A Statistical Technique for Determining Rainfall over Land Employing Nimbus 6 ESMR Measurements

Abstract

At 37 GHz, the frequency at which the Nimbus 6 Electrically Scanning Microwave Radiometer (ESMR 6) measures upwelling radiance, it has been shown theoretically that the atmospheric scattering and the relative independence on electromagnetic polarization of the radiances emerging from hydrometeors make it possible to monitor remotely active rainfall over land. In order to verify experimentally these theoretical findings and to develop an algorithm to monitor rainfall over land, the digitized ESMR 6 measurements were examined statistically. Horizontally and vertically polarized brightness temperature pairs (T_H,T_V) from ESMR 6 were sampled for areas of rainfall over land as determined from the rain recording stations and the WSR 57 radar, and areas of wet and dry ground (whose thermodynamic temperatures were greater than 5°C) over the southeastern United States. These three categories of brightness temperatures were found to be significantly different in the sense that the chances that the mean vectors of any two populations coincided were less than 1 in 100. Since these categories were significantly different, classification algorithms were then developed. Three decision rules were examined: the Fisher linear classifier, the Bayesian quadratic classifier, and a non-parametric linear classifier. The Bayesian algorithm was found to perform best, particularly at a higher confidence level. An independent test case analysis showed that a rainfall area delineated by the Bayesian classifier coincided well with the synoptic-scale rainfall area mapped by ground recording rain data and radar echoes.