Summary: | An algorithmic approach, based on satellite-derived sea-surface (“skin”) salinities (SSS), is proposed to correct for errors in SSS retrievals and convert these skin salinities into comparable in-situ (“bulk”) salinities for the top-5 m of the subpolar and Arctic Oceans. In preparation for routine assimilation into operational ocean forecast models, Soil Moisture Active Passive (SMAP) satellite Level-2 SSS observations are transformed using Argo float data from the top-5 m of the ocean to address the mismatch between the skin depth of satellite L-band SSS measurements (∼1 cm) and the thickness of top model layers (typically at least 1 m). Separate from the challenge of Argo float availability in most of the subpolar and Arctic Oceans, satellite-derived SSS products for these regions currently are not suitable for assimilation for a myriad of other reasons, including erroneous ancillary air-sea forcing/flux products. In the subpolar and Arctic Oceans, the root-mean-square error (RMSE) between the SMAP SSS product and several in-situ salinity observational data sets for the top-5 m is greater than 1.5 pss (Practical Salinity Scale), which can be larger than their temporal variability. Thus, we train a machine-learning algorithm (called a Generalized Additive Model) on in-situ salinities from the top-5 m and an independent air-sea forcing/flux product to convert the SMAP SSS into bulk-salinities, correct biases, and quantify their standard errors. The RMSE between these corrected bulk-salinities and in-situ measurements is less than 1 pss in open ocean regions. Barring persistently problematic data near coasts and ice-pack edges, the corrected bulk-salinity data are in better agreement with in-situ data than their SMAP SSS equivalent.
|