Rain Estimation from Geosynchronous Satellite Imagery—Visible and Infrared Studies

Abstract

A diagnostic method to estimate rainfall over large space and time scales by the use of geosynchronous visible or infrared satellite imagery has been derived and tested. Based on the finding that arms of active convection and rainfall in the tropics are brighter or colder on the satellite visible or infrared photographs than inactive regions, ATS-3 and SMS/GOES images were calibrated with gage-adjusted 10 cm radar data over south Florida. The resulting empirical relationships require a time sequence of cloud area, measured from the satellite images at a specified threshold brightness or temperature to calculate rain volume over a given period. Satellite rain estimates were made for two areas in south Florida that differ in size by an order of magnitude (1.3×104km2 vs 1.1×105km2) and verified by a combined system of gages and radar. Contrary to our expectations, the rain estimates for the smaller area agreed better with the raingage-radar groundtruth than the satellite rain estimates for the larger area. As expected, the accuracy of the rain estimates is a function of the period of rain estimation. The error and scatter of the hourly estimates are relatively large but both decrease as the estimates are accumulated with time. For periods of 6–9 h the mean absolute errors are factors of 1.50 and 1.90 using the visible and infrared imagery, respectively. Additional tests for which groundtruth was available were also made to determine the applicability of this scheme to tropical areas other than the region of derivation. These include nine days over portions of Venezuela, five days over Honduras during the passage of Hurricane Fifi, and one day each of Hurricane Belle (1976) over the East Coast of the United States and Hurricane Agnes over Florida. The potential of this method as a forecasting tool for hurricane-caused flooding is investigated in a study of selected Atlantic hurricanes between 1969 and 1977. Storms were ranked by satellite-calculated total daily rain volume and daily area-averaged rain depth. Relatively wet storms could be distinguished from the relatively dry storms. No correlation was found between volumetric storm rain output or area-averaged rainfall and storm intensity, which suggests that the location of latent heat releases and not its magnitude determines storm intensity. Computer automation of the technique for real time and diagnostic estimates is discussed briefly. Satellite-inferred rains for the Big Thompson (Colorado) flood of 1976 and the Atlantic Ocean during GATE (summer 1974) are presented as examples of the real time and diagnostic computerized schemes, respectively.