Evaluation of Real-Time High-Resolution MM5 Predictions over the Great Lakes Region

Abstract

Real-time high-resolution mesoscale predictions using the fifth-generation Pennsylvania State University?NCAR Mesoscale Model (MM5) over the Great Lakes region are evaluated for the 2002/03 winter and 2003 summer seasons using surface and upper-air observations, with a focus on near-surface and boundary layer properties that are important for applications such as air quality and fire weather predictions. The summer season predictions produce a cold bias in maximum daily temperature and a warm bias in minimum temperature that together lead to a good prediction of daily mean temperature but a smaller-than-observed diurnal temperature cycle. In winter, the predicted near-surface temperatures are lower both day and night, yielding good agreement with the observed amplitude of the diurnal temperature cycle but relatively large cold bias in daily mean temperature. The predicted temperatures in the boundary layer are also systematically lower than the observed temperatures in the two seasons. The cold bias is consistent with the wetter-than-observed lower atmosphere in the model prediction, which in turn can be attributed to an inadequate specification of soil moisture. In both seasons, the model produced substantially more precipitation in all categories, especially in the heavy precipitation category, and the overprediction is primarily associated with more widespread area coverage in the model prediction. The chances of producing a false precipitation forecast are substantially higher than missing an observed precipitation event. Small systematic errors are found in the predictions of low-level winds, but above the boundary layer, the predicted winds are predominantly from the west, while the observed winds are from the west-northwest. The model is able to capture the general development and evolution of the lake?land breezes in areas surrounding Lake Michigan during summer, although errors exist in the strengths of the breezes and the timing of their transition. Predicted early morning inversions are slightly stronger than observed in winter and weaker than observed in summer. The weak summer morning inversion results in a rapid inversion breakup followed by an earlier growth of a mixed layer after sunrise. Despite the head start, the predicted mixed-layer heights in late afternoon are lower than those observed, suggesting that either the predicted surface sensible heat flux may be too low or the boundary layer flux divergence may be too high. Decreasing horizontal grid spacing from 12 to 4 km results in little improvement in the predictions of near-surface and boundary layer properties except for precipitation, for which the model bias is significantly reduced by the increase in horizontal resolution. The cold and wet biases and errors in inversion strengths and mixed-layer development call for extra caution when using products from mesoscale forecasts in applications such as air pollution and fire weather prediction.