ANN models for groundwater dynamics in the lower reach of Shiyang River basin, northwest China

摘要

Two BP-ANN models were developed for monthly average groundwater level in two sub-regions, i.e.. Xinhe and Xiqu in the Minqin Oasis of northwest China. The models include seven input factors, which were monthly average groundwater level prior to the month under study, irrigation area, surface water from outer regions, precipitation, evaporation, population, and monthly irrigated water amount. The test results showed that the two models have a high precision, with average absolute errors of 0.29 m and 0 37 m for Xinhe and Xiqu, respectively, A sensitivity analysis was conducted to compare the effect of input factors on groundwater level. The results showed that the main factor affecting groundwater level fall was human activities. The reduction of surface water from outer regions also accelerated the drop of local groundwater level. The results showed that the ANN model for groundwater level provides fairly good modelling capability and is an easier approach in sensitivity analysis, which is conducive to more appropriate groundwater management strategies.