A Reduced-Order Observation Operator for TRMM/TMI based on Canonical Correlation Analysis

In assimilating all-sky radiances, an observation operator suitable for operations requires trade-offs between accurate simulation of upwelling-radiation, computational feasibility, and minimizing the impact of poorly modeled parameters such as microphysical drop-size distribution. We have developed a candidate operator that uses canonical correlation analysis to find the optimal linear operator (in the least-squared sense) that is not sensitive to small changes in microphysical variables, gives surprisingly decent results, and is very computationally inexpensive. We present this method applied to the Hurricane WRF (HWRF) model and the Hurricane WRF Satellite Simulator for the simulation of TRMM/TMI all-sky radiances trained on Earl 2010 data, and present the results and error statistics of this method. We conclude that this approach is in general applicable to any observation operator and can serve as an important tool in the data assimilation toolbox.