This folder contains a Python script for training the U-FLOOD model, as well as the training data used in the associated article. The provided datasets are: ############### #Flood\Y_data.arr Stacked numpy array containing matrices of 3740x4273 pixels that represent simulated maximum water depths in 5m resolution. Water depths were generated in hydrodynamic simulations using MIKE 21 as detailed in the article. Water depth maps for all rain events considered in the study are enclosed. #Datafiles\Flood\events.npz List in numpy npz format that for each entry in Y_data.arr contains the name of the corresponding rain event. Names are aligned with the event indices in the paper ########### The datasets below are in geotiff format and can thus be accessed in GIS. #Rasters\aspect.tif ASP dataset as described in the study #Rasters\curvature.tif CURV dataset as described in the study #Rasters\DEM_5M_WB_WRDb.tif DEM dataset as described in the study. Locations with buildings have been slightly elevated in the dataset. This dataset was used as input to hydrodynamic simulations and as a potential input for the U-FLOOD model. #Rasters\dem_wb_minus_focmean_r100m.tif DEM_L dataset as described in the study. #Rasters\flacc.tif FLACC dataset as described in the study. #Rasters\flacc_imp_weighted.tif FLIMP dataset as described in the study. #Rasters\flacc_slopefocmean_weighted.tif FLSLO dataset as described in the study. #Rasters\impervious_5m.tif IMP dataset as described in the study. This dataset was used as input to hydrodynamic simulations and as a potential input for the U-FLOOD model. #Rasters\sinkdepth.tif SDEPTH dataset as described in the study. #Rasters\slope_focmean_r100m.tif SLOPE dataset as described in the study. #Rasters\TWI.tif TWI dataset as described in the study. ########### Watersheds\wwatersheds_marked_edited_dissolved5.tif Outline of the study area. ########################################## ########################################## To train the model, perform the following steps: 1. Run Execute A_LoadData.py The script reads the spatial input data from the GEOTIFF source files and creates the required input files for the CNN script. 2. Run B_TrainCNNModel.py Reads the input data, creates training and validation datasets, creates a neural network configuration with k=3,bf=64 and d=4, trains the model and computes some scores for validating performance on the validation dataset. Note that it would be extremely time consuming to train U-FLOOD on a machine that is not GPU enabled. The GPU should have a memory of 16GB, and 32GB is preferred.