Benchmarking classical, ensemble, and deep learning models for soil temperature prediction: A multi-criteria evaluation approach, Türkiye
Journal of Atmospheric and Solar-Terrestrial Physics, cilt.285, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 285
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.jastp.2026.106881
- Dergi Adı: Journal of Atmospheric and Solar-Terrestrial Physics
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Artic & Antarctic Regions, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Anahtar Kelimeler: Deep learning, Ensemble methods, Machine learning, Soil temperature prediction, Türkiye
- Erciyes Üniversitesi Adresli: Evet
Özet
This study aims to comprehensively evaluate the performance of different machine learning and statistical methods for estimating soil temperature in the modeling of complex environmental time series. Soil temperature is a critical variable in terms of agricultural productivity, ecosystem dynamics, and climate change processes, and its accurate estimation is of great importance for sustainable environmental management. Accordingly, twelve different approaches, ranging from classical regression methods to deep learning models, were compared on the same dataset. Model performances were evaluated using multiple statistical measures such as RMSE, MAE, NSE, KGE, and R2, and visual validation was performed using Taylor diagrams and graphical analysis. In addition, Kruskal–Wallis and Wilcoxon signed-rank tests were applied to test the statistical significance of the obtained results. The findings showed that the Recurrent Neural Network model was the most successful method, exhibiting the lowest error values and the highest accuracy measures. However, Gaussian Process Regression, Gated Recurrent Unit, and Bagging Regressor models also demonstrated similarly high performance. Statistical test results revealed no significant difference between these high-level models, but linear methods such as Lasso and Ridge showed significantly lower performance. In conclusion, this study demonstrates that deep learning and ensemble methods offer more reliable and robust results for soil temperature estimation, providing a comprehensive comparative framework for the literature.