Seasonal predictions of precipitation in the Aksu-Tarim River basin for improved water resources management

- Water scarcity threatens various aspects of life in the Tarim River basin.
- Seasonal predictions of precipitation facilitate early warnings for precipitation deficits.
- Two seasonal precipitation forecast models are developed for the Aksu River subbasin.
- The neural network model's performance is slightly better than that of the multiple linear regression model.

Since the 1950s, the population in the arid to hyperarid Tarim River basin has grown rapidly concurrent with an expansion of irrigated agriculture. This threatens the Tarim River basin's natural ecosystems and causes water shortages, even though increased discharges in the headwaters have been observed more recently. These increases have mainly been attributed to receding glaciers and are projected to cease when the glaciers are unable to provide sufficient amounts of meltwater. Under these circumstances water management will face a serious challenge in adapting its strategies to changes in river discharge, which to a greater extent will depend on changes in precipitation. In this paper, we aim to develop accurate seasonal predictions of precipitation to improve water resources management.Possible predictors of precipitation for the Tarim River basin were either downloaded directly or calculated using NCEP/NCAR Reanalysis 1 and NOAA Extended Reconstructed Sea Surface Temperature (SST) V3b data in monthly resolution. To evaluate the significance of the predictors, they were then correlated with the monthly precipitation dataset GPCCv6 extracted for the Tarim River basin for the period 1961 to 2010. Prior to the Spearman rank correlation analyses, the precipitation data were averaged over the subbasins of the Tarim River. The strongest correlations were mainly detected with lead times of four and five months. Finally, an artificial neural network model, namely a multilayer perceptron (MLP), and a multiple linear regression (LR) model were developed each in two different configurations for the Aksu River subbasin, predicting precipitation five months in advance. Overall, the MLP using all predictors shows the best performance. The performance of both models drops only slightly when restricting the model input to the SST of the Black Sea and the Siberian High Intensity (SHI) pointing towards their importance as predictors.