LUCID’s Land Use Change Analysis as an Approach for Investigating Biodiversity Loss and Land Degradation Project An Approach to Dual Land Use and Land Cover Interpretation of 2001 Satellite Imagery of the Eastern Slopes of Mt. Kenya LUCID Working Paper Series Number: 16 by Bilal Butt and Jennifer M. Olson Department of Geography Michigan State University East Lansing, MI 48824 USA October 2002 Address Correspondence to: LUCID Project International Livestock Research Institute P.O. Box 30709 Nairobi, Kenya E-mail: lucid@cgiar.org Tel. +254-20-630743 Fax. +254-20-631481/ 631499 An Approach to Dual Land Use and Land Cover Interpretation of 2001 Satellite Imagery of the Eastern Slopes of Mt. Kenya The Land Use Change, Impacts and Dynamics Project Working Paper Number: 16 by Bilal Butt and Jennifer M. Olson Department of Geography Michigan State University East Lansing, MI 48824 USA October 2002 Address Correspondence to: LUCID Project International Livestock Research Institute P.O. Box 30709 Nairobi, Kenya E-mail: lucid@cgiar.org Tel. +254-20-630743 Fax. +254-20-631481/ 631499 Copyright © 2002 by Michigan State University Board of Trustees, International Livestock Research Institute, and United Nations Environment Programme/Division of Global Environment Facility Coordination. All rights reserved. Reproduction of LUCID Working Papers for non-commercial purposes is encouraged. Working papers may be quoted or reproduced free of charge provided the source is acknowledged and cited. Cite working paper as follows: Author. Year. Title. Land Use Change Impacts and Dynamics (LUCID) Project Working Paper #. Nairobi, Kenya: International Livestock Research Institute. Working papers are available on www.lucideastafrica.org or by emailing lucid@cgiar.org. LUCID Working Paper 16 ii TABLE OF CONTENTS List of Tables ..........................................................................................................................iv List of Figures .........................................................................................................................iv List of Appendices ..................................................................................................................iv A. INTRODUCTION .............................................................................................................1 B. THE LAND USE AND LAND COVER CLASSIFICATION APPROACH .....................1 C. IMAGE INTERPRETATION ............................................................................................3 1. Study site characteristics affecting interpretation .........................................................3 2. Automated vs. visual interpretation ..............................................................................3 3. Stages of interpretation .................................................................................................4 a. Georectification .........................................................................................................4 b. Initial land cover interpretation ..................................................................................5 c. Ground truthing ..........................................................................................................7 d. Correction of land cover attributes and generation of land use attributes ..................8 e. Calculation ..............................................................................................................10 D. CONCLUSIONS AND RECOMMENDATIONS ..........................................................10 E. REFERENCES ................................................................................................................16 Appendices .............................................................................................................................17 LUCID Working Paper 16 iii LIST OF TABLES 1. Land use and land cover classes for the Eastern Mt. Kenya study site ...............................2 2. Calculated area and perimeter for each of the land use and land cover categories ...........12 LIST OF FIGURES 1. Vector digitising of the Shamba System polygon at a scale of 1:100,000 ...........................6 2. Map catalogue with the digitised polygons overlaid onto satellite imagery .........................7 3. Attribute data recorded from ground truthing with the use of the design template and the GPS unit ..........................................................................................................................8 4. Island polygon overlaying another polygon and after intersection, two separate polygons created ...........................................................................................................................10 5. Land cover map of the Eastern Mt. Kenya LUCID study site ...........................................13 6. Land use map of the Eastern Mt. Kenya LUCID study site ...............................................14 7. Land cover draped over a 250m resolution DEM ..............................................................15 LIST OF APPENDICES 1. Definitions of Land Use, Land Cover and Land Use/ Cover Change ...............................17 2. Waypoint Sheet for Ground Truthing ...............................................................................18 LUCID Working Paper 16 iv A. INTRODUCTION A dual land use/ land cover mapping exercise was undertaken to identify, interpret and analyse the landscapes of Mt. Kenya and the surrounding lowlands using Landsat ETM+ imagery acquired for 11 March, 2001. Mt. Kenya and its surrounding lowlands are characterized by extreme variations in both