基于多角度微波辐射亮温数据反演冬小麦光学厚度

基于高级积分方程模型(Advanced Integrated Emission Model,AIEM),构建了包含宽范围土壤参数的C波段(6.925GHz)多角度裸露土壤发射率模拟数据库,利用该模拟数据分析了不同观测角度的裸露土壤发射率极化差之间的关系。在此基础上,结合ω-τ零阶辐射传输模型发展了C波段低矮植被光学厚度反演算法,并利用地基微波辐射计观测数据开展了冬小麦的光学厚度反演。结果显示,冬小麦光学厚度反演结果与实测冬小麦LAI在变化趋势上具有较好的一致性,反演算法具有一定的可行性。     

模拟配准误差对结构函数法气溶胶反演的影响

为了确定清晰图像和待反演图像间的配准误差对结构函数法气溶胶光学厚度反演结果的影响程度,该文基于辐射传输模型反演了不同配准误差条件下的气溶胶光学厚度。在已知地表反射率的基础上,利用辐射传输模型得到一定气溶胶光学厚度下的表观反射率,线性插值得到不同配准误差下的表观反射率,利用这些表观反射率和地表反射率反演气溶胶光学厚度;通过不同配准误差下的反演结果分析配准误差的影响,并通过统计反演误差小且不随配准误差改变的稳定像元地表反射率结构函数值和均值等特征得到稳定像元的地表特征。结果表明,配准误差结构函数法的反演结果显著影响,当配准误差小于0.2个像元时,反演误差小于0.2的像元占总像元的比例达63%。此外,研究发现,地表反射率结构函数值与相应计算窗口地表反射率的标准差相关性强,当标准差大于0.02时,反演结果受配准误差的影响相对较小,相对误差最大不超过40%;反之,受配准误差的影响较大。     

基于Landsat ETM+遥感数据的组分温度反演方法研究

地表温度在干旱监测和模拟地表热通量中有重要作用。在干旱半干旱地区,双源能量平衡模型(TSEB)通常用于计算地表的热通量。以黑河中游典型灌区为研究区域,选取4个时相的Landsat-7 ETM+遥感影像,通过植被指数与TSEB模型结合的方法反演土壤表面温度和植被冠层温度,并重点讨论土壤表面温度和植被冠层温度的分解算法。结果表明:土壤表面温度和植被冠层温度具有较好的时空一致性;土壤表面温度与植被冠层温度的反演精度通过地表净辐射与地表热通量得到了间接验证。地表净辐射与地表热通量的计算值与观测值相关性好,相关系数大于0.92。地表净辐射与地表热通量的线性回归分析表明拟合精度高。通过地表温度分解的方法获得的土壤表面温度和植被冠层温度,对监测典型区域的干旱和模拟地表热通量是可行的。     

基于Sentinel-1与FY-3C数据反演植被覆盖地表土壤水分

基于新一代的Sentinel-1SAR数据与FY-3C的MWRI数据,研究植被覆盖地表土壤湿度反演方法。为消除植被对土壤湿度反演影响,首先利用FY-3C/MWRI的微波极化差异指数MPDI,建立植被含水量反演模型;然后,结合植被含水量反演模型和水—云模型,发展一种主被动微波联合反演植被覆盖地表土壤含水量模型;最后,在江淮地区开展反演试验,利用观测的土壤湿度数据进行反演结果的精度验证。结果表明:(1)对于植被覆盖地表土壤湿度反演,由FY3C/MWRI提取的MPDI对于去除植被影响效果较好;(2)相比于VH极化哨兵1号卫星数据,VV极化数据更适用于土壤含水量的反演,能够得到更高的土壤湿度反演精度;(3)哨兵1号卫星数据能够获得较高精度的土壤含水量反演结果,试验反演的土壤湿度值与实测值相关系数为0.561 2,均方根误差为0.044cm~3/cm~3。     

