Enhancing monthly precipitation forecasting by integrating multi-source data with machine learning models: a study in the Upper Blue Nile Basin

This study developed a novel framework to improve monthly precipitation forecasts by integrating multi-source data with machine-learning techniques. The framework used a blend of gauge, satellite, and reanalysis data to train the model with high-quality spatial–temporal precipitation information. Predictors such as local meteorological variables, large-scale climate indices, soil moisture data, and lagged precipitation data were used for one-month ahead precipitation forecasting. Five machine learning algorithms, including Artificial Neural Networks (ANN), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGboost), and Random Forest (RF), are evaluated in the framework for their forecasting performance across five monitoring stations in the upper Blue Nile basin. The results demonstrate that the novel forecasting framework integrating multi-source data with machine learning techniques significantly improved the monthly precipitation forecasts. All models exhibited higher R², closer agreement with observed data measured by Wilmot’s index (d), and lower MAE and RMSE values. During testing phase, the models recorded 0.64–0.87 R ², 0.62–2.95 RMSE, 0.18–2.42 MAE and 0.7–0.85 d values across stations. Among the models evaluated in the framework ANN, SVM, and KNN consistently showcased better performance both in the training and testing phases, with ANN exhibiting the most promising results overall. However, Extreme Gradient Boosting and Random Forests exhibited comparatively weaker performance across stations. Additionally, the models' performance under different climate categories (dry, normal, and wet months) was also explored. The models excel during normal and dry months, but, during wet months their performance diminishes relatively. Furthermore, the proposed framework forecast skills compared with NCEP and ECMWF global precipitation prediction and showed a better performance. Overall, the findings highlight the potential of the developed framework to enhance monthly precipitation forecasts, providing valuable insights for decision-makers in various sectors reliant on accurate precipitation information.