By S. Vicneswary,
Senior Communications and Outreach Advisor,
Global Asia Insurance Partnership
27 December 2023
In the realm of natural disasters, floods pose significant threat to communities worldwide. Effectively understanding and quantifying flood risk is paramount for the insurance sector to ensure appropriate pricing of coverages. To address this challenge, a study by GAIP's academic partner Nanyang Technological University (NTU) has developed a physics-based hierarchical deep learning framework for flood risk modelling.
This study represents a fusion of high-resolution meteorological and hydraulic data, leveraging a hierarchical deep learning model structure tailored to geographical locationsm and the frameworkdeveloped demonstrated outperformance over conventional deep learning models in flood risk assessment.
The research was led Mr. Yanbin Xu, a Ph.D. Candidate under the GAIP-NTU Ph.D. scholarship.
![](https://www.gaip.global/wp-content/uploads/2023/12/20231117_162518-1024x614.jpg)
"Our target was to provide a new deep learning structure that can produce a justifiable and transferable physics-based model for flood risk pricing. Through comprehensive evaluations, we were able to demonstrate the performance of the proposed framework. It surpassed benchmarks set by traditional deep learning models, signifying its ability in forecasting and assessing flood risks."
Traditional modelling approaches depend very much on historical data. However, with climate change, the historical data might not be ideal for future projections of risk incidences.
"In the current environment of the increasing uncertainties brought about by climate change, many of the traditional approaches may are no longer ideal. What sets the new model apart is its ability to integrate high-resolution meteorological and hydraulic data, effectively capturing spatial and temporal information critical for quantification," said Min Hung Cheng, Deputy CEO, GAIP.
A technical workshop was held to introduce the framework and deep dive into the framework in terms of the model structure, as well as demonstrate the performance of the model on actual policy and claims datasets. About 44 representatives from GAIP partner organisations participated in the workshop which was held on 17 November 2023.
View the slides presented during the workshop here. (Members only)