the physical landscape, ranging from glaciers to dryland savannahs, and in the human landscape, ranging from large paddy rice schemes to small crop/livestock farms. Due to this heterogeneity, automated supervised and unsupervised classification resulted in the misclassification of much of the study area. As a result, the LUCID team adopted a mixed approach using supervised classification and visual interpretation. The land use and land cover mapping activities were largely based on visual interpretation and transformed into a GIS through the vector digitisation of polygon features. The interpretations were corrected after ground truthing and consulting secondary sources. The team assigned separate land use and land cover classes to each polygon to better link the spatial patterns of use/ cover changes to socio-economic and biophysical driving forces and to record the vegetative characteristics for associated analyses of land degradation, biodiversity and carbon sequestration. This working paper was written as a general guide to conducting land use/cover interpretation of satellite imagery in heterogeneous areas where the results of automated classification systems are unsatisfactory. The paper details the conceptual approach to the classification and interpretation strategies, outlines the stages of interpretation and analysis and provides recommendations on how to adapt the approach for similar studies in Africa. The IGBPLUCC definitions of land use, land cover, land use change and land cover change, which the team used to develop its classification scheme, are listed in Appendix 1 (IGBP, 1997). B. THE LAND USE AND LAND COVER CLASSIFICATION APPROACH The land use/cover interpretation of the March 11, 2001 ETM+ satellite image of Mt. Kenya is part of the wider LUCID research project that is examining the relationship between changing land use, biodiversity and land degradation. As such, it was critical that the interpretation of the image provides an accurate rendition of the location and distribution of land use and cover types, including as much information about the type of agricultural cover. The cover information will be important for examining, for example, the relationship between changing vegetative classes, fauna, and soil characteristics. It will also be required for future examination of above ground carbon storage. The wider project is also identifying the underlying socio-economic and biophysical driving forces of land use/cover change in order to better project future changes. For this, it is vital to differentiate areas under different land ownership and management types, a differentiation that could be captured in a land use classification. Both land cover and land use classifications, therefore, are required for the purposes of this study. A hierarchical system of nomenclature was utilized for the land use/ cover classification scheme and is listed in Table 1. The land cover scheme has been adapted from the Food and Agricultural Organization of the United Nations (FAO) (Latham, 2001) and the Biosphere-Atmosphere Transfer Scheme (BATS) classifications (Dickinson et al. 1996). The FAO land use classification scheme was refined to include land ownership and management variations. LUCID Working Paper 16 1 Table 1: Land use and land cover classes for the Eastern Mt. Kenya study site Land Use Code General Land Use Type 1000 1100 1110 1120 1130 1140 1200 1300 2000 2100 2200 2300 2400 2500 3000 3100 3200 3300 4000 4100 4200 4300 5000 6000 7000 7100 7200 8000 8100 8200 Agriculture - Small Scale Land Cover Code General Land Cover Type 1000 2000 2100 2200 2300 2400 2500 2600 2700 3000 4000 4100 4110 4120 4130 4140 4150 4200 4210 4220 5000 6000 6100 6200 7000 Tundra/Mooorland/Glaciers/Grasses Forest Specific Land Use Type Sub-Specific Land Use Type Rainfed Cropping Tea Maize Dominant Mixed Bush/Crops Coffee Irrigated Cropping (Horticulture Dominant) Grazing Land (Bush and Grassland) Agriculture - Large Scale Rainfed Cropping (Wheat Dominant) Irrigated Cropping (Horticulture Dominant) Shamba System (Mix of Crops and Tree Plantations) Ranches Wheat and Grazing Pasture) Protected Areas National Parks National Reserves Forest Reserves Institutional Land Uses University Research Plot KenGen Land Don Bosco Farms Tree Plantations Urban and/or Built-up Areas Water Bodies Dams Lakes Non-Protected Forest Areas Non-Degraded Degraded Specific Land Cover Type Sub-Specific Land Cover Type Bamboo Forest Afro-Montane Forest Woodland (Open Canopy, mostly Dryland Forests) Tree Plantations Shamba System Degraded Forest Degraded Woodland Bush Cultivated Land Rainfed Cultivation Tea Maize Dominant Mixed Bush/Cultivation (Grains Dominant) Wheat and Pasture Coffee Irrigated Crops Rice Horticulture Urban and/or Built-up Area Water Bodies Dams Lakes Grassland In this dual land use/ land cover classification system, every polygon is assigned both a land use and land cover attribute, resulting in separate use and cover spatial layers. More than one polygon may share the same land use code but may have different land cover codes. For example, the land use polygon ‘national park’ on Mt. Kenya includes two land cover polygons: the tundra/moorland/glaciers zone and the afro-montane and bamboo forest zone. Similarly, areas under the same land cover may have different land use designations; such forest cover areas with land use codes of national park, national forest reserve or unprotected forest. This dual land use/ cover coding allows for flexibility in the spatial analyses. LUCID Working Paper 16 2 C. IMAGE INTERPRETATION C.1. Study site characteristics affecting interpretation Mt. Kenya is the second highest mountain