Times are displayed in (UTC-06:00) Central Time (US & Canada) Change

77-12 Deep Learning Approach for Groundwater Level Prediction in Vietnam

Session: Groundwater and Sustainability: Integrating Science, Technology, and Policy

Presenting Author:

Linh Bui

Authors:

Bui, Linh¹, Bui, Duong², Bhattacharya, Prosun³, Winkel, Lenny⁴, Bui, Nuong⁵, Nguyen, Hoa⁶, Vu, Anh⁷, Bui, Dan⁸, Bui, Du⁹

(1) Research Department, Southeast Asia Union for Water, Environment and Geosciences – SEAGU, Hanoi, (None), Vietnam; The Vietnam Chapter of the International Society of Groundwater for Sustainable Development, Hanoi, (None), Vietnam, (2) Research and International Cooperation Department, National Center for Water Resources Planning and Investigation - NAWAPI, Hanoi, (None), Vietnam; Southeast Asia Union for Water, Environment and Geosciences – SEAGU, Hanoi, (None), Vietnam, (3) Department of Sustainable Development, KTH ROYAL INSTITUTE OF TECHNOLOGY, Stockholm, Sweden, (4) Department of Environmental Systems Science, ETH Zurich, Zurich, Switzerland, (5) Faculty of Environment, Hanoi University of Natural Resources & Environment (HUNRE), Hanoi, Vietnam, (6) Department of Environmental Engineering, Hanoi University of Mining and Geology, Hanoi, Vietnam, (7) Hanoi University of Mining and Geology, Hanoi, Vietnam, (8) Southeast Asia Union for Water, Environment and Geosciences – SEAGU, Hanoi, Vietnam, (9) Southeast Asia Union for Water, Environment and Geosciences – SEAGU, Hanoi, Vietnam,

Abstract:

This study investigates the application of a Long Short-Term Memory (LSTM) model for multi-day-ahead groundwater level forecasting. Utilizing precipitation data from GM_force and temperature data from the ERA5 reanalysis dataset, we developed and trained models for three distinct monitoring stations: Q.176a, C3a, and LK75T. The models were trained on 180 days of historical data to forecast groundwater levels for lead times of 1 to 7 days. Performance evaluation, based on the Nash-Sutcliffe Efficiency (NSE) coefficient, revealed that the C3a station model yielded the best overall performance, with NSE values between 0.715 and 0.652 across the forecast horizon. While the Q.176a model demonstrated the highest accuracy for a 1-day lead (NSE = 0.724), its performance substantially decreased at the 7-day lead time (NSE = 0.222). Conversely, the LK75T model exhibited moderate yet consistent performance, with NSE scores ranging from 0.540 to 0.399 for all lead times. The findings confirm the utility of ERA5 meteorological data for multi-day groundwater forecasting and suggest that the LK75T model provides the most reliable and stable performance for operational applications among the tested sites.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-8301

Deep Learning Approach for Groundwater Level Prediction in Vietnam

Description

Session Format: Oral

Presentation Date: 10/20/2025

Presentation Start Time: 11:05 AM

Presentation Room: HBGCC, 210AB

Back to Session