Detail Abstract

An Application of Long-Short Term Memory (LSTM) Neural Network for Groundwater Forecasting. A Case Study in Red River Delta

Abstract

Groundwater (GW) plays an important role in supplying water for household activities, Industry and Agriculture. However, GW is being explored without being aware about both quantity and quality reduction under the impact of economic development and urbanization. In this study, authors apply Mann Kendal method to assess groundwater trend and Long-Short Term Memory (LSTM) Neural Network to forecast groundwater level. As a result, there is a station at lower Red River Delta (RRD) is chosen to analysis the GW trend detection for both Holocene Unconfined Aquafer (HUA) and Pleistocene Confined Aquafer (PCA) from 1995-2020 and predict GW level for 5 lead days. At lower RRD, GW tends to downward significantly at PCA whereas HUA performs seasonal change. For forecasting, Coefficient Nash Sutcliffe Efficiency (NSE), and Root Mean Squared Error (RMSE) are the criteria indexes is used to assess the correlation between in-situ data and the predict results. It indicates that at both training and testing section, NSE and RMSE show the impressive performance with around 1 and 0.1m at training section, and 0.99 and 0.12m at testing part, respectively.