Towards Operational Automatic Flood Detection Using EOS/MODIS Data

Abstract

This study investigates how to derive water fraction and flood map from the Moderate-Resolution Imaging Spectroradiometer (MODIS) using a Regression Tree (RT) approach, which can integrate all predictors. The New Orleans, Louisiana floods in August 2005 were selected as a case study. MODIS surface reflectance with matched water fraction data were used for training. The tree-based regression models were obtained automatically through learning process. The tree structure reveals that near-infrared reflectance is more important than the difference and ratio between near-infrared and visible channels for water fraction estimate. Flood distributions were generated using the differences in water fraction values between after and before the flooding. The derived water fractions were evaluated against 30 m Thematic Mapper (TM) data from Landsat observations. Water fractions derived from the MODIS and TM data agree well (R 2 = 0.94, bias = 0.38 percent, and RMSE = 4.35 percent). The results show that the RT approach in dynamic monitoring of floods is acceptable.