Minyoung Kim

Timely and reliable urban flood forecasting is essential for mitigating damage and supporting emergency decision-making. Ensemble data assimilation can improve forecast reliability by updating model states with observations, but real-time use with high-resolution hydrodynamic models is often constrained by computational cost. We propose an integrated forecasting framework that couples a physics-guided AI emulator with data assimilation to enable efficient, high-resolution spatiotemporal inundation prediction. The emulator is trained on high-fidelity hydrodynamic simulations and reproduces key flood dynamics with substantially lower runtime than conventional solvers, allowing large ensembles generated by perturbing initial conditions and meteorological forcings to quantify uncertainty. Real-time inundation-depth observations are assimilated to update evolving flood states, using both synthetic data for controlled testing and ground-based depth information derived from surveillance-camera imagery for real-event conditions. The framework is applied for an urban drainage basin in Seoul, South Korea. The presentation will discuss key challenges for real-time urban flood assimilation, including observation uncertainty and representativeness, intermittent availability and latency, and the balance between ensemble size and update frequency. We also examine how emulator design affects physical consistency during assimilation and outline remaining limitations for operational deployment.

Accurate flood risk assessments are critical for mitigating the impacts of pluvial flooding in densely populated urban areas. However, conventional flood modeling approaches often face limitations due to the lack of measurement information. To address this challenge, we develop, implement, and evaluate a novel framework that integrates crowdsourced data, such as citizen observations, with process-based modeling to enhance the accuracy of urban flood risk assessments. The proposed method utilizes indicator co-kriging techniques to merge citizen-sourced data with auxiliary variables, including inundation maps generated from 1D-2D urban flood models driven by high-resolution radar rainfall estimates. The framework is applied to the Oncheon River catchment in Busan, South Korea, a region highly vulnerable to pluvial flooding due to its urbanization and complex hydrological conditions. To evaluate the method, synthetic citizen observation data were generated based on inundation maps. These synthetic experiments assess the influence of the spatial distribution and quality of citizen observations on urban flood risk predictions. This study examines the integration of citizen observations into urban flood modeling workflows to address uncertainties in models and observations. In particular, we investigate the extent to which distributed citizen observations enhance prediction accuracy and analyze the effects of model bias on the reliability of flood risk assessments. The study quantitatively evaluates the effects of citizen data quality and spatial distribution on the accuracy of urban flood risk mapping. Furthermore, a sensitivity analysis is conducted for co-kriging parameters, focusing on semivariogram model selection and its influence on prediction accuracy.

As climate change intensifies hydrologic extremes, the need for near real-time urban flood prediction becomes critical. Pluvial flooding occurs when intense rainfall overwhelms urban drainage systems, involving complex hydrodynamic interactions between surface runoff and subsurface sewer flow—known as dual drainage. Capturing the spatiotemporal evolution of these processes requires detailed representations of flow patterns, inundation propagation, and runoff accumulation. However, physics-based hydrodynamic models, while effective at resolving the fine-scale dynamics of flood events, face significant computational limitations, particularly for large urban areas or high-resolution domains. To address this challenge, we propose a deep learning-based urban flood prediction model that integrates surface runoff dynamics with sewer network interactions. The model is developed using training data generated from physics-based 1D-2D hydrodynamic simulations that capture interactions between 2D surface flow and 1D sewer network flow. The Oncheoncheon River catchment in Busan, South Korea—a region frequently impacted by urban flooding—serves as the study area. Various synthetic rainfall scenarios are used to train the model, ensuring its ability to generalize across different extreme rainfall events. Model validation against historical flood events shows that the deep learning model accurately predicts flood evolution patterns while significantly reducing computational time compared to traditional hydrodynamic models. This study demonstrates the potential of deep learning-based approaches to enhance real-time urban flood prediction and provides valuable insights for developing efficient, data-driven disaster management strategies.