Evaluation of Performance of the GPM/DPR Rain Retrieval Algorithm

The GPM core satellite was launched successfully on 28 February 2014 with the Dual-frequency Precipitation Radar (DPR) and the GPM Microwave Imager (GMI) on it. This paper reviews the performance of the DPR rain retrieval algorithm by looking at more than a year of data and by comparing them with other rain data that include TRMM/PR's estimates of rainfall. The DPR algorithm has several new challenges because expectations and requirements for the DPR rain estimates are more demanding than those for the PR. Continuity with the PR estimates is an important requirement. Detection of snow and quantification of its rate at high latitudes is a new challenge. Better classification of storm types and higher accuracy of rainfall estimates by DPR than those by PR are expected because of the use of dual-frequency radar. The paper will describe the at-launch version of the algorithm that was expected to handle these challenges. One year of accumulated data, however, revealed issues that turned out to be more severe than we had expected. One important example is the effects of non-uniform beam filling (NUBF) to the attenuation estimates in both the surface reference technique (SRT) and the DSD retrieval algorithm. Intrinsic fluctuations of radar echoes are magnified in algorithms that use dual-frequency ratios. A small mismatch between the Ku- and Ka-band radar beams also creates a problem of defining the surface range bin. A few minor differences in hardware between the TRMM/PR and the Ku-band radar (KuPR) of GPM prevent us from running exactly the same rain retrieval algorithm with data from these radars so that we cannot separate hardware effects from software effects by examining the differences between the products. Validation of the DPR products is another difficult issue. Since the frequency of DPR overpasses of Japan is substantially less than the PR's, we cannot obtain enough data for a statistically meaningful comparison with ground measurements. The reasons for the small but persistent differences among estimates obtained from the DPR, Ka-only, and Ku-only algorithms need to be found. The paper will discuss our ideas for the solution to or mitigation of these issues.