Meteorological drought modeling using recurrent neural network and long short-term memory

Danandeh Mehr A., Alkubaisi H., TÜR R.

Drought Dynamics: Modeling and Assessment to Confront a Changing Climate, Elsevier, ss.281-294, 2026 identifier

  • Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1016/b978-0-443-34205-9.00020-2
  • Yayınevi: Elsevier
  • Sayfa Sayıları: ss.281-294
  • Anahtar Kelimeler: drought, Iraq, LSTM, Machine learning, RNN, SPEI
  • Akdeniz Üniversitesi Adresli: Evet

Özet

This study aims to model meteorological drought phenomenon using recurrent neural network (RNN) and compare its efficiency with long short-term memory (LSTM). These models were trained using global SPEI data extracted for the central region of Iraq. We used 74 years of monthly data (1950-2024), 70% to train and 30% to test the models in which effective historical observations were considered as the forecasting inputs. The results prove both models’ efficiency for month ahead SPEI forecasting with a slightly superior performance of RNN. Numerical metrics showed that both models achieved acceptable accuracy (Nash--Sutcliffe Efficiency [NSE]>0.78) in predicting SPEI-6 values, with the three-layer RNN model achieving slightly higher accuracy (test NSE = 0.8214) than the LSTM model. The results suggest that simpler RNN architectures may perform better when the training data length is limited.