Abstract
In light of previous theoretical calculations, an empirical-statistical analysis using satellite multifrequency dual polarized passive microwave data to detect rainfall areas over land was performed. The addition of information from a lower frequency channel (18.0 or 10.7 GHz) was shown to improve the discrimination of rain from wet ground achieved by using a single frequency dual polarized (37 GHz) channel alone.
The algorithm was developed and independently tested using data from the Nimbus-7 Scanning Multichannel Microwave Radiometer (SMMR). Horizontally and vertically polarized brightness temperature pairs (TH, TV) at 37, 18, 10.7 GHz were sampled for rain areas over land (determined from ground-based radar), wet ground areas (adjacent and upwind from rain areas determined from radar), and dry land regions (areas where rain had not fallen during a previous 24 h period) over the central and eastern United States. Surface thermodynamic temperatures were both above and below 15°C. An examination of the data from each separate channel indicated that the probability (using the F test) for the mean vectors of any two populations being identical is less than 0.01 for classes sampled with surface thermodynamic temperatures ⩾15°C except for the rain over land and wet ground classes observed with the SMMR 37 GHz channel. For the classes sampled with surface thermodynamic temperatures <15°C, none of the classes were significantly different.
Since most of the categories were significantly different for the warmer (≥15°C) land surface cases, a Fisher linear discriminant classifier was then developed for each channel and independently tested. The results from one test case showed that for areas of large-scale heavy rainfall, the lower frequency SMMR channels were better able to delineate rain from wet ground than the 37 GHz channel. However, in areas of light rain and/or where the rain area did not fill the lower frequency instantaneous field of view these channels were not able to differentiate, rain from wet ground.