利用MERIS和AATSR资料估算黄土高原塬区植被含水量时空变化

通过利用2005年黄土高原塬区夏季地表过程野外观测试验期间收集的地面观测的植被含水量、中分辨率影像光谱仪(Medium Resolution Imaging Spectrometer,MERIS)和高级沿轨迹扫描辐射计(Advanced Along-Track Scanning Radiometer,AATSR)卫星遥感资料,分别对归一化差值植被指数(Normalized Different Vegetation Index)和归一化差值水分指数(NormalizedDifferent Water Index)与植被含水量(Vegetation water content)变化关系进行了分析比较,得到了两种不同的植被指数在作物生长背景影响下的异同。并分别利用MERIS的观测资料计算了NDVI,利用AATSR观测资料计算了NDWI,通过分析得出:随着作物的生长或生物量的增加,归一化差值植被指数变化趋于饱和,而归一化差值水分指数仍然继续增加。进一步通过同步地面野外观测植被含水量与卫星遥感观测资料的对比,建立了归一化差值植被指数、归一化差值水分指数和实际野外测量植被含水量的关系,并且得到由两种植被指数反演植被含水量的方法和地面观测之间的误差分别为1.030 64 kg·m-2和0.940 45 kg·m-2。最后通过分析后总结出,利用归一化差值水分指数来反演黄土高原塬区夏季玉米冠层的含水量优于利用归一化差值植被指数。     

土壤—植被混合地表红外辐射温度场的耦合建模方法*

目前混合地表温度场的数学模型忽略了地表内部各组分间的热通量交互,致使温度场模拟结果不够准确、真实.针对已有独立求解模型的不足,结合土壤-植被混合地表的材质组成及空间分布特点,建立温度场的耦合求解模型,通过在热平衡方程中引入组分间热通量的交互项,对土壤-植被-大气耦合的能量平衡过程进行描述.实验结果表明,利用耦合模型求解的混合地表的温度场分布和热图像特征与自然地表红外辐射特性的真实分布规律具有更好的一致性,从而验证了温度场耦合建模方法的有效性.     

黑河综合遥感联合试验研究进展:水文与生态参量遥感反演与估算

介绍了“黑河综合遥感联合试验”在水文和生态变量与参数反演、估算和模型应用方面取得的进展。在水文变量遥感方面,利用车载双偏振多普勒雷达在黑河上游和中游分别开展了高精度降水观测,获取了后向散射系数和极化信息与降水强度之间的定量关系。在综合利用多源观测信息,改进和发展蒸散发估算模型方面取得了实质性的进展。发展了利用K和Ka波段机载微波辐射计数据反演山区积雪深度的方法。针对SAR观测数据反演土壤水分中地表粗糙度的显著干扰,发展了消除粗糙度影响的反演方法。在生态过程遥感参量估算方面,提出了一种基于机载激光雷达和高分辨率光学影像的高精度地物信息分类方法。发展了从高光谱航空遥感提取植被自然光照下的荧光,并与NDVI结合的C3/C4植被分类方法。发展和改进了使用多角度、多光谱观测反演叶面积指数的方法,挖掘了激光雷达在植被垂直结构探测上的潜力,探索了叶面积指数遥感中的尺度转换规律。发展了利用高光谱数据中的荧光信息反演光能利用率的新方法;建立了考虑土壤反射率、冠层结构等因素的光合作用有效辐射比率反演模型;改进了利用遥感估计生态系统生产力的模型。发展了利用高光谱遥感数据提取叶绿素含量和叶绿素荧光强度的方法。     

半干旱区土壤热通量遥感估算模型的适应性对比分析

提出了一种基于地表温度的土壤热通量遥感估算模型,结合另外两种应用比较广泛的遥感估算模型,分别是Moran(1989)提出的基于归一化植被指数NDVI、净辐射通量的模型和Bastiaanssen(1998)的基于NDVI、地表反照率、地表温度、净辐射通量的模型,利用MODIS遥感数据对这3种土壤热通量的模型进行了试验分析。参照半干旱区退化草地和农田地面站点实测的土壤热通量数据,3种遥感模型的试验结果表明:提出的基于MO-DIS地表温度的模型得到的土壤热通量精度最高;Bastiaanssen(1998)模型也能得到精度相当的土壤热通量,特别是它得到的可利用能量精度最高;Moran(1989)模型反演的土壤热通量误差最大。     

地中海生态系统中土地退化,土壤侵蚀和沙漠化遥感监测

简要介绍了国外应用遥感图像进行土地退化制图和监测的一种新方法。遥感图像上像元亮度值代表土地表面多种物质（土壤、植被、阴影、母岩）混合光谱辐射特征，通过多光谱图像辐射校正和典型区野外光谱测试，利用线性光谱混合模拟可以将ＡＶＩＲＩＳ和ＴＭ图像混合光谱中差异明显的几个组分进行分离，进而在各组分丰度分析的基础上，比较准确地进行植被盖度、土壤条件与土壤侵蚀的判别与制图。     

基于Landsat８影像的额济纳荒漠绿洲植被覆盖度估算方法对比研究

荒漠绿洲是维持当地人类生存和社会发展的主要依托,但其地表植被稀疏,生态系统极其脆弱,而植被覆盖度是反映荒漠生态环境信息的重要指标之一.以黑河下游额济纳荒漠绿洲为例,基于Landsat８影像和野外实测植被覆盖度数据,对比和分析现有的适宜于干旱荒漠区的３类植被覆盖度提取方法(经验模型法、像元二分法和三波段梯度差法)在该区域的应用效果,并尝试利用基于转换型土壤调整植被指数(TSAVI)的像元二分模型法和修正的三波段梯度差法(MTGDVI)进行植被覆盖度估算,以期找到计算额济纳荒漠绿洲植被覆盖度的最佳模型. 研究结果表明:用TSAVI像元二分模型法的反演精度高而且能够较好地估算额济纳荒漠区域和绿洲区域的植被覆
盖度,适用于估算额济纳荒漠绿洲的植被覆盖度.     

ENVISAT/AATSR derived land surface temperature over a heterogeneous region

In this paper a method for evaluating land surface temperature (LST) algorithms over heterogeneous areas is presented. The evaluation was made for a set of 12 algorithms derived by using the split-window (SW) and dual-angle (DA) techniques for estimating sea and land surface temperature (SST and LST) from Advanced Along-Track Scanning Radiometer (AATSR) data. A validation of the proposed algorithms was carried out over a heterogeneous region of Morocco in the framework of the WATERMED (WATer use Efficiency in natural vegetation and agricultural areas by Remote sensing in the MEDiterranean basin) project. AATSR data and in situ measurements over this heterogenous region were compared by implementing a classification based strategy over a higher spatial resolution Landsat image. Three reference classes were considered when performing the classification from the Landsat image. Ground based measurements where then used to assign an effective surface radiometric temperature to each of these three classes. Finally, an averaging procedure based on class proportion was implemented for deriving surface radiometric temperature at the AATSR pixel scale. For this heterogeneous site, the results showed that LST can be obtained with a root mean-square error (RMSE) lower than 1.7 K from the split-window algorithms. Dual-angle algorithms, on the other hand, provided greater RMSE due to the different surfaces observed in the nadir and forward views. The results suggest that to retrieve LST from 1 km pixels over heterogeneous surfaces spatial averaging is required to improve accuracy on temperature estimation.     

Thermal infrared inverse model for component temperatures of mixed pixels

Multiangular remote sensing data can be used to retrieve land surface component temperatures, which will have a broad application in the future. For higher resolution pixels of satellite radiometers, the component temperatures may be separated adequately by some methods. However, for coarse resolution pixels that contain a mixture of vegetation and bare soil, the component temperatures may not be retrieved robustly by traditional inversion methods. In this study, a thermal model-based algorithm was developed for mixed pixels. A simulation method was implemented to assess the performance of the algorithm. The method consisted of extensive radiative transfer simulations under a wide variety of Leaf Area Index (LAI) values in the vegetation part of directional thermal infrared (TIR) radiation, vegetation and soil emissivity, vegetation and soil temperatures, bare soil area ratio and downwelling longwave atmospheric radiation. The results indicate that the inversion error of the component temperatures does not exceed 0.5° when LAI values are less than 6.0. A field experiment was also conducted to assess the accuracy of the model. The experimental results indicate that even if the differences between the nadir and off-nadir radiative temperatures over a mixed pixel are small, the model can still determine the component temperatures accurately. A sensitivity analysis shows that an accuracy of less than 10% for LAI in the vegetation part and the bare soil area ratio is required to achieve a precision of 1 K for the component temperatures derived. An error of 1 K in the radiometric temperature leads to an error of 1 K in the component temperatures retrieved.     

