int. j. remote sensing, 2000, vol. 21, no. 6 & 7, 1139–1157 Satellite estimation of tropical secondary forest above-ground biomass: data from Brazil and Bolivia M. K. STEININGER Department of Geography, University of Maryland, College Park, Maryland 20742, USA Abstract. This paper reports on a test of the ability to estimate above-ground biomass of tropical secondary forest from canopy spectral reflectance using satellite optical data. Landsat Thematic Mapper data were acquired concurrent with field surveys conducted in secondary forest fallows near Manaus, Brazil and Santa Cruz de la Sierra, Bolivia. Measurements of age and above-ground live biomass were made in 34 regrowth stands. Satellite data were converted to surface reflectances and compared with regrowth stand age, biomass and structural variables. Among the Brazilian stands, significant relationships were observed between middle-infrared reflectance and stand age, height, volume and biomass. The canopy reflectance-biomassrelationshipsaturatedat around 15.0kg m2, or over 15 years of age (r> 0.80, p< 0.01). In the Bolivian study area, no significant relationship between canopy spectral reflectance and biomass was observed. These contrasting results are probably caused by a low Sun angle during the satellite measurements from Bolivia. However, regrowth structural and general compositional differences between the two study areas could explain the lack of a significant relationship in Bolivia. The results demonstrate a current potential for biomass estimation of secondary forests with satellite optical data in some, but not all, tropical regions. A discussion of the potential for regional extrapolation of spectral relationships and future satellite imagery is included. 1. Introduction The proliferation of human settlements in the Amazon basin results in the conversion of large areas of mature tropical forest to landscapes primarily consisting of pasture, agriculture and secondary forest. Most land use in the Amazon is characterized by high rates of land abandonment and re-clearance, yielding mosaics of land under cultivation and secondary vegetation of different stages of regrowth (Uhl et al. 1988, Turner et al. 1993, Moran et al. 1994). Current estimates of the annual carbon release to the atmosphere by tropical land use are around 1.6 gigatons (Schimel et al. 1996, Schimel 1998). Fearnside (1996) estimated that in 1990, 0.27 gigatons of carbon were released to the atmosphere from the burning of forest and cerrado in the legal Brazilian Amazon alone. An unknown amount of carbon is re-sequestered in secondary forests growing in fallowed or abandoned agricultural lands. Debate continues over whether the many abandoned lands in the Amazon are actually supporting rapid secondary forest regrowth (Lugo 1988), or whether most do not have a chance to revert to forest (Myers 1991). There are also uncertainties about fallow periods and rates of biomass accumulation during regrowth. Because of these uncertainties, we are unable International Journal of Remote Sensing ISSN 0143-1161 print/ISSN 1366-5901 online © 2000 Taylor & Francis Ltd http://www.tandf.co.uk/journals/tf/01431161 .html 1140 M. K. Steininger to determine the net impact of land use on carbon sequestration in the Amazon. While current remote sensing efforts such as the NASA Landsat Pathfinder Tropical Deforestation Project (Lawrence and Chomentowski 1992, Lambin 1994) seek to provide precise estimates of tropical deforestation, they will not provide information on the subsequent patterns of land cover and utilization. Analysis of optical satellite data, such as those from the Thematic Mapper (TM) sensor on the Landsat platform, permits accurate estimation of deforestation areas and rates. Most local studies of deforestation report accuracy levels of over 95% (Roy et al. 1991, Sader et al. 1991, Steininger 1996), and several global deforestation monitoring programs have been implemented during the past decade (Lawrence and Chomentowski 1992, Lambin 1994). Progress has also been made in estimating the area of secondary forests with high resolution optical data. The estimation of regrowth biomass over large areas with satellite imagery would, however, enable many additional questions about the ecological functioning of natural and human-modified landscapes to be addressed, including the net carbon exchange associated with tropical agriculture. Three independent studies using TM observations have reported similar changes in canopy spectra with age over the first 15 years of forest regrowth in the Brazilian Amazon (Foody and Curran 1994, Moran et al. 1994, Steininger 1996). Typically, visible (TM channel 3, 0.63-0.69 mm) reflectance is low for all ages of regrowth while near-infrared (TM channel 4, 0.76-0.90 mm) reflectance increases for four to eight years and then decreases for the following five to ten years of forest regeneration. Reflectances in both middle-infrared TM channels (channel 5, 1.55-1.75mm; channel 7, 2.08-2.35 mm) gradually decrease throughout the first 15 years of regrowth. However, the most reliable estimates of regrowth age are made when images from many years are available (Kimes et al. 1998). These common trends in regrowth canopy reflectance with age have been attributed to the changes in canopy leaf area and geometry as secondary forests regrow. Field studies of secondary forests in the Amazon and elsewhere show that, for recovery from light to moderate use, leaf area rapidly increases, with light capture often over 80% within five years of regrowth (Jordan 1989, Saldarriaga and Luxmoore 1991). Thus visible reflectance is low for most secondary forests over