Applicability of recurrent neural networks to retrieve missing runoff records: challenges and opportunities in Turkey


Alsavaf Y., Teksoy A.

ENVIRONMENTAL MONITORING AND ASSESSMENT, cilt.194, sa.1, 2022 (SCI-Expanded) identifier identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 194 Sayı: 1
  • Basım Tarihi: 2022
  • Doi Numarası: 10.1007/s10661-021-09681-z
  • Dergi Adı: ENVIRONMENTAL MONITORING AND ASSESSMENT
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, ABI/INFORM, Agricultural & Environmental Science Database, Aqualine, Aquatic Science & Fisheries Abstracts (ASFA), BIOSIS, CAB Abstracts, Compendex, EMBASE, Environment Index, Food Science & Technology Abstracts, Geobase, Greenfile, MEDLINE, Pollution Abstracts, Public Affairs Index, Veterinary Science Database, Civil Engineering Abstracts
  • Anahtar Kelimeler: Artificial neural networks, Runoff, Deep learning, River, Missing data, SERIES
  • Bursa Uludağ Üniversitesi Adresli: Evet

Özet

Acquiring river flow records is the primary precondition for providing optimal water resource management practices and preserving the ecohydrological balance. In Turkey, some river gauging stations go intermittently out of service due to some technical problems or unexpected difficulties. Consequently, river flow records are lost, and this is especially true in the rural and remote areas of the nation. In this regard, we investigated the ability of recurrent neural networks (RNN) as a supportive approach for retrieving daily river flow data for some stations, namely, Akcasehir and Dagguney located on the Susurluk basin in Bursa, Turkey. To meet the study goal, flow records from nearby stations were collected. In addition, a RNN with two hidden layers was developed. Initially, the model was trained and validated; then, we tried to predict missing records. The findings showed the potential of RNN in providing good predictions during low and mid flows with acceptable uncertainties rate (less than 30%) even with limited number of input data. Moreover, we discussed the current challenges and opportunities of this issue in remote areas of Turkey. Overall, our findings suggested that RNN may be considered as a practical method to predict the likely periods of floods and droughts in remote areas and interpolate missing records instead of using classical approaches.