in Africa and is located in central Kenya. Its wellwatered slopes provide critical high potential agricultural conditions in the predominately semi-arid nation, and the mid-slopes have been intensely farmed for many years. The mountain is surrounded by a semi-arid lowland plateau. The project’s study site consists of the eastern slopes of Mt. Kenya and encompasses a steep ecological gradient from the glaciers on the mountaintop at 5,199 metres above mean sea level (AMSL) to dryland grasslands at 600 metres AMSL elevation. The site covers 11,670 km2, or approximately one third of an ETM+ Landsat satellite image. The heterogeneity of the landscape of the site is extreme both between and within land use and cover classes. For example, the natural vegetation ranges from sparse tundra vegetation and afro-montane rainforest, to sparse grasslands in the lower elevation, dryland area. The human managed landscape includes irrigated paddy rice, tea and tree plantations, coffee and maize farms and scattered fields of millet and sorghum within bush. Most of the landscape is heavily influenced and closely managed by humans. Land managers include small-scale farmers whose farms are typically less than 2 hectares each, private wheat farms and sheep ranches of up to 300 hectares in size, large agricultural parastatals and parks and reserves managed by local and national governments. C.2. Automated vs. visual interpretation The heterogeneity in land uses and covers led to our inability to rely on automated classification schemes. Traditional supervised and unsupervised classification techniques tended to produce either too many classes differentiating areas that were actually similar, or, when the number of classes was reduced, to join radically different areas such as irrigated areas with mixed bush and crops. An attempt to reduce the noise by dividing the image into separate elevation bands and classifying within those bands, was useful in defining the boundaries of some classes (e.g., forest types on the mountain, grasslands within a park, tea), but was not helpful for most of the study area such as those characterized by small scale agriculture. In the drylands, which are a mosaic of small and large cultivated fields, fallowed fields, grassland, and bush of varying heights, the automated classification schemes produced speckled results without differentiating broader regions, for example areas with medium versus low intensity agriculture. In the drylands, it was necessary to interpret high resolution aerial photographs (1:20,000) in sample sites to obtain estimates of the area under cultivation and under other uses (Olson, 1998). At the lower resolution (1:100,000) of the ETM+ satellite imagery, it was possible to delineate only a class of mixed bush and farms with no finer detail. The scale of 1:100,000 was arrived at by zooming to the raster resolution. To better understand the societal restrictions and processes in order to project future land use/cover change, it was necessary to differentiate land tenure and management types. Differentiating between large and small-scale tea producers, or between government and community managed bush land areas, for example, is critical in predicting how the use of the land will change, and whether the area under that class will expand or contract. This information was gathered from a variety of sources including ground truthing, group interviews, maps and GIS layers from different sources and by consulting experts knowledgeable of the area. The land use class information was saved as a separate variable from the land cover information. Each polygon was thus assigned a land use as well as a land cover attribute. Where the land cover and use boundaries differed (e.g., due to agricultural incursion into a protected forest), the land cover boundary has been used as the initial land use LUCID Working Paper 16 3 boundary (protected area and other ownership boundaries will later be added). The basic approach adopted, therefore, was to conduct visual interpretation of the image and to identify both the land use and land cover of each polygon. The LUCID team also identified the boundaries of some land covers, such as between forest types using an automated classification technique known as ‘seeding’. The vectorized digitising of the raster image file (also known as heads-up digitising) is similar to manual digitising of paper sources in that lines or polygons are traced by hand, but the interpreter works directly on the computer screen using the image as backdrop. With the help of the display tools of ArcView GIS, such as zoom in and out, the operator can work at the resolution of the raster data and thereby digitise at a higher accuracy level. However, the accuracy is still highly dependent on the interpreter. The tracing method automates the process by creating one line or polygon at a time on the image displayed on the computer screen. This is a significant improvement in accuracy and speed over manual digitising of interpretations placed first onto paper. The improvement is especially pronounced when fully automatic raster to vector conversions cannot be applied in cases such as low image quality or complex layers. These include, but are not limited to instances of cloud cover, or when a segment of the image contains a number of different land use or cover classes such as Shamba system interspersed with plantation and afro-montane forests. C.3. Stages of interpretation The various stages of interpretation that were utilized as part of the land use and land cover change mapping assessment included georectification of the satellite image, initial stages of visual