a few years old. Increases in leaf area during early (< 5 years old) regrowth cause an increase in near-infrared reflectance via multiplicative scattering. Middle-infrared radiation is not absorbed by plant pigments and, due to lower leaf transmissivity, is less subject to multiplicative scattering than is near-infrared radiation. The gradual decrease in middle-infrared reflectance throughout the first 14 years of regrowth and the decrease in near-infrared reflectance in older (> 8 years old) regrowth has been attributed to increased canopy shading caused by an increasingly complex canopy geometry. Applications of spectral mixture modelling, where multispectral images are decomposed into estimates of proportions of leaf cover, background cover and shade, have been applied to demonstrate changes in canopy shade among tropical forest types (Shimabukuro and Smith 1991, Adams et al. 1995). This has also been applied to Advanced Very High Resolution Radiometer (AVHRR) imagery, where similar patterns of reflectances among Amazonian forest types have been observed in its middle-infrared channel 3 (Shimabukuro et al. 1994, Lucas et al. 1999). Leaf ageing has also been shown to reduce near-infrared reflectance for both crops and several Amazonian tree species (Gausman et al. 1973, Lin and Ehleringer Global and regional land cover characterization 1141 1982, Rock et al. 1994, Roberts et al. 1998). Roberts et al. (1998) have reported lowered near-infrared reflectance for leaves from Amazonian caatinga trees which have epiphyll growth or have been damaged. Leaf reflectances in the near-infrared of both Protium heptaphyllum and Pradosia schomburgkiana with heavy epiphyll growth were nearly half those with little to moderate epiphyll growth. Roberts et al. argue that such differences in leaf spectral properties account for spectral differences among different types of mature forest. Such effects on leaf spectra could also play a role in the differences in canopy reflectances among early regrowth dominated by short-lived shrubs and trees and older regrowth with forest species with significant leaf necrosis and epiphyll growth. This poses a challenge to the assumption of non-variant spectral end members in the mixture modelling approach to image decomposition. A brightness-based distinction of regrowth appears to be possible only for stages of regrowth when the canopy changes drastically, namely the first 14 years of regrowth during which the canopy progresses from low shrubs to a tall multi-layered forest. Yet within these stages, it is logical that if infrared canopy reflectance is related to regrowth stand parameters such as leaf area, canopy height and complexity, as well as leaf parameters such as leaf age, necrosis, epiphyll growth, then canopy reflectance should be related to stand age. Since older and taller regrowth is usually of higher biomass, then infrared canopy reflectance should also be related to stand biomass. While one study has reported a poor relationship between near-infrared reflectance and stand biomass for a series of Puerto Rican regrowth (Sader et al. 1989), there are no comparisons of regrowth biomass with respect to middle-infrared reflectance. The research reported in this paper is a test of the potential for estimation of tropical secondary forest above-ground biomass with Landsat TM images using data from a series of stands near Manaus, Brazil and Santa Cruz de la Sierra, Bolivia. Three study areas included samples of secondary forests regrowing in abandoned pasture and agricultural fields, mostly manioc in Brazil and rice and corn in Bolivia. The two study areas in Manaus have an average annual temperature of 25.6C, precipitation of 2400 mm yr 1 and a dry season from July to September (Sioli 1984, Leemans and Cramer 1991). The soils at AM10, the northern study area along the AM10 and BR174 highways ( 6 0 W 2 5 0 N), are mostly clayey Oxisols (EMBRAPA-CPAC 1981, Dias and Nortcliff 1985). The second Brazilian study area is Lago Janauaca (60 15 W, 330 S), 40 km south of Manaus, where a large caboclo community practices rotational agriculture on soils classified as Plinthic Dystric Podsols (EMBRAPA-CPAC 1981). The Bolivian study area lies in the Yapacani and Surutu basins in western Santa Cruz, along the new Santa Cruz-Cochabamba highway. The area, centred at 6330 W 1 7 3 0 S, receives around 1600mm of rain and has an average annual temperature of 2 3 C (Roche and Rocha 1985). The winter is dry, with seven months with less than 100 mm of rain, and experiences strong, cool winds originating in the South Atlantic and the Pacific and lasting three to five days (Ronchail 1986). The soils at Surutu can be grouped into dissected terraces and alluvial plains. The soils of the terraces are mostly Quartzipsammentic Haplothorps on dissected terraces to the north and Typic Paleudults on the piedmont to the southwest (Montenegro Hurtado 1987). 2. Methods 2.1. Field surveys. 