Using artificial neural network to forecast rainfall and discharge to support the drought preparedness and mitigation measures in the central highlands - Vietnam

Abstract

     Variations of monthly rainfall and river discharge over the Vietnamese central highlands (VCH) have a strong impact on water use for both livelihoods and agricultural production in the region. Improvement of the rainfall/discharge forecast over the VCH would contribute significantly to water resources planning and management in terms of, e.g., improved reservoir operation, agricultural practice and mitigation of drought effects. Artificial neural network is employed to estimate monthly rainfall and discharge in three river basins within the VCH. The results reveal how the rainfall in different parts of the VCH is influenced by the Pacific Ocean sea surface temperature and local climatic patterns. The quality of the forecast results varies spatially and results improve southward for rainfall. For discharge, the quality of the forecasts varies among the sites and depends on the location and the characteristics of the catchment. The best results are obtained from the artificial neural network models at sites where the rainfall is coherently influenced by the large-scale circulations, i.e., impacted by El Niño Southern Oscillation (ENSO) as reflected in the sea surface temperature (SST) variations, and where the rainfall contributes more directly to the discharge due to the characteristics of the catchments. The highest correlation coefficient between observed and validated time series is found at stations in the south of the VCH with values up to 0.82 for precipitation and 0.88 for discharge.