Comparative evaluation of flood season water level prediction accuracy in the Yudeungcheon and Gapcheon streams based on LSTM training data configuration methods

DongHo Kang; Il Hong; JiSung Kim

doi:10.3741/JKWRA.2026.59.7.725

All Issue

2026 Vol.59, Issue 7 Preview Page Next Page

Comparative evaluation of flood season water level prediction accuracy in the Yudeungcheon and Gapcheon streams based on LSTM training data configuration methods LSTM 기반 학습데이터 구성 방법에 따른 유등천, 갑천 홍수기 수위예측 정확도 비교·평가

31 July 2026. pp. 725-738

PDF XML

Abstract

As extreme rainfall events intensify due to climate change, the need for rapid and accurate flood-stage water level prediction in small and medium rivers has become increasingly critical. This study develops Long Short-Term Memory (LSTM)-based deep learning models for four water level stations (Boksugyo, Inchanggyo, Mannyeongyo, and Wonchongyo) located along Yudeungcheon and Gapcheon, tributaries of the Geum River, and comparatively evaluates prediction accuracy according to different training data set construction methods. Three training data configurations were examined: complete observed water level records spanning the entire observation period (Case 1), records excluding 12-hour dry periods (Case 2), and an adjusted data set with reduced upstream station data (Case 3). While Case 1 yielded higher NSE values over the full 2025 evaluation period, Case 2 demonstrated superior performance during peak flood events. For Mannyeongyo, Case 2 achieved an NSE of 0.94 and RMSE of 0.06 m at a 1-hour lead time, compared to NSE of 0.86 and RMSE of 0.09 m for Case 1. For Boksugyo, Case 2 attained an NSE of 0.99 and RMSE of 0.03 m at a 1-hour lead time. Furthermore, the adjusted training data set (Case 3) at Mannyeongyo reduced peak water level error by up to 0.35 m relative to Case 2, achieving peak errors within 0.04 m even at a 4-hour lead time. The findings provide practical guidance on training data selection for LSTM-based flood forecasting models and are expected to contribute to the development of real-time AI-based flood warning systems.

Keywords

Machine learning

LSTM

Water level prediction

Training data types

기후변화로 인한 극한 강우 사상의 증가로 하천에서의 신속하고 정확한 홍수 수위 예측의 중요성이 높아지고 있다. 본 연구에서는 LSTM (Long Short-Term Memory) 기반 딥러닝 모형을 활용하여 금강의 지류하천인 유등천과 갑천에 위치한 4개 수위관측소(복수교, 인창교, 만년교, 원촌교)를 대상으로 학습데이터 구성 방법에 따른 홍수기 수위 예측 정확도를 비교 분석하였다. 학습데이터는 전체 관측 기간 수위 자료(Case 1), 12시간 무강우 기간을 제외한 수위 자료(Case 2), 상류 관측소 데이터를 조정한 수위 자료(Case 3)로 구성하였다. 2025년 전체 기간을 대상으로 한 분석에서는 Case 1의 NSE가 가장 높게 나타났으나, 최대 홍수 사상을 대상으로 한 분석에서는 Case 2가 우수한 예측 성능을 보였다. 만년교의 경우 1시간 후 예측 기준 Case 2의 NSE가 0.94, RMSE가 0.06 m였으며, Case 1의 NSE 0.86, RMSE 0.09m보다 개선되었다. 복수교의 경우 1시간 후 예측 기준 Case 2의 NSE가 0.99, RMSE가 0.03 m로 나타났다. 또한 만년교에 대해 상류 학습데이터를 조정한 Case 3은 Case 2 대비 Peak Error가 최대 0.35 m 개선되어, 4시간 후 예측에서도 0.04m 이내의 첨두 수위 오차를 달성하였다. 본 연구의 결과는 LSTM 기반 홍수 수위 예측 모형 개발 시 학습데이터 구성 방법의 선택 기준을 제시하는 기초 자료로 활용될 수 있으며, 향후 데이터 기반 수위 예측 모형의 정확도 향상 및 적용성 확대를 위한 연구에 기여할 수 있을 것으로 기대된다.

키워드

머신러닝

LSTM

수위예측

학습데이터 유형

References

An, S.W., Cho, H.M., Lee, B.H., and Kim, B.S. (2025). “Development of an Bi-LSTM model for flood level prediction of Hangang River Jamsoo Bridge considering weather climate data and dam discharge.” Crisisonomy, Vol. 21, No. 6, pp. 101-113.

10.14251/crisisonomy.2025.21.6.101

Bak, G.M., Oh, S.R., Park, G.H., and Bae, Y.C. (2021). “Groundwater Level Prediction using ANFIS Algorithm.” The Journal of The Korea Institute of Electronic Communication Sciences, Vol. 16, No. 6, pp. 1239-1248.

10.13067/JKIECS.2021.16.6.1239

Cho, W.J., and Kang D.S. (2020). “Estimation method of river water level using LSTM. Proceedings of the Korean Information Science Society Conference, Vol. 2020, No. 12, pp.439-441.

Hochreiter, S., and Schmidhuber, J. (1997). “Long Short-Term Memory.” Neural Computation, Vol. 9, No. 8, pp. 1735-1780.

10.1162/neco.1997.9.8.1735

Jung, J.W., Mo, H.L., Lee, J.H., Yoo, Y.H., and Kim, H.S. (2021). “Flood stage forecasting at the Gurye-Gyo station in Sumjin River using LSTM-Based deep learning models. Journal of the Korean Society of Hazard Mitigation,. Vol. 21, No. 3, pp. 193-201.

