Evaluation of TRMM PR retrieved rain products using data from the ground-based polarimetric weather radar of Darwin

Analyzing algorithms results, and exploring possible improvements based on suggested modifications and/or changes in operational versions, are sound works for the TRMM-related science. In this context, the present paper deals with the PR standard (std) rain-profiling 2A-25 algorithm. The previous version-4 of the 2A-25 algorithm was changed to the present version 5 by the end of 1999. Tests reported some underestimation of the rain rate (R) by the version-4 std compared to other ground-based or space-based estimates. Possible improvements relying on different adjustments of the initial rain relations (Ferreira et al., 2000) led us to define two alternatives to the std estimate (Rstd-V4): RN0 that exploits the concept of normalized drop-size distribution (DSD) to scale the rain relations via a relevant parameter (N0*) retrieved from PR data; and RkR computed from a R-k (attenuation coefficient) instead of R-Z (reflectivity) relation as in the std. Change brought by the std version-5 estimate (Rstd-V5) with respect to the std version-4, is also tested.

The reliability of the PR-derived rain estimates is evaluated by comparison with a “reference” rain estimation from the ground based C-band polarimetric radar of Darwin (Australia) on the TRMM ground validation (GV) site. Involved data were acquired mainly over land on 27 January 1998 during a good TRMM overpass (orbit #952). Two reference products are used: i) “Dar-S” products from TRMM GV-std algorithms that provide 3-D Z field, and surface rain rate; and ii) “Dar-P” products from the Z-Phi polarimetric algorithm that provides 3-D Z- and R-fields, and unique estimates of the DSD scaling parameter N0*. Rain-type dependent point-to-point comparisons in the selected space domain were performed after sizing the reference products at the “low” PR beam resolution.

The mean differences in Z, for the Dar-S or Dar-P products, are close to theoretical prediction (Ku- versus C-band). Residual offsets lie within the margin of errors in the calibration of both radars. N0* retrievals from the version-4/version-5 std 2A-25 and the Z-Phi algorithm are in fair/good agreement. Comparisons of rain estimation depend on the involved reference product. For the total rain, the PR mean near-surface rain rates largely exceed the GV-std mean surface rain rate (Dar-S), especially in convective rain. The PR and polarimetric (Dar-P) mean rain rates agree much better: the best agreement is found for RkR (alternative version 4), and Rstd-V5, which differ from RDar-P by 4%, and 8%, respectively; while Rstd-V4 (resp. RN0) underestimates (resp. overestimates) RDar-P by 24% (resp.15%). Improvement in the PR/Darwin-radar rain rate comparison, which is observed when using polarimetric instead of standard product, is likely due to the specific use of N0*-scaled R(Z) relation in the Z-Phi polarimetric algorithm.