Risk analysis to support early action protocols (EAPs) for tropical cyclones
Reliable forecasts and predictions of the impact of tropical cyclones (TCs) are crucial to support humanitarian action and for triggering early action. This research focuses on damage to municipal buildings in the Philippines due to TCs. Multiple linear regression, logistic regression and random for...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2022
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/156715 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Reliable forecasts and predictions of the impact of tropical cyclones (TCs) are crucial to support humanitarian action and for triggering early action. This research focuses on damage to municipal buildings in the Philippines due to TCs. Multiple linear regression, logistic regression and random forest machine learning techniques are used to train a damage prediction model using typhoon specific hazard metrics from 26 historical typhoons, the municipal damage, vulnerability indicators and geographical metrics from every municipality as input. We found that the random forest model performed best with the highest correlation of 0.701 between predicted and observed values. The damage model, together with the Holland wind field model, was then applied to a separate set of 169 historical TC tracks and their corresponding forecast tracks with missing damage values. Evaluation of the forecast quality using the newly obtained forecasted damage and wind values against those of the historical tracks showed an increasing MAPE and SMAPE with increasing lead time, indicating less accurate forecasts with increasing lead time. RMSE, MAE and MBE also peaked at the 72-hour time scale for the wind speed errors. We found that TC intensity peaks with higher wind speeds at the 72-hour lead time, potentially resulting in higher MAE and RMSE. A risk analysis conducted on these damages also found that the shorter the lead time and the lower the trigger threshold, the more optimal the early actions are initiated.
Keywords: Damage assessment model, machine learning, typhoon forecast, lead time, Early Action Protocol, Early Warning System, risk analysis |
|---|