Abstract
Twelve-hour quantitative precipitation forecasts from the National Weather Service National Meteorological Center's nested-grid model are evaluated for 12 stations in the upper Midwest for the period April–July 1987. Statistics for threat score, bias, post agreement, over and under forecasts, and “quantitative correctness” are determined from frequency distributions for precipitation forecasts and verifications categorized into four quantitative amount levels: none, low, medium, and high. The analysis is performed for all cases as a group and for subpopulations representing six different categories for the associated synoptic situation. The synoptic situation descriptors involve proximity to surface frontal or trough positions or lack thereof. It was found that the warm and occluded frontal situations had better forecast performance than the other synoptic situations reflecting the better handling of grid-scale in contrast to convective-scale precipitation by the model. Results provide an example of the aid that can be given to forecasters by suggesting relative levels of reliance to be assigned to specific model forecasts.