interpretation, ground truthing, and finally the correction of land use/land cover attributes. C.3.a. Georectification The first stage in the interpretation of the image was to geographically rectify the raster ETM+ image so that it conforms to existing spatial data. This was conducted in Erdas © Imagine software (ERDAS, 2001) using the following parameters: the datum was set to WGS 84 and referenced to the Universal Transverse Mercator (UTM) Zone 37 South. The image was referenced to a number of Ground Control Points (GCP’s) taken from a 1:50,000 topographic maps produced by the Survey of Kenya (1962-1997). The entire Landsat ETM+ Scene was georectified using approximately 90-100 GCP’s distributed across the image. The GCP’s were selected to be features that were visible on both the image and the topographic map sheets, such as roads, forest boundaries, towns, and other key features. The resulting output was saved as an Erdas Imagine *.img file and then opened in ESRI © ArcView as an image data source. ArcView GIS version 3.2 was used throughout most of this project (ESRI, 1999) The next stage was to overlay selected vector GIS data layers to assist in the land use/cover interpretation of the image. The team consulted layers such as roads, towns and market centres, rivers, administrative and protected area boundaries, and agro-ecological zones (ILRI, 2002, Jaetzold and Schmidt 1983). These data sets had been originally prepared at the national scale and, for this project, clipped to a bounding polygon of the study area. Following the projection conversion from decimal degrees to UTM Zone 37 South, most of the layers became spatially incorrect with errors ranging from as little as 100 metres to as much as 1 kilometre. A possible method by which this error could be reduced would be to transform the polygon with the boundary of the study area to geographic decimal degrees and a WGS 84 datum, and then use this boundary to clip the additional GIS data layers. The clipped layers should then be transformed to UTM coordinates. In this project, the supplemental GIS data layers were only used a guide to help in identification and LUCID Working Paper 16 4 interpretation. The scale that the supplemental data was displayed and interpreted was a minimum of 1:100,000 to maintain consistency with the scale of interpretation of the image. Only features recognizable at that scale are thus mapped. As a result, any features that are too small to clearly visualize at that scale were not digitised, resulting in a de facto minimum mapping unit of 30 hectares. C.3.b. Initial land cover interpretation The next stage of the mapping activity was to label each polygon with its land cover identifiers. A new attribute record was edited to the attribute table and labelled as LC_Code. The new fields were added as a numeric variable with sufficient character spaces for the land cover type labels. Each general land cover type was assigned a 4 digit numeric code, for example, a land cover class of cultivated land was assigned a code of 4000, within this land cover class there are a number of specific land cover types such as rainfed cultivation. Rainfed cultivation also has specific land covers such as tea. Therefore a land cover code for tea was thereby assigned a code of 4100. The full code list is given in Table 1. A variety of combinations of the 30 metre spatial resolution imagery bands were used to assist in the identification and interpretation process. The combinations that were most commonly used were bands [4,3,2] [5,4,3] and [7,4,2] [R,G,B]. These were used in combination with the 15-meter panchromatic band, which was added as a separate layer (typically the 15- meter panchromatic is viewed in ArcView as bands 9,9,9 unless the band has been created as a separate file as in other remote sensing software packages). The 4,3,2 band combination detects vegetation through chlorophyll content, while bands 5,4,3 reflect moisture content and bands 7,4,2 reflect irrigated surfaces. The 4,3,2 band combination was commonly used to differentiate forests and degraded forests, the tea and coffee zones, and the large farms and ranches. The combination 5,4,3 was used to examine dry woodland and riverine forests. Finally the combination 7,4,2 was used to identify large and small-scale irrigated crops such as rice and horticulture. The 15 metre panchromatic band was especially useful for identifying features distinguished by texture or shape such as the boundaries of square fields indicating the presence of farms, or rectangles indicating buildings in urban areas. The next stage of the mapping exercise was to visually interpret and digitise the boundaries of the land cover polygons. Figure 1 illustrates how a feature representing the Shamba systems (a government scheme where a mix of forests and cash crops are grown together) on the upper slopes of Mt. Kenya was digitised at a scale of 1:100,000. The polygon that has been digitised is then given a label identifier (ID) with a generic name representing the land cover type and a land cover code. LUCID Working Paper 16 5 Figure 1. Vector digitising of the Shamba System polygon at a scale of 1:100,000 This same process was repeated until the entire image was interpreted and a new layer created of unique polygons each with their own label identifier. Each polygon was thus surrounded by other unique ID polygons. The general and interactive snapping tolerances were enabled and set at 50 meters to permit adjacent polygons’ vertexes to be joined. Snapping vertexes to nodes is an important procedure to minimize errors and to avoid polygon area miscalculations. The initial interpretation did not