IBM's customized weather forecasting solution, known as Deep Thunder, has been producing 84-hour forecasts for the New York metropolitan area for over three years at high spatial and temporal resolution. Operationally, Deep Thunder runs twice daily at 00 UTC and 12 UTC and covers southeastern New York State and portions of New Jersey and Connecticut at 2-km horizontal resolution, with output every 10 minutes. There are three domains (18-, 6-, and 2-km), with each grid mesh being 76x76 and incorporating 42 vertical levels. The Advanced Research Weather (ARW) core of the WRF-3.1.1 model is utilized and incorporates WSM-6 microphysics, the Grell-Devenyi ensemble cumulus scheme, and the NOAH land surface model (LSM), coupled with a 3-category urban canopy model to better represent the characteristics of the highly urbanized portions of the domain. The model is initialized with 12-km output from the North American Model (NAM), which is also used for the boundary conditions. NCEP 9.25-km sea surface temperatures also drive the initial conditions; and USGS topography data is used for model terrain.
To further study the December 2010 blizzard, Deep Thunder was used in hindcast mode and initialized with NAM output, with the following configuration specifics and incorporated data sets: (1) WRF-ARW 3.3.1 was used as a replacement for WRF-ARW 3.1.1 and will be utilized for operations in the near future; (2) domains and physics packages remained the same with the exception of the microphysics scheme, for which the double-moment WDM-6 scheme was used; (3) NASA 90-m Shuttle Radar Topography Mission (SRTM) data was used as input for model terrain; (4) to better represent the coastal characteristics, NASA 1-km sea surface temperatures were used as part of the initial conditions. Three hindcast simulations were conducted, initialized at 00 UTC and 12 UTC on 25 December and 00 UTC on 26 December. Typical output included 2-m temperature and dew point, sustained wind speeds/direction at 10m, snow depth, visibility, and snowfall rate based on NOAH LSM algorithms.
We will primarily focus on the hindcast initialized at 00 UTC on 26 December. We will discuss the event synopsis, model results and performance, comparisons, and transition to WRF-3.3.1 for operational use.