Drought projection based on a hybrid drought index using Artificial Neural Networks

Abstract

AbstractThe Tarim River Basin is a special endorheic arid drainage basin in Central Asia, characterized by limited rainfall and high evaporation as common in deserts, while water is supplied mainly by glacier and snow melt from the surrounding mountains. The existing drought indices can hardly capture the drought features in this region as droughts are caused by two dominant factors (meteorological and hydrological conditions). To overcome the problem, a new hybrid drought index (HDI), integrating the meteorological and hydrological drought regimes, was developed and tested in the basin in the work. The index succeeded in revealing the drought characteristics and the ensemble influence better than the single standardized precipitation index or the hydrological index. The Artificial Neural Network approach based on temperature and precipitation observations was set up to simulate the HDI change. The method enabled constructing scenarios of future droughts in the region using climate simulation of the GCMs under four RCP scenarios from the latest CMIP5 project. The simulations in the study have shown that the water budget patterns in the Tarim River Basin are more sensitive to temperature than to precipitation. Dominated by temperature rise causing an accelerating snow/glacier melt, the frequency of drought months is projected to decrease by about 14% in the next decades (until 2035). The drought duration is expected to be shortened to 3 months on average, with the severity alleviated. However, the region would still suffer more severe droughts with a high intensity in some years. The general decrease in drought frequency and intensity over the region in the future would be beneficial for water resources management and agriculture development in the oases. Copyright © 2014 John Wiley & Sons, Ltd.