Retrieval of Near-Surface Temperatures from Satellite Data

Abstract

Now that satellite retrieval are being used in global forecasts, there is increased interest in using satellite retrievals for use such as mesoscale forecasts and inputs to crop models. These uses require more accurate retrievals near the ground where satellite retrievals tend to be affected by the large temperature difference between the lower atmosphere and the earth's surface. Standard regression techniques tend to produce a normal difference and low level atmospheric retrievals tend to be too warm over a hot surface and too cold over a cold surface. Because radiances are measured at several wavelength regions with very different relationships between temperature and radiance, these differences can be detected and used to stratify the data for which a given set of coefficients is applied. Data were stratified into three groups based an differences between radiances in the 4.3 and 15 ?m regions. The stratification increased the accuracy of retrievals in the lower atmosphere by up to 0.3 K depending on the level and the error. This is an improvement of about 10% of the existing error. As part of the study, retrievals based on hourly surface observations were compared with retrievals based on radiosondes for the surface temperature. Retrievals based on hourly surface observations and compared with hourly surface observations produced ret6cvals with rms differences less than 2.0 K while retrievals based on radiosondes produced differences over 3.0 K. This indicates that satellite soundings are more accurate near the surface than the values indicated by comparisons with radiosondes.