A study by Joseph Tenerelli from CLS/brest
Modeling studies conducted by several teams prior to SMOS launch indicated that the downwelling celestial radiations at L-band that are scattered back by the ocean surface toward the upper hemisphere can be a source of brightness contamination affecting the quality of sea surface salinity retrieval.
For sun-synchronous polar-orbiting satellite measurements of upwelling L-band radiation over the ocean, like with SMOS, this so-called sky noise depends strongly on pass direction (ascending or descending), time of year and surface roughness (wind speed).
Based upon the modeling studies for SMOS sensor, the impact is expected to be strongest for descending passes in September because for these passes the reflections of the instrument viewing directions over the field of view tend to lie along the galactic equator where L-band galactic emission is maximum.
If the contribution to the brightness temperatures measured by SMOS from celestial sky noise is underpredicted, the residual brightness temperatures will be overpredicted and the resulting sea surface salinity will be underpredicted (because brightness temperature increases as surface salinity decreases).
As illustrated in the figures below, recent SSS retrievals produced by DPGS show along-track strips of low SSS in descending orbits that are aligned with the expected maximum in galactic noise contribution to the brightness temperatures.
Left: Image of a SMOS Level 2 SSS retrieval for one 1/2 orbit in Descending passes the 25th of september. Note the blue (too fresh) along-track strips due to bad correction of the galactic signal reflection on the sea surface. Right: Model prediction of the specularly reflected galactic equator radiations for that period (Kelvins).
The most convincing evidence that the galactic noise model is the origin of these low SSS streaks is obtained by collecting the SMOS data onto celestial dwelllines and subtracting from the SMOS brightness temperatures ((Tx+Ty)/2 all but the galactic noise model.
What remains should be the reflected celestial sky noise,and indeed that is what we find. Below is a gif animated (click on it to see the animation) of the median of the differences between alias free FOV SMOS brightness temperatures and the radiative transfer forward model without skynoise projected on to the celestial sphere (here we assumed SSS to be equal the world Ocean Atlas climatology). The first image in the animation is showing the model for the specularly reflected sky image as shall be seen by SMOS if the ocean surface were perfectly smooth, the following images are showing the SMOS data classified as function of the surface wind speed values by 3m/s wide bins (from low to higher winds).
Click on the image above to see the gif-animation
To make these maps we have used many Pacific ascending and descending passes from April through September,although the signal appears clearly even in singlepasses. As expected, the ocean surface reflected images of the L-band sky radiations seen by SMOS are blurred by the roughness of the sea surface. The stronger the surface winds, the weaker the scatteredsignal from the galactic equator and the more spread the associated signal.
Examination of the SMOS brightness temperatures and comparison with the scattering model reveals deficiencies in the modeling of the celestial sky noise that are at the origin of these strips of spuriously low SSS: in general we underestimate the scattered energy from the galactic equator, meaning that our sea surface scattering model is overpredicting the roughness effect. An adapted correction is understudy.
Clearly the stability of the galactic sources and the SMOS imaging capacity of these sky sources is a very promising tool for correctly adjusting our rough-sea surface scattering model at L-band. More details can be found in a detail presentation here