LSTM ve Hibrit CNN–LSTM Derin Öğrenme Yaklaşımları ile Isparta İli İçin Zaman Serisi Tabanlı Sıcaklık Tahmini

Öz

Günlük hava sıcaklığı, tarım, enerji, sağlık ve su yönetimi gibi pek çok sektörü doğrudan etkileyen kritik bir meteorolojik parametredir. Küresel iklim değişikliğiyle birlikte sıcaklık rejimlerindeki dalgalanmalar ve uzun vadeli artışlar, bölgesel tahmin modellerinin geliştirilmesini daha da önemli hale getirmiştir. Bu çalışmada, Isparta ili için günlük hava sıcaklığı tahmini amacıyla dört farklı modelin (XGBoost, LSTM, CNN ve CNN–LSTM) karşılaştırmalı analizi gerçekleştirilmiştir. Geçmiş yıllara ait günlük sıcaklık verileri kullanılarak yapılan tahminlerde, modeller hem eğitim hem de test setleri üzerinde MAE, MSE, R², NSE ve Willmott d gibi istatistiksel ölçütlerle değerlendirilmiştir. Elde edilen sonuçlar, test verisi üzerinde LSTM modelinin en yüksek doğruluğu sunduğunu göstermiştir (MAE 1,139, MSE 2,334, R² 0,966, NSE 0,965, Willmott d 0,991). CNN–LSTM modeli ise LSTM’e çok yakın değerlerle (MAE 1,236, MSE 2,661, R² 0,962, NSE 0,960, Willmott d 0,990) istikrarlı bir alternatif olarak öne çıkmıştır. CNN modeli rekabetçi performans sergilemiş (MAE 1,228, MSE 2,688, R² 0,961, NSE 0,960, Willmott d 0,989), XGBoost modeli ise diğer modellere kıyasla daha zayıf kalmıştır (MAE 2,526, MSE 10,63, R² 0,855, NSE 0,843, Willmott d 0,960). Genel olarak, LSTM modeli uzun vadeli bağımlılıkları yakalama başarısıyla öne çıkarken, CNN–LSTM modeli kısa dönemli örüntüleri de dikkate alarak güvenilir ve kararlı bir tahmin yaklaşımı sunmuştur. Bu bulgular, derin öğrenme modellerinin yerel ölçekli sıcaklık tahminlerinde karar vericilere daha doğru öngörüler sağlayabileceğini ortaya koymaktadır.

Anahtar Kelimeler

• CNN
• Derin öğrenme
• LSTM
• Sıcaklı Tahmini
• XGBoost
• Zaman Serisi

Time Series–Based Temperature Forecasting for Isparta Province Using LSTM and Hybrid CNN–LSTM Deep Learning Approaches

Öz

Temperature and its variations play a crucial role in agriculture, energy consumption, and urban planning. Accurately forecasting air temperature is essential for sustainable resource management and climate adaptation strategies. In this study, four different models—XGBoost, LSTM, CNN, and the hybrid CNN–LSTM—were applied for daily temperature prediction in Isparta, a region sensitive to climatic fluctuations. The performance of the models was evaluated using statistical metrics including MAE, MSE, R², NSE, and Willmott’s d. The results show that the LSTM model achieved the highest accuracy on the test data, with a mean absolute error of 1.14 °C, mean squared error of 2.33, coefficient of determination of 0.966, Nash–Sutcliffe efficiency of 0.965, and Willmott’s d index of 0.991. The CNN–LSTM model provided very close results, with a mean absolute error of 1.23 °C, mean squared error of 2.66, coefficient of determination of 0.962, Nash–Sutcliffe efficiency of 0.960, and Willmott’s d index of 0.990, demonstrating a stable predictive ability. The CNN model also produced competitive results, while the XGBoost algorithm showed weaker performance with higher error values and lower generalization capability. These findings highlight that LSTM is most effective in capturing long-term dependencies, whereas the CNN–LSTM hybrid offers a robust alternative capable of modeling both short-term patterns and long-term dynamics. Thus, both models can be considered valuable tools for local meteorological forecasting and climate-aware decision-making.

Anahtar Kelimeler

• CNN
• Deep learning
• LSTM
• Temperature forecasting
• XGBoost
• Time series

Kaynakça

1. Aksay, E., Çoban, E., & Güçlü, Y. S. (2025). Normalized innovative trend analysis model and Mann-Kendall test for solar data. Modeling Earth Systems and Environment, 11(5), Article 347. https://doi.org/10.1007/s40808-025-02550-5
2. Alizamir, M. (2025). Daily soil temperature prediction using hybrid deep learning and SHAP for sustainable soil management. International Journal of Green Energy. Advance online publication. https://doi.org/10.1080/19942060.2025.2541686
3. Astsatryan, H., Hakobyan, G., Harutyunyan, A., Poghosyan, M., & Kostanyan, H. (2021). Air temperature prediction using artificial neural networks for the Ararat Valley. Earth Science Informatics, 14, 2359–2372. https://doi.org/10.1007/s12145-021-00583-9
4. Brownlee, J. (2019). XGBoost With Python: Gradient Boosted Trees with XGBoost and scikit-learn. Machine Learning Mastery.
5. Beloev, H. I., Saitov, S. R., Filimonova, A. A., Chichirova, N. D., Babikov, O. E., & Iliev, I. K. (2025). Short-Term Electrical Load Forecasting Based on XGBoost Model. Energies, 18(19), 5144. https://doi.org/10.3390/en18195144
6. Buratto, W. G., Muniz, R. N., Nied, A., Barros, C. F. D. O., Cardoso, R., & Gonzalez, G. V. (2024). Wavelet CNN‐LSTM time series forecasting of electricity power generation considering biomass thermal systems. IET Generation, Transmission & Distribution, 18(21), 3437-3451.https://doi.org/10.1049/gtd2.13292
7. Chen, T., & Guestrin, C. (2016, August). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785-794).https://doi.org/10.1145/2939672.2939785
8. Cifuentes, J., Marulanda, G., Bello, A., & Reneses, J. (2020). Air temperature forecasting using machine learning techniques: a review. Energies, 13(16), 4215. https://doi.org/10.3390/en13164215

