Improving the quality of Sentinel-3A data with a hybrid mean sea surface model, and implications for Sentinel-3B and SWOT

摘要

In this paper we compute a new local mean sea surface (MSS) model along the Sentinel-3A ground track. This so-called hybrid mean profile (HMP) blends the content of an average of 18 months of Sentinel-3A data for wavelengths ranging from 15 to 100 km, and the CNES/CLS 2015 gridded MSS model for larger and shorter scales. The improvement observed on Sentinel-3A sea level anomalies (SLA) is significant: the residual error is 0.2 cm~2, i.e. 17% of the SLA variance between 15 and 100 km, or 57% less than the gridded MSS model error. The highest error reduction is observed for wavelengths ranging from 20 to 80 km. From a geographical point of view, the improvement is mainly located along geodetic features that are not completely resolved in the gridded MSS models. It can locally be as high as 1 cm~2, i.e. very large when compared to the variance of the small scale SLA. Similarly, in coastal regions where the gridded model is known to exhibit higher errors, the HMP exhibits a very stable behavior that is on average 4 times more accurate. To understand the implications for future datasets and mission, we also develop a simple prediction model for the leakage of noise and small scale SLA into the MSS model. The model was validated with Sentinel-3A and ENVISAT data. Using the HMP strategy on the 21-day phase of the SWOT mission would be attractive for two reasons: I/this methodology would reduce the small scale error of gridded MSS models thanks to SWOT's unprecedented 2D topography coverage and noise level, and 2/the gridded MSS models provides a more trustworthy reference for the larger temporal scales that cannot be not averaged out by SWOT alone. After 3 years of nominal mission, the residual SWOT HMP error should be less than 2% of the sea level anomaly variance. In contrast, the so-called fast-sampling (or 1-day repeat) phase of SWOT is slightly more challenging because of the temporal correlation of the SLA (1-day samples are not independent). Depending on the accuracy of the pre-launch gridded MSS (i.e. upper wavelength limit where SWOT data must be used), the decorrelation scales could range from 1 to 5 days. The resulting HMP error would ranges from 5 to 12% of the SLA variance at the end of the fast-sampling phase. These results emphasize the need to keep improving the smaller scales of gridded MSS models as they will remain a major altimetry asset, at least until 2023, when SWOT has collected enough 21-day samples to provide a very robust HMP model.