As the derecho propagated, it produced a leading-edge gust front, which was associated with multiple downbursts and surface divergence. At 18:30 EDT, thought to dissipate due to the complex terrain and forthcoming loss of heat fluxes due to diurnal heating, the storm system crossed the Appalachians, weakened slightly, and then came into contact with another unstable air mass, resulting in a re-intensification. At this point, the NOAA Storm Prediction Center (SPC) issued a severe thunderstorm watch for Washington, D.C., and by 21:00 EDT a moderate derecho risk was extended into the Mid-Atlantic States. The storm passed through the Baltimore-Washington metropolitan area at approximately 22:00 EDT and across the Chesapeake Bay and out to sea around 02:00 EDT on 30 June. The overall damage was extensive and widespread: 22 deaths across seven states and 3.7 million people left without power, with one million in Virginia and one million in Maryland and Washington, DC.
IBM Research developed and operates a high spatial and temporal resolution weather forecasting capability known as Deep Thunder, which can be utilized to predict the business impact of severe weather events in order to optimize resources and mitigate negative effects. Deep Thunder was deployed for the Baltimore-Washington area in an attempt to capture the 29 June 2012 derecho via hindcast analysis. For this purpose, the Advanced Research Weather (ARW) core of the Weather Research and Forecasting (WRF) model (version 3.3.1) was used to produce several 84-hour forecasts prior to the event, with output every ten minutes. The domain was based on the current operational Deep Thunder configuration for the New York metropolitan area but was fine-tuned for the characteristics of the Baltimore-Washington region, which resulted in the use of more sophisticated cumulus and microphysics schemes. The model consisted of three, two-way feedback nests at 15.75-, 5.25-, and 1.75-km horizontal resolution, each with forty-two vertical levels. Background fields were derived from 12-km output from the regional NOAA/NCEP North American Model (NAM), along with NASA 1-km sea surface temperatures to account for the Chesapeake Bay influence along the Atlantic Coast. The grid mesh sizes were the following: 76x76 (at 15.75-km), 154x154 (at 5.25-km), and 178x178 (at 1.75-km). Because of the complex topography of the Appalachians, 90-m terrain data were ingested for the outer two domains, while 30-m terrain data were used for the innermost domain, with both sets available from the Shuttle Radar Topography Mission (SRTM) archive.
A three-dimensional variational data assimilation technique was also employed to perturb the initial conditions for each domain via the ingestion of surface observations from the dense mesonet (WeatherBug) operated by EarthNetworks. Near-real time surface measurements are available every five minutes. For the assimilation, WRFDA, version 3.3.1, was utilized and able to process temperature (K), dew point (calculated, K), pressure (Pa), and horizontal wind components (m/s). The total number of stations in each domain were: 3900 (outermost), 2849 (intermediate), and 1053 (innermost).
We will primarily focus on one of the hindcasts, initialized at 00 UTC on 29 June 2012, which would have provided an 18-hour lead time for the event's impact in the Baltimore Washington metropolitan areas. The model was able to effectively capture characteristics of the bow-echo structure, multiple downbursts and low-level divergence, resulting in the propagating gust front, and associated surface wind gusts. Mid-level analysis revealed the presence of graupel and ice, as well as the release of latent heat, both of which likely contributed to the simulated downbursts. We will discuss the research objectives and challenges, model results and comparisons with observations, and potential future work.