Mapping urban flooding risk using machine learning

Combining Convolutional Neural Network (CNN) and geospatial data for mapping flooding risk areas

Climate change is increasing the intensity of storm events and hurricanes. Assessments of potential flood hazards can promote rapid emergency response and facilitate coastal adaptation. This study combined image-based machine learning CNN and geospatial data (i.e. elevation, slope, topographic wetness index, soil texture, surface runoff, land use/cover) for mapping flooding risk areas based on Hurricane Matthew imagery data. CNN demonstrated the highest overall accuracy (90.8%), precision (94.1%), recall (93.6%), and F-measures (93.8%). While further work should seek to establish the general applicability of this approach to other contexts, it has the potential to facilitate community planning for the increasingly extreme precipitation and intense storm events forecast by climate models. (#hurricane #machinelearning #flooding #climatechange)