Optimizing Precipitation Prediction at Dukan Dam through a Compartive Climate and Temporal Modeling with Multi-Factor Statistical Analysis"

Abstract

Rainfall forecasting serves as a vital tool for mitigating droughts and floods and managing water resources in areas where rainfall predictability is prone to being inadequate. This study evaluates the effectiveness of the precipitation forecast models for the Dukan Dam catchment area in northern Iraq by combining historical rainfall data with key meteorological factors, including temperature, humidity, and wind speed. To achieve this objective three modeling approaches were applied: SARIMA, Prophet, and XGBoost, using monthly data from 2015 to 2025. With feature engineering and lag factors included, the XGBoost model performs superior to the others, obtaining the lowest RMSE (normalized 0.0416; 1.13 mm) and MAE (normalized 0.0296; 0.80 mm). The forecast accuracy is found to be higher when using multi-factor climatic inputs. In areas with limited resources, the suggested framework serves as a scalable tool for climate adaptation, planning with XGBoost significantly outperforming statistical models.