differentiate land uses from land covers, but instead determined just the cover types, as the mapping exercise was to provide a general identification of landscape boundaries. The initial interpretation contained twelve land cover classes, including large scale farms, small scale maize farming, small scale tea/coffee farming, tundra, afro-montane forest, riverine woodlands, shrub land, bush, deforested or bare soil, urban, and water. An additional category labeled ground truth contained polygons in which the visual interpretation was difficult and required ground truthing and/or consulting additional sources. A critical component of the initial land use/cover classification involved the use of the ‘seed tool’, a procedure available in the image analysis extension of ArcView GIS 3.2. The seed tool identifies a contiguous area of an image with spectral characteristics similar to a training area that the interpreter selects. The seed tool was mostly used to delineate the boundary between land covers. Examples of how the seed tool was used include differentiating the tea, coffee and maize small-scale agriculture zones, and bounding forests and water bodies. LUCID Working Paper 16 6 C.3.c. Ground truthing Once the polygons had been digitised and each assigned a land cover code, the resultant maps were prepared for ground truthing. The study area was divided into blocks, each corresponding to the extent of a 1:50,000 scale topographic map. The blocks, with their land cover interpretation displayed over the image, were printed and compiled into a field notebook. See Figure 2. Figure 2. Map catalogue with the digitised polygons overlaid onto satellite imagery. Each map zone was printed separately for ground truthing. Two key points that enhance the utility of the ground truthing maps are: 1) grid coordinates should be displayed in the correct coordinate system (e.g., in UTM they should be in metres and tic lines spaced at 5,000-metre intervals), and 2) the polygons should be displayed such that they are ‘transparent’ with only their boundaries visible so that the satellite image remains visible underneath the polygon boundaries. During ground truthing, the observer is then able to easily identify and correct the boundaries. Each printed map was placed in plastic sheets and compiled into folder. This permitted changes to the boundaries and identified fields to be drawn or written directly on the maps while in the field. During the ground truthing, Global Positioning System (GPS) units were used to identify where we were, to document the location of waypoints and to track line features such as roads or tracks. The correct parameters need to be programmed into the GPS unit, ideally the same parameters as those used to geo-reference the image, to ensure that the GPS recordings correspond to the GIS data layers. For example, the datum was set at WGS 84 and projection parameters set to UTM/UPS. When recording waypoints in the GPS, it is important to LUCID Working Paper 16 7 average the recordings to reduce the x and y coordinate errors by remaining at the same place for approximately 60-90 seconds. The point and line data recorded on the GPS was downloaded onto a laptop computer every night to prevent the accidental overwriting of the data in the GPS. The downloading was done using OziExplorer GPS Software (downloadable at www.oziexplorer.com). This easy-to-use software exports data to ESRI format shape files, which can be read by ArcView. The coordinates should first be exported in decimal degrees or latitude/longitude coordinates, and then if necessary transformed into UTM co-ordinates using either the projection utility tool. Data sheets (see Appendix 2) were completed that documented each waypoint’s surrounding land covers and uses, including details such as plant species and degree of deforestation. Interviews with people near the waypoints and in nearby towns helped to clarify land ownership, management and causes of use/cover change. The team also used still photography to document various observations. The information from the data sheets and the roll and frame numbers of the photographs were recorded as attribute data of the waypoints shape file as shown in Figure 3. C.3.d. Correction of land cover attributes and generation of land use attributes Information from the field, including notes on the notebook maps, interviews, photographs and the attribute data from the GPS, as well as secondary sources, were used to verify and correct the original land cover interpretation and use these secondary sources to determine the general, specific and sub-specific land use categories. Secondary sources included the 1:50,000 scale topographic maps, national and regional GIS layers for Kenya and its provinces, and the Mt. Kenya Aerial Survey Report and GIS dataset (SoK, 162-1997, ILRI, 2002, Gathaara 1999). By switching between the attribute table and the shape file (overlaid onto the image) the interpreter could identify and correct the land use and land cover codes. Figure 3: Attribute data recorded from ground truthing with the design template and the GPS LUCID Working Paper 16 8 Although few corrections in the location of boundaries between polygons were necessary, several polygons had originally been assigned incorrect land covers. Examples of omission and commission resulting in misclassification included: Several large farms and institutional land in the semi-arid area had been misclassified as small scale agriculture; A large region in a forest reserve had been classified as tree plantation but was found to be under the Shamba system (a governmental scheme of rotating planted trees and crops); A forest