EGU24-121, updated on 08 Mar 2024
https://doi.org/10.5194/egusphere-egu24-121
EGU General Assembly 2024
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

A process-data duality driven hybrid model for improving flood forecasting

Chengjing Xu1, Ping-an Zhong1,2, Feilin Zhu1, Bin Xu1,3, Yiwen Wang1, Luhua Yang1, Sen Wang1, and Sunyu Xu1
Chengjing Xu et al.
  • 1College of Hydrology and Water Resources, Hohai University, Nanjing, China (xucj1997@gmail.com)
  • 2National Engineering Research Center of Water Resources Efficient Utilization and Engineering Safety, Hohai University, Nanjing, China
  • 3Cooperative Innovation Center for Water Safety & Hydro Science, Nanjing 210098, China

Floods are the most destructive events among natural disasters that restrict national economic development and threaten the safety of human lives. Accurate and efficient flood forecasting plays an important role in flood warning, flood risk analysis, and reservoir operation. Traditional flood forecasting tools provide fixed-value predictions. However, due to the complexity of reality and the limitations of human cognition, many inherent uncertainties are inevitably ignored. Therefore, it is of great significance to improve the existing hydrological forecasting models based on the full consideration of the uncertainty information input and migration transformation law. Probabilistic flood forecasting breaks through the conventional thinking of "single point, single value", and provides the probability distribution function of the forecast target variable.

Process-driven hydrological models (HMs) are limited to simplified hydrological processes and have difficulty dealing with complex non-linear relationships between environmental variables and runoff. Data-driven models (DDMs) are good at capturing complex nonlinear relationships, but are overly dependent on data and lack consideration of physical mechanisms. Therefore, a hybrid model for probabilistic flood forecasting that couples the process-driven HM and DDM is proposed. HM can transfer the physical process information of observed runoff to the DDM, while DDM can extract additional nonlinear information not captured by HM, thus giving full scope to their respective advantages.

The hybrid model treats the DDM as a residual model, that is, it corrects the residuals produced by the HM simulation, and the corrected values are added to the original hydrological simulation results to obtain the final runoff predictions. In order to quantify the uncertainty information in the forecasting process, the uncertainty in the HM parameters is used as the source of error, and the resulting input, parameter, and structural uncertainties in the DDM are investigated to construct a hybrid modelling framework that takes into account multiple sources of uncertainty. In addition to deterministic forecasts, this framework simultaneously provides interval forecasts and probabilistic forecasts for quantitative uncertainty assessment, which can provide more abundant and complete risk information for subsequent flood warning and reservoir operation.

How to cite: Xu, C., Zhong, P., Zhu, F., Xu, B., Wang, Y., Yang, L., Wang, S., and Xu, S.: A process-data duality driven hybrid model for improving flood forecasting, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-121, https://doi.org/10.5194/egusphere-egu24-121, 2024.