Analysis Of Cost Functions For Retrieving Rough Sea Surface Parameters

Generally, it is difficult to apply a theoretical model to accurately retrieve SSSfrom the TBbecause of the influences of the oceanic, atmospheric, and cosmicparameters. Recently, a semi-empirical model called model-2P including the effectsof both significant wave height and wind speed has been derived with a highinversion precision. Based on the model-2P, a Bayesian-based cost functionminimized by the least square method was built to accurately retrieve sea surfacesalinity and other ocean parameters.For brightness temperature of L band (TB), two kinds of Bayesian-based costfunctions (i.e. the unconstrained cost function and parameter-constrained costfunction) are investigated for retrieving the sea surface salinity (SSS). In low SSSregions, we have analyzed the sensitivities of the two cost functions to geophysicalparameters. The results show that the unconstrained cost function is valid forretrieving several parameters (including SSS, wind speed and significant waveheight), and the constrained cost function, which largely depends on the accuracy ofreference values, may lead to large retrieval biases. Furthermore, it is found that, as aretrieval parameter, the sea surface temperature (SST) can result in the divergence ofother geophysical parameters in an unconstrained cost function due to the strongsensitivity of brightness temperature to SST. In addition, using the unconstrainedcost function and the simulated brightness temperature TBwith white noises, theretrieval biases of SSS are discussed with the following two procedures. Procedurea), the simulated TBvalues are first averaged, and then SSS is retrieved. Procedure b),the SSS is directly retrieved from the simulated TB, and then the retrieved SSSvalues are averaged. The results indicate that, for low SSS and SST distributions, theSSSretrieval by procedure a) has less biases compared with that by procedure b),while the two procedures give almost the same retrieval results at high SSS and SSTsea regions.In low SSS and SST regions, because of large biases when retrieving sea surfacesalinity by unconstrained cost function, the cost function sensitivities of vertical andhorizontal polarizations for both L and C bands are analyzed, the optimum polarization model is built for retrieving sea surface parameters. The results show asfollows:(1) The vertical polarizations of both L and C bands can be used tosimultaneously retrieve SSS and SST.(2) The vertical polarization of L band andhorizontal polarization of C band is adapted to retrieve SSS and U10simultaneously.(3) The vertical polarization of L band combining with either vertical or horizontalpolarization of C band is suitable for retrieving SSS, SST, and U10.(4) Because ofthe low sensitivity of L band brightness temperature to SWH parameter, the costfunction of both L and C bands can not be used to estimate SWH parameter.Furthermore, with the simulated brightness temperature of additive noises, theoptimum multi-polarization models are used to retrieve SSS, SST and U10, and theretrieval results show that these optimum models are powerful for a higher noiselevel.