Forecasting Monthly Water Deficit Based on Multi-Variable Linear Regression and Random Forest Models
WATER(2023)
摘要
Forecasting water deficit is challenging because it is modulated by uncertain climate, different environmental and anthropic factors, especially in arid and semi-arid northwestern China. The monthly water deficit index D at 44 sites in northwestern China over 1961-2020 were calculated. The key large-scale circulation indices related to D were screened using Pearson's correlation (r). Subsequently, we predicted monthly D with the multi-variable linear regression (MLR) and random forest (RF) models at certain lagged times after being strictly calibrated and validated. The results showed the following: (1) The r between the monthly D and the screened key circulation indices varied from 0.71 to 0.85 and the lagged time ranged from 1 to 12 months. (2) The calibrated and validated performance of the established MLR and RF models were all good at the 44 sites. Overall, the RF model outperformed the MLR model with a higher coefficient of determination (R-2 > 0.8 at 38 sites) and mean absolute percentage error (MAPE < 50% at 30 sites). (3) The Pacific Polar Vortex Intensity (PPVI) had the greatest impact on D in northwestern China, followed by SSRP, WPWPA, NANRP, and PPVA. (4) The forecasted monthly D values based on RF models indicated that the water deficit in northwestern China would be most severe (-239.7 to -62.3 mm) in August 2022. In conclusion, using multiple large-scale climate signals to drive a machine learning model is a promising method for predicting water deficit conditions in northwestern China.
更多查看译文
关键词
monthly water deficit,circulation indices,random forest,multi-variable linear regression,northwestern China
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要