20 secondary forest stands in Brazil were visited from September to November 1995, and 14 stands in Bolivia were visited from May to June 1996. The secondary 1142 M. K. Steininger forest stands were all cases of regrowth in abandoned small (< 10 ha) agricultural fields and pastures (table 1). Most of the stands had bordering mature or secondary forest, and all stands were dominated by woody vegetation. Plots ranging from 0.07 to 0.10 ha in size were surveyed in each stand, and structural measurements of all trees with a diameter at breast height (dbh, 1.3 m above ground) greater than 5 cm were made. Volumes were estimated using a tree form factor of 0.62 (Brown and Lugo 1990, Brown et al. 1995). Canopy height was estimated as the mean plus two standard deviations of the heights of all trees in the stand. This estimate approximates the height of the taller canopy trees but not unusually tall trees in the stand. All structural and biomass data were calculated and compared for pioneer and climax genera. Following Uhl (1987), Swaine and Whitmore (1988) and Faber-Langendoen and Gentry (1991), the terms ‘pioneer’ and ‘climax’ are used here to refer to trees of different life history strategies. The term pioneer refers to trees which rapidly colonize disturbed areas and rapidly senesce when shaded by over-storey trees. Based, in part, on Uhl (1987), Saldarriaga et al. (1988) and Faber-Langendoen and Gentry (1991), genera recorded as pioneer in the Brazilian stands for this study are Bellucia and Miconia (MELASTOMATACEAE), Vismia (GUTTIFERAE), Cecropia (MORACEAE), and Isertia (RUBIACEAE). Genera recorded as pioneer in the Bolivian stands are Miconia (MELASTOMATACEAE), Vernonia and Senecio (COMPOSITAE), Piper (PIPERACEAE), Psidium (MYRTACEAE), Guadua (GRAMINEAE), Psychotria (RUBIACEAE), Cecropia and Pourouma (MORACEAE), Pseudobombax, and Ochroma (BOMBACACEAE), and Heliocarpus (TILIACEAE). All other trees recorded are considered climax, and thus this is a broad group which includes many late successional taxa. A series of allometric equations derived from destructive samples of other neotropical secondary forests were applied to estimate the biomass of each measured tree (table 2). The equations were chosen because they are based on similar genera to those in the surveyed stands and because they enabled a specification of wood density. The allometric equations used were from the works of Uhl (1987) for trees shorter than 16m, Scatena and Silver (1993) for softwooded pioneer trees taller than 16 m, and Saldarriaga et al. (1988) for hardwoods taller than 16 m. The latter group was further divided into five wood density classes, ranging in specific density from 0.42 to 0 . 8 5 g m m 3 (dry weight/green volume). The majority of sampled trees was identified to genus, thus allowing for a general grouping based on wood density and tree form. Taxonomic identification was to either genus or family, referring to Gentry (1993) and Killeen et al. (1993). Tree wood density values for common genera were obtained from Hidayat and Simpson (1994), Brown et al. (1995) and Finegan (1996). Above-ground standing live biomass density (hereafter referred to as stand biomass) was calculated as the sum of tree biomass estimates divided by plot area. Sampling errors for stand biomass were estimated from two stands in Bolivia, one five-year old and one 15 year-old, following the approach used by Brown et al. (1995). In these stands, 0.28 and 0.30ha plots were surveyed, marking each 0.01 ha sub-plot. Based on the running means and the coeffcient of variation of the standard errors of the sub-plot biomass estimates, sampling errors for both stands were estimated to be 9% for plot sizes of 0.70 ha and larger. This estimate was taken as the error for all sampled stands. Further discussion of stand structure and composition and field estimation of stand biomass is provided in Steininger (1998). 2.2. Image geo-registration Prior to conducting the field surveys, TM images from 1991 for Manaus and 1995 for Santa Cruz were obtained from the NASA Landsat Pathfinder Humid Global and regional land cover characterization 1143 Table 1. Stands surveyed in all study areas. S.T. =short-term agriculture, L.P.= long-term pasture, M.F.=medium-fallow agriculture, S.F.=short-fallow agriculture (one case in Brazil), and L.T.= long-term agriculture (one case in Brazil). S/B indicates clearing by the slash-and-burn method. Stand Stand Stand Sampled Land Id. age (y) Area (ha) area (ha) use 4 0.09 M.F. Land use history, notes Bolivia, Piedmont S3 10 S4 15 S12 20 S13 4 S14 10 S15 12 Bolivia, Alluvial S5 25 4 2 2 4 5 0.08 0.10 0.10 0.10 0.10 S.T. S.T. S.T. S.T. S.T. 3 0.10 M.F. S6 15 3 0.80 M.F. S7 5 4 0.10 L.P. S8 6 4 0.10 M.F. S10 8 3 0.10 M.F. S16 8 S17 15 S18 5 Brazil, AM10 5 4 4 0.10 0.28 0.30 S.T. S.T. S.T. A2 A3 A5 8 4 2 0.07 0.07 0.08 S.T. L.T. S.T. A6 20 A7 23 A8 20 A9 15 A10 12 A11 12 A12 7 Brazil, Janauaca 6 5 4 4 2.5 3 1.5 0.07 0.09 0.10 0.07 0.10 0.08 0.08 S.T. S.T. S.T. L.P. S.T. S.T. L.P. J1 J2 J3 J4 30 26 15 8 6 4 3 2 0.10 0.10 0.10 0.07 L.P. S.T. S.T. S.F. J5 J6 12 15 4 3 0.08 0.09 S.T. L.P. J7 20 8 0.07 L.P. J9 10 2 0.10 L.P. J10 5 1.5 0.08 L.P. J11 5 2 0.08 S.T. 10 4 12 Forest, 4+ cycles of S/B, 1 year rice, 7 year fallow Forest, S/B, 1 year rice Forest, S/B, 1 year rice Forest, S/B, 1 year rice Forest, S/B, 1 year rice Forest, S/B, 1 year rice Forest, 4+ cycles of S/B, 1 year rice, 5 year fallow Forest, 4+ cycles of S/B, 1 year rice, 5 year fallow Forest, 3 cycles of S/B, 1 year rice, 5 year fallow, 7 year pasture Forest, 4+ cycles of S/B, 1 year rice, 5 year fallow Forest, 3 cycles of S/B, 1 year rice, 8 year fallow Forest, S/B, 1 year manioc Forest, S/B, 1 year manioc Forest, S/B, 1 year rice Forest, S/B, 1 year rice Forest, S/B, 6 year rice Forest, S/B, seeded w/rice only+abandoned, sandy hilltop Forest, S/B, 1 year rice Forest, S/B, 1 year rice Forest, S/B, 1 year pasture Forest, S/B, 7 year pasture Forest, S/no-burn, no cultivation Forest, S/B, 1 year pasture Forest, S/B, 6 year pasture Forest, S/B, 10 years pasture Forest, S/B, 1 year manioc Forest, S/B, 1 year manioc Forest, 4 cycles of S/B, 1 year manioc, 3 year fallow Forest, S/B, 1 year manioc Forest, S/B, 1 year manioc, 5 year fallow, S/B, 5 year pasture Forest, S/B, 1 year manioc, 3 year fallow, S/B, 10 year pasture Forest, S/B, 1 year manioc, 3 year fallow, S/B, 5 year pasture Forest, S/B, 1 year manioc, 3 year fallow, S/B, 6 year pasture Forest, S/B, 1 year manioc 1144 M. K. Steininger Global and regional land cover characterization 1145 Tropical Deforestation Project (Lawrence and Chomentowski 1992). These images were geographically registered using GPS measurements collected during previous site visits. The corrected positional errors of the imagery were estimated to be within 30 m, based on the GPS points used in registration. During the field surveys, additional GPS measurements were collected along roads, clearance edges and canopy gaps within stands. 