Meteorological drought forecasting based on a statistical model with machine learning techniques in Shaanxi province,...

Area of Science:

• Hydrology
• Climate Science
• Data Science

Background:

• Droughts are major natural disasters causing significant socio-economic losses.
• Existing drought prediction models often rely on single-station data, limiting regional accuracy.
• Developing multistation models is crucial for effective drought preparedness and irrigation management.

Purpose of the Study:

• To optimize predictor selection and develop an advanced multistation model for drought prediction.
• To forecast the Standardized Precipitation Evapotranspiration Index (SPEI) using historical drought data, meteorological measures, and climate signals.
• To evaluate model performance across 32 stations in Shaanxi province, China, from 1961 to 2016.

Main Methods:

• Compared cross-correlation function and distributed lag nonlinear model (DLNM) for optimal predictor selection and lag time determination.
• Developed and validated DLNM, artificial neural network, and XGBoost models for SPEI prediction.
• Utilized multistation data, incorporating nonlinear and lag effects of predictors.

Main Results:

• The DLNM outperformed the cross-correlation function in predictor selection and lag effect analysis.
• The XGBoost model demonstrated superior accuracy in predicting SPEI (R² values 0.84-0.95 at 12-month scale) compared to DLNM and artificial neural networks.
• XGBoost achieved high prediction accuracy for overall droughts (89%-97%) and specific drought categories (76%-94%).

Conclusions:

• A novel modeling strategy for drought prediction using multistation data has been established.
• Integrating nonlinear and lag effects within the XGBoost framework significantly enhances SPEI and drought prediction accuracy.
• The developed model offers a robust tool for regional drought forecasting and management.