Assessment of Tropical Cyclone Disaster Damage Based on Learnable Inter-City Interaction GNN

+ Author Affiliations + Find other works by these authors
Funds: 
Supported by the National Key Research and Development Program of China (2020YFA0608000), the Meteorological Joint Funds of the National Natural Science Foundation of China (U2142211), the National Natural Science Foundation of China (42075141, 42341202, 62088101), and the Shanghai Municipal Science and Technology Major Project (2021SHZDZX0100).
Note: This paper has been peer-reviewed and is just accepted by J. Meteor. Res. Professional editing and proof reading are underway. Please use with caution.

PDF

  • Tropical cyclone (TC) is one of the most frequent extreme events in China, causing widespread disaster damage. Traditional approaches oversimplify by treating regions as isolated units and ignore inter-regional interactions. This leads to an underestimation of complex dynamics in disaster damage. In this paper, we create a novel TC disaster damage dataset for the first time, where each sample represents a unique disaster event, incorporating city-specific multi-dimensional features and damage indicators. Then we take provincial administrative devisions in China as examples, innovatively using cities as nodes and constructing inter-city interactions graphs. To align with the physical interactions, a deep learning model named TC-Damage is specifically established. It includes edge building module and backbone. Edge building module aims to construct inter-city interaction features from multiple perspectives. Backbone employs a multi-layer graph neural network based on Graph Sample and Aggregate (GraphSAGE) and Jumping Knowledge Network (JKNet) to learn comprehensive and hierarchical features of inter-city interactions. A loss function combined with focal loss and node-level loss is remarkably proposed to address data imbalance and enforce node representation distribution. Multiple experiments demonstrate that TC-Damage outperforms others and effectively identifies high-contribution factors. Through an explainability analysis of Super Typhoon Lekima, we find key edges are adjacent to cities with high disaster factors and social development levels and significantly overlap with edges exhibiting strong inter-city interactions.

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return