2.3. Image atmospheric correction For this study, the 6S atmospheric correction program (Vermote et al. 1997) was used. This is a multiple-stream simulation of atmospheric absorption and scattering based on user-defined inputs of optical depth and aerosol composition. Estimates of optical depth over dark, closed canopy forested sites were derived from a program developed at the University of Maryland Institute of Advanced Computer Studies (UMIACS) (table 3). This program employs the approach of Kaufman and Remer (1994) by taking differences in top-of-atmosphere (TOA) reflectances in the blue and middle-infrared channels for forest sites to estimate optical depth in a given satellite image (Fallah-Adl 1996). Aerosol composition was defined as 80% water-soluble, 15% soot, 5% dust-like and 0% oceanic (Holben 1996 personal communication). A comparison of estimated surface reflectances for dark water and closed-canopy primary forest, surface features that can be assumed to have constant reflectance, in images from different years is shown in figure 1. The 6S corrected spectra produced between-year differences of one to two per cent reflectance in all channels except channel 7. 2.4. Analysis of canopy spectra Surveyed stands were identified in the geo-referenced imagery. Means of the corrected stand reflectances in each of the TM channels were calculated and plotted against stand above-ground biomass and structural variables. Linear regressions of each channel reflectance against biomass were performed using a general linear model (Neter et al. 1990). An F test was used to test for an effect of study area on the regression. Finally, step-wise regression was used to compare biomass estimation using single versus multiple band reflectances. Based on the correlation matrix, stand canopy reflectances were plotted against the other stand structural characteristics to assist interpretation of the biomass-canopy reflectance relationship. Table 3. Summary of Landsat 5 TM images and aerosol optical depth estimates. Optical depth was estimated with the University of Maryland Instituteof Advanced Computer Studies correction program (UMIACS O.D.). Study area Scene date AM10, Janauaca 15 August 1988 Surutu 8 August 1991 20 September 1995 25 July 1986 9 July 1992 5 August 1996 Path/ Solar zenith row () 0.49mm 0.66mm 231/62 43 0.16–0.28 < 0.01–0.19 231/72 44 38 68 69 66 0.24–0.33 0.33–0.52 0.12–0.21 0.16–0.24 0.38–0.46 0.15–0.32 0.32–0.51 0.01–0.19 0.16–0.23 0.36–0.45 UMIACS O.D. 1146 Figure 1. M. K. Steininger Means of Thematic Mapper digital numbers (DN) and atmospherically-corrected surface reflectances for dark targets. 3. Results 3.1. Patterns of biomass and canopy reflectance in Brazil Most of the early regrowth stands surveyed in Brazil were dominated by a mix of Vismia, Miconia and Bellucia (Steininger 1998). None of the stands surveyed in this study were dominated by Cecropia, in contrast to the early regrowth stands surveyed by Foody and Curran (1994). The most common taxa in the Brazilian stands older than 10 years of age were Inga (MIMOSOIDEAE), Goupia (CELASTRACEAE), Didymoponax (ARALIACEAE), and genera of the ANNONACEAE, LAURACEAE and BURSERACEAE. The range of total aboveground live biomass among the stands in Brazil was from 0.2 to 20kgm 2 . The highest correlations among stand variables were those among stand age, basal area, volume and canopy height (table 4). These variables were all inversely correlated to canopy reflectance in all three infrared bands. The highest correlations between canopy reflectances and structural variables were between middle-infrared reflectance and stand basal area (r2 = 0.715) and biomass (r2 = 0.701). Reflectances in the visible channels rapidly decreased from over 0.06 in pastures Global and regional land cover characterization 1147 1148 Figure 2. M. K. Steininger Means of estimated canopy reflectances in Thematic Mapper channels 3, 4, 5 and 7 versus above-ground biomass for all regrowth stands surveyed in Brazil. and agricultural fields to below 0.03 in low to high biomass regrowth stands (figure 2(a)). Reflectances in the near-infrared were around 0.30 for young, low biomass stands and decreased thereafter to around 0.27 for the older, high biomass stands (figure 2(b)). Reflectances in both middle-infrared channels decreased linearly with both stand age and biomass (figure 2(c),(d), table 4). The best spectral estimator of biomass among the Brazilian stands was an exponential relationship with channel 5 reflectance for all stands and mature forest (r = 0.714, p< 0.05, n=18). For all stands less than 1 5 k g m 2 , a linear relationship best predicted stand biomass from channel 5 reflectance (r = 0.854, p< 0.01, n = 12; figure 3, table 5). Step-wise regression only slightly increased this relationship by incorporating channels 4 and 3 in the regression model. Because of this, the linear model based on channel 5 alone was used to estimated regrowth biomass in the two Brazilian study areas (figures 4 and 5). The estimated average biomass of secondary forests at Janauaca is 7.2kgm2 and at AM10 is 7 . 