Systems engineering is performed to optimize a system or its response given limited resources. With any system, it is important to understand which subcomponents or inputs are most important and which are less important so that appropriate resource allocations may be made. An example of a complex system is the Tropical Rainfall Measuring Mission (TRMM). Its subsystems include the satellite vehicle, the precipitation radar (PR), the ground validation (GV) sites, and the retrieval algorithms. In the context of systems engineering, this paper describes a method of performing sensitivity analysis on a TRMM-like retrieval algorithm to better describe how the uncertainty in the model output can be apportioned to the uncertainty in the input factors and gain greater understanding as to the relative importance of each of the input factors. With specific information about the importance of each factor, more effort can be applied to increase the knowledge of specific factors and reduce the output variance from the model.
The sensitivity analysis described in this paper is based on variance decomposition using the method of Sobol. A description is given of the TRMM-like (TL) retrieval algorithm, its assumptions, and deviations from the actual TRMM algorithm. Probability density functions are given for each of the input factors. Analyses are done and results are presented for factor importance for cases over ocean. Results show that at low rain rate, the a and b coefficients in the R = a Ze b relationship contribute the greatest amount to the output variance. At higher rain rates, above about 5 mm/hr, the error from Ds° is the greatest contributor to error in algorithm output. Sensitivity analysis methods are applied to a GPM dual-frequency profile-optimization retrieval method and results are presented showing that the error in the Ka-band (35.6 GHz) reflectivity profile dominates the output variance.