A New Classifier for Flood Hazard Mapping over Large Regions

Abstract:

The knowledge of the position and the extent of the areas exposed to the flood hazard is essential to any strategy for minimizing the risk. Unfortunately, in ungauged basins the use of traditional floodplain mapping techniques is prevented by the lack of the extensive data required. The main aim of the present work is to overcome this limitation by defining an alternative simplified procedure for a preliminary, but efficient, floodplain delineation. To validate the method in a data-rich environment, eleven flood-related morphological descriptors derived from DEMs have been used as linear binary classifiers over the Ohio River basin and its sub-catchments. Their performances in identifying the floodplains have been measured at the change of the topography and the size of the calibration area, and the best performing classifiers among those analysed have been applied and validated across the continental U.S. The results suggest that the classifier based on the index ln(h_r/H), named the Geomorphic Flood Index (GFI), is the most suitable to detect the flood-prone areas in ungauged basins and for large-scale applications, providing good accuracies with low requirements in terms of data and computational costs.