Combining Landsat ETM+ and Terrain Data for Scaling Up Leaf Area Index (LAI) in Eastern Amazon: an Intercomparison With MODIS Product
Luiz
Eduardo Oliveira Cruz de
Arag�o, INPE, aragao@ltid.inpe.br
(Presenting)
Yosio
Edemir
Shimabukuro, INPE, yosio@ltid.inpe.br
Mathew
Williams, University of Edinburgh, mat.williams@ed.ac.uk
Fernando
Del Bom
Esp�rito Santo, INPE, fernando@ltid.inpe.br
The aim of this study was to produce a continuous field of LAI to evaluate LAI surface (MOD15 product) retrieved from moderate resolution imaging spectroradiometer (MODIS), for the Tapaj�s region, eastern Amazonia. Our method consisted in generating regression models combining spectral data derived from Enhanced Thematic Mapper Plus (ETM+) sensor (07/30/2001) and terrain slope and altimetry information extracted from a digital terrain model. We built three different models, the first (Model 1) used only spectral variables, the second (Model 2) combined spectral and terrain data, and the last (Model 3) accounted only with terrain data. We compared the mean LAI estimated by the three models (270m spatial resolution) with the 8 days LAI composition (08/13/2001) derived from MODIS sensor (1km spatial resolution) and also with field data. The three models generated by our method fitted the field LAI data within the confidence level of 95%. The model 2 had the best fit (R2 = 0.91), an improvement of 11% over the estimation using a purely spectral model (Model 1). The mean LAI and the standard deviation found for the whole region varied from 3.50 � 0.83 (Model 2) to 5.46 � 0.60 (MODIS). The mean LAI for the field measurements was 3.84 � 0.89, with a minimun LAI of 1.56 and a maximum of 5.10. The MODIS LAI surface for the Tapaj�s region showed a more homogenous surface and little information about land cover. when we visualy compared with our estimations. This fact occurs due to the 1km resolution from de MODIS LAI against the 270m resolution used in our approach. The mean LAI derived from MODIS sensor is significantly overestimated (P<0.05) in relation to both field and modelled data. We conclude that this approach seems to be promising for generating LAI surfaces over the Tapaj�s region. The integration of terrain and ETM+ spectral information improved the purely spectral model. In addition, terrain-spectral model captured more realistically the LAI hetereogeneity in the Tapaj�s region than did the spectral and MODIS estimations. The spectral-terrain model also reproduced the mean, standard deviation and minimum and maximum LAI values found in the field data, while MODIS overestimated LAI in the Tapaj�s.
Submetido por Luiz Eduardo Oliveira Cruz de Arag�o em 18-MAR-2004
Tema Cient�fico do LBA: CD (Armazenamento e Trocas de Carbono)