The dynamic fine-scale urban flood risk map based on insurance data, machine learning, and GPU-WRF

Abstract:

Flood insurance has developed rapidly in recent years, aiming to mitigate flood disasters. For cities, urban flooding usually can cause unpredictable losses, and thus, establishing corresponding P&C insurance is of great importance for the society. However, due to the lack of related data, the traditional hydrodynamic model could not be utilized well to all regions for the risk simulation and assessment. To solve this issue, we built an urban flood risk forecasting service based on the insurance data, weather data, DEM, building outline, and etc. The statistical analysis and a fine-scale GPU WRF were employed as well. Based on data from historical insurance claims caused by typhoon and rainstorm since 2010, we generated a static 1-km resolution flood risk map that is built on the kernel density estimation. We also built a risk level forecasting model by employing the machine learning algorithm and the fine-scale numerical precipitation forecast. Our trial runs within Guangdong Province, China indicate that the predicted risk zone covers more than 70% of the occurred claims and the result of risk level forecasting reaches an AUC score above 0.8. Additionally, the nested 9-km and 3-km grid spacing WRF model has been run in GPU to provide finer, more accurate and efficient precipitation forecast. Results show that the GPU WRF can increase the calculation speed of CPU WRF by 10 times on average, and this increased calculating efficiency will help to earn more time ahead of risks. The flood risk map we built would benefit the risk management. The risk map could have a penetration of both static as well as dynamic flooding risk and quantify the detailed risk exposures, improving risk-tackling strategies. In short, this study shows the possibility of building an efficient flood risk forecasting service based on big data technology and numerical simulations. This service would be more applicable than pure hydrology model in the scenario of insurance, pointing out a new direction for flood risk assessment.