A novel framework for multi-hazard loss assessment of tropical cyclones
A county-level interpretable machine learning model
Bibliographic Data
| ID | 22026376 |
|---|---|
| Authors | Jinli Zheng (Beijing Normal University), Weihua Fang (0000-0002-6433-7281, Beijing Normal University, corresponding author), Jinyan Shao (0009-0000-8820-4072, Beijing Normal University) |
| Year | 2025 |
| Volume | 117 |
| Pages | 105204 |
| Publication date | 2025-02-01 |
| Open Access | Yes |
| Type | ARTICLE |
| Venue | International Journal of Disaster Risk Reduction (JOURNAL) |
| Journal identifiers | ISSN: 2212-4209 |
| Publisher | Elsevier BV (PUBLISHER) |
| DOI | 10.1016/j.ijdrr.2025.105204 |
| OpenAlex | W4406240684 |
| Language | EN |
| Citations received | 2 |
| References cited | 50 |
Reliable loss assessment of tropical cyclones (TCs) is essential to effective disaster emergency response. Currently, most existing methods do not consider the joint effects of multiple hazards of TCs, which may decrease loss estimation accuracy. To address this issue, a novel assessment framework is proposed that incorporates the effects of wind, rainfall, storm surge, wave, and flood, brought by TC, to improve disaster loss modeling. First, the multi-hazard features of TC are selected, including maximum gust wind (3s), total rainfall, daily rainfall, hourly rainfall, surge heights, significant wave heights, and daily runoff. The model was then trained on 1341 county-level records, optimizing four machine learning algorithms, i.e., Categorical Boosting (CatBoost), Transformer, Backpropagation Neural Network (BPNN), and Support Vector Machine (SVM). Next, the impact of feature variables and training samples was analyzed based on the best-performing model. Finally, Shapley additive explanations (SHAP) were employed to interpret the model. The results demonstrated that CatBoost outperformed the other machine learning algorithms, achieving an accuracy of 0.8196. Notably, the model incorporating all feature variables achieves a maximum performance improvement of 19.06% compared to scenarios involving single, double, or triple hazards. Furthermore, the county-level model at the national scale demonstrates good applicability in coastal and inland scales, maintaining an accuracy exceeding 0.79. The inclusion of SHAP provides interpretable insights into factor weights and the interrelationship between multiple hazards. This approach not only improves the reliability of loss assessments but also reduces the 'black box' limitations of machine learning, offering vital insights for better disaster response. • A novel multi-hazard assessment framework of tropical cyclone loss grade is developed using an interpretable machine learning model. • The CatBoost-SHAP model outperforms other models, delivering the best performance at the county level. • The multi-hazard model's accuracy is improved by approximately 19.06% compared to single, double, or triple-hazard scenarios. • The county-level model at the national scale demonstrates good applicability in coastal and inland scales, maintaining an accuracy exceeding 0.79
Forensic engineering · Geography · Hazard · Hazard analysis · Machine learning · Meteorology · Reliability engineering · Tropical cyclone · Computer Science · Engineering · Environmental Science · Flood Risk Assessment and Management · Hydrological Forecasting Using AI · Tropical and Extratropical Cyclones Research · Artificial Intelligence
Learning representations by back-propagating errors
Support-Vector Networks
Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead
Smote
Induction of decision trees
Definitions, methods, and applications in interpretable machine learning
Suitability estimation for urban development using multi-hazard assessment map
Greedy function approximation
Assessing Influential Factors on Inland Property Damage from Gulf of Mexico Tropical Cyclones in the United States
How could students be safe during flood and tsunami events
Empirical estimation based on remote sensing images of insured typhoon-induced economic losses from building damage
Predicting the spatial distribution of direct economic losses from typhoon storm surge disasters using case-based reasoning
A multi-hazard framework for spatial-temporal impact analysis
High-accuracy estimation method of typhoon storm surge disaster loss under small sample conditions by information diffusion model coupled with machine learning models
Climate costs of tropical cyclone losses also depend on rain
A Data Augmentation-Based Evaluation System for Regional Direct Economic Losses of Storm Surge Disasters
| Citations per year | 2 |
|---|---|
| Citation span | 2025 - 2026 (2) |
| Highly cited | No |