Hisam et al. (2025) Precipitation downscaling with the integration of multiple precipitation products, land surface data and gauge stations using explainable machine learning algorithms: A case study in the Mediterranean region of Turkiye

Identification

• Journal: The Science of The Total Environment
• Year: 2025
• Date: 2025-09-27
• Authors: Enes Hisam, Elif Sertel, Dursun Zafer Şeker
• DOI: 10.1016/j.scitotenv.2025.180540

Research Groups

• Istanbul Technical University, Graduate School, Geomatics Engineering Program
• Istanbul Technical University, Faculty of Civil Engineering, Geomatics Engineering Department

Short Summary

This study downscaled monthly gridded precipitation data to a 0.04° spatial resolution in the Mediterranean region of Türkiye by integrating multiple precipitation products and land surface characteristics using explainable machine learning, finding Random Forest to be the most accurate model.

Objective

• To downscale monthly gridded precipitation data to a high spatial resolution (0.04°) over the Mediterranean region of Türkiye by integrating ground-based, satellite-based, and reanalysis precipitation datasets with land surface characteristics (topography, Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), and distance from the sea) using explainable machine learning algorithms.

Study Configuration

• Spatial Scale: Mediterranean region of Türkiye, downscaled to 0.04° spatial resolution (approximately 4.4 km at the equator). Original coarse resolutions ranged from 0.1° to 0.25°.
• Temporal Scale: Monthly precipitation data from 2017–2021 for model training and validation, and 2016 for independent evaluation.

Methodology and Data

• Models used: Cubist (rule-based), Random Forest, XGBoost, LightGBM, CatBoost (machine learning algorithms). SHAP (SHapley Additive exPlanations) values were used for model explainability.
• Data sources:
  • Gridded Precipitation Products: PERSIANN-CCS, PERSIANN-CDR, PDIR-Now, CHIRPS, GSMaP MVK v7, GSMaP Gauge v7, IMERG v6, ERA5 (satellite-based and reanalysis).
  • Ground-based Observations: Monthly precipitation data from 193 meteorological stations.
  • Land Surface Characteristics: Topography, Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), and distance from the sea.

Main Results

• Monthly precipitation was successfully downscaled to a 0.04° spatial resolution by integrating multiple precipitation products and land surface data.
• Among the evaluated rule-based and machine learning methods, Random Forest demonstrated the best overall accuracy.
• The inclusion of land features such as NDVI, LST, and topography significantly improved monthly rainfall estimates.
• Combinations of precipitation products (Comb1 and Comb2) consistently outperformed individual products, achieving Pearson Correlation Coefficients (PCC) greater than 0.79, Root Mean Square Errors (RMSE) less than 39 mm, and Mean Absolute Errors (MAE) less than 24 mm.
• SHAP values effectively explained the feature importance within the models, providing insights into their decision-making processes.
• The generated monthly precipitation maps for the independent year 2016 showed robust statistical and visual agreement with meteorological station data.

Contributions

• Developed a novel approach for high-resolution (0.04°) monthly precipitation downscaling in the Mediterranean region of Türkiye by comprehensively integrating multiple ground-based, satellite-based, and reanalysis precipitation datasets with diverse land surface characteristics.
• Evaluated and compared the performance of various explainable machine learning algorithms (Random Forest, XGBoost, LightGBM, CatBoost) and a rule-based algorithm (Cubist) for precipitation downscaling in a complex topographical region.
• Demonstrated the significant improvement in downscaling accuracy achieved by merging multiple precipitation products and incorporating relevant land surface features.
• Utilized SHAP values to enhance the interpretability of the machine learning models, providing valuable insights into the drivers of precipitation downscaling.

Funding

• Not specified in the provided text.

Citation

@article{Hisam2025Precipitation,
  author = {Hisam, Enes and Sertel, Elif and Şeker, Dursun Zafer},
  title = {Precipitation downscaling with the integration of multiple precipitation products, land surface data and gauge stations using explainable machine learning algorithms: A case study in the Mediterranean region of Turkiye},
  journal = {The Science of The Total Environment},
  year = {2025},
  doi = {10.1016/j.scitotenv.2025.180540},
  url = {https://doi.org/10.1016/j.scitotenv.2025.180540}
}