In the days before a hurricane landfall, the insurance industry pays close attention to location and intensity forecasts – a logical concern as these represent the key drivers of uncertainty in estimating risk.
As the system moves inland, we typically get very good guidance on what just occurred for both these metrics thanks to quick reporting from agencies such as the U.S. National Hurricane Center (NHC). There are however other drivers of hurricane risk for which the level of uncertainty is still very high, even days after an event.
A good knowledge of the likely wind field size and shape are, for instance, needed to provide post-landfall guidelines on expected wind speeds, and these are rarely available with any great level of certainty. Any slight change in these metrics can have a large effect on measures of peak gust winds as the influence of the terrain (e.g. changes in topography and roughness) can vary sharply over small distances.
Our immediate post-landfall event response tool (HindCyc) is designed to explicitly account for that uncertainty. In the current (beta) version of the product, we simulate 100 scenarios where track trajectory and intensity are fixed to agencies estimates but all other parameters are sampled from modelled distributions. This includes:
- The location where maximum winds occur around the storm centre (radius and angle of maximum winds relative to storm heading).
- The shape of the wind field (size of the eye, wind field asymmetries and radial decay in wind magnitude).
All models responsible for estimating these distributions are Machine Learning based and have been trained using our proprietary database InCyc. Details on the methods are available in Loridan et al. (2017).
To fully account for the downstream impact of these variations in wind field characteristics on expected peak gusts, HindCyc simulates the influence of changes in terrain topography and roughness at site and up to 3 km upwind. In other words, all 100 footprints are terrain corrected.
This process allows us to generate 100 gust footprints (1 km resolution) consistent with reported track trajectories and intensities as well as the experience from our InCyc database. By ranking the 100 footprints in terms of the gusts simulated at site, we can assemble probabilistic estimates of gust wind risk.
The end product of this beta release consists of a 50th percentile map, available immediately after landfall (see the example map above for Hurricane Laura 2020) and we are offering free access until the end of the 2020 hurricane season – simply subscribe to our list here:
Subscribe Now!
Reference
Loridan, T., R.P. Crompton, and E. Dubossarsky, 2017: A Machine Learning Approach to Modeling Tropical Cyclone Wind Field Uncertainty. Mon. Wea. Rev., 145, 3203–3221,