A Stochastic Methodology for Regional Wind-Field Modeling

Abstract

Three-dimensional, regional scale (≈1000 km) air-quality simulation models require hourly inputs of U and V wind components for each vertical layer of the model and for each grid cell in the horizontal. The standard North American meteorological observation network is used to derive the wind-field inputs for the U.S. Environmental Protection Agency's Regional Oxidant Model (ROM) and other regional models. While a fairly dense surface network with hourly observations exists, upper-air data are obtained only twice per day at monitoring sites typically separated by distances of 300?500 km. Using these data to derive the more spatially and temporally resolved gridded wind fields needed by the ROM introduces uncertainties and errors into the model. We present a method of developing gridded wind fields for the ROM that accounts for these nondeterministic features. The method produces a family of potential gridded wind fields that allows for the stochastic nature of the interpolation process. Examples of the derived wind fields are given for the northeastern United States. Potential differences between wind fields, in terms of their effects on air-quality modeling, are inferred from following multiday flow trajectories using various members of the wind-field family. After 72 h of travel time, a trajectory spread of 100?200 km was not uncommon. The sensitivity of results to the density of surface observational data is also presented. Results of the sensitivity analysis showed that, as more stations were eliminated from the analysis, the calculated flow followed the domain mean flow more closely, showing fewer local influences.