HR: 1330h
AN: H23A-15    [Abstracts]
TI: The Validation of TRMM TMI and PR Precipitation Estimates at Climatological Scales
AU: * Fisher, B L
EM: fisher@radar.gsfc.nasa.gov
AF: NASA Goddard Laboratory for Atmospheres, Greenbelt Road, Greenbelt, MD 20771 United States
AB: The Tropical Rainfall Measuring Mission (TRMM) has been collecting data for over eight years. One of the original goals of TRMM was to ascertain uncertainties in satellite-derived monthly precipitation estimates from the TRMM Microwave Imager (TMI) and the Precipitation Radar (PR). The accuracy of TRMM monthly estimates, however, are limited by the sampling frequency of the satellite, which varies as a function of latitude from one to three samples per day within a latitudinal range of 40 N and 40 S. Monthly integrations of data collected by TRMM sensors are determined statistically from the total number of samples collected each month. Sampling errors in the range of ñ8-12% month over the tropical oceans were expected, in addition to the retrievals errors associated with the actual measurement and estimation of instantaneous areal rainfall from space. Consequently, whereas sampling errors result from a lack of information about the state of the atmosphere when the satellite is not overhead, retrieval errors are mostly attributed to the physical modeling of an instantaneous areal rate from the real time data collected by the TRMM sensors. This climatological validation study, based on a methodology developed by Fisher (2004), decomposes the sampling and retrieval errors associated with TRMM TMI and PR monthly estimates into two distinct error distributions. This method uses high-resolution ground data, sub-sampled at satellite overpass times. The sub-sampled rain estimate is then assumed to contain a sampling error equivalent to the satellite. The sampling error variance can then be partly parameterized based on the statistical differences between the rain estimate, R0, computed at all times and sub-sampled rain parameter, RS. Similarly, the retrieval error distribution is statistically parameterized in terms of the statistical variance between the satellite estimate, S and the sub-sampled ground estimate, RS. An annual bias factor is also computed for both sampling and retrievals that is weighted by the annual climatology, computed from the validation parameter, R0. The formulation of the bias used in this study represents a recent modification on the methodology of Fisher (2004). The TRMM GV site in Melbourne Florida was used as a regional test of this proposed methodology. The TMI and PR monthly precipitation estimates from version 5 and 6 are validated over a four-year period (1998-2001) using both the Melbourne NEXRAD radar and a large network of 97 rain gauges distributed within the radar domain that extends out 150 km from the radar. A 2 x 2 deg. gridded region was selected for this study, which only considered 0.5 deg. grid boxes where there existed rain gauges. Monthly rainfall was first computed at 0.5 deg. resolution and then averaged over the full grid space. Error statistics were computed at both 0.5 and 2.0 degrees for each year of the study. Random errors were characterized by the coefficient of variation (CV=std/mean). At the 2x 2 deg. scale, these statistics showed the PR CVsam about 25% higher than the TMI, whereas PR CVret were about 30% lower than the TMI. Version 5 shows a large positive summertime bias. Another interesting result showed a negative bias between the sub-sampled radar and gauge, which may be attributable to spatial sampling differences between the two sensors.
DE: 1836 Hydrologic budget (1655)
DE: 3354 Precipitation (1854)
DE: 3360 Remote sensing
SC: Hydrology [H]
MN: 2005 Joint Assembly