Comparing High Resolution Weather Forecasts to Observations

The Advanced Research version of the Weather Research and Forecasting model (WRF) is a mesoscale numerical weather prediction (NWP) system, with a horizontal grid spacing of several kilometers to several hundred kilometers. WRF can create forecasts of finer horizontal resolution by embedding a smaller domain inside the parent domain, a process called nesting. A nest may be embedded simultaneously within a coarser-resolution (parent) model run, or run independently as a separate model forecast. Army operations require weather forecasts on a scale of one kilometer or less, an area of weather modeling known as 'terra incognita' between which large eddy simulation and traditional mesoscale NWP models are applied with most confidence. Complex terrain leads to differences in surface temperature, moisture gradients, and wind speed /wind direction, and these differences are not always well-characterized by mesoscale WRF forecasts. Differences in land surface characteristics produce air flows such as mountain/valley breezes, and sea breezes that are of vital importance to Army and Air Force operations. Atmospheric effects on commercial as well as military air platforms and any associated subsystems is of critical concern, whether for commercial flight planning or for military mission execution. The traditional model verification techniques currently used aggregate the error statistics over an entire domain (such as on the order of 100km x 100km to 500km x 500km in size), techniques which produce results that often appear smoothed and may not show the value added of higher resolution WRF output at grid resolutions of 1km or less. Point verification methods can also be ineffective due to 'double counting' errors of phase and spatial nature, and failing to capture model skill in resolving mesoscale structure. More in-depth analysis of the forecast errors are needed to deduce the various sub-regimes and temporal and spatial trends which may govern the statistics in a way which cannot be seen when the statistics are aggregated over the entire domain. This study uses a geographical information system (GIS) to compare WRF forecasts against gridded observations from multiple data sources.