2. Evaluating and optimizing Impact-based Forecasting models: Evaluating Sensitivity Using Historical, Synthetic, and Projected Hazard Data

Evaluating and optimizing Impact-based Forecasting models: Evaluating Sensitivity Using Historical, Synthetic, and Projected Hazard Data

M-GEO
M-SE
4D-EARTH
Staff Involved
Norman Kerle
Marc van den Homberg
Sahara Sedhain
Topic description

Anticipatory actions (AAs) are pre-emptive measures taken to mitigate the impacts of disasters based on forecasted hazards. Trigger models using Impact-based forecast (IBF), play a critical role in determining when these actions should be initiated by defining specific thresholds for activation. However, in data-scarce regions, limited historical data hinders comprehensive evaluation and optimization of these models.

This research focuses on evaluating the sensitivity of trigger models to changes in thresholds, using historical data where available, and generating synthetic hazard datasets for regions with limited data. By statistically analyzing the frequency and accuracy of trigger activations, this study aims to improve the design and reliability of IBF trigger models, enhancing their effectiveness in guiding anticipatory actions.

Topic objectives and methodology

The primary objective of this research is to assess how trigger models respond to varying activation thresholds by analyzing their sensitivity and frequency of activation under different conditions. This includes examining the balance between false positives and missed activations to optimize Anticipatory efforts. Where historical data is unavailable or insufficient, synthetic hazard data will be developed and incorporated into the analysis, offering an alternative to fill data gaps.

The methodology involves collecting and preparing historical hazard and vulnerability data. Where needed, synthetic hazard datasets will be generated to mimic historical patterns of key parameters. Statistical models will then simulate trigger activations across varying thresholds, and sensitivity analyses will measure the performance and reliability of these models. The approach will be applied to a specific case study, such as tropical cyclone trigger models in Philippines, to validate findings and assess scalability to other hazards or regions.

References for further reading