The Impact of Spatial Resolution Enhancement of SSM/I Microwave Brightness Temperatures on Rainfall Retrieval Algorithms

Abstract

The impact of spatial resolution enhancement on estimates of tropical typhoon rainfall based on SSM/1 (Special Sensor Microwave/Imager) measurements is evaluated with six different microwave precipitation retrieval algorithms. Passive microwave estimates of rainfall are susceptible to errors from nonhomogeneous beam filling. The SSMIX ground footprints for the 19-, 22-, and 37-GHz channels have considerable overlap, and thus deconvolution techniques can be applied to enhance spatial resolution of measurements at those frequencies. The authors utilize a Backus-Gilbert matrix transform approach to accomplish the deconvolution so as to minimize noise amplification, as suggested by Stogryn. The deconvolution scheme is evaluated in terms of its impact on rain rates throughout the life cycles of seven tropical cyclones that occurred during the 1987 hurricane and typhoon season. The evaluation was performed on a single-frequency emission-based algorithm, a single- frequency scattering-based algorithm, two multiple-frequency statistical regression algorithms, and two physical inversion-based profile algorithms. While rainfall patterns detected by all algorithms were qualitatively enhanced by accentuating rainfall gradients and other smaller-wale features, quantitative responses to the deconvolution process were quite different for each algorithm. Furthermore, each of the algorithms, which uses its own distinct scientific approach, exhibits its own distinct properties in retrieving the rainfall patterns and in recovering the storm domain-averaged rain rates. The rain rates derived from the single-frequency emission algorithm were consistently increased by application of the deconvolution procedure. Time-and space-averaged rain rates were elevated by approximately 5%?6% due to the nonlinear relationship of rain rate to brightness temperature. On the other hand, rain rates from the single-frequency scattering algorithm were consistently reduced, with the time-space-averaged reduction between 10% and 20%. This effect is not algorithm related but is due to alteration of noise properties of the two polarized 37-GHz channels introduced during the deconvolution process. The multiple-frequency algorithms have more complex responses to deconvolution. Although instantaneous rain rates can be changed quite significantly by these methods, differences between deconvolved and raw time-space-averaged rain rates are small compared to the single-channel algorithms because the pixel-scale differences tend to be of a more random nature (positive and negative changes instead of consistent bias). However, it appears that the profile methods can undergo the greatest improvement to instantaneous rain rates after deconvolution is applied because they use perturbative inversion procedures rather than fixed brightness temperature-rain rate relationships.