests of the Former Soviet Union. Global Biogeochemical Cycles (in press) Kolchugina, T.P. and T.S. Vinson, 1993b. Equilibrium Analysis of Carbon Pools and Fluxes of Forest Biomes in the Soviet Union. Canadian Journal of Forest Research, Vol. 23, pp. 81-88 Kolchugina, T.P. and T.S. Vinson, 1993c. Comparison of Two Methods to Assess the Carbon Budget of Forest Biomes in the Former Soviet Union, Journal of Water, Air and Soil Pollution. (in press) Loveland, T., J.W. Merchant, D.O. Ohlen, and J.F. Brown, 1991. Development of a Land-Cover Characteristics Data Base for the Conterminous United States, Photogrammetric Engineering and Remote Sensing, Vol. 57, No. 11, pp. 1453-1463 Norwin, J. and D.H. Greegor, 1983. Vegetation Classification Based on AVHRR Satellite Imagery. Remote Sensing of Env., Vol. 13, pp. 69-87 Panel on Policy Implications of Greenhouse Warming (PPIGW), 1992. Policy Implications of Greenhouse Warming - Mitigation, Adaptation, and the Science Base, Committee on Science, Engineering, and Public Policy, Natl. Academy of Sciences, Natl. Academy of Engineering, Inst. of Medicine, National Academy Press, Washington, D.C., Post, W.M., W.R. Emanuel, P.I. Zinke, and A.G. Stagenberger, 1982. Soil Carbon Pools and World Life Zones. Nature, Vol. 298, No. 5870, pp. 156-159. Post, W.M., Tsung-Hung Peng, W.R.Emanuel, A.W. King, V.H. Dale, and D.L. De Angelis, 1990. The global carbon cycle. American Scientist No. 78 pp. 310-326. Raich, J.W. and W.H. Schlesinger, 1992. The Global Carbon Dioxide Flux in Soil Respiration and its Relationship to Vegetation and Climate, Tellus, Vol. 44B, pp. 81 -99. Runyon, J., R.H. Waring, R.W. McCrieght, 1991. Assessing the Impact of Climatic Stress on Forest Production. in; Proceedings of the International Workshop on Carbon Cycling in Boreal Forest and Sub-Arctic Ecosystems. T.S. Vinson and T.P. Kolchugina eds. Corvallis, Oregon. Ryabchikov, A.M. ed., 1988. Map: Geographical Belts and Zonal Types of Landscapes of the World. School of Geography, Moscow State University, Moscow. Schneider, S.H., 1990. Global Warming. Vintage Books, New York. 343 p. Sokolovskiy, V.G. ed., 1989. Report on the State of the Envorinment in the USSR:1988. State Committee for the Protection of Nature. Moscow, 184 pp. Spanner, M.A., L.L. Pierce, D.L. Peterson, and S.W. Running, 1990. Remote Sensing of Temperate Coniferous Forest Leaf Area Index: The Influence of Canopy Closure, Understory Vegetation and Background Reflectance. International J. of Remote Sensing, Vol. 11, No. 1, pp. 95-111 State Forestry Committee, USSR, 1990. Forest Fund of the USSR. Vol.! 1005 pp. Moscow Tans, P.P., I.Y. Fung, and T. Takahash. 1990. Observational Constraints of the Global Atmospheric CO2 Budget. Science, No. 247 pp. 1431-1438 Townshend, J.R.G., C.O. Justice, and V. Kalb, 1987. Characteristics and Classification of South American Land Cover Types Using Satellite Data. International J. of Remote Sensing, Vol. 8, No. 8, pp. 1189-1207 Townshend, J.R.G. and C.O. Justice, 1986. Analysis of the Dynamics of African Vegetation Using the Normalized Difference Vegetation Index, International J. of Remote Sensing, Vol. 7, No. 11, pp. 1435-1445 Townshend, J.R.G. and C.J. Tucker, 1984, Objective Assessment of AVHRR Datafor Land Cover Mapping, International J. of Remote Sensing. Vol. 5, No. 2,pp. 497-504 Tucker, C.J., J.R.G. Townshend, T.E. Goff, 1985, African Land Cover Classification Using Satellite Data. Science. Vol. 223, No. 4685, pp. 369-375 Tucker, C.J. and P.J. Sellars, 1986, Satellite Remote Sensing of Primary Production. International J. of Remote Sensing, Vol. 6, No. 1, pp. 1395-1415 Tucker, C.J., C. Van Praet, E. Boerwinkle, and A. Gaston, 1983. Satellite Remote Sensing of Total Dry Matter Production in the Senegalese Sahel. Remote Sensing of Env., Vol. 13, pp. 461-474 Tucker, C.J., B.N. Holben, J.H. Eglin, and J.E. McMurtrey. 1981. Remote Sensing Total Dry Matter Accumulation in Winter Wheat. Remote Sensing of Env., No.11 pp. 171-189. U.S. Dept. of Agriculture (USDA), 1990. USSR Trade and Agriculture Report: Situation and Outlook. USDA, RS-90-1, May 1990 Vinson, T.S., and T.K. Kolchugina, 1993. Pools and Fluxes of Biogenic Carbon in the Former Soviet Union. Journal of Water, Air and Soil Pollution. (in press). Vorobyov, G.I., ed., 1985. Forest Encyclopedia. Sovetskaya Ecyclopedia Press, Moscow. Class 1. Water / No Vegetation Well Identified class found in the Aral Sea, Lakes Balkash,Onega,Lagoda,and Baikal. Of note are the Pamir mountains, and Lake Ala-Kul in South Central Asia. This class also appears along the eastern shore of the Caspian Sea, almost completely surrounding the Kara-Bogaz-Gol. This area according to Suslov (p.485) "... The shore is covered with hill sand amid which there are numerous salt lakes..." Coincidence with Vegetation Map Classes: Water 38.81% Desert Marshes 10.20% Class 1 (water % Arable Lands: NA Overall Mean: 28.409 Area under TNRC: SO March-October 190.8 M-k Apd M., . Max April-September 126.38 J . Std Dev My Average A.y bfin Closely Related to Class: 13 Desert Summary Statistics for Class: Minimum v Maximum Mean Std Dev -------------------------------------------------------------__________ Variable MAR APR MAY JUN JUL AUG SEP OCT 1588 1588 1588 1588 1588 1588 1588 1511 17.0000000 11.0000000 0 0 1.0000000 0 0 17 nnnnnnn 54.0000000 52.0000000 53.0000000 56.0000000 52.0000000 53.0000000 50.0000000 99 nnnnnnn 36.8520151 27.9628463 24.4534005 22.5510076 24.0724181 25.7424433 29.5642317 19 0741071 6.0265232 5.1165871 7.9601066 9.7661655 9.5135188 10.3045514 7.1338119 1014575 ti Class 2. High Mountain Tundra / Arctic Desert Well identified class seen in Burranga Mountains, Putoran Mountains, Verkohansk, Chersky, and North Ural Mountains. Also located in the Pamir Mountains in the south. This class also appears in the Lena River delta, and along the Arctic coast. The confusion between low elevation Arctic tundra and mountain tundra should not be critical for carbon quantification. Coincidence with Vegetation Map Classes: Northern Mountain Tundra 43.27% Arctic Tundra 12.91% Class 2 % Arable Lands in Class: NA Overall Mean: 39.398 lm F Area under TNRC: too March-October 270.717 April-September 195.477 A1.d Closely Related to Class: M., y.. Max ,®. ,d, - -.- Std Dev _,_ Average -.0- Min 2-4-6 in the mountains 3-5-8 on the arctic coasts Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 2745 2745 2745 2745 2745 2745 2745 2745 29.0000000 19.0000000 11.0000000 8.0000000 28.0000000 26.0000000 28.0000000 33.0000000 59.0000000 39.0000000 40.0000000 61.0000000 69.0000000 73.0000000 58.0000000 51.0000000 45.5253188 30.7697632 24.2848816 30.5737705 49.6069217 51.2255009 39.7854281 43.4163934 5.4315409 2.6801349 4.3795519 7.7115762 7.7735082 8.5273829 4.1948105 2.6775891 Class 3. Arctic Tundra/Mountain Tundra Closely related to class 2 and 5; tundra vegetation, more dense vegetation than class 2. Located on the lower slopes of the Arctic mountains. Patterns in the Yamal Peninsula do not fit very well mapped distribution. Coincidence with Vegetation Map Classes: Northern Mountain Tundra 30.84% Arctic Tundra 19.55% Typical Tundra 12.05% Class 3 % Arable Lands in Class: NA 230 Overall Mean: 46.073 200 150 Area under TNRC: :ao March-October 323.525 April-September 248.308 so Closely Related to Class: Apnl MW -.