Dataset for CNN-based Bayesian Calibration of TELEMAC-2D Hydraulic Model

This dataset contains both raw and processed data used for the Bayesian calibration of a 2D hydraulic model (TELEMAC-2D) using a convolutional neural network (CNN)-based emulator.The dataset is organized into three ZIP files:<b>DEPTHS.zip (319.5 MB)</b><br>This archive contains high-resolution GeoTIFF raster files representing water depth outputs from TELEMAC-2D simulations. These were used as target variables in CNN training and correspond to various Manning’s friction scenarios..<b>FRICTION.zip (35.7 MB)</b><br>Includes the corresponding friction zone maps (Manning’s n coefficients), also in GeoTIFF format, used as spatially distributed input conditions in the simulations. These maps define the roughness zones used in TELEMAC-2D.<b>NPY_SPLITS.zip (364.7 MB)</b><br>Contains the NumPy arrays (<code>.npy</code> format) derived from the TIFF rasters, used to train and validate the CNN emulator. The structure is as follows:Files starting with <code>X_part</code> are flattened input arrays representing friction information extracted from the friction zone rasters.Files starting with <code>y_part</code> are flattened output arrays representing corresponding water depth values.The <code>.npy</code> files were loaded and processed using the following approach in Python:# Load the input and output numpy arraysinput_path = "../data/depths/NPY_SPLITS"x_files = sorted([f for f in os.listdir(input_path) if f.startswith("X_part")])y_files = sorted([f for f in os.listdir(input_path) if f.startswith("y_part")])X_data = np.concatenate([np.load(os.path.join(input_path, f)) for f in x_files], axis=0)y_data = np.concatenate([np.load(os.path.join(input_path, f)) for f in y_files], axis=0)# Split into training and validation setsX_train, X_val, y_train, y_val = train_test_split(X_data, y_data, test_size=0.2, random_state=42)This dataset enables reproducibility of the emulator training process and supports future developments in physically informed machine learning for hydraulic model calibration.<br>