S29 Comparisons of flood affected area derived from MODIS and Landsat imagery

Sunday, 23 January 2011
Kevin W. Van Leer, University of Illinois, Urbana-Champaign, Urbana, IL; and J. F. Galantowicz

During the first few weeks of June 2008, the Midwest experienced a weather system that dropped large amounts of rain across the region. In southern Indiana, the Wabash and White Rivers went several feet above their flood stage and many people were displaced from their homes and businesses. This study uses the event as a test case for comparisons of resolutions and flood area estimates from the Moderate-Resolution Imaging Spectroradiometer (MODIS) and the Landsat 5 Thematic Mapper (TM) satellite sensors. A k-means classification scheme was used to identify the affected flood region by separating water from vegetation and other surface covers. This classification allowed for an estimate of flood area for each resolution provided by the two sensors. A statistical study was then performed to analyze false positive and false negative ratios using the Landsat 5 TM imagery as “ground truth”. The area estimates and statistical analysis support a claim that coarse resolutions, one and two kilometers, provide the most accurate estimate of area in large scale flood events, but the overall location of the fine details of the flood are lost. The finer resolutions , 500 and 250 meters, while more accurate in determining locations of fine details, have higher false positive and false negative ratios for area estimates. From these conflicting results, this study raises further questions about what resolutions could be effectively used to gain both an accurate map and an area estimate of the inundation's spatial extent.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner