PNNL

Abstract

Diagnostic wind field models continue to play indispensable roles in air quality studies and emergency response. The wind fields provided by these models are interpolations of surface and upper-air observations within constraints such as mass conservation. The inputs for diagnostic wind models are generally obtained from operational weather monitoring networks, which are often established for purposes other than to provide data to diagnostic models. As a result, the data available may not be optimum for calculating accurate wind fields. In particular, measurement locations may be poorly distributed. In this case, it is desirable to add monitoring stations to supplement existing measurements. We have developed an objective approach to identify the best location for one or more additional stations. The placement of new monitoring stations can be cast as an optimization problem. The best new locations should result in the forecasted wind fields being closest to the actual wind fields. Such an optimization task is not trivial because the response surface of the cost function on the location parameters to be optimized could be very complicated. The complication results from the complex nature of meteorological processes and the nonlinear behavior of location parameters affecting the evaluation of the cost function. This complexity usually precludes an analytical solution. For this work we used a genetic algorithm (GA) as the optimization tool to search for the optimum locations of new monitoring stations to be added to an existing meteorological network located in the city of Chicago, Illinois. Because complicated wind fields pose the greatest challenge for diagnostic models, we identified dates on which typical examples of such wind fields occurred in the city. These wind fields were modeled with the Pennsylvania State University/NCAR mesoscale model (MM5) using four-dimensional data assimilation. The MM5 results were used as surrogate observations. Wind components at the candidate station locations and at the locations of existing stations were sampled from MM5 and treated as actual observations. These “observations” were provided as inputs to an EPA-recommended diagnostic model, CALMET, to produce diagnostic wind fields. The calculated difference between the diagnosed wind fields and the MM5 wind fields was used to create the cost function in the optimization. The simulation and optimization approach was illustrated by the selection of a single additional surface station to improve the diagnosed wind field for a typical lake breeze episode near the city of Chicago.