On the Relative Accuracy of Satellite and Raingage Rainfall Measurements over Middle Latitudes during Daylight Hours

Abstract

Empirical relationships between visible and/or IR data and rainfall rate are derived by comparing gage-calibrated radar data with colocated satellite information over Montreal, Canada. The accuracy of 1739 point gage-satellite measurements from 14 sequences of summertime rainfall during daylight hours is evaluated. The absolute difference, defined as ? |GI ? Si|/? Gi where Gi and Si are the corresponding gage and satellite estimates, is 85%. The Critical Success Index (CSI), the Probability of Detection (POD), and the False Alarm Ratio (FAR) at the 2 mm level are of the order of 50, 70 and 35%, respectively, and the cross-correlation coefficient ? is computed to be 0.56. The relative accuracy in the rainfall estimation of four empirical methods based on point satellite readings is determined. The scores of the visible-IR and ?visible only? methods am adequate (? = 0.56 and 0.50, respectively), but the scores of the ?IR only? method are judged inadequate (? = 0.30) because of its serious overestimation. A rain/no-rain estimate scores nearly as high (? = 0.50) as a continuous rain estimate. It is found that satellite estimates using our objective techniques are better than gage-interpolated estimates at locations where the nearest gage is farther than 40 km. The usefulness of these statistical satellite rainfall measurements is thus limited to the data sparse regions of the world.