Bilingual
En/Fa
Geographical Research is Published in both Persian and English Full-text.
Volume 40, Issue 3 (2025)
GeoRes 2025, 40(3): 1001-1014 |
Back to browse issues page
Article Type:
Subject:
History
Received: 2025/10/27 | Accepted: 2025/11/8 | Published: 2025/09/1
Received: 2025/10/27 | Accepted: 2025/11/8 | Published: 2025/09/1
How to cite this article
Babaeian A. Daily Minimum Temperature Forecasting at Mashhad Station Using Ensemble Learning Algorithms and Comparing the Role of Surface and Upper-Air Data. GeoRes 2025; 40 (3) :1001-1014
URL: http://georesearch.ir/article-1-1857-en.html
URL: http://georesearch.ir/article-1-1857-en.html
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Rights and permissions
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Abstract (2 Views)
Aims: Accurate prediction of daily minimum temperature (Tmin) plays a crucial role in agricultural management, frost prevention, and energy consumption. Despite advances in machine learning methods, systematic comparisons of surface and upper-air data performance for Tmin prediction in arid and semi-arid regions such as Mashhad remain limited. This study evaluates five ensemble learning algorithms—CatBoost, XGBoost, LightGBM, AdaBoost, and Random Forest—under three data scenarios: upper-air, surface, and combined, for next-day Tmin prediction.
Methodology: Daily meteorological data from the Mashhad synoptic station and ERA5 reanalysis at 300, 500, and 700 hPa levels were utilized for the period 2000–2023. All predictors were incorporated with a one-day lag relative to the target Tmin. The algorithms were trained using cross-validation. Multicollinearity among predictors was controlled using the VIF, and the optimal subset of features was determined through the Best Subset Selection (BSS) method based on the coefficient of determination (R ²
Findings: Integrating surface and upper-air data significantly improved the accuracy and stability of the models. In the combined scenario, the LightGBM algorithm achieved the best performance on the test set (R 2 =93.90 MAE = 1.63 ∘ C RMSE = 2.10 ∘ C KGE =0.93
Conclusion: Results show that combining surface and upper-air data within the LightGBM framework, along with systematic feature selection, provides the most accurate approach for short-term Tmin prediction.
Methodology: Daily meteorological data from the Mashhad synoptic station and ERA5 reanalysis at 300, 500, and 700 hPa levels were utilized for the period 2000–2023. All predictors were incorporated with a one-day lag relative to the target Tmin. The algorithms were trained using cross-validation. Multicollinearity among predictors was controlled using the VIF, and the optimal subset of features was determined through the Best Subset Selection (BSS) method based on the coefficient of determination (
Findings: Integrating surface and upper-air data significantly improved the accuracy and stability of the models. In the combined scenario, the LightGBM algorithm achieved the best performance on the test set (
Conclusion: Results show that combining surface and upper-air data within the LightGBM framework, along with systematic feature selection, provides the most accurate approach for short-term Tmin prediction.