In our continuing work focused on providing weather sensitive business solutions, IBM's Deep Thunder service provides operational forecasts twice daily for areas of southeastern New York State and northern New Jersey. With an operational history that spans more than a decade, producing one- to three- day model-based forecasts at one to two kilometer resolution, the overall model configuration has evolved and improved over time to reflect improvements in NWP model capability as well as computational efficiency. Over the past several years, the system has focused on producing 84-hour predictions updated every 12 hours. The NWP component is derived from a configuration of the WRF-ARW (version 3.1.1) community model. It operates in a nested configuration, with the highest resolution at two km, utilizing 42 vertical levels. The configuration also includes parameterization and selection of physics options appropriate for the range of geography within the domain from highly urbanized to rural. This includes WSM-6 microphysics (explicit ice, snow and graupel), Yonsei University non-local-K scheme with explicit entrainment layer and parabolic K profile in the unstable mixed layer for the planetary boundary layer, NOAH land-surface modelling with soil temperature and moisture in four layers, fractional snow cover and frozen soil physics, Grell-Devenyi ensemble cumulus parameterization, and the 3-category urban canopy model with surface effects for roofs, walls, and streets.
Given the model length and frequency, the system produced six operational forecasts that covered the period prior to landfall and the impact in New York and New Jersey. The system exhibited proficient skill in forecasting regional as well as local scale impacts of Irene with significant lead time.. In particular, with the model run initialized at 12 UTC on 26 August 2012 and those produced afterwards forecasted Irene to weaken and make landfall as a tropical storm.
In order to evaluate the quality of the forecasts produced by Deep Thunder at a storm-scale and its potential skill, we compare the model results with observational data and other available forecasts as well as the operational availability of specific forecast products. Such performance is examined by considering forecast timing, locality, and intensity of the storms impacts as well as through the utilization of traditional and spatial verification methodologies.