Prediction of the standardized precipitation index based on the long short-term memory and empirical mode decomposition-extreme learning machine models: The Case of Sakarya, Türkiye

Coşkun Ö., ÇITAKOĞLU H.

Physics and Chemistry of the Earth, cilt.131, 2023 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 131
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1016/j.pce.2023.103418
  • Dergi Adı: Physics and Chemistry of the Earth
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Aquatic Science & Fisheries Abstracts (ASFA), Artic & Antarctic Regions, Chimica, Communication Abstracts, Compendex, Geobase, INSPEC, Metadex, Pollution Abstracts, Civil Engineering Abstracts
  • Anahtar Kelimeler: Drought prediction, Empirical mode decomposition (EMD), Extreme learning machine (ELM), Long short-term memory (LSTM), Meteorological drought, Standardized precipitation index (SPI), Türkiye
  • Erciyes Üniversitesi Adresli: Evet

Özet

This research predicted the meteorological drought of Sakarya province in northwest Türkiye using long short-term memory (LSTM). This deep learning algorithm has gained popularity in prediction studies. The standardized precipitation index (SPI), which can only be derived using precipitation data, was utilized for 1, 3, and 6-month time scales. SPI-1, SPI-3, and SPI-6-month time scales drought data calculated from the monthly precipitation data of the Sakarya Meteorology Station between 1960 and 2020 were taken as input data in the LSTM model. SPI drought data were used between 1960 and 2005 as training data and 2006–2020 as test data. Drought at t+1 output time was predicted using SPI values at t, t-1, t-2, and t-3 lag times as input variables. In addition, the results were compared with the empirical mode decomposition (EMD)-extreme learning machine (ELM) hybrid model to understand the capabilities of the standalone LSTM prediction model. The LSTM model yielded the best results (MAE = 0.11, NSE = 0.97, R2 = 0.97) for the SPI-1-month time scale and the best results (MAE = 0.18, NSE = 0.92, R2 = 0.94) for 3-month time scale. The EMD-ELM hybrid model yielded the best results (MAE = 0.22, NSE = 0.95, R2 = 0.96) for the SPI-6-month time scale. Due to the high performance of this study's proposed standalone LSTM model, it was concluded that drought time series do not need to be subjected to pre-processing techniques.