- Max J- I* A{oa sue. t Std Dev -&-Average .¢ Min 2 and 5 Summary Statistics for Class: Variable N Minimum Maximum Mean MAR 3213 APR MAY JUN JUL AUG 3213 3213 3213 3213 3213 SEP 3213 OCT 3213 34.0000000 23.0000000 11.0000000 11.0000000 40.0000000 52.0000000 32.0000000 37.0000000 56.0000000 38.0000000 32.0000000 48.0000000 91.0000000 101.0000000 74.0000000 50.0000000 45.9203237 30.1596639 21.8972923 32.4749455 71.9489574 75.5144725 46.4718332 44.1954560 Std Dev 4.3843093 2.1355886 2.9577824 6.1020594 7.3666989 7.0718730 6.0648914 2.1078674 Class 4. Mountain Tundra Very closely related to class 2. Located in small amounts below class 2 in the Siberian high country. With the exception of a core area of class 1, the Pamir mountains are characterized by class 2 and 4. Coincidence with Vegetation Map Classes: Northern Mountain Tundra 45.95% Class 4 % Arable Lands in Class: 1.7 % Overall Mean: 47.11 = Area under TNRC: March-October 332.965 April-September 216.897 M.W M.y Apnl . Max I- t Std Dev Idy Angoe sq-.ce Average .¢. Min Closely Related to Class: 2 and 6 (Classes 2-4-6 form a mountain tundra complex) Summary Statistics for Class: N Obs Variable Minimum 1903 MAR APR 26.0 22.0 MAY 12.0 JUN JUL AUG SEP OCT 28.0 38.0 29.0 30.0 38.0 Maximum 59.0 45.0 58.0 75.0 86.0 85.0 75.0 55.0 Mean 43.97 32.56 31.08 50.47 65.93 61.06 47.8 44.17 Std Dev 6.24 3.30 6.8 7.38 7.93 7.53 6.63 2.91 136 Class 5. "Typical" Tundra A more dense tundra type vegetation, located primarily in the Yamal and Tamir peninsulas with smaller areas scattered across the Arctic coast. Coincidence with Vegetation Map Classes: "Typical Tundra" 29.49% Northern Mountain Tundra 16.66% Arctic Tundra 15.67% Class 5 % Arable Lands in Class: NA Overall Mean: 51.522 Area under TNRC: March-October 366.562 April-September 290.8 AW9 Closely Related to Class: . Max M" I®. Ian _ Std Dev -,Average Min 3 and 8 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT v Minimum Maximum Mean Std Dev 4291 4291 4291 4291 4291 4291 4291 4291 34.0000000 24.0000000 11.0000000 17.0000000 70.0000000 74.0000000 39.0000000 36.0000000 56.0000000 36.0000000 30.0000000 51.0000000 105.0000000 117.0000000 66.0000000 51.0000000 46.6047541 30.1479842 21.0109532 36.7720811 89.0426474 93.8673969 50.1069681 44.6222326 3.8073559 1.9812072 2.8764977 5.6530845 5.6012652 6.0429224 4.3267451 2.0256306 Class 6. Shrubby Tundra of Siberian Mountains Part of the 2-4-6 class complex that identifies the Siberian mountain ranges, Burranga, Verkohansk, etc. Higher biomass suggests shrubby tundra vegetation. Also located in Boggy areas of western Siberia. Coincidence with Vegetation Map Classes: Northern Mountain Tundra 40.96% Northern Mountain Scattered Larch 15.11% Bogs 7.52% Class 6 % Arable Lands in Class: NA 2" Overall Mean: 52.832 I x4 150 Area under TNRC: 100 March-October 377.927 4 April-September 301.235 Apa Closely Related to Class: Mq _41- Max ,m. Jay _._Std Dev -*.-Average Aopa ..Min 2-4 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT t: Minimum Maximum Mean Std Dev 3322 3322 3322 3322 3322 3322 3322 3322 29.0000000 23.0000000 12.0000000 28.0000000 67.0000000 57.0000000 33.0000000 36.0000000 56.0000000 40.0000000 43.0000000 83.0000000 109.0000000 97.0000000 77.0000000 53.0000000 44.8642384 31.9659843 27.9936785 58.4840458 85.6035521 76.9981939 52.1550271 44.5888019 3.7693103 2.7399130 4.6112619 8.7860524 6.0554220 5.5933670 6.8493744 2.5755845 Class 7. Bunch Grass Steppe (Solenchuck) Generally located east of Ural mountains, very similar to class 17 except peak in NDVI occurs in July rather than in June. There is no good information available right now to positively identify the vegetation/landscape represented by this class. However, considerable evidence points to Solenchuck soils with high salt content. Coincidence with Vegetation Map Classes: Grassy Steppe 19.18% Grassy Steppe w/Herbaceous plants 12.0% % Arable Lands in Class: Class 7 11.25% Po Overall Mean: 59.462 Area under TNRC: March-October 433.378 IOD April-September 352.695 Closely Related to Class: 0 ApI M.y y.. Max Jr 1dy _. Std Dev -t- Average Ao{ar SW--b- Min 17 and 23 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean 2212 2212 2212 2212 2212 2212 2212 2212 26.0000000 21.0000000 29.0000000 46.0000000 50.0000000 40.0000000 39.0000000 35.0000000 63.0000000 58.0000000 78.0000000 104.0000000 117.0000000 101.0000000 96.0000000 63.0000000 38.5682640 38.3801989 53.9945750 77.4534358 87.2373418 74.5212477 59.4882459 46.0547016 _- Std Dev 6.8958358 4.9103797 8.7925469 8.7188407 10.7727313 9.3051685 7.9007495 3.9578501 Class 8. "Typical" / Shrubby Tundra This class represents part of the transition from Arctic deserts to northern Taiga forest. This image class covers large areas of the Tamir Peninsula and the plains west of the Yamal peninsula. This class appears to identify a region of mixed "typical" and shrubby tundra. Representing a transition from class 5 to classes 10-12. Coincidence with Vegetation Map Classes: "Typical" Tundra 28.66% Shrubby Tundra 20.68% Forest Tundra Larch 16.61% % Arable Lands in Class: NA Class 8 780 Overall Mean: 56.615 700 Area under TNRC: ISO March-October 408.680 1W April-September 334.939 50 Closely Related to Class: Fwd t Max 5, 10 and 12 w, 7®e 7WY Au{= srymes _.s._ Std Dev -A.-Average -.0- Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT Std Dev N Minimum Maximum Mean. 3231 3231 3231 3231 3231 3231 3231 3231 33.0000000 25.0000000 11.0000000 23.0000000 85.0000000 91.0000000 40.0000000 37.0000000 53.0000000 36.0000000 33.0000000 62.0000000 131.0000000 125.0000000 73.0000000 50.0000000 44.6771897 2.7915346 29.4986072 2.2150445 4.0486505 21.3249768 45.2732900 104.1770350 108.0922315 56.0718044 43.8074899 - 6.4105101 7.8135537 5.4056075 4.8469882 1.9694461 Class 9. Dwarf Shrub Pine and Willow (Riparian grass and willows in lower Ob) This is a problem class as it is seen both in the Ob river plain and more extensively on the Pacific coast near the Anadyr depression. Evidently, the dwarf shrub Pine/Alder in a shrubby/forest tundra type environment described by Suslov (pg 224) in the Anadyr-Penzhina depression has a temporal /NDVI signature that overlaps riparian grass and willow of the lower Ob. Carbon content evaluation will split this class calculating the carbon contents for each separately. Coincidence with Vegetation Map Classes: Northern Mountain Tundra 29.94% Pinus pumila 21.11% Class 9 % Arable Lands in