Estabelecimento de um modelo de detecção de inundação no Pantanal Norte a partir de produtos derivados de imagens LANDSAT TM

Abstract

Drought and flood cycles determine the dynamics of ecological processes in the Pantanal. Therefore, the identification of flooded and non-flooded areas is critical for understanding spatial linkages between landscape compartments at different water levels, as well as for evaluating biodiversity patterns and appropriately managing the activities developed within the ecosystem. This study is mainly aimed at establishing a flood mapping model in the Northern Pantanal system from products derived from Landsat TM images. Using the logistic regression method, the flood conditions measured in situ (2007) were used as a dichotomous variable and spectral vegetation and humidity indices (NDVI, EVI, SAVI, NDWI, LSWI), as well as the land cover classification derived from LANDSAT TM images, were used as independent variables. In the selection of independent variables to compose the model, the combination of the EVI and LSWI indices and the land cover classification resulted in the best model fit, which showed an overall accuracy of 84.97% and a Kappa index of 0.691. The model performance varied according to the flood cycle period, with best performance during rising and receding waters. The model also presented a tendency of super classification of the flooded area in regions with dense riverine vegetation. Overall, the flooding classification in the Northern Pantanal from products derived from LANDSAT TM images presented a satisfactory mapping performance and can therefore be considered an important alternative for flood monitoring in large tropical floodplains, which have limited access during most of the year.