Many approaches can be used to reduce the amount of noise present in a given set of data (observed or retrieved). In the SMOS processing chain, weighted averages are used to reduce the noise present in the sea surface salinity (SSS) data retrieved from brightness temperature measurements. This is the rationale of the existence of the higher production levels (Levels 3 and 4) of sea surface salinity and soil moisture.
Data assimilation (i.e., the process of combining observations and numerical models) is a technique that can also be used to reduce the amount of noise in SSS fields. In a recent study we have shown that the salinity field resulting from assimilating SMOS Level 3 (binned) data into an ocean model has less noise than the input data. Moreover, by using the ocean model as a dynamical interpolator, the resulting fields have no data-voids and increased geophysical coherence.
The region of study has been the Macaronesian Region (the Northeast subtropical Atlantic Gyre), where SMOS data is prone to large errors (about 0.50 in the practical salinity scale, Figure 1) due to the vicinity of the large continental masses and the presence of artificial Radio Frequency Interferences (RFI).
Figure 2 shows a SSS zonal transect (26oN) from the Level 3 SMOS data, the output of model simulation without assimilation (FREE-run), the output from various data assimilation experiments, and the co-located Argo data. Model (and Argo) diverges from SMOS data in the eastern part of the basin. This expected effect is linked to the closeness of the continental land. Salinity output from assimilation experiments weighting observations too much (as the so called EXP0) provides unrealistic results (when compared with independent Argo data) in the eastern part of the domain. Appropriate weighting (as for example in experiments EXP1 and EXP2) provides more realistic solutions. In the western part of the transect, the resulting surface salinity moves away from the FREE-Run and gets closer to the SMOS data, illustrating the positive impact of assimilating SMOS data. Also, the salinity from the assimilation experiments demonstrates the noise filtering role by reducing the amplitude of oscillations present in the SMOS product.
The Macaronesian Region is one of regions where data assimilation will be the most challenging due to the large noise in the observations. However, assimilation of monthly-binned SMOS data (1/4 degree resolution) fulfills its objectives as the assimilation results are closer to the Argo in situ data than the original SMOS data. Moreover, assimilation also beats the model when the noise in the input data is small enough (e.g. the April – November period shown in Figure 3).
Details of the work and further results can be found in N. Hoareau et al. (2013), On the potential of data assimilation to generate SMOS-Level 4 maps of sea surface salinity, Remote Sensing of Environment (DOI: 10.1016/j.rse.2013.10.005).