Class: NA Overall Mean: 57.550 Area under TNRC: March-October 416.598 April-September 343.245 April M.) -a- Max Closely Related to Class: J My A"- t Std Dev -&- Average _0.. Min Statistically related to class 8 and 11 Geographically related to class 19 and 22 Summary Statistics for Class: MAR APR MAY JUN JUL AUG SEP OCT 1788 1788 1788 1788 1788 1788 1788 1788 li- maximum Mean 29.0000000 54.0000000 22.0000000 41.0000000 12.0000000 45.0000000 21.0000000 89.0000000 53.0000000 120.0000000 69.0000000 123.0000000 52.0000000 102.0000000 37.0000000 62.0000000 - - - - - - - --- - - - 40.9882550 29.5503356 25.7063758 53.1923937 94.0369128 96.3199105 73.9893736 46.6168904 minimum -Variable = Std Dev 3.0620110 3.2698509 5.7152565 12.5710030 10.1287615 9.0020520 6.8451085 3.1452943 14 Class 10. Shrubby Tundra This class falls between 8 and 12-20; this class is part of the transition towards more trees. In the Far East, this class (without class 10) reaches almost to the Arctic ocean. According to Suslov (pg 181) ..."In eastern Siberia, tundra meadows become more scarce. Under the hot, dry summer conditions, woody vegetation reaches to within 39 miles of the shore".... "the sharply continental climate (east of the Lena River) contributes to the northward expansion of woody plants". Coincidence with Vegetation Map Classes: "Typical" Tundra 15.63% Northern Taiga Larch 11.78% Class 10 % Arable Lands in Class: NA 2O Overall Mean: 56.363 2W Area under TNRC: 13D March-October 406.380 Im April-September 331.283 Closely Related to Class: AFQ IMP -a-MU IStd Dev My Aug- _r Average Sq-.b. on Min 12 and 20 on the southern edge 8 to the northern edge Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Devi 3597 3597 3597 3597 3597 3597 3597 3597 35.0000000 25.0000000 12.0000000 38.0000000 87.0000000 73.0000000 40.0000000 38.0000000 55.0000000 39.0000000 35.0000000 75.0000000 118.0000000 44.1887684 - 3.4229492 2.0812813 3.5111002 6.0404215 5.2017977 111.0000000 70.0000000 52.0000000 30.5752016 24.3914373 57.5376703 101.4931888 93.2571587 54.6035585 44.8554351 5.8294048 5.7743698 2.3149431 Class 11. Sparse Northern Taiga Larch Light crowned Taiga forest. Very sparse trees predominantly located in the plateaus of eastern Siberia. In this region, the first of the forest classes. Very closely related to class 15 and 16. Coincidence with Vegetation Map Classes: Northern Taiga Larch 49.8% Northern Mountain Tundra 14.11% % Arable Lands in Class: class 11 NA Overall Mean: 58.704 X0 Area under TNRC: 150 March-October 424.669 ou April-September 346.603 50 0 M-b Closely Related to Class: ApN - Max Nry J.. -r_ Std Dev Icy A.Cm Average Sqembc . Min 16 and 15 Summary Statistics for Class: -Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 5996 5996 5996 5996 5996 5996 5996 5996 37.0000000 26.0000000 18.0000000 63.0000000 82.0000000 64.0000000 37.0000000 37.0000000 54.0000000 40.0000000 46.0000000 95.0000000 119.0000000 103.0000000 70.0000000 54.0000000 45.6976318 33.1032355 31.2815210 76.7510007 100.9929953 86.4683122 51.1089059 44.2284857 2.3958125 1.9803496 3.9873192 5.3122868 5.8193503 5.8559434 6.0408104 2.4723652 Class 12. Transitional area of tundra Increasing percentage of trees concentrated mainly in favorable microclimates. See comments for class 10 regarding the distribution of this class east of the Lena River. Coincidence with Vegetation Map Classes: Northern Taiga Larch 16.39% N.Mt. Larch on Slopes 14.01% N.Mt. Tundra 14.9% Forest Tundra 10.6% Class 12 % Arable Lands in Class: NA iso 200 Overall Mean: 61.901 Area under TNRC: mo March-October 451.055 50 April-September 374.823 M-h Ap. M.y Max Closely Related to Class: - I Joy AogM Std Dev -d, Average -.0.- Min 11 and 20 in the northern forest classes 8 and 10 shrubby tundra Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT 3642 3642 3642 3642 3642 3642 3642 3642 Minimum Maximum Mean Std Dev+ 36.0000000 25.0000000 16.0000000 47.0000000 98.0000000 80.0000000 43.0000000 38.0000000 53.0000000 42.0000000 40.0000000 87.0000000 138.0000000 129.0000000 80.0000000 55.0000000 43.6861614 32.0826469 27.2803405 70.5345964 115.0024712 101.2583745 60.7468424 44.6139484 2.9079196 3.0080969 3.9001818 7.0919364 5.9518109 6.6403647 6.7547086 2.6068599 Class 13. Desert The geographic extent of this class matches very well with the limits of the deserts in South Central Asia as described by both Berg and Suslov.. This class is well identified, examining the NDVI/time plot. The spring "bloom" of vegetation shows clearly in April-May. Coincidence with Vegetation Map Classes: Desert (all subdivisions) 81.3% Class 13 (desert) % Arable Lands in Class: 0.68% Overall Mean: 43.303 Area under TNRC: I5D March-October 305.226 April-September 218.027 30 Closely Related to Class: 0 Apf Max f® ,Wr w, AMM t Std Dev .. Average -0-Min see. 1 (in shape of TNRC) Summary Statistics for Class: Variable N Minimum Maximum Mean Std Dev MAR 4389 40.0968330 4.9456906 4389 4389 4389 4389 4389 4389 4389 25.0000000 26.0000000 32.0000000 26.0000000 25.0000000 24.0000000 26.0000000 30.0000000 62.0000000 APR MAY JUN JUL AUG 71.0000000 67.0000000 46.0009114 49.5026202 45.3533835 39.9346093 40.7664616 42.4700387 42.2998405 6.1673708 7.0079916 SEP OCT 61.0000000 52.0000000 54.0000000 54.0000000 Si 0000000 5.8698064 4.7753124 4.5814721 3.7499714 3.4898410 Class 14. This class is well described by Berg and Suslov as sparse Artemisia with widely scattered bunch grass. Coincidence with Vegetation Map Classes: Semi-desert Bunchgrass 21% Desert, Artemisia 14% Desert, Wazzi 25% % Arable Lands in Class: Class 14 0.89% Overall Mean: 53.438 Area under TNRC: Ib March-October 385.180 IOD April-September 289.859 M Closely Related to Class: 0 Apd t Max 28 Ji ,w AYP-0. Std Dev -. Average -0. Min M., sue. Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT 3063 3063 3063 3063 3063 3063 3063 3063 Minimum Maximum Mean Std Dev 24.0000000 29.0000000 34.0000000 44.0000000 38.0000000 38.0000000 36.0000000 32.0000000 66.0000000 80.0000000 101.0000000 84.0000000 74.0000000 71.0000000 69.0000000 71.0000000 36.3251714 52.9951028 69.5563173 63.7502449 53.9249102 51.5710088 51.0564806 48.3258244 8.2512024 7.8130701 9.6877783 6.5478178 5.4309011 4.6699110 4.3694216 4.4570974 Class 15. Scattered Larch on Mountain Slopes The lower slopes of the Trans-Baikal mountains are spotted with this class, this class is also located in the North Ural mountains and significantly in the identified wetland area of western Siberia. Coincidence with Vegetation Map Classes: Northern Mountain Larch 16% Northern Mountain Larch on Slopes 14% Northern Mountain Tundra 13% % Arable Lands in Class: Class 15 2.9% __T Overall Mean: 64.403 Area under TNRC: 150 March-October 469.255 ioo April-September 387.119 Closely Related to Class: 10 0 H., Aped Max Jim. t Std Dev l.y A.p m Average s gombv Min 11 and 16 Summary Statistics for Class: -Variable MAR