Deep learning approach for prediction of water level in rivers

Takehiko Ito, Ryo Kaneko, Tomoya Kataoka, Shiho Onomura, Yasuo Nihei

Research output: Contribution to conferencePaperpeer-review

Abstract

In recent years, climate change has been responsible for many flood disasters. It is essential that protective measures should be developed against these. One of these measures is a flood forecasting system in which rising water-levels in rivers are predicted ahead of time. Although several researchers have applied artificial intelligence, and especially deep learning technology, to flood prediction, there is a lack of clarity regarding which deep learning approach is most effective in flood prediction. This study aimed to investigate the prediction of floods from water-level data by using a deep learning approach with data collected from the Kinu River. We adopted the LSTM (long short-term memory) algorithm, which is a type of recurrent neural network that readily reflects time-series data. In this study, we collected water-level data from five stations on the Kinu River, a branch of the Tone River, Japan. The results indicate that it is possible to utilize water-levels to predict flood events with a high level of accuracy.

Original languageEnglish
Publication statusPublished - 2020
Event22nd Congress of the International Association for Hydro-Environment Engineering and Research-Asia Pacific Division: Creating Resilience to Water-Related Challenges, IAHR-APD 2020 - Sapporo, Virtual, Japan
Duration: 14 Sep 202017 Sep 2020

Conference

Conference22nd Congress of the International Association for Hydro-Environment Engineering and Research-Asia Pacific Division: Creating Resilience to Water-Related Challenges, IAHR-APD 2020
Country/TerritoryJapan
CitySapporo, Virtual
Period14/09/2017/09/20

Keywords

  • Deep learning
  • Flood prediction
  • LSTM
  • Recurrent neural network
  • Water-level

Fingerprint

Dive into the research topics of 'Deep learning approach for prediction of water level in rivers'. Together they form a unique fingerprint.

Cite this