DeepCyc

DeepCyc is Reask's global tropical cyclone hazard model for long-term and climate-adjusted risk assessment. Event sets calibrated only on the historical record struggle to show how risk shifts with ocean temperatures, wind shear and steering patterns, and generalise poorly to basins where few storms have been observed. DeepCyc instead uses machine learning driven by ERA5 reanalysis and NCAR-CESM ensemble members to generate more than 100 million physically realistic stochastic events conditioned on past, present and future climates, with event sets spanning 1900 to 2100, 20,000- and 100,000-year catalogues, and climate scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5) to 2080.

Wind fields draw on Reask's InCyc 1 km WRF simulations, and a machine-learning terrain correction converts them into local 3-second gusts over real terrain, accounting for roughness and topography at the site and up to 3 km upwind. Reask's published description of this terrain model builds its global grids from SRTM 90 m elevation (Shuttle Radar Topography Mission) and land-cover maps: the 30 m US National Land Cover Database where available and the 500 m GLCNMO product elsewhere.

DeepCyc comes as Maps (return-period hazard maps), Tracks (stochastic tracks and event characteristics) and Events (gridded wind footprints) at 1 x 1 km resolution across all tropical cyclone basins. It is delivered through Reask's REST API, by FTP for on-premise storage, and through distributors such as EigenRisk and Geosite. Reask also uses it to price parametric covers that Metryc then triggers. Published users include MS Amlin, which used DeepCyc to model US hurricane losses under 2 °C of warming, and Inigo, which uses Reask's climate-conditioned stochastic tracks in one of the two methods behind its 2025 seasonal hurricane outlook.