Applications of the unified retrieval technique: Land surface temperature

A unified retrieval(UR) technique has been used to retrieve geophysical parameters over land using SSM/I data. In this paper we focus on the land surface temperature(LST) retrievals. For the time period October 1-6 1995 we have compared retrieved LST to surface observation temperatures for the continental U.S. The surface observation temperatures are readings taken hourly at various locations at about 2m above the surface. All seven SSM/I channels are used in the retrieval with the higher resolution channels averaged over the 19 GHz footprint. For the morning, descending, passes the RMS difference between retrieved LST and surface observations is about 3.0K with a 0.0K bias. For the evening, ascending, passes we obtain a RMS of about 3.8K and a -2.3K bias. The magnitude of the differences is regional. In the Great Plains the RMS is about 1.9K while in the Midwest the RMS is about 4.0K. This contrast can partially be explained by the fact that in the Plains there is little variation in the topography within an SSM/I footprint. But in the Midwest the surface observations are usually collected in river valleys where as the SSM/I footprint encompasses areas with potentially large height variations. In deserts the expected large bias between LST and surface observations in the morning is achieved. Time sequences of differences between LST and surface observations for individual observation locations match well. Comparisons have also been made between LST retrievals and surface measurements collected from May 1987 to August 1989 during the FIFE campaign. The retrieved LST match the measurements quit well. Also, we have compared the UR method with regression techniques.