Geohazard assessment
DOME
Regional debris-flow susceptibility assessment combining dynamic learner selection, metaheuristic optimization, stacked ensembles and SHAP interpretation.
Ensemble learningInterpretation
Small-sample prediction
TabPFNv2.5-DebrisFlow
Small-sample debris-flow volume prediction with TabPFN-2.5, sediment-source information, machine-learning comparisons and regional holdout evaluation.
Tabular learningBenchmarking
Landslide forecasting
RSRL-HU
One-day-ahead landslide velocity forecasting combining robust motion-state estimation, groundwater forecasts, hydrology-informed residual learning and sequential prediction intervals.
Hydrology-informed learningUncertainty quantification
Early-warning support
Hydrological Readiness Forecasting
Calibrated probabilities of warning-relevant hydrological states over the next three days in small catchments, with chronological evaluation and optional SHAP attribution.
Probabilistic forecastingHydrological states
Catchment hydrology
BWSD
Daily Basin Water Storage Degree calculation and hydrological-state classification from the Fengshuwan catchment water balance.
Water storageCase study
Hydrological prediction
HydroPIML-RS
Physics-informed residual stacking for short-lead runoff and water-stage forecasting, with prediction replay and documented revision evidence for the Fengshuwan catchment.
Physics-informed learningRunoff and water stage