The Evolution of the Goddard Profiling Algorithm (GPROF) for Rainfall Estimation from Passive Microwave Sensors

Christian Kummerow and D.-B. Shin

Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado

Y. Hong

The Aerospace Corporation, Los Angeles, California

W. S. Olson and S. Yang

Joint Center for Earth Systems Technology, University of Maryland, Baltimore County, Baltimore, Maryland

R. F. Adler

NASA Goddard Space Flight Center, Greenbelt, Maryland

J. McCollum and R. Ferraro

NOAA/NESDIS, Camp Springs, Maryland

G. Petty

University of Wisconsin—Madison, Madison, Wisconsin

T. T. Wilheit

Texas A&M University, College Station, Texas

ABSTRACT

This paper describes the latest improvements applied to the Goddard profiling algorithm (GPROF), particularly as they apply to the Tropical Rainfall Measuring Mission (TRMM). Most of these improvements, however, are conceptual in nature and apply equally to other passive microwave sensors. The improvements were motivated by a notable overestimation of precipitation in the intertropical convergence zone. This problem was traced back to the algorithm's poor separation between convective and stratiform precipitation coupled with a poor separation between stratiform and transition regions in the a priori cloud model database. In addition to now using an improved convective–stratiform classification scheme, the new algorithm also makes use of emission and scattering indices instead of individual brightness temperatures. Brightness temperature indices have the advantage of being monotonic functions of rainfall. This, in turn, has allowed the algorithm to better define the uncertainties needed by the scheme's Bayesian inversion approach. Last, the algorithm over land has been modified primarily to better account for ambiguous classification where the scattering signature of precipitation could be confused with surface signals. All these changes have been implemented for both the TRMM Microwave Imager (TMI) and the Special Sensor Microwave Imager (SSM/I). Results from both sensors are very similar at the storm scale and for global averages. Surface rainfall products from the algorithm's operational version have been compared with conventional rainfall data over both land and oceans. Over oceans, GPROF results compare well with atoll gauge data. GPROF is biased negatively by 9% with a correlation of 0.86 for monthly 2.5° averages over the atolls. If only grid boxes with two or more atolls are used, the correlation increases to 0.91 but GPROF becomes positively biased by 6%. Comparisons with TRMM ground validation products from Kwajalein reveal that GPROF is negatively biased by 32%, with a correlation of 0.95 when coincident images of the TMI and Kwajalein radar are used. The absolute magnitude of rainfall measured from the Kwajalein radar, however, remains uncertain, and GPROF overestimates the rainfall by approximately 18% when compared with estimates done by a different research group. Over land, GPROF shows a positive bias of 17% and a correlation of 0.80 over monthly 5° grids when compared with the Global Precipitation Climatology Center (GPCC) gauge network. When compared with the precipitation radar (PR) over land, GPROF also retrieves higher rainfall amounts (20%). No vertical hydrometeor profile information is available over land. The correlation with the TRMM precipitation radar is 0.92 over monthly 5° grids, but GPROF is positively biased by 24% relative to the radar over oceans. Differences between TMI- and PR-derived vertical hydrometeor profiles below 2 km are consistent with this bias but become more significant with altitude. Above 8 km, the sensors disagree significantly, but the information content is low from both TMI and PR. The consistent bias between these two sensors without clear guidance from the ground-based data reinforces the need for better understanding of the physical assumptions going into these retrievals.