reserve (Imenti) had been classified as non-degraded forest when actually it was vigorous secondary growth following complete clearance of a mature forest a few years earlier; Some hills in the semi-arid zone had been misclassified as bush when they were actually degraded woodland, having been thinned out by grazing and cutting for charcoal; The area under irrigated agriculture had been underestimated because the interpreters were not expecting to find irrigation in zones where it had been only recently developed. These examples illustrate the importance of ground truthing and consulting supplemental sources, due to the limited information discernable from remotely sensed imagery and errors made by the interpreters due to a lack of complete knowledge. Areas with similar spectral characteristics may have very different covers (e.g., degraded woodland versus bush) or uses (large versus small scale agriculture). Land management systems (e.g., tree plantation versus Shamba system) are not visible on the image yet define how the land is used and how it will change. The next stage was to clean the data layer and correct any ‘island’ polygons. Island polygons are created when digitising a small polygon on top of a larger feature (see Figure 4). They cause problems during interpretation and analysis because that area has been assigned two attributes. There are a variety of methods to clean island polygons. One method is to download the ‘shape clean’ ArcView extension from the Internet and use the ‘intersection’ command. For small areas, an alternative is to digitise the island feature and then the larger polygon. A different method is to convert the layer containing the island polygons to a grid file (provided that the spatial analyst extension is loaded in ArcView) and then re-convert the grid file to a shape file. The island polygons will then be clipped out of the larger polygons. One disadvantage of the latter technique is that the land use or cover string label will be assigned a numeric code and you loose the text label, for example all plantation polygons will be converted to a code of 2. The observer will then need to manually correct the numeric grid codes to be text. Another disadvantage is that island polygons may be lost and need to be redigitised. Through trial and error, the LUCID team found that the output grid cell size specification should be set to about 50 meters to preserve the shape of smaller polygons. Finally, another method is to draw a line splitting the larger polygon, and digitise the island polygon adjacent to the line. The split polygon can then be deleted and a new polygon drawn. Snap the edges of the previously split polygon to the new polygon using the general and interactive snapping. Through any of these procedures, the cleaned island polygons should resemble the left portion of Figure 4. Once all the polygons had been digitised, each polygon assigned a land use and cover code (noting the land use categories had been derived from the initial land cover interpretation and LUCID Working Paper 16 9 subsequently from secondary data sources), and the interpretation corrections completed; the resultant shape file was then built and cleaned. The build command creates or updates the attribute tables, whereas the clean command generates coverages with correct polygon topology. The clean command also edits and corrects geometric coordinate errors, assembles arcs into polygons and creates feature attributes for each polygon. This was accomplished by converting the shape file to an Arc/Info coverage using the SHAPEARC command in Arc/Info (if Arc/Info is not available, the X-Tools ArcView extension may be downloaded from the Arc Scripts menu on the ESRI website at http://www.esri.com). Figure 4. Island polygon overlaying another polygon and after intersection, two separate polygons created Forest Mixed Bush/Cultivation E. Area C.3.e. Calculation The X-Tools extension in ArcView calculates the area of each polygon in square meters, acres or hectares. To match the identification code with the text description, double click the legend and add the appropriate labels. Save the land use and land cover legend files under separate names in the working directory to prevent confusion between the land uses and cover codes. Tables and maps created with these files then include the text descriptors. The calculation of area (in meters squared, acres and hectares) and perimeter (in meters) is conducted in the table properties and the calculate menu is selected. Tables of area and perimeter are then constructed automatically. Table 2 shows the area and perimeters of each land use and land cover code that have been generated from this interpretation. D. CONCLUSIONS AND RECOMMENDATIONS The land use and land cover mapping procedure described above effectively represented the heterogeneous mix of human and natural landscapes of the Mt. Kenya area. The refinement of the land cover classification consisted mostly of adding classes that are important economically (e.g., irrigated agriculture) or for plant biodiversity and land degradation (e.g., degraded versus non-degraded forest). The refinement of the land use classification scheme was based on knowledge of the drivers of land use change in the area derived from previous fieldwork and data analysis. In the land use classification, the major addition was differentiating between land management and ownership types. This information will be critical in the process of projecting how the land use may change in the future. For example, large-scale farmers and agricultural institutions are much more likely to keep their land under pasture, or invest