This study presents a framework to reconstruct long-term daily data for 464 Iranian synoptic stations from 1950 to 2025 using the Extra Trees machine learning algorithm trained on ERA5 reanalysis data. Nine meteorological variables were reconstructed: temperature, maximum and minimum temperature, dew point, surface pressure, wind speed, wind direction, precipitation, and visibility. Reconstruction accuracy varies spatially and by parameter. Temperature-related variables achieve high fidelity, with R² up to 0.99 and RMSE as low as 0.63°C, notably in central and southeastern Iran. Dew point and surface pressure also showed strong performance (RMSE ≈ 0.90°C and 0.41 hPa). In contrast, precipitation and visibility posed greater challenges, with RMSEs reaching 871.27 mm and 9741.01 m, respectively, particularly in northwestern and western mountainous regions, underscoring the difficulty of capturing small-scale processes with global reanalyses. Wind speed and direction exhibited spatially variable accuracy, influenced by complex topography. The reconstructed datasets filled critical data gaps, enabling robust climate trend analyses, agricultural planning, water resource management, and renewable energy assessments in regions where models require high-quality inputs. In error-prone areas, bias correction or local observations were essential. Overall, the study demonstrates the efficacy of machine learning in overcoming historical data limitations and provides a solid foundation for evidence-based climate adaptation strategies in Iran.
Rad et al. (Mon,) studied this question.