Methodology

The Jupiter WindScore Planning application, which is built within the Jupiter ClimateScore Intelligence Platform,  provides probabilistic projections of wind speeds under current and future climate conditions. For a given region, location, or asset, key atmospheric conditions that drive the distribution of wind speeds are identified and analyzed to provide a historical framework upon which the modeling and assessment of future conditions are based.  Machine-learning methods project key drivers of extreme wind speed onto a representative suite of future climate states to define the statistical character of current and future wind speed distributions.  

WindScore Planning probabilistic projections provide metrics that include the number of days per year in which wind speeds exceed specified thresholds, and wind speeds associated with return periods that range from 2 years through 500 years.  All projections contain uncertainty intervals that utilize the probabilistic nature of the WindScore Planning methodology to identify measures of confidence in the chosen metrics.

Extreme winds are governed both by large-scale ocean and atmospheric patterns (e.g., storms and their associated storm tracks) and by highly localized features like mountain breezes and sea-breeze fronts. The methods employed in derivation of the Jupiter WindScore Planning Product capture three key components required to assess future risk to extreme wind events.  The first necessary component is sufficient spatial and temporal resolution to fully represent the physical characteristics associated with perils that lead to extreme wind events at a given location. The second component is the ability to project perils that lead to extreme wind events under viable future climate scenarios. The third component is to convey future risk in a probabilistic manner with relevant measures of uncertainty.  

While global climate models are useful for simulating large-scale changes in the climate system, they have horizontal resolutions that are too spatially coarse to simulate many of the complex, local scale processes that are important for accurately representing physical perils that drive extreme wind events. Very high-resolution numerical models (“downscaling” models) serve to bridge the gap down to the regional scale--allowing us to combine the large-scale features from climate models, with the ability to simulate weather that is realistic at a regional level.

To achieve higher resolution (1 km), Jupiter uses dynamical downscaling in conjunction with climate modeling. At the core of the method is the idea that the frequency and intensity of extreme wind events may change, but core physical characteristics are discernible in the future. Hundreds to thousands of regional simulations of historical and potentially damaging events are completed with the Weather Research and Forecasting (WRF version 4) model, optimized for the local meteorology - e.g., terrain, land-sea gradient, land surface characteristics.  The high-resolution simulations are verified against local surface observations such as the Global Historical Climate Network (GHCN) and other high-quality resources and bias-corrected.

The result is a very large set of calibrated simulations that adequately samples the tail of the wind distributions, which is what is most relevant. That set is the foundation for an analog-based resampling method whereby future climate model extreme wind projections are matched to analogs from the downscaled record.   The annual exceedance probability, and number of days exceeding given thresholds, is computed based on the archived set of records.

Scenarios