Reconstruction of groundwater level at Kumamoto, Japan by means of deep learning to evaluate its increase by the 2016 earthquake

K Yokoo,Y Kawagoshi,T Nagasato,H Ito,K Ishida

doi:10.1088/1755-1315/851/1/012032

K Yokoo, Y Kawagoshi + Show 3 more

Open Access

https://doi.org/10.1088/1755-1315/851/1/012032

Copy DOI

Abstract

Groundwater is a very important water resource at Kumamoto City. Kumamoto City is the capital city of Kumamoto Prefecture, which is located in the Kyushu region, Japan. All domestic water is obtained from groundwater in Kumamoto City. Modeling groundwater is a difficult issue. Conditions under the ground are complex, and difficult to be obtained. Even the delineation of a groundwater basin is frequently unknown. Nowadays, deep learning is a hot topic in many research fields including geoscience. A recurrent neural network (RNN) is a type of deep learning that is suitable for time series modeling. Then, it has been successfully applied for groundwater modeling. Therefore, this study utilized a new type of RNN, Long and Short-Term Memory (LSTM) network, to model groundwater level at a monitoring well within Kumamoto City. The results in this study showed good agreement with the observed groundwater. In addition, it is known that severe earthquakes in April 2016 affected the groundwater level around Kumamoto City. The groundwater level model by LSTM was also utilized to estimate the effects of the severe earthquakes on the groundwater level. The results indicated that the earthquakes may have increased the groundwater level at Kumamoto City by more than 3 m.

Highlights

Groundwater is a very important water resource at Kumamoto City
Long and Short-Term Memory network The Long and Short-Term Memory (LSTM) network is a kind of recurrent neural network (RNN)
LSTM with the four layers showed the best accuracy with respect to Nash-Sutcliffe efficiency (NSE) and room mean square error (RMSE)

Summary

Introduction

Groundwater is a very important water resource at Kumamoto City. Kumamoto City is the capital city of Kumamoto Prefecture, which is located in the Kyushu region, Japan (Figure 1). Approximately 80% of domestic water is obtained from groundwater in Kumamoto Prefecture. The developed model is utilized to reconstruct groundwater level after the earthquakes. 2. Study area and Data This study focuses on groundwater level at a monitoring well within Kumamoto City. The groundwater around Kumamoto City clearly increased, which was caused by water release due to ruptures at the surrounding mountain aquifers accordint to Hosono et al (2020) [21]. It is expected that a deep learning method can reflect effects of air temperature on the groundwater level thorugh evapotranspiration. These data were obtained at the Kumamoto meteorological station from a website operated Japan Meteorological Agency (https://www.jma.go.jp/jma/). The meteorological data are available from 1891 to the present

Long and Short-Term Memory network The Long and Short-Term

Application and model configuration

Results and Discussions

Conclusions