Sea Surface Salinity (SSS) has a strong impact in many climate-related processes. For instance, SSS reflects the balance of evaporation and precipitation at ocean surface, and provides valuable information about water vapor and energy fluxes between the ocean and the atmosphere that drive the long-term evolution of climate. SSS samplings by in situ devices are quite scattered and can only resolve processes well above the oceanic mesoscale, which is insufficient in particular for the characterization of horizontal and vertical fluxes. On the other hand, SSS derived by remote sensing means provides adequate space and time sampling, but the time series are relatively short for some climate applications, and in particular to detect and characterize climate trends. It is possible to use the existing series of remote sensing SSS observations to reconstruct SSS maps at times prior to the advent of satellite SSS observations. One option is to assimilate existing SSS maps in a numerical model of the ocean and to integrate backwards in time. This approach does not benefit from all the information of SSS series but just of the relatively short period that the data assimilation scheme can influence numerical integration – of the order of weeks, at most. For the same reason, and due to the approximative nature of parametrizations, the influence of data assimilation rapidly fades away, and therefore, the quality of the reconstructed SSS data relies on the quality of the physics implemented in the numerical model. A completely different approach was introduced in Droghei et al., Front. Mar. Sci. (2018). In that reference, the authors build a multivariate optimal interpolation (OI) combining Sea Surface Temperature (SST) and SSS, then they exploit the connection between both variables to reconstruct past SSS using existing SST maps and the known OI matrices. The limitations of such approach are that the statistical relations between SSS and SST are assumed to be stationary, the linearization of the relations between SSS and SST, and that the requirements for constructing statistically significant OI matrices degrade the effective resolution of the time-extrapolated SSS maps. Nevertheless, the quality of the reconstructed SSS is very reasonable when validated with independent in situ data. In this work, we propose to go a step forward. We use the multifractal fusion approach first proposed by Umbert et al., Rem. Sen. Env. (2014), to derive the maps of scalar fusion parameters that relate SSS and SST over a given overlap period, with the goal of using those fusion parameters to reconstruct SSS back on time. Multifractal fusion parameters generalize the concept of T-S curves characteristic to water masses, providing a more accurate, geographically localized description of the link SST-SSS: those maps of multifractal fusion parameters have the same resolution as SST maps, thus providing a very detailed geographical description of the dynamic links between SST and SSS and also an improved resolution of the reconstructed maps. Besides, due to the special characteristics of the multifractal fusion parameters (their spatial regularity is granted by construction), they evolve very slowly in time, providing a very stable scheme for reconstructing the sea state of the past. With the multifractal fusion parameters derived from the period of SST-SSS period, we used fixed versions of those parameters applied to long series of SST maps to construct semi-climatologies of SSS. We denote this reconstruction of past states of SSS by “semi-climatologies” because the dynamic compensation between SSS and SST, which is contained in the map of multifractal fusion parameters, is assumed fixed, but the map of SST used to reconstruct SSS corresponds to the specific date at which we want to reconstruct SSS. The SSS maps generated that way contain all the dynamic of SSS than can be explained by its typical dynamic adjustment with SST, but not any transient event which departs significantly from those “standard conditions”. In summary, those maps have variable conditions introduced by SST, and are thus dynamic, but assuming climatologic adjustment: thus the name “semi-climatological”. We have derived SSS semi-climatologies, using SST from ESA CCI+ SST project to construct a decades-long series of SSS maps. We will show how those semi-climatological maps compare to standard SSS maps, analyzing and discussing the differences from a climate perspective. We finally discuss the skill of semi-climatological SSS to describe climate-relevant events such as ENSO or IOD

European Space Agency’s 2019 Living Planet Symposium, 13-17 May 2019, Milan, Italy

Peer reviewed