in irrigation technology, than small-scale farmers. Small-scale farmers may respond more quickly than parastatals to changes in the market, for example by switching from tea to maize when prices change. Similarly, the Kenya Wildlife Service (KWS) and the Forest Department (FD) have different policies regarding the harvesting of trees in forests. LUCID Working Paper 16 10 This type of land use and cover interpretation and analysis requires information about the area that can be only obtained from a variety of supplemental sources including maps, literature, interviewing people and ground truthing. Interpretation based only on the image’s spectral characteristics is fraught with limitations and the resultant errors would be compounded during a change analysis. Automated classification based on the spectral characteristics was, in this image, not helpful due to the heterogeneous landscape and very small land management units. The LUCID team recommends that for future land use and land cover analysis of such heterogeneous landscapes: A Mix of automatic classification and visual interpretation o Automatic is useful for delimiting homogeneous vegetation zones, such as forests and tree or agricultural plantations. o Visual interpretation is useful to reduce interpretation errors in heterogeneous natural landscapes as well as in complex human-managed landscapes. In human managed landscapes, supplemental information is often required to differentiate, for example, between extensive areas of grain crops versus natural savannah, or to correctly identify zones of intense agriculture. In our area, for example, much of the landscape was covered by tiny fields under a variety of crops (perennial crops such as tea and coffee interspersed with seasonal maize and horticultural crops, with fields separated by planted trees). Adopting a dual land use and land cover classification scheme, to provide critical information on land use drivers and constraints in projecting future changes in use, and to provide information on the biophysical characteristics of the landscape for a variety of environmental analyses. Fuzzy boundaries exist between some of the largest and most important land use/cover classes in tropical agricultural settings, but it is nevertheless important to attempt their rough delimitation. The transition from tea/coffee to cropping/maize dominant to mixed maize/ bush, for example, is gradual and not necessarily visible on imagery. The biophysical and socio-economic differences, however, are significant and important to recognize. Identifying changes between in their spatial extent using imagery, however, may not be possible. LUCID Working Paper 16 11 Table 2: Calculated area and perimeter for each of the land use and land cover categories LAND USE Land Use Code 1110 1120 1130 1140 1300 2100 2200 2300 2400 2500 3100 3200 3300 4100 4200 5000 6000 7100 7200 8100 8200 Land Cover Code 1000 2100 2200 2300 2400 2500 2600 3000 4110 4120 4130 4140 4150 4210 5000 6100 6200 7000 Count 3 5 303 2 41 1 11 6 1 7 9 6 59 1 1 8 6 4 1 21 5 Count 1 1 15 30 9 6 35 35 3 5 313 6 2 11 6 4 1 18 Area 634,837,037.25 1,337,136,352.55 5,236,709,395.65 868,175,221.14 316,942,599.31 18,825,909.06 110,116,218.05 19,233,075.62 63,168,376.27 161,425,273.62 1,079,733,572.68 486,881,765.78 1,137,655,210.34 7,646,988.50 66,618,737.93 51,393,392.80 3,796,692.52 29,624,850.02 418,595.76 32,651,974.09 7,456,017.06 Perimeter Acres Hectares 269,630.10 375,565.50 1,674,895.80 356,518.33 675,841.16 29,288.54 116,727.57 55,240.79 43,979.53 106,145.51 241,700.36 269,872.65 832,009.54 15,718.63 74,201.41 111,540.50 18,831.15 101,257.87 2,455.74 179,893.92 31,228.70 156,871.02 330,412.27 1,294,013.90 214,529.91 78,317.91 4,651.97 27,210.20 4,752.58 15,609.18 39,888.89 266,806.91 120,310.62 281,119.60 1,889.60 16,461.78 12,699.53 938.18 7,320.43 103.44 8,068.45 1,842.42 63,483.70 133,713.64 523,670.94 86,817.52 31,694.26 1,882.59 11,011.62 1,923.31 6,316.84 16,142.53 107,973.36 48,688.18 113,765.52 764.70 6,661.87 5,139.34 379.67 2,962.49 41.86 3,265.20 745.60 LAND COVER Area Perimeter 720,922,999.34 149,074.84 445,076,764.09 229,313.51 807,686,784.55 524,897.35 117,681,520.30 272,327.48 51,538,752.74 113,374.11 19,233,075.62 55,240.79 238,246,065.85 278,933.42 628,878,990.84 773,095.77 634,837,037.25 269,630.10 1,337,136,352.55 375,565.50 5,256,982,095.66 1,721,025.30 161,279,913.68 107,979.12 868,175,221.14 356,518.33 110,116,218.05 116,727.57 3,796,692.52 18,831.15 29,624,850.02 101,257.87 418,595.76 2,455.74 238,815,871.04 222,968.45 Acres 178,143.24 109,980.42 199,582.95 29,079.62 12,735.45 4,752.58 58,871.65 155,398.76 156,871.02 330,412.27 1,299,023.37 39,852.98 214,529.91 27,210.20 938.18 7,320.43 103.44 59,012.45 Hectares 72,092.30 44,507.68 80,768.68 11,768.15 5,153.88 1,923.31 23,824.61 62,887.90 63,483.70 133,713.64 525,698.21 16,127.99 86,817.52 11,011.62 379.67 2,962.49 41.86 23,881.59 LUCID Working Paper 16 12 Figure 5. LUCID Working Paper 16 13 Figure 6. LUCID Working Paper 16 14 Figure 7. Land cover draped over a 250m resolution DEM with location of study transect LUCID Working Paper 16 15 E. REFERENCES ESRI (Environmental Systems Research Institute). 1999. ArcView GIS version 3.2. Redlands, CA, USA. ERDAS Inc. 2001. Erdas Imagine 8.5 Software - Atlanta, GA, USA. Gathaara, Gideon N.. 1999. Aerial Survey of the Destruction of Mt. Kenya, Imenti and Ngare Ndare Forest Reserves. Kenya Wildlife Service (KWS): Nairobi, Kenya. Geist, H. and Lambin, E. 2002. Proximate Causes and Underlying Driving Forces of Tropical Deforestation. BioScience 52(2): 143-150. Jaetzoldt, R. and Schmidt, H. 1983. Farm Management Handbook of Kenya: Part A, Eastern Kenya. Ministry of Agriculture, Government of Kenya: Nairobi, Kenya. IGBP (International Geosphere-Biosphere Project). 