The inter-comparison of AATSR dual-view aerosol optical thickness retrievals with results from various algorithms and instruments

The Advanced Along-Track Scanning Radiometer (AATSR) dual-view (ATSR-DV) aerosol retrieval algorithm is evaluated for a single scene over Germany (49–53? N, 7–12? E) on 13 October 2005 by comparison of the aerosol optical thickness (AOT) at 550 nm with products from Multiangle Imaging SpectroRadiometer (MISR), Moderate Resolution Imaging Spectroradiometer (MODIS) and Medium Resolution Imaging Spectrometer (MERIS), in addition to the Atmospheric Aerosol Retrieval using Dual-View Angle Reflectance Channels (AARDVARC) algorithm developed at Swansea University. The AOT was retrieved from the AATSR using the ATSR-DV algorithm, for the pixel size of 1 km × 1 km (at nadir). Then these values were meshed to be consistent with the sampling of the products from the other satellite instruments. The ATSR-DV results compare favourably with the products from orbiting optical instruments dedicated to aerosol retrieval, such as MODIS and MISR, which leads to the conclusion that AATSR is well suited for aerosol retrieval over land when the dual view is used with the ATSR-DV algorithm.     

Aerosol optical depth retrieval over snow using AATSR data

Aerosol observations over the Arctic are important because of the effects of aerosols on Arctic climate, such as their direct and indirect effects on the Earth's radiation balance and on snow albedo. Although information on aerosol properties is available from ground-based measurements, passive remote sensing using satellite measurements would offer the advantage of large spatial coverage with good temporal resolution, even though, due to light limitations, this is only available during the Arctic summer. However, aerosol optical depth (AOD) retrieval over the Arctic region is a great challenge due to the high reflectance of snow and ice and due to the high solar zenith angle. In this article, we describe a retrieval algorithm using Advanced Along-Track Scanning Radiometer (AATSR) data, a radiometer flying on the European Space Agency (ESA) Environmental Satellite (ENVISAT), which offers two views (near nadir and at 55° forward) at seven wavelengths in the visible thermal-infrared (VIS-TIR). The main idea of the Dual-View Multi-Spectral (DVMS) approach is to use the dual view to separate contributions to reflectance measured at the top of the atmosphere (TOA) due to atmospheric aerosol and the underlying surface. The algorithm uses an analytical snow bidirectional reflectance distribution function (BRDF) model for the estimation of the ratio of snow reflectances in the nadir and forward views, as well as an estimate of the atmospheric contribution to TOA reflectance obtained using the dark pixel method over the adjacent ocean surface, assuming that this value applies over nearby land surfaces in the absence of significant sources across the coastline. An iteration involving all four AATSR wavebands in the visible near-infrared (VIS-NIR) is used to retrieve the relevant information. The method is illustrated for AATSR overpasses over Greenland with clear sky in April 2009. Comparison of the retrieved AOD with AErosol Robotic Network (AERONET) data shows a correlation coefficient of 0.75. The AODs retrieved from AATSR using the DVMS approach and those obtained from AERONET data show similar temporal trends, but the AERONET results are more variable and the highest AOD values are mostly missed by the DVMS approach. Limitations of the DVMS method are discussed. The pure-snow BRDF model needs further correction in order to obtain a better estimation for mixtures of snow and ice.     

Analytical parameterization of canopy directional emissivity and directional radiance in the thermal infrared. Application on the retrieval of soil and foliage temperatures using two directional measurements

This work is aimed at deriving canopy component (soil and foliage) temperatures from remote sensing measurements. A simulation study above sparse, partial and dense vegetation canopies has been performed to improve the knowledge of the behaviour of the composite radiative temperature and emissivity. Canopy structural parameters have been introduced in the analytical parameterization of the directional canopy emissivity and directional canopy radiance:namely, the leaf area index (LAI), directional gap fraction and angular cavity effect coefficient. The parameterization has been physically defined allowing its extension to a wide range of Leaf Inclination Distribution Functions (LIDF). When single values are used as leaves and soil temperatures, they prove to be retrieved with insignificant errors from two directional measurements of the canopy radiance (namely at 0 and 55 from nadir), provided that the canopy structure parameters are known. A sensitivity study to the different parameters shows the great importance of the accuracy on LAI estimation (an accuracy of 10 per cent is required to retrieve the leaves temperature with an accuracy better than 0.5 degK, the same requirement being 5 per cent for the retrieval of soil temperature). The radiometric noise is important too, but its effects may be limited by using very different angles for the measurements: for 0 and 55, the effect of a Gaussian noise (NEDeltaT 0.05deg K) is lower than 0.5degK on the retrieved soil and foliage temperatures). Uncertainties on the leaf and soil emissivities (Delta epsilon 0.01) cause little errors in the retrieval (lower than 0.5degK). If the inclination dependence of the leaves temperature is considered, a 1 degK error is observed in the retrieved soil and foliage temperatures. This error is due to the fact that the effective foliage temperature varies with the view angle (a few 10 -1 deg K at 55 ), which implies errors in the inversion scheme. This effect may be corrected for by using an angular corrective term delta depending only on the off-nadir angle used.     