10.9798/KOSHAM.2021.21.3.193

Jung, S.H., Lee, D.E., and Lee, K.S. (2018a). “Prediction of river water level using deep-learning open library.” Journal of the Korean Society of Hazard Mitigation, Vol. 18, No. 1, pp. 1-11.

10.9798/KOSHAM.2018.18.1.1

Jung. S.H., Cho, H.S., Kim, J.Y., and Lee, G.H. (2018b). “Prediction of water level in a tidal river using a deep-learning based LSTM model.” Journal of Korea Water Resources Association, Vol. 51, No. 21, pp. 1207-1216.

Kim, H.I., Jang, S.D., Choi, H., Kim, T.H., and Kim, B. (2025). “Prediction of flood level using LSTM and watershed hydrological data.” Journal of Flood Risk Management, Vol. 18, e70123.

10.1111/jfr3.70123

Lee, M.J., Kim, J.S., Yoo, Y.H., Kim, H.S., Kim, S.E., and Kim, S.J. (2021). “Water level prediction in Taehwa River basin using deep learning model based on DNN and LSTM.” Journal of Korea Water Resources Association, Vol. 54, No. S-1, pp. 1061-1069.

Li, J., Wu, G., Zhang, Y., and Shi, W. (2024). “Optimizing flood predictions by integrating LSTM and physical-based models with mixed historical and simulated data.” Heliyon, Vol. 10, e33669.

10.1016/j.heliyon.2024.e3366939040386PMC11262550

Li, S., Wang, Z., Huang, B., Chen, D., Chen, L., Chen, J., Zhang, L., and Peng, H. (2025). “Does more data always help? Input configuration impacts on LSTM-based water level prediction.” Research Square, Preprint, doi:10.21203/rs.3.rs-7597625/v1.

10.21203/rs.3.rs-7597625/v1

Lim, Y.J., Kang, B.S., and Kim, H.K. (2025a). “Lead-time-dependent predictive performance of a MIMO-LSTM model for urbna river flood forecasting.” Korea Water Resources Association, Vol. 58, No. S-4, pp. 1629-1643.

Lim, Y.J., Kim, D.G., Ji, Y.E., Sung, J.Y., and Kang, B.S. (2025b). “Optimization of physical and hyperparameters for enhancing the response performance of an LSTM-based virtual flood level sensor.” Korea Water Resources Association, Vol. 58, No. 3, pp. 225-239.

Martel, J.L., Arsenault, R., Turcotte, R., Castañeda-Gonzalez, M., Brissette, F., Armstrong, W., Mailhot, E., Pelletier-Dumont, J., Lachance-Cloutier, S., Rondeau-Genesse, G., and Caron, L.P. (2025). “Exploring the ability of LSTM-based hydrological models to simulate streamflow time series for flood frequency analysis.” Hydrology and Earth System Sciences, Vol. 29, pp. 4951-4968.

10.5194/hess-29-4951-2025

Ministry of Land, Infrastructure and Transport (MOLIT) and Daejeon Regional Construction and Management Administration (DROCM) (2021). Basic river plan for the Gapcheon and Yudeungcheon River basin. No. 11-1613217-000053-01, pp. 51-52.

Moriasi. D.N., Arnold, J.G., Van Liew, M.W., Bingner, R.L., Harmel, R.D., and Veith, T.L. (2007). “Model evaluation guidlines for systematic quantification of accuracy in watershed simulations.” Transactions of the American Society of Agricultural and Biological Engineers, Vol. 50, No. 3, pp. 885-900.

10.13031/2013.23153

Oddo, P.C., Bolten, J.D., Kumar, S.V., and Cleary, B. (2024). “Deep Convolutional LSTM for improved flash flood prediction.” Frontiers in Water, Vol. 6, 1346104.

10.3389/frwa.2024.1346104

Presidential Commission on Climate Crisis Response (PCCCR) (2023). Press Release, 4th Plenary Meeting of the Carbon Neutral Green Growth Committee held, accessed 22 June 2023, <https://www.pcccr.go.kr/base/board/read?boardManagementNo=3&boardNo=2191&page=13&searchCategory=&searchType=&searchWord=&menuLevel=2&menuNo=7>.

Song, S.Y., Choi, Y.Y., Lee, B.J., Lim, Y.S., and Seo, Y.M. (2025). “Application of machine learning methods for groundwater level forecasting in the estern of Jeju Island.” Crisisonomy, Vol. 21, No. 2., pp. 13-23.

10.14251/crisisonomy.2025.21.2.13

Yang, H.S., Choi, Y.D., Kim, S.H., and Lee, J.S. (2026). “Development and performance evaluation of LSTM-based multi-points water level forecasting model for downstream region of Yongdam Dam.” Korea Water Resources Association, Vol. 59, No. 2, pp. 173-186.

10.3741/JKWRA.2026.59.2.173

Information

Publisher :KOREA WATER RESOURECES ASSOCIATION
Publisher(Ko) :한국수자원학회
Journal Title :Journal of Korea Water Resources Association
Journal Title(Ko) :한국수자원학회 논문집
Volume : 59
No :7
Pages :725-738
Received Date : 2026-04-03
Revised Date : 2026-05-19
Accepted Date : 2026-05-29
DOI :https://doi.org/10.3741/JKWRA.2026.59.7.725

Journal of Korea Water Resources Association ISSN:2799-8746(Print) 2799-8754(Online) 한국수자원학회 논문집

All Issue