9. Çoban, E. (2024). Makine öğrenmesi algoritmaları ile yaz sezonu ortalama akım değerlerinin tahmini. Journal of Innovations in Civil Engineering and Technology, 6(2), 73–81. https://doi.org/10.60093/jiciviltech.1497771
10. Fang, Z., Yang, S., Lv, C., An, S., & Wu, W. (2022). Application of a data-driven XGBoost model for the prediction of COVID-19 in the USA: A time-series study. BMJ Open, 12(2), e056685. https://doi.org/10.1136/bmjopen-2021-056685
11. Gao, P., Li, D., Liu, Z., Li, W., & Yang, X. (2023). NDVI prediction based on time series decomposition and CNN–LSTM combination. Water Resources Management, 37, 1089–1106. https://doi.org/10.1007/s11269-022-03419-3
12. Henderi, H., Wahyuningsih, T., & Rahwanto, E. (2021). Comparison of Min-Max normalization and Z-Score Normalization in the K-nearest neighbor (kNN) Algorithm to Test the Accuracy of Types of Breast Cancer. Indonesian Journal of Information and Information Systems, 4(1), 1–6. https://doi.org/10.47738/ijiis.v4i1.73
13. Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
14. Hou, J., Li, L., Wang, R., & Wang, Z. (2022). Hourly temperature prediction using CNN–LSTM. Geomatics, Natural Hazards and Risk, 13(1), 1138–1157. https://doi.org/10.1080/19475705.2022.2102942
15. Hou, W., Zhang, X., & Wang, Y. (2022). Hourly air temperature prediction using CNN–LSTM hybrid deep learning model: A case study in Yinchuan, China. Atmosphere, 13(8), 1269. https://doi.org/10.1080/19475705.2022.2102942
16. Kalkınç, H. Y., Çoban, E., & Güçlü, Y. S. (2025). Wilcoxon testi ve saçılma diyagramı kullanılarak hidrometeorolojik verilerin trend analizi. Turkish Journal of Civil Engineering, 36(5), 1-31. https://doi.org/10.18400/tjce.1630037
17. Karevan, Z., & Suykens, J. A. K. (2020). Transductive LSTM for time-series prediction: An application to weather forecasting. Neural Networks, 125, 1–9. https://doi.org/10.1016/j.neunet.2019.12.030
18. Kim, T. Y., & Cho, S. (2019). Predicting residential energy consumption using CNN-LSTM neural networks. Energy, 182, 72–81. https://doi.org/10.1016/j.energy.2019.05.230
19. LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278–2324. https://doi.org/10.1109/5.726791
20. Mutlu, A. Comparative evaluation of NASA, ERA5, and observational data for accuracy and reliability. Theor Appl Climatol 156, 367 (2025). https://doi.org/10.1007/s00704-025-05605-w
21. Nelson, D. M. Q., Pereira, A. M., & de Oliveira, R. A. (2017). Stock market's price movement prediction with LSTM neural networks. In 2017 International Joint Conference on Neural Networks (IJCNN) (pp. 1419–1426). https://doi.org/10.1109/IJCNN.2017.7966019
22. Nielsen, D. (2016). Tree boosting with XGBoost: Why does XGBoost win “every” machine learning competition? Master’s thesis, NTNU. http://hdl.handle.net/11250/2433761
23. Oord, A. V. D., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., ... & Kavukcuoglu, K. (2016). WaveNet: A generative model for raw audio. arXiv preprint arXiv:1609.03499. https://arxiv.org/abs/1609.03499
24. Saplıoğlu, K., & Çimen, M. (2010). Yapay sinir ağlarını kullanarak günlük yağış miktarının tahmini. Mühendislik Bilimleri ve Tasarım Dergisi, 1(1), 14–21.
25. Utku, A. (2024). CNN–LSTM hibrit modeli ile Bingöl ve Tunceli için hava sıcaklığı tahmini. International Journal of Pure and Applied Sciences, 10(2), 550–562. https://doi.org/10.29132/ijpas.1548698
26. Xiao, C., Liu, X., & Zeng, Z. (2019). Sea surface temperature prediction using LSTM-AdaBoost combination with time series satellite data. Remote Sensing of Environment, 233, 111358. https://doi.org/10.1016/j.rse.2019.111358
27. Yan, H., He, Z., Gao, C., Xie, M., Sheng, H., & Chen, H. (2022). Investment estimation of prefabricated concrete buildings based on XGBoost machine learning algorithm. Advanced Engineering Informatics, 54, 101789. https://doi.org/10.1016/j.aei.2022.101789
28. Yilmaz, M. (2023). Accuracy assessment of temperature trends from ERA5 and ERA5-Land. Science of The Total Environment, 856, 159182. https://doi.org/10.1016/j.scitotenv.2022.159182
29. Yu, Y., Si, X., Hu, C., & Zhang, J. (2019). A review of recurrent neural networks: LSTM cells and network architectures. Neural computation, 31(7), 1235-1270. https://doi.org/10.1162/neco_a_01199
30. Zhang, P., Jia, Y., & Shang, Y. (2022). Research and application of XGBoost in imbalanced data. International Journal of Distributed Sensor Networks, 18(6), 15501329221106935. https://doi.org/10.1177/15501329221106935