6 kg m 2 . Observation of reflectance changes over a 4- and 7-year time period showed a decrease in middle-infrared reflectance for almost all of the stands surveyed, although the magnitude of this decrease was only from near zero to 0.045 reflectance over 7 years (figure 6). Global and regional land cover characterization 1149 Figure 3. Above-ground standing biomass of secondary forest regrowth versus Thematic Mapper channel 5 reflectance based on stands surveyed in Brazil. The linear regression model is based on stands up to 15 kgm2. Table 5. Linear and exponential regressions of regrowth above-ground standing biomass versus Landsat TM channel 5 reflectance based on all the stands surveyed in Brazil. The linear model is based on stands up to 15.0k g m 2 . Equation Y = 4.166e+5e( 73.936X) Y=50.77 287.62X Slope n 18 13 p < < 0.01 0.01 % error r 0.10 0.08 0.714 0.854 Y= above-ground standing dry weight (kgm2). X = Landsat TM channel 5 atmospherically corrected reflectance. e = natural logarithm. 3.2. Patterns of canopy reflectance in Bolivia In Bolivia, stands were generally of lower stature and biomass. Stand canopy heights ranged from 8 m in an 8-year old to almost 30 m in a 15-year old stand. Stand total above-ground biomass in Bolivia ranged from 2.4 to 13.4kgm2. The youngest stands in Bolivia were dominated by a shrub layer of Vernonia, Senecio, Psychotria and Miconia (Steininger 1998a, b). Stands up to 10 years old were dominated by Ochroma, Cecropia, Heliocarpus and Inga. Cecropia was the most common genus among all the stands in Bolivia. Many of the older stands in Bolivia had an upper canopy composed of the same softwooded pioneer tree taxa which dominated the younger stands. These were mostly Cecropia spp., Pourouma spp., Heliocarpus spp., and Ochroma pyrimidale. In stands that were not dominated by these genera, Anadenanthera spp. and Swartzia spp. were most common in more open canopies over a dense shrub layer. These stands had relatively lower infrared reflectances among the sample of Bolivian regrowth than similarly aged stands dominated by the softwood trees. The correlation coeffcients among stand structural variables reported in table 4 reveal further differences between the regrowth in Brazil and Bolivia. In the Brazilian study sites, all height variables for both pioneer and climax trees were highly correlated with stand total biomass (r2>0.79), particularly tree height variables (r2>0.89). This strong relationship between stand height variables and biomass was not found in Bolivia. Among the stands surveyed at Surutu, the correlation between 1150 M. K. Steininger 60°05'W 00' **'- .a 59°55'W » t.. 59°50'W kgm"2 10 0 2° 35'S 1 fUTW"* ' * ....... , . . . . , . - ■ ' ""■■ i*f f; «?**■«&. f 0 km 3 8 2'40'S 1 ......................... i-H ^ ■-■■■■/, .- J ts*- 45' 50' 50' i ■-,* ... ~'K~-%r^ ■%£,■*>., -\'~ «.' ■ ■ ^^C^'FT' * r i i •*•!►■• ..-t, ' ■ i •*yi' ','. S0°O5'W G/L&T, HGS 84 (GPS) 00' 59°55*W 59°50'W Figure 4. Distribution of secondary regrowth above-ground standing biomass in 1995 at AM10 (Amazonas, Brazil). Cleared land is shown in white, mature forest is shown in grey, colours are estimates of regrowth above-ground biomass. This image was produced using the linear regression model reported in table 5. stand height variables and biomass ranged from 0.34 to 0.73. Furthermore, stand biomass was most highly correlated with the pioneer canopy height, revealing the greater importance of pioneer trees in the biomass of the Bolivian regrowth stands. The canopy spectral reflectances of the secondary forests in Bolivia were Global and regional land cover characterization 60D20'W 1 » Ui m. 60°10'W 15' w ISfe* I p ^SR-^fj -'v J •if-?:-" & ^ >•'-■ . :l iff ■>'-' '■'..•/,-;'.-:¥ ,K'.'". "\ ; -,.r % 0 ),Oi G/L& T, 60°20' W WGS 84 (GPS) , ''.•>'* >1M -:■.■* 4>j w 30' CO If". ■ 3|] <n ■'■■*■' w kg rrf2 10 1 km ■ *^g ;-* % : .> 0 6 0 ° 0 5 'W - ^fKr :•>. ? b 30' 1151 | W G m K*"* H hs 15' ... : . _w 60°10'W Figure 5. Distribution of secondary regrowth above-ground standing biomass in 1995 at Lago Janauaca (Amazonas, Brazil). Cleared land is shown in white, mature forest is shown in grey, colours are estimates of regrowth above-ground biomass. This image was produced using the linear regression model reported in table 5. Figure 6. Changes in regrowth canopy reflectance versus age in 1995 for all stands surveyed in Brazil. Open triangles are reflectance differences between 1995 and 1988, full triangles are reflectance differences between 1995 and 1991. 1152 Figure 7. M. K. Steininger Means of canopy reflectances in Thematic Mapper channels 3, 4, 5 and 7 for all regrowth stands surveyed in Bolivia. consistently higher than those in Brazil. Among stands of up to 7 years old in Bolivia, visible reflectance was from 0.04 to 0.06. Near-infrared reflectance was from 0.45 to 0.70 and middle-infrared reflectance was from 0.22 to 0.30 (figure 7). Moreover, the patterns of change in canopy spectra with age reported in the 1995 Brazilian data in this study, and in other studies, were not observed in the Bolivian data. Only near-infrared reflectance showed a slight decline with stand age and biomass density. 