Application of penalized linear regression and ensemble methods for drought forecasting in Northeast China

Effective drought prediction can be conducive to mitigating some of the effects of drought. Machine learning algorithms are increasingly used for developing drought prediction models due to their high efficiency and accuracy. This study explored the ability of several machine learning models based on penalized linear regression and decision tree (DT)-based ensemble methods to predict drought conditions represented by the Standardized Precipitation-Evapotranspiration Index (SPEI) in Northeast China. We compared the forecasting performance of the penalized linear regression models based on ridge regression (RR) and lasso regression (LR) with the ordinary least squares (OLS) regression model. In addition, the AdaBoost and Random Forests (RF) models were also used to explore the suitability of ensemble methods for improving the forecasting performance. The SPEI was forecast at the different timescales of 3, 6, 12, and 24 months using the aforementioned machine learning models and the indices were used to predict short-term and long-term drought conditions. The prediction results indicated that the penalized linear regression models provided better prediction results and the ensemble methods consistently outperformed the DT model. Overall, the LR models were the optimum models for forecasting the SPEI at different timescales in Northeast China.