An adaptive learning framework for forecasting seasonal water allocations in irrigated catchments

This paper describes an adaptive learning framework for forecasting end-season water allocations usingclimate forecasts, historic allocation data, and results of otherdetailed hydrological models. The adaptive learning frameworkis based on artificial neural network (ANN) method,which can be trained using past data to predict future waterallocations. Using this technique, it was possible to developforecast models for end-irrigation-season water allocationsfrom allocation data available from 1891 to 2005 basedon the allocation level at the start of the irrigation season.The model forecasting skill was further improved by the incorporationof a set of correlating clusters of sea surface temperature(SST) and the Southern oscillation index (SOI) data.A key feature of the model is to include a risk factor for theend-season water allocations based on the start of the seasonwater allocation. The interactive ANN model works in a riskmanagementcontext by providing probability of availabilityof water for allocation for the prediction month using historicdata and/or with the incorporation of SST/SOI informationfrom the previous months.