High-Resolution Temperature Downscaling via Machine Learning: A Comparative Study of ERA5-Land and MSWX Datasets

Articles in Press

Document Type : Original Research Article

Authors

1 Department of Environmental Sciences and Engineering, Faculty of Natural Resources, Behbahan Khatam Alanbia University of Technology, Behbahan, Iran

2 Department of Forestry and Cellulose Industries, Faculty of Natural Resources, Behbahan Khatam Alanbia University of Technology, Behbahan, Iran

Abstract

High-resolution temperature data are essential for climate-sensitive decision-making. This study compares the downscaling performance of ERA5-Land and MSWX reanalysis datasets using three ensemble-based machine learning models—Random Forest, XGBoost, and LightGBM—over the topographically complex Karun and Jarahi Basins in Iran. A two-stage framework involving spatial downscaling and error correction was implemented with auxiliary predictors. To address class imbalance, SMOGN and KNNOR-Reg resampling techniques were tested. Results show that LightGBM consistently achieved the best predictive accuracy (RMSE ≈ 1.85°C–2.31°C), with MSWX yielding superior results overall due to its finer spatial resolution. KNNOR-Reg preserved the physical plausibility of temperature distributions better than SMOGN. Visual assessments confirmed that downscaled outputs effectively captured terrain-driven microclimates. This comprehensive evaluation highlights the importance of dataset choice, model selection, and balancing strategy in downscaling workflows and underscores the potential of machine learning in improving the utility of coarse-resolution climate data for localized applications. These advancements ultimately enhance the reliability of climate information needed to support environmentally sustainable planning, resource management, and climate-resilience strategies in vulnerable regions.

Keywords

Main Subjects

Sustainable Earth Trends

Articles in Press, Accepted Manuscript
Available Online from 08 January 2026

Files

Share

How to cite

Statistics