Multi-model Tropical Cyclone Wind Field Forecasting

Hurricanes threatening and impacting areas of high insurance exposure can affect global financial markets, and even in areas with no insurance take-up, those falling into the protection gap, the losses still exist but have to be covered by government and international aid agencies.

Knowing where a hurricane will go and how strong it will be when it gets there, and how much of the area will be flooded by storm surge, allows preparations that make sure adequate capital exists to be able to cover claims, set up claim centers, secure adjusters, and see if there are areas where losses are concentrated. While deterministic forecasts perform well at short lead times, ensemble models account for the uncertainty of the atmospheric flow in which the storm is embedded. Multi-model forecasts that include both deterministic and ensemble approaches can help provide a full range of the potential storm impact locations.

Each of the major weather forecasting centers have invested significantly in global numerical weather prediction. Massive amounts of observations from conventional and remote sensing sources contribute to defining the initial conditions as represented in the model’s 3D grid. Global ensemble forecast models account for uncertainty by perturbing the initial conditions of the atmosphere. The models are integrated in time out to periods of 16 days or more, but for tropical cyclone purposes, the most accurate information is limited to about a 5 day forecast.

Global model grid resolutions are relatively coarse, in order to achieve computational efficiency to run all the member forecasts out to periods over a week, globally. A higher resolution, limited-domain, deterministic version of the model is used as the control member, the initial conditions of which are perturbed to create the ensemble members. Ensemble model grids are generally too coarse to resolve a hurricane, so “trackers” have been developed to follow vorticity centers that serve as proxies for incipient tropical cyclone circulations. Control member models have finer grid spacing and more precise physics, allowing tropical cyclone circulations and warm core thermodynamics to be simulated to the extent that sea level pressure and maximum surface winds (intensity) may be depicted realistically. The finer scale deterministic control forecasts can estimate peak winds and minimum sea-level pressures of hurricanes but lack information on the uncertainty of the forecast.

Ultimately, a large number of forecasts, usually well over 100, are available on a 12 hour cycle, and the meteorologist’s job is to find a way to interpret such information into one or more likely scenarios. For insurance and financial industry applications, we have developed methods to distill the numerous forecasts into a handful of plausible scenarios, while also correcting for wind field biases caused by the relatively coarse grid cells used in global ensemble models. The consequent wind fields provide forcing to run wave and surge models, resulting in multi-hazard footprints that can drive damage and financial loss models. This presentation will describe this effort and include recent examples from the 2017-2019 Atlantic hurricane seasons.