Effects of sub-pixel heterogeneity on the retrieval of soil moisture from passive microwave radiometry

A simple formulation relating the L-band microwave brightness temperature detected by a passive microwave radiometer to the near surface soil moisture was developed using MICRO-SWEAT, a coupled microwave emission model and soil-vegetation-atmosphere-transfer (SVAT) scheme. This simple model provides an ideal tool with which to explore the impact of sub-pixel heterogeneity on the retrieval of soil moisture from microwave brightness temperatures. In the case of a bare soil pixel, the relationship between apparent emissivity and surface soil moisture is approximately linear, with the clay content of the soil influencing just the intercept of this relationship. It is shown that there are no errors in the retrieved soil moisture from a bare soil pixel that is heterogeneous in soil moisture and texture. However, in the case of a vegetated pixel, the slope of the relationship between apparent emissivity and surface soil moisture decreases with increasing vegetation. Therefore for a pixel that is heterogeneous in vegetation and soil moisture, errors can be introduced into the retrieved soil moisture. Generally, under moderate conditions, the retrieved soil moisture is within 3% of the actual soil moisture. Examples illustrating this discussion use data collected during the Southern Great Plains '97 Experiment (SGP97).     

Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS data

An experimental site was set up in a large, flat and homogeneous area of rice crops for the validation of satellite derived land surface temperature (LST). Experimental campaigns were held in the summers of 2002-2004, when rice crops show full vegetation cover. LSTs were measured radiometrically along transects covering an area of 1 km². A total number of four thermal radiometers were used, which were calibrated and inter-compared through the campaigns. Radiometric temperatures were corrected for emissivity effects using field emissivity and downwelling sky radiance measurements. A database of ground-based LSTs corresponding to morning, cloud-free overpasses of Envisat/Advanced Along-Track Scanning Radiometer (AATSR) and Terra/Moderate Resolution Imaging Spectroradiometer (MODIS) is presented. Ground LSTs ranged from 25 to 32 °C, with uncertainties between ± 0.5 and ± 0.9 °C. The largest part of these uncertainties was due to the spatial variability of surface temperature. The database was used for the validation of LSTs derived from the operational AATSR and MODIS split-window algorithms, which are currently used to generate the LST product in the L2 level data. A quadratic, emissivity dependent split-window equation applicable to both AATSR and MODIS data was checked as well. Although the number of cases analyzed is limited (five concurrences for AATSR and eleven for MODIS), it can be concluded that the split-window algorithms work well, provided that the characteristics of the area are adequately prescribed, either through the classification of the land cover type and the vegetation cover, or with the surface emissivity. In this case, the AATSR LSTs yielded an average error or bias of − 0.9 °C (ground minus algorithm), with a standard deviation of 0.9 °C. The MODIS LST product agreed well with the ground LSTs, with differences comparable or smaller than the uncertainties of the ground measurements for most of the days (bias of + 0.1 °C and standard deviation of 0.6 °C, for cloud-free cases and viewing angles smaller than 60°). The quadratic split-window algorithm resulted in small average errors (+ 0.3 °C for AATSR and 0.0 °C for MODIS), with differences not exceeding ± 1.0 °C for most of the days (standard deviation of 0.9 °C for AATSR and 0.5 °C for MODIS).     

