Background and justification of selected algorithm

There exists a wide range of retrieval algorithms for PMW SST using different approaches and channel combinations. Among these, two main approaches can be identified; statistical algorithms and physically-based algorithms. Statistically-based retrieval algorithms relate the information from satellite measured TBs and additional data, such as e.g. atmospheric and oceanic reanalysis data, to in situ observations [Alerskans et al., 2020, Chang et al., 2015, Gentemann et al., 2009, Shibata, 2006, Wentz and Meissner, 2007]. The physically-based retrieval algorithms, on the other hand, use an RTM to simulate the top of the atmosphere TBs given instrument information, such as e.g. azimuth and earth incidence angles, frequency and polarization, and atmospheric and oceanic information, such as e.g. SST, SSS, WS, water vapor density, liquid water density and atmospheric profiles of temperature. Among these, one frequently used approach is OE, in which the RTM is inverted given apriori atmospheric and oceanic information (and associated uncertainties) and satellite measured TBs to retrieve SST [Nielsen-Englyst et al., 2018].

The advantage of the OE approach is that it allows for indication of measurement errors, such as e.g. imperfect calibration and channel contamination [Minnett et al., 2019]. However, this also means that the performance of OE algorithms depends on the accuracy of the RTM and the representativeness of the observations and prior error covariances [Merchant et al., 2020]. Furthermore, ad-hoc corrections to the retrievals are necessary to deal with measurement errors [Meissner and Wentz, 2012, Nielsen-Englyst et al., 2018]. Statistically-based retrieval algorithms, on the other hand, partly includes the effect of measurement errors during the coefficient derivation process. However, a drawback of this type of methods is that they are constrained by established statistical relationship between the geophysical variables considered in order to simulate the processes that influence the surface emissivity and radiative transfer through the atmosphere. Therefore, both the physical and statistical retrieval algorithms make assumptions about the nature of the radiative transfer process.

The performance of these types of retrieval algorithms is well-documented for retrieval of PMW SST from AMSR-E and AMSR2. For statistically-based algorithms, independent validation using in situ observations have reported biases up to \(\pm\)0.05 K and standard deviations in the range 0.45-0.55 K [Alerskans et al., 2020, Gentemann, 2014, Gentemann and Hilburn, 2015, O’Carroll et al., 2008] for AMSR-E and AMSR2 retrieved PMW SSTs. Nielsen-Englyst et al. [2018] reported a bias and standard deviation of 0.02 \(\pm\) 0.47 K for AMSR-E PMW SSTs using a physically-based retrieval algorithm.

The algorithm approach chosen for retrieval of PMW SSTs for the CIMR Level 2 product is a statistically-based model based on the work done by Alerskans et al. [2020].