| description abstract | The second intercomparison project of the Global Precipitation Climatology Project examined the estimation of midlatitude, cool-season precipitation. As part of that effort, the authors report here on the results of two microwave techniques the Goddard scattering algorithm and the physical retrieval algorithm of Kummerow. Results from the estimation of instantaneous rain rate for five overpasses of the Special Sensor Microwave/Imager (SSM/I) are presented in a case study mode to illustrate both the strong and weak points of each technique. These five cases represent a sampling of the various types of precipitating systems observed. Results for the complete set of 20 swaths chosen by the United Kingdom Meteorological Office are then categorized by scatterplots and statistics of instantaneous radar versus microwave-estimated rain rate, rain/no-rain contingency tables, and scatterplots of arch coverage of rainfall. Neither algorithm produced a good statistical correlation with the radar data, yet in general, both did well at determining rainy areas. Two reasons are suggested for the low correlation coefficients between both algorithms and the radar data. Time differences between the SSM/I overpass and the radar observations can occasionally account for some of the differences. The primary reason for the low correlations, however, appears to be the predominance of very light rain in the area of interest during the winter. Both algorithms are in good spatial agreement with the radar when the radar data are restricted to rates above 1 mm h?1. When all radar rain rates are included, the radar areal coverage increases by as much as a factor of 10 in some cases. Because the Kummerow algorithm does not handle such low rain rates over land very well, and because the Goddard scattering algorithm uses 1 mm h?1 as the minimum reliably detectable rain rate, regimes that contain large arm of very fight rain present inherent difficulties for these retrieval methods. Therefore, the proliferation of low rain rates observed during the experiment is the main contributor to low correlation coefficients and high root-mean-square differences. Misidentification of cold surface (e.g., snow cover) as precipitation was also a problem in several instances. | |
