Monsoon Flooding Response: a Multi-scale Approach to Water-extent Change Detection

This paper has the aim of illustrating an automatic and speditive way for retrieving inundation extent from multispectral and multitemporal satellite data, together with land-cover changes caused by flooding events, which is a fundamental issue for managing a reconstruction plan after the event. A straightforward method to map inundated areas was applied in the North-Eastern region of Bangladesh, heavily struck by monsoonal rains in September 2000. This method in based on the Principal Components Transform (PCT) of multispectral satellite data, in its Spectral-Temporal implementation, followed by logical filtering and image segmentation, in order to reach the needed coherency of the results. The use of multiresolution data (28.5-meters ground resolution Landsat-7/ETM+ and 1,100-meters ground resolution NOAA-14/AVHRR) makes possible to evaluate hazard affected areas at different scales. Comparison to RADARSAT-derived water extension maps assessed an Overall Accuracy between 86.4% (for the flood map derived with NOAA-14/AVHRR data over the whole Bangladesh) and 90.6% (for the flood map derived with Landsat-7/ETM+ data over the North-East part of the country).