Sea surface temperature prediction model for the Black Sea by employing time-series satellite data: a machine learning approach

Aydinli H. O., EKİNCEK A., Aykanat-Atay M., Saritas B., Ozenen-Kavlak M.

APPLIED GEOMATICS, cilt.14, ss.669-678, 2022 (ESCI) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 14
  • Basım Tarihi: 2022
  • Doi Numarası: 10.1007/s12518-022-00462-y
  • Dergi Adı: APPLIED GEOMATICS
  • Derginin Tarandığı İndeksler: Emerging Sources Citation Index (ESCI), Scopus, Aerospace Database, Communication Abstracts, Geobase, Metadex, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.669-678
  • Anahtar Kelimeler: Sea surface temperature, Prediction, Black Sea, Satellite images, Machine learning, TROPICAL PACIFIC, VARIABILITY, SST, FORECASTS
  • Anadolu Üniversitesi Adresli: Evet

Özet

High temporal resolution remote sensing images provide continuous data about the marine environment, which is critical for gaining extensive knowledge about the aquatic environment and marine species. Sea surface temperature (SST) is one of the basic parameters that can be obtained with the help of remote sensing. Long-term alterations in the SST can affect the aquatic environment and marine species, such as the life expectancy of anchovies in the Black Sea. Forecasting the dynamics of SSTs is crucial for detecting and eliminating the SST-oriented impacts. The goal of the current study is to construct a predictive model to estimate the daily SST value for the mid-Black Sea using a machine learning approach by employing time-series satellite data from 2008 to 2021. Turkey's mid-Black Sea coastal line, comprising Ordu, Samsun, and Sinop stations, was chosen as the study area. The SST predictive model was represented by applying the recurrent neural network (RNN) long- and short-term memory (LSTM). Adam stochastic optimization was used for validation, and the mean square error (MSE) for each location was found to be 0.914, 0.815, and 0.802, respectively. The findings indicate that our model is significantly promising for accurate and effective short- and midterm daily SST prediction.