Hybrid CNN-LSTM surrogate modeling for rapid urban flood depth prediction

Mohammad, Suleiman; Vasudevan, Asokan; Hunitie, Mohammad Faleh Ahmmad; Erdonov, Bekmurod; Gafarova, Aziza; Turaev, Bobir

doi:10.66224/NMCE.2608.1170

Hybrid CNN-LSTM surrogate modeling for rapid urban flood depth prediction

Document Type : Research

Authors

Suleiman Mohammad ¹

Asokan Vasudevan ²

Mohammad Faleh Ahmmad Hunitie ³

Bekmurod Erdonov ⁴

Aziza Gafarova ⁵

Bobir Turaev ⁶

¹ Department of Business Administration, Business School, Al al-Bayt University, Mafraq 25113, Jordan. Research follower, INTI International University, 71800 Negeri Sembilan, Malaysia

² Faculty of Business and Communications, INTI International University, Persiaran Perdana BBN Putra Nilai, 71800 Nilai, Negeri Sembilan, Malaysia. School of Management, Department of Management Studies, Vel Tech Rangarajan Dr. Sagunthala RandD Institute of Science and Technology, University in Morai, Tamil Nadu 600062, India

³ Department of Public Administration, School of Business, University of Jordan, Jordan

⁴ Department of Information Technology and Exact Sciences, Termez University of Economics and Service, Termez, Uzbekistan

⁵ Department of Use of Hydromelioration Systems, Tashkent Institute of Irrigation and Agricultural Mechanization Engineers, National Research University, 100000 Tashkent, Uzbekistan

⁶ PhD, Associate Professor, Samarkand State Medical University, 140103 Samarkand, Uzbekistan

10.66224/NMCE.2608.1170

Abstract

This study was designed to propose and validate an alternative data-driven approach for urban flood depth prediction using machine learning technology. The target variable of the model is the flood depth at each observation point, and the model produces a spatial flood depth map of the urban watershed. The developed approach relies on a hybrid CNN-LSTM model that takes into account both spatial (topography, land use, and distance to channel) and temporal (precipitation hydrograph) patterns. In the process of training the model, 850 HEC-RAS models were used for 12,800 observation points in the Chirchik River basin of northern Tashkent, Uzbekistan, totaling 10.88 million flood depth data points. However, based on the calculated values for the coefficient of determination (0.938) and Nash-Sutcliffe (0.921), it can be noted that the hybrid model demonstrated better results than the basic CNN and LSTM, by 6% and 11.4%, respectively. The false negative value of the hybrid model is 5.8%, which is twice lower than 13.5% in case of the standalone LSTM. The inference time is 0.85 seconds for generating a complete flood depth map of the entire watershed under a single scenario, compared to 55 minutes when using HEC-RAS, results in a speedup of 3,882 times. This inference time includes the forward pass of the network and the extraction of the predicted flood depths at the 12,800 observation points by bilinear interpolation. Preprocessing and postprocessing steps are excluded. In the case of transferability experiments, it has been shown that the performance of the transfer learning algorithm was 0.861 in the adjacent watershed after fine-tuning with just 50 flood simulations, while the fully trained model had a Nash-Sutcliffe value of 0.902. Sensitivity analysis shows the significance of the influence of the training dataset size and observation point density on the performance. In conclusion, this study presents a conceptual model for the prediction of flood depths in developing cities using physical modeling and machine learning. However, some of the shortcomings of this model include the fact that it is trained and evaluated at 12,800 fixed observation points, and its accuracy at arbitrary locations has not been independently validated; it is based on synthetic data from HEC-RAS rather than actual flood events; and the fact that its transferability has been tested for one nearby watershed with similar properties.

Keywords

Surrogate model

flood risk

energy efficiency

Hybrid machine learning

CNN-LSTM

Subjects

Hydraulic Engineering

[1] Tellman, B., Sullivan, J.A., Kuhn, C., Kettner, A.J., Doyle, C.S., Brakenridge, G.R., Erickson, T.A., & Slayback, D.A. Satellite imaging reveals increased proportion of population exposed to floods. Nature, 596 (2021) 80–86. https://doi.org/10.1038/s41586-021-03695-w

[2] Babamuratov B., Jumaniyazov F., Sadullaev S., Yokutoy O., Saydullayeva S., Ulug'bek E., o'g'li A., Turgunov D., Global climate change and its impact on water resources: sustainable management strategies, Procedia Environmental Science, Engineering and Management, 12(3) (2025) 969-977.

[3] Zahura, F.T., Goodall, J.L., Sadler, J.M., et al. Training Machine Learning Surrogate Models from a High-Fidelity Physics-Based Model: Application for Real-Time Street-Scale Flood Prediction in an Urban Coastal Community. Water Resources Research, 56 (2020) e2019WR027038. https://doi.org/10.1029/2019WR027038

[4] Kim, H.I., Han, K.Y. Inundation Map Prediction with Rainfall Return Period and Machine Learning. Water, 12 (2020) 1552. https://doi.org/10.3390/w12061552

[5] An, H., Ouyang, C. A hybrid framework for real-time flash flood forecasting in small ungauged catchments: Integrating hydrodynamic simulations with LSTM networks. Journal of Hydrology, 661 (2025) 133688. https://doi.org/10.1016/j.jhydrol.2025.133688

[6] Xu, C., Zhong, P.A., Zhu, F., et al. A hybrid model coupling process-driven and data-driven models for improved real-time flood forecasting. Journal of Hydrology, 638 (2024) 131494. https://doi.org/10.1016/j.jhydrol.2024.131494

[7] Chen, G., Hou, J., Liu, Y., et al. Urban Inundation Rapid Prediction Method Based on Multi-Machine Learning Algorithm and Rain Pattern Analysis. Journal of Hydrology, 633 (2024) 131059. https://doi.org/10.1016/j.jhydrol.2024.131059

[8] Dang, T.Q., Tran, B.H., Le, Q.N., et al. Application of machine learning-based surrogate models for urban flood depth modeling in Ho Chi Minh City, Vietnam. Applied Soft Computing, 150 (2024) 111031.