4. Discussion The patterns of canopy reflectance among the regrowth stands surveyed in Brazil agree with those previously reported for regrowth near Manaus and in eastern Para (Foody and Curran 1994, Moran et al. 1994, Steininger 1996). In this and earlier studies, most secondary forest stands of 15 years and older appeared spectrally similar to nearby mature forest. The optical band most sensitive to canopy changes during succession was the middle-infrared. Visible reflectance was low throughout the chronosequence. Nearinfrared reflectance was high for all stands, presumably because of high canopy leaf cover. In Landsat TM data, the 1.55-1.75/am band (channel 5), is more sensitive than the 2.1-2.3/am band (channel 7). Global and regional land cover characterization 1153 Stands surveyed in Brazil demonstrated a rapid progression from a canopy dominated by shrubs to an even tree canopy dominated by softwooded pioneers, to an uneven canopy dominated by later successional trees. The most common pioneer trees in the Brazilian stands are Cecropia spp., which have a planophile distribution of large leaves. These trees rapidly senesce once later successional species such as Miconia spp., Inga spp., Bellucia spp., and Goupia glabra reach the canopy. Spectral canopy models suggest that it is primarily the changes in the upper canopy structure, i.e. canopy height, upper-canopy leaf area density and angle distribution, during these stages of regeneration that cause the observed increases in canopy shading and decreases in middle-infrared reflectance. Since the scenes were all acquired from nearly the same time of year (table 3), Sun angle variation is an unlikely explanation of between year differences in the retrieved surface reflectances. Soil moisture is also an unlikely explanation since all images were from the late dry season and since canopy cover was high for most stands. Between-year differences in rainfall may have caused between-year differences in leaf area, which could vary near- and middle-infrared canopy reflectances among regrowth stands of similar age and biomass stocks. Figure 6 shows that the estimated changes in middle-infrared reflectance over 4 and 7 years are around 0.01 to 0.02, and the pattern of reflectance versus stand biomass in figure 2 agrees with this. Thus between-year errors of 0.01 to 0.02 reflectance in atmospheric correction remain a limit to accurate estimation of biomass changes using multi-date optical imagery. There are at least three possible explanations for the significant relationship between stand age, biomass and middle-infrared reflectance in the two study areas in Brazil and the lack of one in the Bolivian study area. The first is the low Sun angle during the 1996 satellite pass over Santa Cruz (table 3). A second hypothesis is that a relationship between canopy infrared reflectance and stand variables such as age, height and biomass is not valid for regrowth in areas of tropical deciduous forest, as in Santa Cruz, Bolivia. Measurements of leaf area were not collected in this study, although it is possible that canopy leaf senescence during the dry season reduces the effects of upper-canopy structure on surfaces reflectances. In Maraba, mature forest showed increases in nearinfrared reflectance during the dry season, becoming nearly as bright as forest regrowth and suggesting a leaf flush of under-storey vegetation in the dry season (Bohlman 1998). Another possible cause for the different spectral relationships to regrowth biomass in Brazil and Bolivia is a difference in the development of regrowth canopy structure between the two areas. Among the stands surveyed in Bolivia, the stage of succession dominated by pioneer trees persists through the first 20 years of regrowth, considerably longer than in Brazil. Other than Cecropia spp., these trees included Pourouma spp., Ochroma pyrimidale and Heliocarpus spp., all which typically have large leaves and Cecropia-like planophile leaf angle distributions (Uhl 1987, Finegan 1996). Canopy heights of the Bolivian regrowth were lower than those in Brazil. Furthermore, canopy height was strongly correlated to stand biomass in Brazil and poorly correlated to biomass in Bolivia (table 4). Thus, in Bolivia, while the stands increased in height and biomass with age, the structural changes of the upper canopy probably had a lesser effect on canopy shading and thus were poorly related to stand age and biomass. The compositional differences between the Brazilian and Bolivian regrowth suggest that differences in canopy structure could account for the contrasting patterns of infrared reflectance and its relationship to vegetation properties in the two study areas. 1154 M. K. Steininger The potential for monitoring changes with multi-date imagery will most certainly improve with data from the sensors onboard the soon to be launched Earth Observing System (EOS-1) and Landsat-7 satellites. Specifically, improved radiometric sensitivity and atmospheric correction based on simulation models and data internal to the images will allow more precise comparisons of multi-date imagery. The Moderate Resolution Imaging Spectroradiometer (MODIS), to be onboard EOS-1, has 36 channels in visible to thermal bands. Atmospheric properties and their effects on imagery can be precisely modelled using the thermal channels to estimate columnar water vapour and the visible and middle-infrared channels to estimate aerosol concentration (Justice et al. 1998). The Multi-Angle Imaging Spectroradiometer (MISR), also to be onboard EOS-1, is expected to improve on inputs to aerosol optical models by estimating particle size distribution and discriminating spherical from non-spherical particles (Diner et al. 1988). These corrections apply not only to MODIS and MISR imagery, but the estimated atmospheric water vapour and aerosols can also be used to correct higher resolution images from alternative sensors. Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), also on EOS-1, will provide 15 to 30 m resolution imagery in the visible and near-infrared bands (Yamaguchi et al. 1998). Landsat-7 will have the same spectral channel configuration as Landsat-5, although is planned to follow the same orbit as EOS-1 within a half-hour (Goward and Williams 1997). Thus, in most instances the correction of a Landsat-7 image based on atmospheric properties estimated from virtually synchronous image data from EOS-1 will be possible. Many research questions continue to drive us to produce better regional estimates of tropical secondary forest distribution andbiomass. Observations of high resolution optical imagery to date have demonstrated consistent patterns of change in canopy spectra with tropical regrowth age, with the exception of the Bolivian results in this study. The results of this study suggest that the potential for use of optical satellite data for above-ground biomass estimation of tropical secondary forests is similar to that for age estimation. With data from EOS-1, isolation of atmospheric effects from estimates of canopy reflectance will greatly increase the potential for extrapolations of spectral relationships over large areas or over multiple dates. This will require a better understanding of the possible variability of canopy structure for similarly aged patches of tropical regrowth. References Adams, J. B., Sabol, D. E., Kapos, V., Almeida Filho, R., Roberts, D. A., Smith, M. O,. and Gillespie, A. R., 1995, Classification of multispectral images based on fractions of end-members: application of land cover change in the Brazilian Amazon. Remote Sensing of Environment, 52, 137-154. Bohlman, S, 1988, Seasonal foliage changes in eastern Amazon detected from Landsat Thematic Mapper images. Biotropica, 30, 376-391. Brown, S, and Lugo, A. E, 1990, Tropical secondary forests. Journal of Tropical Ecology, 6, 1-32. Brown, I. F., Martinelli, L. A., Thomas, W. W., Moreira, M. Z., Ferreira, C. A. C, and Victoria, R. A., 1995, Uncertainty in the biomass of Amazonian forests: an example from Rondonia, Brazil. Forest Ecology and Management, 75, 175-189. Dias, C. C. P., and Nortcliff, S, 1985, Effects of two land clearing methods on the physical properties of an Oxisol in the Brazilian Amazon. Tropical Agriculture, 62, 207-212. Global and regional land cover characterization 1155 Diner, D. J. et al, 1998, Multi-angle Imaging Spectroradiometer (MISR) instrument description and experiment overview. IEEE Transactions on Geoscience and Remote Sensing, 36, 1072-1087. EMBRAPA-CPAC, 1985, Land systems map: physiognomy, climate, vegetation, topography, and soil of the central lowlands of tropical South America. Map at a scale of 1:5 000 000 (Bogota, Colombia: Centro Internacional de Agricultura Tropical/EMBRAPACPAC). Faber-Langendoen, D., and Gentry, A. H., 1991, The structure and diversity of rain forests at Bajo Calima, Choco region, western Colombia. Biotropica, 23, 2-11. Fallah-Adl, H., JaJa, J., Liang, S., and Townshend, J., 1996, Fast algorithms for removing atmospheric effects from satellite images. IEEE Computer Science and Engineering, 3, 66-77. Fearnside, P. M, 1996, Amazonian deforestation and global warming: carbon stocks in vegetation replacing Brazil’s Amazon forest. Forest Ecology and Management, 80, 21-34. Finegan, B., 1996, Patterns and process in neotropical secondary rain forests: the first 100 years of succession. Tree, 11, 119-124. Fo¨lster, H., de las Salas, G. et al, 1976, A tropical evergreen forest site with perched water table, Magdalena valley, Colombia: biomass and bioelement inventory of primary and secondary succession. Oecologia Plantarum, 11, 297-320. Foody, G, and Curran, P., 1994, Estimation of tropical forest extent and regeneration stage using remotely sensed data. Journal of Biogeography, 21, 223-237. Gausman, H. W., Allen, W. A., Cardenas, R., and Richardson, A. J., 1973, Reflectance discrimination of cotton and corn at four growth stages. Agronomy Journal, 65, 194-198. Gentry, A. H., 1993, A Field Guide to the Families and Genera of Woody Plants of Northwest South America (Colombia, Ecuador, Peru) with Supplementary Notes on Herbaceous Taxa (Washington, D.C: Conservation International). Goward, S. N, and Williams, D. L, 1997, Landsat and earth systems science: development of terrestrial monitoring. Photogrammetric Engineering and Remote Sensing, 63, 887-900. Hidayat, S., and Simpson, W. T, 1994, Use of green moisture content and basic specific gravity to group tropical woods for kiln drying (Madison, Wisconsin: U.S. Department of Agriculture). Jordan, C. F, 1989, An Amazonian Rainforest: The Structure and Functioning of a Nutrient Stressed Ecosystem and the Impact of Slash and Burn Agriculture (Paris: UNESCO). Justice, C. O. et al, 1998, The Moderate Resolution Imaging Spectroradiometer (MODIS): land remote sensing for global change research. IEEE Transactions on Geoscience and Remote Sensing, 36, 1228-1249. Kaufman, Y. J., and Remer, L. A., 1994, Detection of forests using mid-IR reflectance: an application for aerosol studies. IEEE Transactions on Geoscience and Remote Sensing, 32, 672-683. Killeen, T. J., Emilia Garcia, E., and Beck, S. G, 1993, Gu´a de Arboles de Bolivia (St. Louis: Missouri Botanical Garden). Kimes, D. S., Nelson, R. F., Skole, D. L., and Salas, W. A., 1998, Accuracies in mapping secondary tropical forest age from sequential satellite imagery. Remote Sensing of Environment, 65, 112-120. Lambin, E. F, 1994, Modelling Deforestation Processes (Luxembourg: ECSC-EC-EAEC). Lawrence, W. T, and Chomentowski, W, 1992, Data base project to use nearly 3000 satellite scenes. Earth Observation Magazine, December, 28-30. Leemans, R., and Cramer, W. P., 1991, The IIASA Database for Mean Monthly Values of Temperature, Precipitation and Cloudiness on a Global Terrestrial Grid (Laxenburg, Austria: International Institute for Applied Systems Analysis). Lin, Z. F, and Ehleringer, J., 1982, Changes in spectral properties of leaves as related to chlorophyll content and age of papaya. Phytosynthetica, 16, 520-525. Lucas, R. M., Honzak, M., Curran, P. J., Foody, G, Milne, R., Brown, T., and Amaral, S, 2000, Mapping the regional extent of tropical forest regeneration stages in the Brazilian legal Amazon using NOAA AVHRR data. International Journal of Remote Sensing, in press. 