1997. LUCC Report Series No. 3. 1997. LUCC Data Requirements Workshop: Survey of Needs, Gaps and Priorities on Data for Land Use/Land Cover Change Research. Organized by IGBP/IHDP-LUCC and IGBP-DIS. Barcelona Spain, 11-14 September 1997. ILRI (International Livestock Research Institute). 2002. ILRI GIS Database. Available from: http://www.ilri.cgiar.org/gis/. Latham, J. 2001. AFRICOVER East Africa. LUCC Newsletter (7):15-16 Olson, Jennifer M. 1998. A Conceptual Framework of Land Use Change in the East African Highlands. Paper read at Earth's Changing Land: Joint Global Change and Terrestrial Ecosystems and Land Use and Land Cover Change Open Science Conference on Global Change, at Barcelona, Spain. Survey of Kenya (SoK). 1962-1997. 1:50,000 Scale Topographic Map Sheet Index Chuka, Embu, Embu North, Gatunga, Irereni, Ishiara, Ithanga, Karatina, Kiambere, Kimangau, Kindaruma, Makuyu, Marania, Masinga, Maua, Meru, Mitunguu, Mt. Kenya, Muranga, Mwingi, Nanyuki, Naro Moru, Nkubu, Siakago, and Tseikuru. Ministry of Lands and Settlement, Government of Kenya. Government Printer: Nairobi, Kenya. LUCID Working Paper 16 16 APPENDICES Appendix 1. Definitions of Land Use, Land Cover and Land Use/ Cover Change Below are the land use and land cover definitions adopted by LUCC-IGBP-IHDP (quoted from LUCC Report Series No. 3, 1997: 19-20). Land cover refers to the physical characteristic of earth’s surface, captured in the distribution of vegetation, water, desert, ice, and other physical features of the land, including those created solely by human activities such as mine exposures and settlement. Land use is the intended employment of and management strategy placed on land cover type by human agents, or land managers. Forest, a land cover, may be used for selective logging, for resource harvesting, such as rubber tapping, or for recreation and tourism. Shifts in intent and/or management constitute land-use changes. Land-cover and land-use changes may be grouped into two broad categories: conversion or modification. Conversion refers to changes from one cover or use type to another. For instance, the conversion of forests to pasture is an important land-use/land-cover conversion in the tropics, while abandonment of once permanently cultivated land and the regeneration of forests is taking place in parts of the mid-latitudes. In contrast, modification involves maintenance of the broad cover or use type in the face of changes in its attributes. Thus a forest may be retained while significant alterations take place in its structure or function (e.g., involving biomass, productivity, or phenology). Likewise, slash-and-burn agriculture, a use, may under-go significant changes in the frequency of cropping, and use capital and labor inputs while retaining the rotation, cutting, and burning that constitute such uses. Land-cover conversion operates through many pathways, the constellations of which form specific processes. For instance, deforestation leads to many types of land cover, but one common conversion process entails cutting, burning, and even planting of grass to create a pasture. In turn, site abandonment may lead in succession to a secondary forest. These pathways and processes, such as deforestation, desertification, wetland drainage, or agricultural intensification mediate the conversion or modification of land cover. Thus they can be envisioned as forcing functions, which have direction (forest to pasture or pasture to forest), magnitude (amount of change), and pace (rates of change). In turn, these pathways are typically triggered by changes in the use of the land, specific operating strategies (e.g., labor, capital, crops), which are linked to changes in the purpose of land management (e.g., for subsistence, market, occupation, or recreation). Thus changes in the controlling land agents or the context in which they operate affect land use and, ultimately, land cover. It is important to recognize that many land-use/land-cover change pathways exist and are differentiated globally and over time. The study of LUCC focuses much of its effort and emphasis on understanding the specific conditions and controls - both biophysical and social which determine these pathways. LUCID Working Paper 16 17 Appendix 2. Waypoint Sheet for Ground Truthing Collector: __________ Map Catalogue #: _______ Sheet Name: ____ Date: __/__/02 Northing:_________ Easting: _______ EPE:_____ Photographic Roll #: ______ Frame #: ______ Slope: ___________ Photographic Description: __________________________________ Cover Type at Present Position:______________________________ _______________________________________________________ Dominant Plant Species at Present Position:____________________ _______________________________________________________ Status of Vegetation Degradation:____________________________ _______________________________________________________ Cover Type Looking East: __________________________________ _______________________________________________________ Cover Type Looking West: _________________________________ _______________________________________________________ Cover Type Looking North: _________________________________ _______________________________________________________ Cover Type Looking South: _________________________________ _______________________________________________________ Additional Notes: _________________________________________ _______________________________________________________ _______________________________________________________ LUCID Working Paper 16 18