APR MAY JUN JUL AUG SEP OCT 4121 4121 4121 4121 4121 4121 4121 4121 Minimum Maximum Mean 28.0000000 21.0000000 22.0000000 61.0000000 78.0000000 73.0000000 47.0000000 37.0000000 58.0000000 52.0000000 61.0000000 108.0000000 127.0000000 119.0000000 89.0000000 56.0000000 45.1778694 36.1681631 40.9012376 84.5964572 102.0412521 92.5020626 67.0778937 46.7573405 - Std Dev 4.2217255 3.3851641 7.1099343 7.0640603 7.0166063 6.6404584 7.1674154 2.7442500 Class 16. Larch Forest (Light Crowned Conifer Forest) Restricted to large areas of Central and Eastern Siberia. From the literature, almost pure stands of Dahurian and Siberian larch. The TNRC of this class is virtually identical to class 11. Coincidence with Vegetation Map Classes: Northern Mountain Larch on Slopes 37% Northern Taiga Larch 14% Northern Mountain Tundra 13% % Arable Lands in Class: NA class 16 2O Overall Mean:64.770 Area under TNRC: 2W March-October 473.250 170 April-September 395.044 Im 50 Closely Related to Class: 0 April 11, 15, 25, and 32 -}- Max 1® 1* Agsa UP-4.- Std Dev _ Average -0- Min M., Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 5679 5679 5679 5679 5679 5679 5679 5679 37.0000000 27.0000000 18.0000000 77.0000000 97.0000000 79.0000000 42.0000000 38.0000000 52.0000000 42.0000000 49.0000000 113.0000000 133.0000000 114.0000000 73.0000000 54.0000000 44.8265540 33.2923050 33.5323120 92.4602923 114.4627575 97.8688149 56.7198450 45.0005283 2.0793588 1.9781643 5.1864806 6.2690751 5.2643382 5.9944096 5.6441573 2.4142569 Class 17. Brush and Bunch Grass: Transition from Semi-Desert to Grassy Steppe This class is located adjacent to class 14; semi-desert. It appears to represent a zone of increasing grass, predominantly bunch grass, in the desert shrub zone. Coincidence with Vegetation Map Classes: Grassy Steppe 35.49% Semi-desert: Artemisia and Bunch Grass 32.39% Semi-desert: Bunch Grass 12.86% Class 17 % Arable Lands in Class: 8.26% 270 Overall Mean: 61.659 Area under TNRC: March-October 452.433 too April-September 362.551 0 Closely Related to Class: Apil nt., -a- Max i. -4- Sid Dev toy Average AupW s.p-e. Min 7, 23, and 26 forms the northern border of class 14 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 3102 3102 3102 3102 3102 3102 3102 3102 25.0000000 26.0000000 52.0000000 62.0000000 49.0000000 40.0000000 38.0000000 35.0000000 52.0000000 84.0000000 110.0000000 115.0000000 94.0000000 82.0000000 74.0000000 70.0000000 32.3913604 49.0415861 80.2295293 87.5119278 72.0228885 64.2163121 58.5699549 49.2901354 3.8775555 7.8630169 8.2340124 8.2427899 7.7609266 6.4980799 4.5685971 4.6863190 Class 18. Northern Taiga Spruce / Larch Forest This class is located in a rough band from the White Sea near Archanglsk to the wetlands east of the Ob River. Class 18 is also located in a band south of class 16 near the Vilyuy River in Central Siberia. Coincidence with Vegetation Map Classes: Middle Taiga Larch 31.4% Northern Taiga Spruce 18% Class 18 % Arable Lands in Class: 1.96% 750 Overall Mean: 71.269 700 Area under TNRC: ISO March-October 523.424 too April-September 435.015 50 0 Closely Related to Class: Apa Ma -... Max M., J _. Std Dev ,m, A SWebr onobw Average --o-Min statistically to 22 and 26 geographically to 20, 23, 29, 16, and 25 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean 5175 5175 5175 5175 5175 5175 5175 5175 27.0000000 21.0000000 36.0000000 80.0000000 85.0000000 79.0000000 44.0000000 38.0000000 59.0000000 59.0000000 70.0000000 119.0000000 148.0000000 126.0000000 84.0000000 55.0000000 47.1426087 41.6805797 51.8021256 101.0564251 115.8977778 100.6823188 65.5764251 46.3142029 - Std Dev 3.9528227 4.9504112 6.0035001 6.2663705 7.4006357 6.2587200 5.9890301 2.1868243 1 Class 19. Mountain shrubs of the Far East: Pinus pumila and Scrub Alder Seen in the Kamchatka and Kolyma ranges. The southern tip of Kamchatka is class 19. This area is described by Suslov (pg 397) as moss-scrub alder/ash and Pinus pumila. This class is also seen in the middle region of the Ob River. Apparently, riparian grass and willows in the flood plain have signatures that correspond to class 9 in the north and class 19 in the mid-course of the Ob. Upstream from class 19, class 27 appears to correspond to the course of the river. Coincidence with Vegetation Map Classes: v Pinus pumila 40% Northern Mountain Tundra 20% Cotton Grass Tundra 10.27% Class 19 % Arable Lands in Class: NA 270 Overall Mean: 68.398 200 Area under TNRC: 170 March-October 501.535 I00 April-September 425.950 30 Closely Related to Class: Apd Mq _..- Max ,m. Jm, Ao{m _*. Std Dev t Average spec Min -9, 22, and, 27 Summary Statistics for Class: Variable Minimum maximum Mean Std Dev ----------------------------------------------------------------------MAR APR MAY JUN JUL AUG SEP OCT N 1299 1299 1299 1299 1299 1299 1299 1299 31.0000000 24.0000000 16.0000000 23.0000000 63.0000000 62.0000000 56.0000000 38.0000000 51.0000000 42.0000000 59.0000000 102.0000000 162.0000000 181.0000000 143.0000000 71.0000000 41.0084681 29.9368745 29.1593533 66.4965358 116.9022325 120.4603541 92.9314858 50.2871440 ----------------- 2.4740419 2.8312617 7.7700986 12.4715002 12.6902345 15.0692472 13.3373008 4 71SA4Q1; Class 20. Northern Taiga Spruce Forest This class is located in a discontinuous band from near the Kanin Peninsula to the wetlands east of the Ob River. This class picks up in Siberia south of the Putorana mountains. This class is closely related to class 12, the forest tundra zone, and class 18. The eastern end of this class occurs in central Siberia. This class is not located in eastern Siberia. Coincidence with Vegetation Map Classes: Northern Taiga Spruce 29.58% Northern Taiga Larch 19.79% Middle Taiga Larch 12.25% % Arable Lands in Class: Class 20 NA Overall Mean: 69.472 Area under TNRC: March-October 509.998 April-September 425.218 Closely Related to Class: 12, 18, and, 24 Apd y- Max M -Y J- -4.. Std Dev t J* Average A- s,pembv . Min Summary Statistics for Class: Variable N Minimum Maximum Mean = Std Dev ----------------------------------------------------------------------MAR 3190 30.0000000 64.0000000 47.1225705 4.0912711 APR 3190 27.0000000 57.0000000 39.0018809 4.9105511 MAY 3190 17.0000000 55.0000000 36.7542320 5.6774744 JUN 3190 58.0000000 114.0000000 87.8426332 9.1019624 JUL 3190 102.0000000 155.0000000 126.2473354 6.6970892 AUG 3190 87.0000000 147.0000000 110.5304075 7.3319054 SEP 3190 46.0000000 83.0000000 63.8435737 5.7936434 OCT 3190 38.0000000 51.0000000 44.4338558 1.7128024 Class 21. Irrigated Agriculture / Grasslands? A most interesting class. This class appears along the Amu-Darya and SyrDarya in south central Asia. A detailed view shows this lass in the Fergana valley, along the Iii River. This class also appears in the Caucuses mountains and east of the Urals. Two small patches appear in the Kopet Dag along the Kara Kum canal. This class also appears