The Use Of MODIS BRDF/Albedo Products To Recover Land Surface Properties

MODIS BRDF/Albedo products use an operational kernel driven RossThick - LiSparse Reciprocal (RTLSR) BRDF model as main algorithm to provide global albedo and BRDF products for user community. This paper assesses MODIS BRDF products including MODIS BRDF parameters and MODIS BRDF shape indicator products using 20 ground BRDF data sets and satellite sample from MODIS, explores the capability of MODIS BRDF features to improve classification accuracy as additional information by adding these BRDF features to an decision tree algorithm (c4.5) together with MODIS spectral signatures from Nadir BRDF-Adjusted Reflectance (NBAR) and Enhanced Vegetation Index (EVI), which have already been used as input features for operational MODIS land cover products, and initially validates MODIS albedo product based on ground campaign at ShunYi, Beijing. This paper pursues research in following three aspects:This paper assesses MODIS BRDF products using 20 ground BRDF data sets collected by numerous investigators, and satellite samples from MODIS, focusing on analysis of the physical mechanism of spectral BRDF effects according to the relative magnitude of volumetric effects and geometric effects. Among MODIS BRDF indicators, the Anisotropic Flat Index (AFX) indicates the BRDF shape in terms of its value around unit. AFX < 1.0 indicates a dome shape anisotropic pattern with strong backscatter peak reflectances, and potential retro solar view angle (hotspot) maximum reflectances; AFX = 1.0 approximately indicates a Lambertian surface; AFX > 1.0 corresponds to a bowl shape anisotropy pattern where BRF values near nadir are lower than for larger scattering angle. we also analysize MODIS BRDF shape indicators by Principal Component Analysis (PCA) and correction analysis techniques, and empirically calculate the directional variance (BRVF) to analysize the variance variation within class.This paper examines the utility of including MODIS BRDF products as additional input features to a decision tree classifier (c4.5). Our results show that supplementing MODIS BRDF parameters (f_vol and f_geo) with Nadir BRDF-Adjusted Reflectances (NBAR) and Enhanced Vegetation Indexes (EVI) increases overall classification accuracy by 3.02% to 4.72%, and reduces misclassification rates by 15% to 22%, depending on how many BRDF parameters are added (fourteen vs. four), and if these BRDF parameters are normalized by their isotropic parameters (f_iso). The greatest improvements are seen for Wetland shrub with user and producer's accuracy increased by up to 15.05% and 8.18% respectively. Increases on the order of 5% to 10% are encountered for the Wetland tree, Coniferous dense and Coniferous open with no detriments to other candidate classes. However, adding MODIS BRDF shape indicators produces little improvements in classification accuracy in this study. We also examine the capability in classification owned by spectral nadir reflectance (NBAR) and directional information contained in f_vol and f_geo, and find that spectral information is dominant in identifying and discriminating land cover types in this study.This paper also provides a method on initial validation of MODIS albedo products by comparing it with the albedo of ground measurements and that retrieved by air-borne multiple angular data (AMTIS).Finally, we expect further application of these BRDF shape indicators especially use of AFX to retrieve the vegetation structure such as clumping index based on model or computer simulation in near future.