Possible improvements in the standard algorithms for TRMM/PR

The tropical rainfall measuring mission (TRMM) satellite has been producing more than nine years of precipitation data over the globe. In particular, the precipitation radar (PR) is a unique sensor to measure three dimensional information of the precipitation. The rain retrieval algorithm for TRMM/PR consists of several parts: rain/no-rain classification, surface echo measurement for the surface reference technique, rain type(convective/stratiform) classification, and rain retrieval itself. Since the PR can obtain limited information of the rainfall, the rain retrieval algorithm needs to assume several atmospheric parameters, such as precipitation structure, drop size distribution (DSD), rainfall distribution within a PR's footprint, and so on. The comparison of rain products between PR and TMI (TRMM Microwave Imager) shows reasonably close to each other in the latest version (Version 6) of the standard algorithms. On the other hand, the comparison of rainfall between PR and the gauge network shows PR still underestimates rain rates (over land) by more than 20%. Currently, there are no clear explanation for this difference. In order to reveal the inconsistency between the satellite observation and the gauge data, several assumptions in the TRMM/PR algorithm are examined in this paper, such as DSD model, rain type classification, non-uniform beam filling (NUBF) effect, and so on. One of the most important items is the DSD model parameter. This model parameter is obtained from PR's standard algorithm (2A25) but it is biased because 2A25 adopts hybrid approach (simple Z-R relationship which does not change the DSD model parameter and surface reference technique which tries to estimate the appropriate DSD model parameter) for the rain rate estimation. A new DSD model parameter is estimated by statistical way by using only observable information (not use of algorithm itself). The rain type classification algorithm mainly depends on the bright band detection algorithm that uses level one data. Comparison using ground based multiparameter radar shows misjudgment of bright band frequently occurs for heavier rainfall cases because the algorithm uses observed reflectivity factor (Zm) profile data in which Zm decreases toward the ground in heavy rainfall and tend to form a peak near the freezing level because attenuation of snow is quite small. The 2A25 algorithm uses an a priori attenuation model which defines the relationship between the attenuation coefficient (k) and Ze. The k-Z model for convective rainfall gives almost constant value from the rain bottom to 1 km above the freezing level and linearly decreases toward the top of the echo where the attenuation coefficient of snow is used. In other words, the algorithm assumes mixed phase (ice and rain) precipitation between 1 km above the freezing level and echo top level. Even though the echo reach above the homogeneous freezing level, the algorithm uses attenuation coefficient larger than the actual condition. The preliminary test result shows that the rainfall increases up to 5% if the top of the mixed phase region is defined at -20 degree C level. The most tough issue is how to introduce the NUBF effect to the algorithm. The NUBF effect is examined by looking at the relationship between path integrated attenuation (PIA) obtained from surface echo data and rain rate, and the result shows NUBF clearly appears if the rain rate exceeds 10 mm/h. Another issue in NUBF effect is in PIA estimation itself. If the rainfall is non-uniform, the PIA in the footprint should be non-uniform and it is reflected at the surface echo profile. Since the current algorithm searches a surface echo peak and compares the non-raining surface echo peak to estimate the PIA, the estimated PIA tends to be smaller (small attenuation) under non-uniform case. A preliminary result shows PIA is underestimated by about 10 dB for a heavy rainfall case.