along the Mongolian border in what Suslov "cereal steppes". A few of these pixels also appear as the blanket bogs on the west coast of the Kamchatka peninsula. This scattering of this class warrants care in assigning carbon values for the carbon budget. This class is obviously irrigated agriculture however in the Amur region class 21 is located on the middle slopes of the interior mts. This pattern appears to occur throughout southern siberia. Coincidence with Vegetation Map Classes: The vegetation map used for this research is a 'potential' natural vegetation map and thus may be inappropriate for use in identifying class 21. (there is no real major class to identify) % Arable Lands in Class: 8% Class 21 iw Overall Mean: 73.710 Area under TNRC: hm March-October 541.856 lh April-September 451.025 0 Closely Related to Class: M.y Apnt - Max J. _ Std Dev !my A gm Average -0-Min Summary Statistics for Class: -Variable MAR APR MAY JUN JUL AUG SEP OCT 2572 2572 2572 2572 2572 2572 2572 2572 Minimum Maximum Mean Std DevI 28.0000000 24.0000000 38.0000000 57.0000000 68.0000000 73.0000000 61.0000000 37.0000000 69.0000000 76.0000000 88.0000000 120.0000000 140.0000000 150.0000000 127.0000000 83.0000000 43.6924572 43.0093313 59.7892691 92.4568429 107.0260498 106.8405910 84.9121306 51.9517885 7.8666377 7.5537173 7.7637667 10.1470367 9.6603794 9.6732024 10.5217106 6.7446053 Class 22. Dwarf Forests of the Pacific Coast (Pinus pumila, Betula ermanii, Larch and Spruce depending on site) Wind sculpted shrubs; short trees of higher elevations in Kamchatka, and along the shore of Okhotsk sea. This area is described by Suslov on pgs 371-372. A significant area of this class is also located at the base of the Kanin Peninsula. Coincidence with Vegetation Map Classes: Scattered Larch on Mountain Slopes 25.97% Larch and Pine (Far East) 19.03% Japanese Stone Pine (Pinus pumila) 15.66% % Arable Lands in Class: Class 22 NA Overall Mean: 70.905 Area under TNRC: ISO March-October 521.293 Itb April-September 441.980 So Closely Related to Class: Ape] . Max 9and19 Mq ,me Mm, Aug- SepemhQ .4- Std Dev -t- Average -o- Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 3368 3368 3368 3368 3368 3368 3368 3368 30.0000000 24.0000000 18.0000000 69.0000000 97.0000000 91.0000000 64.0000000 40.0000000 54.0000000 45.0000000 56.0000000 122.0000000 141.0000000 136.0000000 110.0000000 63.0000000 43.3785629 33.3643112 35.6125297 94.9833729 118.7111045 110.7131829 81.9596200 48.5201900 2.8766510 3.2850875 6.3432963 7.9013646 7.8701578 7.3836070 7.6852338 3.5040232 ---------------------- Class 23. Scattered Forest (Pine / Aspen in the South) (Pine / Birch in the North) One of the few classes that shows a north/south confusion. This class is located both in the Karalean peninsula, the northern Taiga forest, and in the North Urals and in the steppes of Khazaldstan, north of Lake Balkash. The evidence would suggest scattered trees with a strong herbaceous matrix. This north/south confusion may require adjustment in the carbon budget of the FSU for this class. Coincidence with Vegetation Map Classes: Grassy Steppe 21.17% Grassy Steppe w/ herbs 17.96% Northern Taiga Pine 9.23% Middle Taiga Spruce 8.41% % Arable Lands in Class: Class 23 17.46% Overall Mean: 68.398 Area under TNRC: Ib March-October 506.625 too April-September 419.120 50 Closely Related to Class: APB May Max 18 and 29 in the north 7 and 26 in the south J* M. t Std Dev A.{.a see. Average -0-Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 3069 3069 3069 3069 3069 3069 3069 3069 26.0000000 26.0000000 48.0000000 76.0000000 67.0000000 56.0000000 47.0000000 38.0000000 68.0000000 70.0000000 93.0000000 127.0000000 125.0000000 104.0000000 85.0000000 57.0000000 40.6203975 43.5008146 70.3558162 99.3365917 101.1000326 83.2479635 65.0791789 47.3880743 7.8313956 7.0917632 7.4062469 7.8056013 9.3902827 7.4023918 5.9991242' 3.1397691 Class 24. Mixed Conifer Forest (Larch, Pine, Spruce) This class is located in central Siberia. From the head waters of the Taz River to the head waters of the Lena River. The band of class 24 is located between class 20, 15, 18 and on the north, and class 30, 31, and 36-39 on the south. Coincidence with Vegetation Map Classes: Middle Taiga Larch 30.76% Larch-Pine of Baikal zone 22.39 Northern Mountain: Spruce / Fir 15.61% Middle Taiga Spruce 11.89% % Arable Lands in Class: NA Class 24 2O Overall Mean: 79.944 Area under TNRC: March-October 585.846 100 April-September 473.516 Closely Related to Class: Apnl _...Max 25 M" ,®. ,m, A s > tStdDev ..,-Average -.0-Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 2971 2971 2971 2971 2971 2971 2971 2971 48.0000000 44.0000000 45.0000000 72.0000000 88.0000000 80.0000000 54.0000000 40_000nnnn 73.0000000 82.0000000 90.0000000 130.0000000 151.0000000 130.0000000 98.0000000 59.3527432 58.6273982 66.2204645 103.9229216 119.2790306 106.8004039 77.2935039 4A ns717in 4.2026256 6.2949536 7.4876913 8.8053655 9.4541545 7.0779841 6.8913175 7 A97RA19 tin nnnnnnn Class 25. Open Forest: Larch in Vilyuy Basin Aspen / Birch in Central Siberia Located along the Vilyuy and middle Lena Rivers surrounding class 32. A narrow strip of this class is located south of class 32 in central Siberia. In central siberia This class is representative of forest meadow steppes. Coincidence with Vegetation Map Classes: Middle Taiga Larch 33.8% Larch and Pine of Baikal zone 18% % Arable Lands in Class: Class 25 5.2 % Overall Mean: 77.176 Z!M Area under TNRC: Iw March-October 570.599 loo April-September 482.554 50 Closely Related to Class: M- Apnl _._Max 32 wr ,_ tStdDev my flog- Sq-.be .Average -_Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 4361 4361 4361 4361 4361 4361 4361 4361 27.0000000 24.0000000 32.0000000 88.0000000 102.0000000 92.0000000 50.0000000 40.0000000 63.0000000 58.0000000 73.0000000 134.0000000 156.0000000 132.0000000 108.0000000 54.0000000 46.5441413 41.2371016 55.1595964 113.6489337 127.2098143 112.9701903 73.5652373 47.0726898 5.5598032 4.5360138 5.8826623 6.8499902 6.8338241 5.8273344 8.4272615 2 n1R4Ani Class 26. Grassy Steppe This class is located only west of the Ural monutains, between class 35 and class 17. This area appears to represent the classic grassy steppes of central asia. East of the Urals the equivalent area is represented by a complex of class 23 and 7. Coincidence with Vegetation Map Classes: Grassy Steppe 22.1% Semi-desert: Sage and grass 16.7% Grassy Steppe w/ forbs 14.7% % Arable Lands in Class: Class 26 36 % Overall Mean: 72.565 Area under TNRC: UD March-October 532.685 April-September 423.446 70 Closely Related to Class: Mq Apnl Max I Jd, Apmbs M$ _.,..Std Dev ..i.. Average oc.ow Min 23 and 7 Summary Statistics for Class: 4 Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 1712 1712 1712 1712 1712 1712 1712 1712 27.0000000 26.0000000 59.0000000 58.0000000 52.0000000 51.0000000 42.0000000 39.0000000 77.0000000 106.0000000 129.0000000 136.0000000 129.0000000 106.0000000 95.0000000 73.0000000 39.8580607 61.4088785 91.9620327 99.5747664 88.4667056 76.4830607 66.9591121 55.8037383 9.2812572 11.6300811 9.9308695 11.4473978 10.2431030 9.6442722 7.7804900 5.9543390 Class 27. Pine / Birch Forests of Kamchatka Suslov, (pg 395) described this area as "Taiga forest with birch groves". Again like class 9 and 19. By far, the largest area of this class is located in the Far East,Kamchatka. However, in the Ob river basin, the lower reaches of the river are class 9, the middle regions class 19, and the upper reaches of the Ob are identified as class 27. Coincidence with Vegetation Map Classes: Stony Birch (Betula ermanii) 32.3% Pinus pimula 12.9% % Arable Lands in Class: NA Class 27 2O Overall Mean: 86.167 Area under TNRC: tm March-October 641.552 IC0 April-September 559.633 30 Closely Related to Class: 0 Y-b April _.