Source Codes, Trained Models, and Training, Validation, and Test Datasets for "Data-Driven Pluvial Flood Modelling for Urban Regions: An Investigation of Spatial Generalizability"

This repository contains the source codes, trained models, and complete training, validation, and test datasets used in the study “Data-Driven Pluvial Flood Modelling for Urban Regions: An Investigation of Spatial Generalizability”, submitted to Environmental Modelling & Software. The materials are organized as follows: scripts/ Python scripts used to develop and train the three machine learning and deep learning models (Random Forest, XGBoost, and U-Net) for each scenario. models/ Trained model files corresponding to each method and scenario. input data/ The complete training, validation, and test datasets used for model development and performance evaluation. Performance evaluation of all models.py Unified script to reproduce model performance metrics (e.g., RMSE, MAE) and visual comparisons using the provided test datasets. The repository contains two compressed files. The first includes the scripts, trained models, and test datasets and is organized into four subdirectories corresponding to the simulation cases: depth – with depth– without velocity – with velocity – without Within each subdirectory, models, data, and scripts for the three predictive methods (RF, XGBoost, and U-Net) are provided. The second compressed file follows the same subdirectory structure but contains the corresponding training and validation datasets for the four simulation cases. All scripts were developed in Python 3.10.13 using: scikit-learn 1.5.1 for Random Forest xgboost 2.1.4 for XGBoost tensorflow 2.10.0 for U-Net