Beyond the training inventory: transferability and independent validation of national machine-learning flood susceptibility maps for Slovenia

Data deposit for: Arora, A., Petan, S. Beyond the training inventory: transferability and independent validation of national machine-learning flood susceptibility maps for Slovenia (submitted to the Journal of Hydrology). National flood susceptibility maps for Slovenia on a 12.5 m grid (EPSG:3794, D96/TM) from Random Forest, XGBoost, CatBoost, a deep neural network and their ensemble mean, trained on 20 conditioning factors. Contents (one zip per folder; GeoTIFFs are tiled and losslessly DEFLATE-compressed): probability_maps: flood susceptibility probability (0–1) for each model binary_maps_screening: flood-prone / not flood-prone at each model’s screening threshold binary_maps_prevalence: the same at the threshold matching the national flood base rate (5.6 %) model_agreement: number of the four learners flagging each cell area_of_applicability: area of applicability and dissimilarity index (100 m) warning_units: flood-prone share of each ARSO hydrological warning unit conditioning_factors: the 20 predictors and the flood-inventory raster models: the trained national models, scaler and thresholds samples: training and test samples README.md lists every file, the thresholds, nodata values and the citation. checksums_md5.txt gives the MD5 checksum of each file inside the archives. Code: https://github.com/ami01/fhm-slovenia (MIT).