Precipitation Forecasting in the Konya Closed Basin Using LSTM: A Comparative Analysis of Optimization Algorithms

Demir V., Carbas S., Candan B., Sevimli M. F., ÇITAKOĞLU H.

Studies in Systems, Decision and Control, Springer International Publishing Ag, ss.369-388, 2026 identifier

  • Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1007/978-3-032-07738-7_18
  • Yayınevi: Springer International Publishing Ag
  • Sayfa Sayıları: ss.369-388
  • Anahtar Kelimeler: Deep learning, Konya closed basin, Long short-term memory, Optimization algorithms, Precipitation forecasting
  • Erciyes Üniversitesi Adresli: Evet

Özet

Konya Closed Basin (KCB), one of the driest regions in Türkiye, holds significant importance in terms of sustainable water resource management. Accordingly, accurate precipitation forecasting plays a vital role in water budgeting, agricultural planning, and climate adaptation strategies. In this chapter, monthly precipitation data spanning 58 years (1967–2024) from the KCB were used to perform precipitation forecasting using the Long Short-Term Memory (LSTM) model, a deep learning approach specifically designed for time-series data. The analysis was based on data collected from 11 meteorological stations located within the basin. The dataset was divided into 75% for training and 25% for testing. During the modeling process, a comparative analysis was conducted using three optimization algorithms: Adaptive Moment Estimation (ADAM), Root Mean Square Propagation (RMSProp), and Stochastic Gradient Descent with Momentum (SGDM). Model performance was evaluated based on Root Mean Square Error (RMSE), Mean Absolute Error (MAE), the Coefficient of Determination (R2), and the Mean of Min Over Max Error (MMME). The results revealed that the choice of optimization algorithm significantly affects prediction accuracy. The ADAM algorithm yielded the best overall performance, followed by SGDM and RMSProp, both of which also produced comparably reliable results, confirming their applicability for precipitation forecasting tasks.