Subgrid variability and stochastic downscaling of modeled clouds: Effects on radiative transfer computations for rainfall retrieval

The Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) Goddard Profiling (GPROF) rainfall retrieval algorithm is an inversion type algorithm, which uses numerical cloud models and radiative transfer schemes to simulate the brightness temperatures that the TMI would see, thereby allowing one to relate hydrometeor profiles to brightness temperature. The variability in modeled hydrometeor fields is known to have an important effect on simulated brightness temperatures, and while the TMI instrument sees all the variability down to scales of a few meters, cloud models are typically run at resolutions of 1-3 km. This paper is an illustrative investigation into the importance of subgrid variability (scales below 1-3 km), which is ignored when simulating brightness temperatures. Previous studies on the importance of subgrid variability have been based on comparisons of simulated brightness temperatures computed from hydrometeor fields of a high resolution model and spatially aggregated hydrometeor fields from the same model run. It is argued that numerical cloud models have reduced small-scale variability due to model artifacts such as computational mixing, and this may lead to an underestimation of the importance of including subgrid variability. To address this problem, stochastic downscaling developed in a wavelet-based framework is used to reintroduce the variability reduced by computational mixing. In particular, a high resolution model is spatially aggregated (i.e., upscaled) over the scales affected by computational mixing and stochastically downscaled back to the original resolution of the model. The higher degree of variability introduced by the downscaling (which is a closer approximation to the variability observed in hydrometeor concentrations as compared to that produced by high resolution models) is found to result in larger biases in estimated brightness temperature. This points to the potential for a significant source of bias in microwave-sensed precipitation retrievals that requires further study.