River flow forecasting with artificial neural networks using satellite observed precipitation pre-processed with flow length and travel time information: case study of the Ganges river basin

摘要

This paper explores the use of flow length and travel time as apre-processing step for incorporating spatial precipitation information intoArtificial Neural Network (ANN) models used for river flow forecasting.Spatially distributed precipitation is commonly required when modelling largebasins, and it is usually incorporated in distributed physically-basedhydrological modelling approaches. However, these modelling approaches arerecognised to be quite complex and expensive, especially due to the datacollection of multiple inputs and parameters, which vary in space and time.On the other hand, ANN models for flow forecasting are frequently developedonly with precipitation and discharge as inputs, usually without taking intoconsideration the spatial variability of precipitation. Full inclusion ofspatially distributed inputs into ANN models still leads to a complexcomputational process that may not give acceptable results. Therefore, herewe present an analysis of the flow length and travel time as a basis forpre-processing remotely sensed (satellite) rainfall data. This pre-processedrainfall is used together with local stream flow measurements of previousdays as input to ANN models. The case study for this modelling approach isthe Ganges river basin. A comparative analysis of multiple ANN models withdifferent hydrological pre-processing is presented. The ANN showed itsability to forecast discharges 3-days ahead with an acceptable accuracy.Within this forecast horizon, the influence of the pre-processed rainfall ismarginal, because of dominant influence of strongly auto-correlated dischargeinputs. For forecast horizons of 7 to 10 days, the influence of thepre-processed rainfall is noticeable, although the overall model performancedeteriorates. The incorporation of remote sensing data of spatiallydistributed precipitation information as pre-processing step showed to be apromising alternative for the setting-up of ANN models for river flowforecasting.