Flood forecasting for fast responding catchments including uncertainty

点击翻译按钮获取中文摘要

Flood forecasting for fast responding catchments encounters problems especially in terms of short warning periods and a very limited reliability. We envisage tackling these shortcomings by using a symbiosis between physically based stochastic hydrological modelling and computationally highly efficient artificial intelligence techniques which surpasses current deterministic forecast practice and/or high computational burden of hydrologic/meteorological ensemble forecasting. Within a new stochastic decomposition framework based on a rigorous rainfall-runoff modelling, new perturbation and stochastic inference techniques we consider uncertainties of three sources: (i) hydrologic calibration uncertainty, (ii) hydrologic soil data uncertainty, and (iii) the uncertainty of the meteorological rainfall forecast. Mirroring the results of hydrologic stochastic decomposition by a problem specific stochastic Artificial Neural Networks (ANN-S) finally allows the instantaneous computation of the runoff under hydrological uncertainties. Combining the hydrologic uncertainty with the meteorological uncertainty gained from a very large number of ANN-S applications to rainfall scenarios generated by radar based ensemble forecasts allows then a real-time operation for flood forecasting including a realistic uncertainty assessment.