Real-time reliability assessment and RMSE-based correction for LSTM flood forecasting models

Seungho Lee; Sooyoung Kim; Jaeyeon Lim; Kwang Seok Yoon

doi:10.3741/JKWRA.2025.58.12.1501

All Issue

2025 Vol.58, Issue 12 Preview Page

Research Article

Real-time reliability assessment and RMSE-based correction for LSTM flood forecasting models LSTM 홍수예측모형의 실시간 신뢰도 평가 및 RMSE 기반 보정기법 개발

31 December 2025. pp. 1501-1512

PDF XML

Abstract

This study proposes an RMSE-based continuous bias-correction method to evaluate real-time reliability and improve prediction accuracy of LSTM-based flood forecasting models. Although the Nash–Sutcliffe Efficiency (NSE) is effective for evaluating the overall fit of a complete time series, it often over-represents low-flow periods and fails to accurately reflect prediction performance during flood events. To address this limitation, the real-time prediction error RMSE was computed by integrating forecast errors across all lead times (10–360 min) at each prediction time. A real-time performance index was then defined based on the relative magnitude of RMSE compared with the reference error RMSE. In addition, a continuous error-correction equation incorporating lead-time-dependent prediction bias was applied to derive real-time corrected water-level forecasts. Application of the proposed method using observed and predicted water-level datasets demonstrated that RMSE decreased by approximately 1~5% at the Dayeuh Kolot station and by more than 10% during rapid rising-limb conditions at the San Mateo station. These results indicate that the proposed reliability-based correction method effectively improves prediction accuracy under diverse hydrological conditions. Owing to its lightweight structure, the method can be readily integrated as a post-processing module into real-time flood forecasting and warning systems using single-model LSTM architectures.

Keywords

Artificial intelligence

Flood forecasting and warning

Real-time performance index

Error correction

본 연구에서는 LSTM 기반 홍수예측모형의 실시간 신뢰도 평가 및 예측 정확도 향상을 위해 RMSE(Root Mean Squared Error)를 활용한 연속 편의 보정 기법을 제안하였다. 기존의 NSE(Nash–Sutcliffe Efficiency coefficiency)는 전체 시계열의 평균 적합도를 평가하는 데에는 유효하나, 저수위 구간의 영향이 과대하게 반영되어 홍수사상에 대한 예측 성능을 정확히 표현하지 못하는 한계가 존재한다. 이를 보완하기 위해 본 연구에서는 예측시점에서 예측선행시간별(10~360분)의 예측오차를 통합한 RMSE를 산정하고, 이를 활용하여 레퍼런스 RMSE를 산정하고 비교하여 실시간 예측 성능지표를 정의하였다. 또한 예측선행시간별 평균 편의(Bias)를 반영한 연속 오차보정식을 적용하여 실시간 보정 예측값을 산출하였다. 관측 자료와 예측 결과를 활용한 적용성 평가 결과, 인도네시아 Dayeuh Kolot 수위관측소 지점에서는 약 1~5% 정도의 RMSE값 감소가 있었고, 필리핀 San Mateo-1 수위관측소 지점에서는 급격한 수위 상승 구간에서 10% 이상의 감소가 확인되어, 제안된 기법이 다양한 수문학적 조건에서 예측 정확도를 개선시키는데 효과적임을 확인하였다. 본 연구에서 제안한 실시간 성능지표 기반 보정기법은 단일 LSTM 구조에 직접 적용 가능한 경량 후처리 방식으로, 향후 국내·외 홍수예경보 시스템의 실시간 예측 신뢰도 향상에 활용될 수 있다.

키워드

인공지능

LSTM

홍수예경보

실시간 성능지표

오차보정

References

Choi, Y.J., Kim, J.H., and Lee, D. (2023). “Real-time performance- index-based correction for short-term flood forecasting.” Journal of Hydrology, Elsevier, Vol. 620, No. 129, pp. 1-12.

Fang, K., Shen, C., Kifer, D., and Yang, X. (2021). “Flood vulnerability assessment using LSTM-based deep learning frameworks.” Environmental Modelling & Software, Elsevier, Vol. 144, No. 105, pp. 1-12.

