Forecasting Monthly Water Deficit Based on Multi-Variable Linear Regression and Random Forest Models

Forecasting water deficit is challenging because it modulated by uncertain climate, different environmental and anthropic factors, especially in arid semi-arid northwestern China. The monthly index D at 44 sites China over 1961−2020 were calculated. key large-scale circulation indices related to screened using Pearson’s correlation (r). Subsequently, we predicted with the multi-variable linear regression (MLR) random forest (RF) models certain lagged times after being strictly calibrated validated. results showed following: (1) r between varied from 0.71 0.85 time ranged 1 12 months. (2) validated performance of established MLR RF all good sites. Overall, model outperformed a higher coefficient determination (R2 > 0.8 38 sites) mean absolute percentage error (MAPE < 50% 30 sites). (3) Pacific Polar Vortex Intensity (PPVI) had greatest impact on China, followed SSRP, WPWPA, NANRP, PPVA. (4) forecasted values based indicated that would be most severe (−239.7 −62.3 mm) August 2022. In conclusion, multiple climate signals drive machine learning promising method for predicting conditions