Neural networks and non-parametric methods for improving real-time flood forecasting through conceptual hydrological models

摘要

Time-series analysistechniques for improving the real-time flood forecasts issued by a deterministiclumped rainfall-runoff model are presented. Such techniques are applied forforecasting the short-term future rainfall to be used as real-time input in arainfall-runoff model and for updating the discharge predictions provided by themodel. Along with traditional linear stochastic models, both stationary (ARMA)and non-stationary (ARIMA), the application of non-linear time-series models isproposed such as Artificial Neural Networks (ANNs) and the ‘nearest-neighbours’method, which is a non-parametric regression methodology.For both rainfall forecasting and discharge updating, the implementation of eachtime-series technique is investigated and the forecasting schemes which performbest are identified. The performances of the models are then compared and theimprovement in the efficiency of the discharge forecasts achievable isdemonstrated when i) short-term rainfall forecasting is performed, ii) thedischarge is updated and iii) both rainfall forecasting and discharge updatingare performed in cascade. The proposed techniques, especially those based onANNs, allow a remarkable improvement in the discharge forecast, compared withthe use of heuristic rainfall prediction approaches or the not-updated dischargeforecasts given by the deterministic rainfall-runoff model alone. style="line-height: 20px;">Keywords: real-time flood forecasting, precipitation prediction, discharge updating, time-series analysis techniques