Assessing Climatic Drivers of Glacier Retreat in Türkiye Using ERA5 Data and Machine Learning Methods
Pure and Applied Geophysics, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s00024-026-04136-3
- Dergi Adı: Pure and Applied Geophysics
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, L'Année philologique, Aerospace Database, Compendex, Geobase, INSPEC, Academic Search Ultimate (EBSCO), Natural Science Collection (ProQuest), Earth, Atmospheric, & Aquatic Science Collection (ProQuest), Engineering Source (EBSCO), Technology Collection (ProQuest)
- Anahtar Kelimeler: Climate change, ERA5, glacier retreat, machine learning, SHAP analysis, Türkiye
- Akdeniz Üniversitesi Adresli: Evet
Özet
The main objective of this research is to quantify the climatic factors that control solid precipitation (snow) as the primary accumulation component in the largest glacier areas of Türkiye (Ağrı, Kaçkar, Süphan, Cilo, Erciyes, and Mercan) utilizing the fifth generation European Centre for Medium-Range Weather Forecasts atmospheric reanalysis (ERA5) data and a sophisticated hybrid machine learning technique. By modelling solid precipitation and assessing the relative importance of temperature versus precipitation variables, we infer which climatic factors exert the dominant control on the net mass balance and, consequently, on the potential for glacier retreat. The study employed a two-stage modelling framework: first, a CatBoostClassifier classified days with solid precipitation (snow); second, a stacking ensemble of CatBoostRegressor and LightGBM, combined with a Ridge meta-learner, estimated the amount of solid precipitation. SHAP (SHapley Additive ExPlanations) analysis was applied for increasing the interpretability of model predictions as well as determining the relative importance of variables. It was observed that long-term trends in solid precipitation (statistically insignificant in Ağrı and Kaçkar; significant decrease in Süphan) are not as critical as temperature in determining glacier dynamics. The hybrid model built showed robust and reliable performance with high accuracy (> 99%) and explainability (R2 >0.99) and confirmed that variables such as temperatures (lagged temperatures, moving averages) have the most dominating effect on model predictions. The findings unambiguously indicate that the water resources management strategies in Türkiye need to prioritize the reduction of the impacts of the rising temperatures.