Akademik Veri Yönetim Sistemi
Journal of the Faculty of Engineering and Architecture of Gazi University, cilt.41, sa.1, ss.579-594, 2026 (SCI-Expanded, Scopus, TRDizin)
Soil temperature is a critical variable influencing glacier energy balance, mass balance, ecological stability, and agricultural productivity; however, its accurate prediction remains challenging due to complex and nonlinear interactions. Existing methods often fail to adequately capture these complexities, highlighting the need for more advanced modeling approaches. This study proposes a novel transformer-based framework for soil temperature forecasting. Leveraging the superior capability of transformer architectures to model long-term dependencies and temporal patterns, a high-accuracy predictive model is developed using environmental data. Data collected from six FLUXNET stations are employed for model development and evaluation. Comparative analyses are conducted against tree-based methods, conventional deep learning architectures, and five advanced transformer models (Vanilla Transformer, Informer, Autoformer, Reformer, and ETSformer). Experimental results demonstrate that transformer-based models achieve substantially higher predictive accuracy than traditional approaches. Moreover, the proposed method exhibits consistent, robust, and generalizable performance across diverse environmental conditions. These findings indicate that transformer models hold significant promise for environmental forecasting tasks, particularly for soil temperature prediction. The study contributes to advancing scientific understanding of soil temperature dynamics while providing a scalable, reliable, and practically applicable tool. Overall, the proposed approach addresses complex environmental prediction challenges and underscores the transformative potential of transformer-based methodologies for future research and applications.