-Max 19 Mq Jme July _...StdDev _a-Aveage Ao-m Sq--h- .Min Summary Statistics for Class: -variable Minimum Maximum Mean Std Dev --------------------------------------------------------------------MAR APR MAY JUN JUL AUG SEP OCT N 1209 1209 1209 1209 1209 1209 1209 1209 30.0000000 26.0000000 22.0000000 52.0000000 93.0000000 107.0000000 71.0000000 40.0000000 55.0000000 63.0000000 94.0000000 146.0000000 188.0000000 189.0000000 152.0000000 80.0000000 41.5690653 34.1364764 49.1877585 102.1993383 151.4400331 146.1397849 110.6658395 53.9975186 3.4698898 5.2893496 14.0069868 15.4319420 14.8699635 14.1036659 12.2170289 5.6413939 Class 28. Semi-Desert Steppe in Kopet Dag and Pamir Mountains This class is located on the southern border of the desert. Suslov (pg 569) describes this steppe differently from the balance of the semi-desert steppes in central Asia as being more "Iranian" in character. Also reference the video series "Realms of the Russian Bear: the Red Deserts" which describes this region very well. The early spring bloom, the rapid senescence due to moisture deficits etc. In terms of carbon values, it is unlikely that this class is significantly different from class 14. Coincidence with Vegetation Map Classes: Ephemeral Shrubs and Artemisia 11.7% Desert 11.07% Class 28 % Arable Lands in Class: 7.10% Overall Mean: 64.239 :a) Area under TNRC: DSO March-October 454.342 too April-September 306.720 Closely Related to Class: `A 0 M-h Ap-i ...- Max M.y J® -e-Std Dev July Auu. SW-.b- Average .. Min 14 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 840 840 840 840 840 840 840 840 29.0000000 24.0000000 43.0000000 28.0000000 19.0000000 23.0000000 18.0000000 31.0000000 142.0000000 155.0000000 154.0000000 101.0000000 91.0000000 106.0000000 103.0000000 69.0000000 71.7928571 88.0523810 76.5690476 63.2250000 57.0607143 55.7464286 54.1190476 47.3452381 20.6080384 20.4972656 19.8794338 11.7971772 10.6887877 11.1652977 9.2131491 4.6631312 Class 29. Middle Taiga Spruce and Pine This class is located west of the Ob River between class 18 and class 3233. This class appears to be very similar to class 30 located east of the Ob River. Class 23 appears as discontinuous patches in class 29. The location of the Ural mountains is well defined by class 23 in the division of the zonal distribution of class 29. Coincidence with Vegetation Map Classes: Middle and Southern Taiga Pine 25.26% Middle Taiga Picea 16.57% % Arable Lands in Class: 6.2 % Class 29 Overall Mean: 77.599 No Area under TNRC: IM March-October 572.492 100 April-September 475.865 so Closely Related to Class: April 30 and 23 t Max may Jr t Std Dev hur Apa -.r- Average -.0-Min Summary Statistics for Class: Variable N Minimum Maximum Mean Std Dev ----------------------------------------------------------------------MAR 3948 29.0000000 68.0000000 48.2558257 5.5226373 APR 3948 26.0000000 70.0000000 48.3267477 6.2226344 MAY 3948 59.0000000 102.0000000 76.6263931 6.7807969 JUN 3948 85.0000000 133.0000000 110.1790780 6.3739635 JUL 3948 84.0000000 140.0000000 116.0590172 8.3903266 AUG 3948 74.0000000 124.0000000 99.9442756 7.6459491 SEP 3948 49.0000000 91.0000000 73.0564843 6.2664509 OCT 3948 37.0000000 63.0000000 48.3437183 3.3986306 Class 30. Middle Taiga Conifer (Spruce-Fir-Pine) Forest This class is located east of the Lena River and east of Lake Bakail, across the Trans-Bakail mountains to the northern 1/3 of Sakhalin island. A small patch of this class is located in the center of the Kamchatka valley indicating the small pocket of middle taiga that can grow at lower elevation in the peninsula. Coincidence with Vegetation Map Classes: Mountain Larch and Pine 25.6% Middle Taiga Spruce 16.57% Class 30 % Arable Lands in Class: 2.2% Overall Mean: 82.529 M Area under TNRC: March-October 611.345 100 April-September 516.582 so 0 MW& Closely Related to Class: Apn, M., t Max le M, _._Std Dev __ Average M Sq_ftW ooae. Mm 29 in Europe and 36 in southern siberia I Summary Statistics for Class: Variable 1 Minimum Maximum Mean Std Dev ----------------------------------------------------------------------MAR APR MAY JUN JUL AUG SEP OCT N 3918 3918 3918 3918 3918 3918 3918 3918 29.0000000 29.0000000 52.0000000 88.0000000 91.0000000 96.0000000 64.0000000 40.0000000 64.0000000 64.0000000 93.0000000 135.0000000 149.0000000 137.0000000 115.0000000 65.0000000 48.1199592 45.8736600 73.1835120 112.6607963 127.5104645 116.3328229 86.8945891 49.6590097 4.6143492 5.4369095 6.8249766 6.9865877 6.9303230 6.6339808 7.1989040 3.5361815 Class 31. Conifer Forests on Mountain Slopes of Southern Siberia. This class is located only in southern Asia near Lake Bakail. This class is also located in the mountains of the Amur/Primoriski region and appears to be closely related to elevation. Coincidence with Vegetation Map Classes: Spruce-Fir in Central Mountains 33.96% Southern Taiga Larch-Pine 18.37% % Arable Lands in Class: 3.05 % Class 31 Overall Mean: 89.158 :Go Area under TNRC: 170 March-October 652.773 :OD April-September 526.612 !0 Closely Related to Class: n+., Ap d Max 24, 30, and, 36 ,_ ,m,- IWy _._ Std Dev ..r Average s.,. Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT 14 2506 2506 2506 2506 2506 2506 2506 2506 Minimum Maximum Mean 33.0000000 44.0000000 65.0000000 71.0000000 81.0000000 81.0000000 75.0000000 41.0000000 92.0000000 95.0000000 108.0000000 143.0000000 148.0000000 148.0000000 125.0000000 89.0000000 63.8431764 65.6679968 82.6360734 115.3084597 121.8172386 112.4397446 94.4102155 57.1444533 Std Dev! 7.7339846 7.3344884 6.6398693 9.3236026 9.2316884 8.3314904 6.8179140 7.2559070 Class 32. Forest Meadow Steppe / Boggy Larch in Vilyuy Basin This class has two areas of very different structure as part of the same class. The forest meadow steppes east of the Ural mountains are represented as class 32. These have been described as mixed conifer/birch groves in grassland. (Suslov, pg 36; Berg, pg 80). East of Altai mountains, four isolated patches of forest meadow steppe appear to correspond to the isolated patches of forest meadow steppe in south central Siberia described by Suslov (pg 204). The other area of this class is located in the Vilyuy and Lena basins. This corresponds to larch with pothole swamps class on the vegetation map. A coincidence tabulation restricted to this region show 45.24% larch with pothole swamps, plus 33.6% middle taiga larch. Obviously the NDVI signature of swampy larch forest is very similar to that of forest meadow steppe in western Siberia. Coincidence with Vegetation Map Classes: Meadow and Forest Steppe 22.9% Middle Taiga Larch with Pothole Swamps 19.2% Aspen 10.58% Middle Taiga Larch 14.66% % Arable Lands in Class: Class 32 =m 13 % Overall Mean: 82.903 Area under TNRC: IT March-October 618.506 50 April-September 533.117 D. Closely Related to Class: M- APO Mq -a- Max J®r 1A, Aapa Z. Saprbr Ocher t Sid Dev _ Average -0.. Mia 25 and 35 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT sN. 3198 3198 3198 3198 3198 3198 3198 3198 Minimum Maximum Mean Std Dev 26.0000000 22.0000000 42.0000000 110.0000000 120.0000000 101.0000000 56.0000000 40.0000000 60.0000000 60.0000000 85.0000000 159.0000000 165.0000000 137.0000000 104.0000000 57.0000000 42.6932458 40.6751094 64.0434647 128.7141964 140.4878049 120.5819262 79.2892433 46.7357724 4.9601501 4.5797230 6.3618093 7.1575674 6.9336420 5.9018146 7.8275807 2.2629551 Class 33. Coniferous Forest of Russian Plain This class is located across the center of the Russian plain, bordered by class 29-30 on the north, and class 37 on the south. Evidence suggests that this is a region of mixed conifers. Coincidence with Vegetation Map Classes: Middle Taiga Spruce 23.5% Southern Taiga Spruce 22.5% Middle and Southern Taiga Pine 21.8% % Arable Lands in Class: Class 33 14% 210 Overall Mean: 84.167 .00 Area under TNRC: 130 March-October 624.734 IDD April-September 527.372 Closely Related to Class: M., Ap.d Max ,®. -4._ Std Dev Jy Aopr s.,amnc Average ..Min Summary Statistics for Class: Variable N Minimum Mean Std Dev Maximum ----------------------------------------------------------------------MAR 47.9333959 3198 29.0000000 55.0000000 5.7523387 APR 49.0240775 3198 30.0000000 6.1936517 66.0000000 MAY 3198 63.0000000 82.3242652 5.7134345 97.0000000 JUN 3198 110.0000000 152.0000000 128.5056285 6.9709033 JUL 3198 105.0000000 128.5659787 151.0000000 6.7492647 AUG 112.9186992 3198 91.0000000 135.0000000 6.5310514 SEP 75.0572233 3198 52.0000000 5.7335403 91.0000000 OCT 3198 38.0000000 48.7426517 60.0000000 3.3634590 -------------I Class 34. Grassy Steppe on Slopes of Caucasus and Pamir Mountains and Crimea Well described in Crimea by Berg. Closely related to class 26 and classes This class appears to be related to elevation. Examination of detailed maps shows strong correlation with lower slopes of Pamir mountains. 23-7. Coincidence with Vegetation Map Classes: Grassy Steppe w/ herbs 22.51% Grassy Steppe 11.86% % Arable Lands in Class: 44 % Class 34 Overall Mean: 90.192 :m Area under TNRC: March-October 668.128 April-September 533.804 50 Closely Related to Class: A.d . Max 26 and 35 - ,.y ,m. -4-Std Dev A.00 _ Average serer Min n Summary Statistics for Class: Variable Minimum Maximum Mean Std Dev ------------------------------------------------------------------ MAR APR MAY JUN JUL AUG SEP OCT N 942 942 942 942 942 942 942 942 - 30.0000000 46.0000000 92.0000000 75.0000000 68.0000000 59.0000000 55.0000000 44.0000000 105.0000000 142.0000000 169.0000000 189.0000000 159.0000000 140.0000000 123.0000000 82.0000000 49.5169851 80.9161359 121.5254777 123.4957537 111.0721868 97.0774947 80.6337580 57.2972399 11.8971517 15.6294218 12.3815917 16.1139501 15.0261936 12.0787947 10.4475130 5.9133643 Class 35. Forest Meadow Steppe This class is located only in the western FSU, between the grassy steppe of class 26 and the mixed forest of class 38. This class is related to class 32 on the east side of the Ural mountains. The forest groves west of the Urals are predominantly oak; east of the Urals, birch and aspen. According to Berg (pg 83), "The steppe (parts) of the forest steppe are almost entirely under cultivation." From the Arable Lands Map, class 35 has the greatest amount of high percentage arable lands. Coincidence with Vegetation Map Classes: Grassy Steppe w/herbs 43.92% Meadow and Forest Steppe 22.4% % Arable Lands in Class: Class 35 64 % Overall Mean: 79.975 Area under TNRC: IS March-October 595.128 Iro April-September 504.114 m Closely Related to Class: M Apnl ..... Max M" h. 1d, -... Std Dev -. Average Min falls between 38 and 26 the same structure appears to be class 32 east of the Ural Mtns. Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 3179 3179 3179 3179 3179 3179 3179 3179 26.0000000 33.0000000 72.0000000 99.0000000 89.0000000 76.0000000 53.0000000 40.0000000 50.0000000 68.0000000 121.0000000 151.0000000 141.0000000 121.0000000 91.0000000 70.0000000 35.6744259 46.3444479 95.4970116 122.2632903 117.3000944 97.1273986 71.9263919 53.6653036 3.8061108 6.4590453 7.9742885 8.4887410 8.1231419 7.4614049 6.6158356 4.8545192 Class 36. Larch / Pine (Spruce) Forests in Southern Siberia This class is located from the Ob to Sakhalin Island across souther Siberia. This class appears to be closely related to class 30-31 on the north, and class 40 to the south. A correlation with elevation may be present. This connection is not completely clear however. Coincidence with Vegetation Map Classes: Mountain Larch/Pine 17.4% Southern Taiga Larch/Pine 12.21% Middle Taiga Spruce 11.39% % Arable Lands in Class: Class 36 2.58 % 20 Overall Mean: 90.123 Area under TNRC: IM March-October 669.825 April-September 565.875 so Closely Related to Class: A,nl 31 and 40 M" Max J. _*__ Std Dev )m, AoSM i.. Average -.o. Min sser Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT :N' Minimum 30.0000000 33.0000000 53.0000000 97.0000000 112.0000000 103.0000000 74.0000000 41.0000000 - - - --- - - - - 3613 3613 3613 3613 3613 3613 3613 3613 Mean Std Dev 52.1228896 75.0000000 52.0669803 74.0000000 75.7879878 95.0000000 125.3617492 148.0000000 140.9288680 170.0000000 127.2637697 149.0000000 96.5327982 133.0000000 51.6421257 70.0000000 ---------- 5.7132822 5.7826494 6.3598616 7.1951171 8.1692154 6.6464594 6.9618356 4.1937526 Maximum Class 37. Southern Taiga Forests of Southern Russia This class represents very productive spruce forests mixed in some places with oak and hornbeam (Berg, pg 54). In the southern Ural mountains, there is a strong correlation with mid to lower slopes and the distribution of class 37. Coincidence with Vegetation Map Classes: Southern Taiga Spruce 35.3% Mixed Forest 13.5% Broadleaf Forest 11.25% % Arable Lands in Class: Class 37 :50 27 % Overall Mean: 90.991 Area under TNRC: March-October 679.229 April-September 583.781 70 Ma Closely Related to