Weather Research and Forecasting (WRF) Model is a latest-generation numerical weather prediction model developed in an ongoing collaboration involving primarily the National Center for Atmospheric Research and NOAA's National Centers for Environmental Prediction with several other US Government agencies and universities, and including a large user community. It applies to the mesoscale range of weather phenomena: a few kilometers up to 1000 km. It is classed as a “dynamical” model in that it computes an approximate solution to the differential equations governing fluid motion (air, in this case). The equations and the theory which produced them are elegant and powerful. Given that they can describe the whole atmosphere, including all its clouds and storms as well as a gentle breeze on a single leaf, they are remarkably simple. But applying them to practical weather forecasting requires approximations and still becomes an enormous bookkeeping problem. It's keeping track of a budget; only instead of money, the quantities tracked are matter and energy. Like money, these quantities come in, go out, and change hands, but unlike money they are everywhere governed by the simple and reliable rules expressed in the equations. Given the budget's current state, we can forecast with great confidence—at least for the next minute. Stringing these minutes into multi-hour forecasts, of course, brings in the uncertainties for which meteorology is famous. Some sources of uncertainty, particularly turbulence, will never go away, but explosive growth of computing power, observing capability, and communications have brought steady improvements, of which WRF is a recent example.
The implementation at ATDD uses the Advanced Research WRF (ARW) version on a two-stage nested grid. The outer grid covers most of Tennessee and Kentucky and smaller parts of surrounding states with points separated by 10 km in the horizontal and extending in 50 levels from the surface into the stratosphere up to 16 km. An inner grid covers east Tennessee with 3.3 km point spacing on the same 50 levels. The inner grid resolves features down to about 13 km in horizontal size (four grid intervals). In particular, it sees the Cumberland Plateau, Sequatchie Valley, Smoky Mountains, and Blue Ridge, but not the corrugating ridges in the main valley's floor. Likewise it resolves only imperfectly individual thunderstorms and local terrain-related wind patterns. The model can not correctly place individual thunderstorms, but is able (within error) to identify the areas where thunderstorms will form and the density and intensity of thunderstorms in that area.
The environment (initial and boundary conditions) in which ATDD's simulations are embedded is determined from NOAA's Rapid Update Cycle (RUC) Model having a grid spacing of 20 km. The RUC model is restarted every hour. Eight times per day it makes 12 hr forecasts. These provide the environment for ATDD's 12-hr forecasts using WRF. They are currently unaltered except by interpolation to the finer mesh of our outer and inner grids. The WRF system provides a variational (optimizing) mechanism WRF-3DVAR for modifying the initial fields, e.g. from RUC, to include local measurements. We have adapted WRF-3DVAR to incorporate ATDD's RAMAN mesonet and are currently updating WRF and WRF-3DVAR to mutually compatible versions.