Rainfall Prediction Using Machine Learning in Nineveh Governorate, Iraq
TL;DR
An integrated spatiotemporal framework for rainfall forecasting in Nineveh Governorate, Iraq is proposed, combining non-parametric trend detection, multivariate correlation analysis, and ensemble machine learning to improve predictive reliability and structural interpretation of rainfall variability under regional climate stress conditions.
Abstract
Accurate rainfall forecasting is essential in arid and semi-arid regions such as Iraq due to increasing climate variability, water scarcity, and the growing need for sustainable water resource management. This study proposes an integrated spatiotemporal framework for rainfall forecasting in Nineveh Governorate, Iraq, combining non-parametric trend detection, multivariate correlation analysis, and ensemble machine learning. Monthly climate data from seven stations (1994-2024) were analysed using the Mann-Kendall test and Sen’s slope estimator to assess long-term rainfall dynamics. Pearson and Spearman coefficients were applied to examine interrelationships among climatic variables and to mitigate multicollinearity prior to model construction. Four predictive models (XGBoost, CatBoost, Random Forest, and LSTM) were combined through a weighted stacking ensemble to improve robustness and reduce forecasting uncertainty. Results indicate no statistically significant long-term monotonic trend in annual rainfall; however, clear spatial heterogeneity and seasonal variability were observed. The stacking model outperformed the individual models, achieving an R² value of 0.981 (98.1%), indicating high predictive performance., with reduced RMSE and MAE values. Findings highlight the dominance role of temperature-humidity-pressure interactions in shaping rainfall behaviour in semi-arid environments. The proposed framework enhances both predictive reliability and structural interpretation of rainfall variability under regional climate stress conditions.