Hydrological time series prediction using neural architecture search-enhanced dual-stage attention-based Bi-LSTM

Kilinc H. C., Apak S., ÇITAKOĞLU H., Sammen S. S., Yurtsever A.

HYDROLOGICAL SCIENCES JOURNAL, 2026 (SCI-Expanded, Scopus) identifier

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1080/02626667.2026.2637779
  • Dergi Adı: HYDROLOGICAL SCIENCES JOURNAL
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, IBZ Online, Compendex, Geobase, INSPEC
  • Erciyes Üniversitesi Adresli: Evet

Özet

A Neural Architecture Search-enhanced Dual Stage Attention Bidirectional Long Short-Term Memory (NAS-DSA-BiLSTM) model is proposed to capture nonlinear temporal hydrological patterns and is evaluated at three hydrological stations. The NAS-enhanced model achieved the best performance at Kaptanpasa, with the lowest MSE (11.249-16.224) and the highest R2 values, outperforming LSTM and DNN-LSTM models. Additional evaluations using NSE, sensitivity analysis, and Taylor diagrams confirmed stable and accurate predictions, particularly at Ulucami (NSE = 0.91) and Kaptanpasa (NSE = 0.80), demonstrating the model's robustness for streamflow forecasting. The proposed model improved MSE by 15-26%, demonstrating robust and reliable streamflow forecasting performance.