Early loss estimation in the days leading up to a high-category hurricane landfall

In our latest Research Spotlight, we present our approach to quantifying risk uncertainty during that critical 1-3 day pre-landfall window, using a modelling framework that builds up from granular hazard data to loss estimates. 

The cornerstone of Reask’s pre-landfall modelling tool, LiveCyc, is the belief that no single best-guess deterministic forecast can provide a complete view of the risk ahead.

Relying on one (or even a few) deterministic forecasts to understand landfall risk often creates a false feeling of certainty. Instead, LiveCyc generates 1,000 likely realisations of the next three days, based on the best available deterministic forecasts like those provided by the National Hurricane Center (NHC).

Figure 1: NHC forecast with 1,000 LiveCyc realisations for Hurricane Helene (2024) 48 hours before landfall 

Instead of selecting events from a static pre-computed catalogue (i.e. “similar stochastic events”), each of the 1,000 LiveCyc realisations is generated on-the-fly using Reask’s peer-reviewed and proprietary algorithms.

They sample a distribution centred on the NHC forecast and allow explicit quantification of the uncertainty around landfall location and intensity. 

How can we improve stakeholder communication with LiveCyc scenarios?

The 1,000 LiveCyc simulations provide a highly granular view of the distribution of risk at landfall. However, interpreting 1,000 realisations can be cumbersome and may complicate stakeholder communication.

To address this, Reask also offers a consolidated view of landfall risk in the form of LiveCyc scenarios. Here, all realisations making landfall within the same section of the coast (referred to as a“coastal gate”) are aggregated into a single representative event.

This event, or scenario, is assigned a probability of occurrence based on the size of the sub-population contained within that coastal gate (see Fig. 2, where coastal gates are 15 km wide).

Figure 2: LiveCyc scenarios for Hurricane Helene (2024), with probability of occurrence. The dashed black line shows the NHC 48-hour forecast; the blue gate indicates the observed landfall location. 

Depending on the forecast lead time and associated uncertainty, LiveCyc typically generates between 10 and 30 scenarios, covering all coastal regions potentially at risk.

For each scenario, representative high-resolution surface wind footprints are produced using Reask’s peer-reviewed Machine Learning (ML) algorithms. These footprints provide peak surface gust wind estimates for every 1 km cell in the landfall region, accounting for the influence of land use and topography (see Fig. 3).

Figure 3: Surface wind gust footprints for a range of LiveCyc scenarios. 

Vave case study: From granular hazard to daily loss exceedance probability curves

Starting from LiveCyc scenarios and their high-resolution representative footprints, our customers can objectively quantify the size of their potential loss distribution, with daily updates right up to landfall: 

• D-3: Anchor an early estimate of likely location and broad magnitude of impact (best- to worst-case scenarios) 

• D-2: Monitor changes relative to the initial anchor and refine estimates 

• D-1: Gain increased confidence in reliable estimates of immediate impact 

One such example is US MGA Vave, who used LiveCyc scenario footprints to compute evolving loss profiles in the days leading up to the landfall of Hurricane Helene (2024). Thanks to LiveCyc’s granular, daily updates, Vave was able to:   

1. Provide an explicit description of the spread of likely losses 48 hours before landfall, including analysis of worst-case scenarios in relation to exposure at landfall and NHC forecast trends. 

2. Monitor the evolution of the loss distribution 24 hours later, and assess how changes in the NHC forecast track impacted worst case scenarios.  

3. Gain confidence in the expected outcome, as the 24-hour forecast curve converged towards the median estimate from the previous day.

Figure 4: 48- and 24-hour forecast loss distribution for Hurricane Helene (2024)
 

How to implement the above framework using LiveCyc

1. Convert hazard to loss
   For each scenario, transform LiveCyc gust wind footprints into a portfolio loss estimate. This can be done using damage functions (standard or bespoke) to convert wind estimates into expected damage, then applying those metrics to any exposure database of choice. An alternative method is to match each scenario’s characteristics to a pre-computed loss from a catastrophe model.

2. Rank scenarios
   Rank all the scenarios from highest loss causing event to lowest, and compute cumulative probabilities by adding up each scenario probability of occurrence.

3. Generate EP curve
   Assemble Exceedance Probability (EP) curve, and update daily to monitor the evolution of the risk distribution forecast.

Figure 5: From LiveCyc scenario footprints to loss exceedance probability curves

Want a sample of Reask’s pre-landfall insight? 

Our new global tropical cyclone alert service offers a single, high-resolution snapshot of each storm, combining the latest agency forecast with Reask’s propriety 1 km wind gust overlay.

Try it for yourself by signing up here. If you need a fuller picture, like probabilistic scenarios or early loss modelling, our full LiveCyc pre-landfall suite can help. Get in touch via contact@reask.earth.