Hu, H., Li, X., Chen, Q., and Zhang, Y. (2021). “Urban flood level forecasting using a hybrid CNN-LSTM model.” Water Resources Management, Springer, Vol. 35, No. 7, pp. 2315-2332.

Jiang, S., Yao, L., Wang, Z., and Li, Z. (2020). “LSTM-based rainfall- runoff modeling for diverse catchments.” Journal of Hydrology, Elsevier, Vol. 589, No. 125, pp. 1-14.

Jung, Y., Kim, S., and Kim, H. (2021a). “Comparison of LSTM, SVM, and MLP models for water level forecasting at Gurye station in the Seomjin River Basin, Korea.” KSCE Journal of Civil Engineering, Vol. 25, No. 7, pp. 2873-2884.

Jung, Y.J., Park, J.Y., and Kim, D.H. (2021b). “Application of LSTM for water level forecasting at the Gurye station of the Seomjin River.” Journal of Hydroinformatics, Vol. 23, No. 5, pp. 1020- 1032.

Kim, H.S., Lee, S.H., and Kwon, Y. (2022a). “Short-term water level forecasting using LSTM for the Wangsuokcheon basin.” KSCE Journal of Civil Engineering, KSCE, Vol. 26, No. 6, pp. 2531-2543.

Kim, J.H., Park, S.M., and Lee, S. (2023). “Deep learning-based flood forecasting for the CarayCaray watershed in Eastern Visayas, Philippines.” Journal of Hydro-Environment Research, IAHR, Vol. 48, No. 2, pp. 145-157.

Kim, S., Park, D., and Kim, H. (2022b). “Short-term water level prediction using LSTM for urban stream flooding: A case study of Wangsuk Stream, Korea.” Journal of Hydrology, Elsevier, Vol. 612, pp. 1-12.

Kim, S.Y., Han, G.B., and Park, H.J. (2021). “Lead-time dependent correction of water level forecasts in the Seolma Stream.” Journal of Korea Water Resources Association, KWRA, Vol. 54, No. 8, pp. 645-656.

Le, X.H., Ho, H.V., Lee, G., and Jung, S. (2019). “LSTM-based streamflow forecasting using rainfall and runoff time series.” Journal of Hydroinformatics, Vol. 21, No. 3, pp. 422-438.

Lee, K.M., Choi, Y.S., and Han, D. (2021). “Performance comparison between LSTM and a physics-based routing model for the Taehwa River.” Water Resources Management, Springer, Vol. 35, No. 9, pp. 3011-3026.

Olah, C. (2015). Understanding lstm networks, accessed 23 November 2021, <https://colah.github.io/posts/2015-08-UnderstandingLSTMs/>.

Park, S.J., and Kim, Y.J. (2020). “Implementation of LSTM-based real-time upstream water level prediction system.” Journal of Hydrologic Engineering, ASCE, Vol. 25, No. 3, pp. 1-12.

Tran, D., and Song, C. (2017). “Comparison of RNN and LSTM models for river water level prediction in the Trinity River.” Water Science and Technology, IWA Publishing, Vol. 75, No. 9, pp. 2219-2231.

Yoo, J.H., Choi, K.S., and Kim, T.W. (2019). “Water level forecasting at the Hangang Bridge using NARX neural networks.” KSCE Journal of Civil Engineering, KSCE, Vol. 23, No. 4, pp. 1621-1632.

Zhang, L., Wang, P., and Liu, Y. (2022). “Residual learning for bias correction in deep learning-based hydrological forecasting.” Hydrology and Earth System Sciences, EGU, Vol. 26, No. 11, pp. 2799-2813.

Information

Publisher :KOREA WATER RESOURECES ASSOCIATION
Publisher(Ko) :한국수자원학회
Journal Title :Journal of Korea Water Resources Association
Journal Title(Ko) :한국수자원학회 논문집
Volume : 58
No :12
Pages :1501-1512
Received Date : 2025-11-21
Revised Date : 2025-12-10
Accepted Date : 2025-12-11
DOI :https://doi.org/10.3741/JKWRA.2025.58.12.1501

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

All Issue