Comparative assessment of univariate and multivariate LSTM models for hot day prediction
Meteorology and Atmospheric Physics, cilt.138, sa.4, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 138 Sayı: 4
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s00703-026-01155-6
- Dergi Adı: Meteorology and Atmospheric Physics
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Environment Index, Geobase, INSPEC, Natural Science Collection (ProQuest), Earth, Atmospheric, & Aquatic Science Collection (ProQuest), Technology Collection (ProQuest)
- Çukurova Üniversitesi Adresli: Evet
Özet
Estimating the frequency of hot days—defined in this study as days with maximum temperatures exceeding a fixed threshold of 35 °C—has become increasingly important in the context of climate change. This threshold is adopted to ensure consistency in global-scale analysis, although hot day definitions may vary across regions depending on local climatological conditions. This study comparatively evaluates the performance of univariate and multivariate Long Short-Term Memory (LSTM) models for estimating hot day frequency under different climate scenarios. In the multivariate framework, global monthly CO₂ concentrations derived from Shared Socioeconomic Pathways (SSP) projections up to 2100 are incorporated as an exogenous variable. Model performance is assessed using standard statistical metrics, including mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), and the correlation coefficient (R). The results show that the multivariate LSTM model captures the overall data structure more effectively. It achieves a higher correlation coefficient during training. In contrast, the univariate model produces slightly lower point-wise errors during testing. Visualization tools, including scatter plots, box plots, violin plots, and error histograms, reveal that multivariate models provide improved generalization and distributional fidelity. Future projections indicate a significant increase in hot day frequency under high-emission scenarios such as SSP5–8.5, with annual hot day totals surpassing 140 by the end of the century. In contrast, low-emission scenarios (e.g., SSP1–1.9) suggest a relative stabilization around mid-century. These findings underscore the importance of integrating climate forcing variables into projection models and highlight the trade-off between computational simplicity and scenario sensitivity.