Effects of vegetation canopy structure on remotely sensed canopy temperatures

Remote sensing of vegetation temperatures is a promising technique for inferring plant water stress and yield on a large spatial scale. The effects of vegetation canopy structure on thermal infrared sensor response need to be understood before vegetation surface temperatures of canopies with low percentages of ground cover can be accurately inferred. The response of a sensor is a function of vegetation geometric structure, the vertical surface temperature distribution of the canopy components, and sensor view angle. Large deviations between the nadir sensor effective radiant temperature (ERT) and vegetation ERT for a soybean canopy were observed throughout the growing season. The nadir sensor ERT of a soybean canopy with 35% ground cover deviated from the vegetation ERT by as much as 11°C during the mid-day. These deviations were quantitatively explained as a function of canopy structure and soil temperature. Remote sensing techniques which uniquely determine the vegetation canopy temperature(s) from the sensor response need to be studied.     

Influence of forest cover fraction on L-band soil moisture retrievals from heterogeneous pixels using multi-angular observations

Airborne L-band data from the Australian National Airborne Field Experiment 2005 (NAFE '05) field campaign were used to investigate the influence of fractional forest cover on soil moisture retrievals from heterogeneous (grass/forest) pixels. This study is, to our knowledge, the first to use experimental data on this subject and was done in view of the SMOS mission, in order to contribute to calibration/validation studies and the analysis of heterogeneous surfaces. Because the multi-angle observations were contained in swaths, swaths were used instead of pixels as the basic surface unit in this study. Simultaneous retrievals of soil moisture (SM) and vegetation optical depth (τ_NAD) were undertaken by inversion of the L-MEB zero-order radiative transfer model. This was done for two different retrieval configurations, the first consisting of swath-effective values of SM and τ_NAD and the second consisting of values of SM and τ_NAD for the non-forested (i.e. grass) fraction of the swath, with forest emission known from forward modelling. Model inputs for non-retrieved parameters were either default values taken from the literature or site- and time-specific values obtained from observations of nearby homogeneous swaths gathered during the same flight. The main focus of this study was on retrieval behaviour for various soil moisture conditions and forest fractions. Area-averaged retrieval results were generally very reasonable for both retrieval configurations. When retrieving swath-effective values of SM and τ_NAD, τ_NAD showed an increased overestimation with increased forest fraction. Highest retrieved values of SM were found at intermediate values of forest fraction. The results show the difficulty in flagging upper limits of pixel forest fraction during soil moisture retrievals, besides the fact that erroneous parameter values can lead to high errors in retrieved SM, especially in wet conditions. This study is the first to give a realistic idea of the errors and uncertainties involved in soil moisture retrievals from partly forested swaths, and as such will contribute to a better understanding of SMOS calibration/validation issues.     

View angle effects on the discrimination of selected Amazonian land cover types from a principal‐component analysis of MISR spectra

Multi‐angle Imaging Spectroradiometer (MISR) data, collected in four bands and at nine view angles in the Brazilian Amazon region, were used to describe view‐angle effects on the spectral response and discrimination of three forest types; close and open lowland forests, open submontane forest and green/emerging pastures. A principal‐component analysis (PCA) was applied over 450 bidirectional reflectance factor (BRF) MISR spectra (10 pixels, five land covers and nine view angles) to characterize the spectral‐angular variability in the dataset and to identify the best view direction to enhance land cover discrimination. The analysis was extended into the images of the different cameras, which were classified for the presence of the forest covers using the minimum distance of the pixels to the average PC1 and PC2 scores of each forest class calculated from spectra analysis. Results showed an increase in the mean reflectance over the spectral bands (brightness) of the land covers from nadir to extreme viewing, as indicated by the first principal component, especially in the backward direction due to the predominance of sunlit view vegetation components. The transition from the backward (sunlit view surface components) to the forward (shaded view surface components) scattering directions was also characterized by changes in the shape of the BRF spectra, as indicated by decreasing PC2 score or near‐infrared/blue ratio values. The variations in the MISR BRF followed the regularities expected from theory. PCA results also indicated that the best viewing to discriminate the forest types was the backward scattering direction (?26.1° view angle), whereas the less favourable viewing was the forward scattering direction under the view shading condition (e.g. +45.6° view angle). The overall classification accuracy for the three forest types increased from 52.4% at +45.6° view angle to 78.7% at nadir, and to 95.0% at a ?26.1° view angle. From nadir to extreme view angles, directional effects produced a NDVI decrease for the forest types and an NDVI increase for the green and especially emerging pastures. Results demonstrated that data acquisition in off‐nadir viewing may improve the discrimination and mapping of the Amazonian land cover types.