National scientific priorities in regional numerical weather modeling may be divided into two main areas: (1) generation of weather forecasts for "disruptive high impact weather", and (2) use of regional models to provide meteorological input for air quality models. The first general area involving weather forecasting also includes process studies aimed at gaining understanding and improving the models for "bad" weather events such as heavy precipitation, extreme winds, etc.
The United States Weather Research Program (USWRP), which reflects the joint interests and scientific objectives of the National Science Foundation (NSF), the National Oceanic and Atmospheric Administration (NOAA), the National Aeronautics and Space Administration (NASA) and the Office of Naval Research (ONR), emphasizes study in the following three areas: importance and mix of observations for model initialization, quantitative precipitation forecasting (QPF) and hurricane forecasts near landfall. The armed services and the aviation industry would also include accurate forecasts of aviation parameters (e.g., ceiling, visibility, wind shear, icing, etc.) as a national scientific priority because of the obvious effects of bad weather on anyone who flies the sometimes not-so friendly skies. Clearly, there are many valuable uses of weather information provided by regional numerical weather models.
Another application of regional modeling is to provide the meteorological input for the air-quality models. The Clean Air Act requires the U.S. Environmental Protection Agency (EPA) to review and revise, as appropriate, criteria and national ambient air quality standards (NAAQS) for widespread pollutants such as ozone (03) and particulate matter (PM) which pose serious health risks to humans and animals and threaten sensitive ecosystems. Because observations alone inadequately represent the various scales of atmospheric motions, regional models utilizing four-dimensional data assimilation (FDDA) are often used as dynamic-analysis tools to provide the chemistry models with the best possible numerical representation of the atmosphere. Unfortunately, the weather conditions which usually characterize air pollution events -- dry, light wind, weakly forced (i.e., high pressure) conditions -- are very different from those associated with disruptive high-impact weather (e.g., rapidly deepening low pressure systems, thunderstorms, hurricanes, tornadoes, etc.). However, the modeling of fair weather conditions important for the majority of current air quality work receives relatively less attention than that associated with bad weather, and regional modeling research is often divided along these lines.
The work required to address the aforementioned scientific priorities generally falls into one of three main areas: model initialization, model physics and model numerics. For example, data assimilation work is generally focused on dynamic initialization for the weather forecasting problem, where data are assimilated during a "pre-forecast" period to improve the model initial state for a subsequent numerical forecast. On the other hand, for meteorological support of air-quality modeling, a dynamic analysis is often performed where data are assimilated continuously throughout a model simulation period with the regional model serving as an analysis tool.
Another example of the different priorities for regional modeling research is in the area of model parameterization. Parameterization in numerical weather modeling is the use of grid-scale (resolved) information to characterize subgrid-scale (nonresolved) effects. Despite the fact that shallow convection can "pre-condition" the environment for deep (precipitating) convection, there has been much less scientific work on the parameterization of shallow convection than deep convection. This is largely because fair weather cumulus does not disrupt our lives quite as much as a supercell thunderstorm. However, accurate representation of shallow-cloud processes, which can vent polluted boundary layer air into the layer above cloud base, can greatly affect low-level pollutant concentrations.
As computers have become bigger and faster, and the availability and accessibility (e.g., the Internet) of data has increased, it has become more practical to run regional models at finer resolutions. To meet the challenges facing regional numerical weather modeling today and beyond the year 2000, we must adapt and improve our model initialization, physics and numerics packages for grid scales of 1 - IO km for both the forecasting and analysis sides of the problem. This will require new and more innovative use of the data, further development of physical parameterization and gfid-resolved physics, and more accurate AND efficient numerics. As the models become more sophisticated, there is an even greater need to couple models or submodels from related disciplines (e.g., soil hydrology, vegetation, oceans, chemistry, etc.). It should be noted that there is also research needed to adapt or develop physical parameterizations for coarser grid lengths as found in the outer gfids of a nested-gfid system. Furthermore, mesoscale regional models are now being used to aid in the design of physical parameterizations appropriate for coarser-resolution climate models.
Finally, to evaluate the ability of regional numerical weather modeling to achieve our national scientific goals, work must also continue in the areas of visualization and mesoscale verification. New visualization tools are allowing users of numerical weather models to interpret their data in ways never possible before. We all know that a picture is worth one thousand words, but an animation or virtual reality session using model data can be worth thousands of pictures! But how much confidence can we place in the model solution and model-derived products? To answer this question we must do model verification, but the verification must go beyond computations using our same old standard statistics (e.g., threat score). On the mesoscale with its large-amplitude, small wavelength features, there is a significant "zero-overlap" problem which makes standard statistics like threat score meaningless. Therefore, we must develop new and innovative methodologies which also exploit data from our modern observing systems to quantify the "added value" of regional model products at these very fine scales.