نوع مقاله : مقاله پژوهشی

نویسندگان

• ام البنین بذرافشان ¹
• مهرناز یحیی زاده ²
• نوازاله مرادی ³
• حسین زمانی ³
• مهدی بی نیاز ³

¹ دانشیار علوم و مهندسی آبخیزداری گروه مهندسی منابع طبیعی، دانشکده کشاورزی و منابع طبیعی دانشگاه هرمزگان

² دانشگاه هرمزگان-بندرعباس

³ دانشگاه هرمزگان

چکیده

This study evaluates Support Vector Machine (SVM) and Gaussian Process Regression (GPR) models for predicting drought in Iran using the Standardized Precipitation-Evapotranspiration Index (SPEI). Results demonstrate the superior performance of the GPR model with a Laplace kernel, which achieved higher predictive accuracy (R²: 0.85-0.37 in testing) and superior uncertainty quantification (PICP: 1.0) compared to the SVM-RBF model. GPR's flexibility in modeling abrupt climatic shifts and its probabilistic framework provide more reliable drought forecasts, especially in extreme climates. Random Forest analysis revealed key climatic drivers, with temperature and evapotranspiration dominating in arid regions, while oceanic oscillations (ENSO, WHWP) were more influential in humid zones. Although SVM remained competitive in moderate climates, GPR is recommended as the superior choice for operational early warning systems, enabling optimized water resource management and agricultural advisories through its reliable probabilistic forecasts. This research establishes a framework for integrating model accuracy and uncertainty assessment in drought prediction.

کلیدواژه‌ها

• Drought Prediction
• Machine Learning
• Uncertainty Quantification
• Teleconnection Patterns
• UNEEC Method

موضوعات

• خشکسالی