Basis for a Stochastic Convective Storm Model Using Measured Near-Surface Wind Gust and Reanalysis Data

Windstorms, on an annual basis, typically cause the most insured losses of any natural hazard. A subset of windstorms, convective events, cause a significant amount of these losses. In the United States in 2015, $10B in insured losses were attributed to convective events according to NOAA. Perhaps not surprisingly, convectively-generated winds produce the highest recorded near-surface wind speeds (≤ 10 m) at a number of locations across the U.S. and throughout the world.

Despite these facts, extreme near-surface winds from convective events are poorly predicted and understood due to their relatively small spatiotemporal scales. Surface networks in most of the world are of poor resolution and thunderstorm occurrence has been poorly classified, due to a lack of observers/automated procedures, especially in the time period before Doppler radar and lightning array networks. This poor understanding extends to potential future changes in convective windstorms.

Recently, there has been an increase in the number of dense meteorological networks (i.e. anemometry) across the U.S., subsequently increasing the likelihood of an encounter with an extreme wind event. As with radar data, studies of dense observational networks compared to traditional networks (i.e., airport-based stations) used in engineering- based analyses of extreme convective winds have also revealed that these events occur with greater frequency than previously estimated using improved classification techniques. Development of grid-based reanalysis data sets has also given researchers the opportunity to look at the large-scale environments responsible for extreme convective winds with a spatial coverage (usually global or continental) unmatched by observational techniques.

This study links two such datasets for the United States: 1) NOAA Integrated Surface Database (ISD) spanning 1973 to present and 2) North American Regional Reanalysis (NARR) from 1979-2015. The ISD dataset consists of peak gust data and was heavily quality controlled to ensure thunderstorm occurrence could be associated with a near-surface wind speed with high confidence.

Specifically the study focuses on three major tasks: 1) Development of a regional convective climatology. This climatology will be highlighted through an extreme value analysis of the wind speed data from each of these regions; 2) Estimating the likelihood of convective near-surface wind speeds conditional on marginal and/or joint distributions of reanalysis parameters (e.g., likelihood of convective wind gust given CAPE values); 3) Simulation of thunderstorm occurrence and near-surface wind speed magnitude using information gained from the first two tasks.

Regional cluster analysis shows that the United States can roughly be separated into 6-10 regions that have similar convective extreme wind characteristics, including wind directionality. An initial ‘proof-of-concept’ simulation was run and it was determined that it is indeed possible to re-create the occurrence and wind magnitude of convective events with reasonable results. The analysis also revealed, as should be expected, that the higher the CAPE and lower the CIN values, the more likely the occurrence of a thunderstorm. Utilizing this method without surface data present appears promising. The overall goal of this work is to develop a comprehensive stochastic convective storm model that can be applied in any location, even in areas devoid of near-surface measurements.