1156 M. K. Steininger Lugo, A. E, 1988, Ecosystem rehabilitation in the tropics. Environment, 30, 17-20. Montenegro Hurtado, R., 1987, Estimacion de las susceptibilidades a erosion de los suelos y los riesgos de degradacion erosiva bajo diferentes sistemas de cultivo en el area de Huaytu-Yapacani (Santa Cruz de la Sierra, Bolivia: Universidad Autonoma Gabriel Rene Moreno). Moran, E, Brondizio, E, and Mausel, P., 1994, Integrating Amazonian vegetation, land-use, and satellite data. Bioscience, 44, 1-18. Myers, N, 1991, Tropical forests: present status and future outlook. Climate Change, 19, 3-32. Neter, J., Wasserman, W, and Kutner, M, 1990, Applied Linear Statistical Models (Illinois: Irwin Harwood). Roberts, D. A., Nelson, B. W., Adams, J. B., and Palmer, F, 1998, Spectral changes with leaf aging in Amazon caatinga. Trees Structure and Function, 12, 315-325. Roche, M, and Rocha, N, 1985, Precipitaciones Annuales (La Paz, Bolivia: PHICAB-ORSTOM). Rock, B. N, Williams, D. L., Moss, D. M., Lauten, G. N, and Kim, M, 1994, High spectral resolution field and laboratory optical measurements of red spruce and eastern hemlock needles and branches. Remote Sensing of Environment, 47, 176-189. Ronchail, J., 1986, Variations climatiques hivernales a Santa Cruz de la Sierra, Amazonie Bolivienne (La Paz, Bolivia: Programa Climatologico e Hidrologico de la Cuenca Amazonica Boliviana (PHICAB)). Roy, P. S., Ranganath, B. K., Diwakar, P. G., Vohra, T. P. S., Bhan, S. K., Singh, I. J., and Pandian, V. C, 1991, Tropical forest type mapping and monitoring using remote sensing. International Journal of Remote Sensing, 12, 2205-2225. Sader, S. A., Powell, G. V. N, and Rappole, J. H., 1991, Migratory bird habitat monitoring through remote sensing. International Journal of Remote Sensing, 12, 363-372. Sader, S. A., Waide, R. B., Lawrence, W. T., and Joyce, A. T, 1989, Tropical forest biomass and successional age class relationships to a vegetation index derived from Landsat TM data. Remote Sensing of Environment, 28, 143-156. Saldarriaga, J. G., and Luxmoore, R. J., 1991, Solar energy conversion effciencies during succession of a tropical rain forest in Amazonia. Journal of Tropical Ecology, 7, 233-242. Saldarriaga, J. G., West, D. C, Uhl, C, and Tharp, M. L, 1988, Long-term chronosequence of forest succession in the upper Rio Negro of Colombia and Venezuela. Journal of Ecology, 76, 938-958. Scatena, F. N, and Silver, W, 1993, Biomass and nutrient content of the Bisley Experimental Watersheds, Luquillo Experimental Forest, Puerto Rico, before and after hurricane Hugo. Biotropica, 25, 15-27. Shimabukuro, Y. E., Holben, B. N, and Tucker, C. J., 1994, Fraction images derived from NOAA AVHRR data for studying the deforestation in the Brazilian Amazon. International Journal of Remote Sensing, 15, 517-520. Schimel, D., 1998, The carbon equation. Nature, 393, 208-209. Schimel, D., Alves, D., Enting, I., Heimann, M., Joos, F., Raynaud, D., and Wigley, T, 1995, Radiative Forcing of Climate Change. In Climate Change 1994: The Science of Climate Change (IPCC 1995 Vol. 1), edited by J. T. Houghton et al. (Cambridge: Cambridge University Press), pp. 76-86. Shimabukuro, Y. E, and Smith, J. A., 1991, The least-squares mixing models to generate fraction images derived from remote sensing multispectral data. Remote Sensing of Environment, 29, 16-21. Sioli, H, 1984, The Amazon: Limnology and Landscape Ecology of a Might Tropical River and its Basin (Boston: Kluwer). Steininger, M. K, 1996, Tropical secondary forest regrowth in the Amazon: age, area and change estimation with Thematic Mapper data. International Journal of Remote Sensing, 17, 9-27. Steininger, M. K, 1998a, Tropical Secondary Forest in the Amazon: An Analysis of Land Use and Net Carbon Exchange. Ph.D. Dissertation, University of Maryland at College Park, USA. Steininger, M. K, 1998b, Secondary forest structure and biomass following short and extended land use in central and southern Amazonia. Journal of Tropical Ecology, in press. Global and regional land cover characterization 1157 Swaine, M. D., and Whitmore, T. C, 1988, On the definition of ecological species groups in tropical rain forests. Vegetatio, 75, 81-86. Turner II, B. L., Moss, R. H., and Skole, D. L, 1993, Relating Land Use and Global Land-Cover Change: A Proposal for an International Geosphere-Biosphere Program—Human Dimensions of Global Environmental Change Core Project. 24 (Washington, D.C.: International Geosphere-Biosphere Program). Uhl, C, 1987, Factors controlling tropical forest succession. Journal of Ecology, 75, 377-407. Uhl, C, Busbacher, R., and Serrao, E. A. S, 1988, Abandoned pastures in eastern Amazonia I: patterns of plant succession. Journal of Ecology, 76, 663-681. Vermote, E., Tanre, D., Deuze, J. L., Herman, M., and Morcrette, J. J., 1997, Second simulation of the satellite signal in the solar spectrum: an overview. IEEE Transactions on Geoscience and Remote Sensing, 35, 675-686. Yamaguchi, Y., Kahle, A. B., Tsu, H., Kawakami, T., and Pniel, M, 1998, Overview of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER). IEEE Transactions on Geoscience and Remote Sensing, 36, 1062-1071.