Class: A,nl MW -._ Max J fw Std Dev - Sq.". Average -o-Mm 38, 39 and 33 Summary Statistics for Class: Variable Minimum Maximum Mean Std Dev ---------------------------------------------------------------------- MAR APR MAY JUN JUL AUG SEP OCT N 2576 2576 2576 2576 2576 2576 2576 2576 26.0000000 30.0000000 66.0000000 125.0000000 123.0000000 96.0000000 56.0000000 39.0000000 _--_---- 66.0000000 65.0000000 129.0000000 180.0000000 166.0000000 152.0000000 102.0000000 64.0000000 47.1708075 46.7488354 92.7239907 146.7585404 140.6358696 123.1164596 80.5465839 50.2259317 7.1419136 5.5678166 9.0258906 7.8729674 6.9311299 7.8844338 7.7267009 4.1786563 Class 38. Mixed Conifer / Broadleaf Forests This class is located only in the European plains of Russia and the Ukraine. This class represents oak/pine forest mixed with agricultural lands. Coincidence with Vegetation Map Classes: Mixed Forest 27.12% Broadleaf Forest 23.43% % Arable Lands in Class: Class 38 43% Overall Mean: 89.838 Area under TNRC: to March-October 667.716 April-September 562.001 Closely Related to Class: Apd MW Max 35 and 41 1_ . Sod Dev J* A.P. Avcage .4. Min Sepe.C- Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT Maximum Mean Std Dev 30.0000000 63.0000000 34.0000000 80.0000000 102.0000000 109.0000000 88.0000000 67.0000000 80.0000000 128.0000000 158.0000000 151.0000000 132.0000000 109.0000000 75.0000000 43.1688539 54.7305337 103.8932633 130.3753281 129.0282881 112.1073199 86.5969670 58.7999417 5.2484713 8.1084937 6.5464620 8.1712625 6.6211944 N Minimum 3429 3429 3429 3429 3429 3429 3429 3429 ------------ 45.0000000 ---------- 5.8942080 6.3530652 5.4633274 Class 39. Mixed Conifer / Broadleaf (Aspen-Birch) Forest This class is located between class 32 and 36 in the Ob basin. This class is also located in the Altai mountains in the Amur River. Like class 36, this class is located only in souther Siberia from the Tom to the Amur. I suspect this class represents fairly open forests with meadow openings. Class 38 and 39 should be very similar in structure with oak west of the Urals (class 38) and birch/aspen east of the Urals. Class 39 and 42 may be closely related. Coincidence with Vegetation Map Classes: 4L Meadow Forest Steppe 18.6% Aspen 15.10% Southern Taiga Pine 14.44% Class 39 % Arable Lands in Class: 9.25% .00 Overall Mean: 93.686 Area under TNRC: March-October 701.451 50 April-September 608.724 0 Aprd Mey _45_. Max Closely Related to Class: 1me My A"- _*..Std Derv t Average -: Min 36 and 42 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev 2833 2833 2833 2833 2833 2833 2833 2833 29.0000000 24.0000000 38.0000000 105.0000000 139.0000000 115.0000000 76.0000000 41.0000000 68.0000000 62.0000000 94.0000000 177.0000000 189.0000000 163.0000000 124.0000000 63.0000000 46.3490999 44.6911401 68.8545711 140.6156018 162.3780445 137.8736322 99.0024709 49.7214966 5.9096224 6.8449007 7.6820670 10.6211133 7.0482627 6.8815329 7.1667237 3.4701746 Class 40. Mixed Forest (Birch and Spruce/Laich) This class is located near Lake Baikal, and along the Amur River across the southern tier of southern Siberia. The Amur-Primoriski region is basically characterized by class 40, 42, 39, and 31. Coincidence with Vegetation Map Classes: Picea and Abies of Southern Mountains 17.61% Larch/Pine Forests of Southern Taiga 10.69% % Arable Lands in Class: Class 40 2.45 % Overall Mean: 98.087 200 Area under TNRC: 150 March-October 728.115 April-September 614.496 a Closely Related to Class: 0 ApN 36, 39, and 42 Max M., ,.. AI, Aq.. s.p.6r .t Std Dev _ Aveaage ... Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N Minimum Maximum Mean Std Dev ----------=------''-----------------------------------2271 31.0000000 91.0000000 55.5112285 8.6216985 2271 32.0000000 94.0000000 57.0413915 7.8886555 2271 63.0000000 107.0000000 84.0546015 6.9738729 2271 109.0000000 168.0000000 140.9735799 8.3092331 2271 112.0000000 170.0000000 146.6574196 7.8142948 2271 101.0000000 164.0000000 134.1523558 8.2552155 2271 83.0000000 139.0000000 108.6583003 7.8810776 2271 43.0000000 88.0000000 57.6433289 6.7564000 Class 41. Broadleaf (Mixed) Forest of Russia, Beylorruss and the Ukraine This class is located only in the western part of the FSU. This class is almost completely surrounded by class 38, with large islands of class 38 in the middle. Class 41 could almost be called islands inside a class 38 matrix. Coincidence with Vegetation Map Classes: Mixed Forest 34.05% Broadleaf Forest 31.46% Class 41 % Arable Lands in Class: ss 42 % Overall Mean: 97.838 Area under TNRC: March-October 728.974 April-September 614.863 Apnl Closely Related to Class: .y.. Max w, r®. July An. sM-Me> . Std Dev -- Average -.0-Min 38 Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP nrT 3206 3206 3206 3206 3206 3206 3206 R2nr. Minimum Maximum Mean 29.0000000 36.0000000 77.0000000 112.0000000 103.0000000 99.0000000 76.0000000 4F nnnnnnn 59.0000000 108.0000000 148.0000000 173.0000000 168.0000000 142.0000000 136.0000000 RR nnnnnnn 41.4401123 60.3833437 114.4915783 141.5973175 139.5389894 120.8228322 98.4123518 rA ni4rAnn -- ---------------------------- =_- - Std Dev 3.8339155 8.2035697 8.1823079 8.6060084 7.1288701 6.3222206 7.4159013 q 9232299 --- ---------------- Class 42. Highly Productive Mixed Forest This class represents the "greenest" regions in the FSU. These regions are located on the middle slopes of the Carpathians, Caucasus, and Altai mountains. A small amount of this class is also located in the southern Ural mountains. In these regions, we have mixed conifer/broadleaf forest with extensive and complex shrub and herb understory. In the Amur region, class 42 corresponds very well to the"Ussuri taiga". A mixture of Manchurian oak and other broadleaf trees with a complex understory of shrubs and herbaceous plants. Suslov's map (pg 859) shows this correspondence very well. Coincidence with Vegetation Map Classes: Picea and Abies Central and Southern Highlands 20.35% Broadleaf Forest 16.55% Ussuri Taiga (Broadleaf/Conifer Forest) 14.93% Broadleaf Forest of Far East 11.21% % Arable Lands in Class: 7.36 % Class 42 Overall Mean: 109.847 Area under TNRC: March-October 820.074 April-September 701.344 JO Closely Related to Class: 0 Myth Aped M -Y Max 39 and 40 Jnn July ,... Std Dev Augeue Sepem6r Qm6r Average -.0-Min Summary Statistics for Class: Variable MAR APR MAY JUN JUL AUG SEP OCT N 1236 1236 1236 1236 1236 1236 1236 1236 Minimum Maximum Mean Std Dev -----------------------------------------------------28.0000000 87. 0000000 51.1674757 10.1242617 29.0000000 123. 0000000 60.0315534 15.8419259 47.0000000 168. 0000000 105.3244337 19.2602652 1-13.0000000 198. 0000000 164.0121359 13.1211366 103.0000000 201. 0000000 160.9182848 14.9639444 112.0000000 172. 0000000 146.2192557 9.4046066 90.0000000 162. 0000000 124.8697411 13.9678065 43.0000000 107. 0000000 66.2305825 13.6757501 - -------------