[9] Garzón, A., Kapelan, Z., Langeveld, J. Machine learning-based surrogate modelling for Urban Water Networks: Review and future research directions. Water Resources Research, 58 (2022) e2021WR031808.

[10] Lingutla, V. K., Nagamani, N., & Brahmanandam, P. S. Inferences drawn on common scale parameter of two populations using rainfall data. Proceedings on Engineering, 7(1) (2025) 251-266.

[11] Woo, H., Choi, H., Kim, M., Noh, S.J. Physics-Guided Deep Learning for Spatiotemporal Evolution of Urban Pluvial Flooding. Water, 17 (2025) 1239.

[12] Xu, L., Gao, L. A hybrid surrogate model for real-time coastal urban flood prediction: An application to Macao. Journal of Hydrology, 642 (2024) 131863.

[13] Ronnaritivichai, B., Noichan, W., & Juntanintorn, R. The Application of Community Based Disaster Risk Management Approach in Flood-prone Community. Proceedings on Engineering Science, 6(1) (2024) 119-126. 10.24874/PES06.01.014

[14] Song, Y.M., Shen, C., Liu, X. A Surrogate Model for Shallow Water Equations Solvers with Deep Learning. Journal of Hydraulic Engineering, 149 (2023) 04023045.

[15] Karim, F., Armin, M.A., Ahmedt-Aristizabal, D., et al. A Review of Hydrodynamic and Machine Learning Approaches for Flood Inundation Modeling. Water, 15 (2023) 566. https://doi.org/10.3390/w15030566

[16] Oñate-Valdivieso F., Oñate-Paladines A., Rios A., Validation of flood routing techniques in an ANDEAN micro-basin, Proceedings on Engineering Sciences, 08(1) (2026) 445-448, 10.24874/PES08.01B.002.

[17] Xu, Y., Lin, K., Hu, C., et al. Deep Transfer Learning Based on Transformer for Flood Forecasting in Data-Sparse Basins. Journal of Hydrology, 625 (2023) 129956. https://doi.org/10.1016/j.jhydrol.2023.129956

[18] Li, J., Wu, G., Zhang, Y., et al. Optimizing Flood Predictions by Integrating LSTM and Physical-Based Models with Mixed Historical and Simulated Data. Heliyon, 10 (2024) e33669. https://doi.org/10.1016/j.heliyon.2024.33669

[19] Mechentel, E., Dairi, S., Lefkir, A., Eslamian, S., Abida, H., Djebbar, Y. Data-driven flood susceptibility assessment using hybrid machine learning and optimization techniques: case of the Sedrata Watershed, NE Algeria. Scientific Reports, 16(1) (2026) 16238. https://doi.org/10.1038/s41598-026-43262-9

[20] Chen, C.-C., Sung, W.-P., Chan, H.-C., Wang, P.-T. Sustainable Urban Water Management via AI Surrogate Modeling: A Multi-Dimensional Framework for Drainage Resilience and Microclimatic Assessment. Sustainability, 18(15) (2026) 7820. https://doi.org/10.3390/su18157820

[21] Zhang, J., Guo, W., Chang, S.W., Nguyen, D.D., Ngo, H.H. Data-Driven Innovations in Flood Hazard Assessment with Machine Learning. Earth: Environmental Sustainability, 1(1) (2025) 21–41. https://doi.org/10.53941/eesus.2025.100003

[22] Akinsoji, A.H., Adelodun, B., Adeyi, Q., et al. Integrating Machine Learning Models with Comprehensive Data Strategies and Optimization Techniques to Enhance Flood Prediction Accuracy: A Review. Water Resources Management, 38 (2024) 4735–4761. https://doi.org/10.1007/s11269-024-03885-x

[23] Perera, M.D.D., Angeloudis, A., Siripatana, A., Beevers, L. Hybrid methods in flood inundation modeling: a systematic review. Natural Hazards, 122(8) (2026) 357. https://doi.org/10.1007/s11069-026-08078-w

[24] Moishin, M., Deo, R.C., Prasad, R., et al. Designing Deep-Based Learning Flood Forecast Model with ConvLSTM Hybrid Algorithm. IEEE Access, 9 (2021) 50982–50993. https://doi.org/10.1109/ACCESS.2021.3065939

[25] Ni, C., Fam, P.S., Marsani, M.F. A Data-Driven Method and Hybrid Deep Learning Model for Flood Risk Prediction. International Journal of Intelligent Systems, 2024 (2024) 3562709. https://doi.org/10.1155/2024/3562709.

[26] Brunner, G. W. HEC-RAS River Analysis System – HEC-RAS User's Manual Version 6.6. US Army Corps of Engineers, Hydrologic Engineering Center (HEC), 2024. https://www.hec.usace.army.mil/software/hec-ras/documentation/HEC-RAS_Users_Manual_v6.6.pdf

[27] Brunner, G. W. HEC-RAS River Analysis System – 2D Modeling User's Manual Version 6.6. US Army Corps of Engineers, Hydrologic Engineering Center (HEC), 2024. https://www.hec.usace.army.mil/software/hec-ras/documentation/HEC-RAS_2D_Users_Manual_v6.6.pdf

[28] McKay, M. D., Beckman, R. J., & Conover, W. J. (1979). Comparison of Three Methods for Selecting Values of Input Variables in the Analysis of Output from a Computer Code. Technometrics, 21(2), 239–245. https://doi.org/10.1080/00401706.1979.10489755