Spatiotemporal evaluation of inundated areas using MODIS imagery at a catchment scale

Accurate spatiotemporal analysis of surface water is important for managing water resources. The main objectives of this study were: 1) to assess different remote sensing algorithms’ ability to accurately determine flood extent using MODIS imagery, and 2) to study the temporal dynamics of inundated areas and corresponding return periods in a large catchment in New South Wales, Australia. The different water detection alternatives used included the Open Water Likelihood (OWL) algorithm, the Brakenridge algorithm for single images (BRAK) and composite images (BRAKC), the Islam methodology (ISL), a combination of Normalized Difference Indices (NDIS), and an ensemble of these. Landsat Thematic Mapper (TM) images were used as a reference to benchmark the performance of the algorithms. Occurrence probability maps derived from inundation images and estimated flood extent frequency curves were compared against the Global Surface Water (GSW) dataset of the Joint Research Centre. The ISL methodology indicated the best performance compared with “ground truth” inundated images derived from Landsat, while the temporal performance of the BRAK algorithm indicated significant noise based on daily imagery. The remaining approaches generally matched stream discharge peaks. Maximum cross-correlation coefficients between daily discharge and inundated areas were calculated for different time lags, indicating high spatial variability across the catchment. Accumulated rainfall accounted for up to 40 and 60% of the variation in inundation extent in daily and composite images, respectively. The ensemble technique outperformed the other algorithms to describe the dynamics of flooding, followed by the NDIS and ISL algorithms. The occurrence probability images and flood frequency curves indicated a higher inundation extent than the GSW product for return periods greater than two years, which can be explained by the different temporal resolutions. The results of this study can be applied in water resources planning and flood mitigation efforts.