Is ocean wind stress from scatterometers for ocean forcing and NWP model initialization underexploited?
by
The golden age of wind scatterometry is upon us with three launches this year and the benefits from scatterometer winds are hence rising. However, do we really optimally profit from these accurate data at the atmosphere-ocean interface? A first concern exists in large and persistent local biases, i.e., of the size of the standard deviation of o-b, as documented in Belmonte and Stoffelen (2019) for ERA5. After scatterometer wind assimilation, the model dynamics restore the local biases after a few time steps and hence scatterometer data assimilation is ineffective. Following the Best Linear Unbiased Estimate (BLUE) paradigm of data assimilation, a local VARBC may be helpful to better exploit the dynamical mesoscale information content of wind scatterometers for NWP data assimilation (Stoffelen and Vogelzang, 2021).
The model wind stress errors are associated with lacking mesoscale turbulence, PBL parameterization errors, tropical moist convection and too weak air-sea interaction. Moreover, ocean currents will also cause a difference as scatterometer winds are relative to the moving ocean surface. A map of average scatterometer-model biases illustrates the different artefacts and may be used as basis for attribution of model errors. Moreover, such biases appear persistent over several days, seasons and years and hence might be useful as a correction to NWP/ERA model winds before ocean forcing. This is now developed as a Copernicus Marine Environment Monitoring Service (CMEMS) Level-4 product, based on ERA5 and OPS. Besides biases, also variances will be compared, both for wind, stress and spatial derivatives and eventually their role evaluated in ocean forcing and air-sea interaction.
Scatterometer winds have spatially uniform quality and 25-km resolution, hence thinning the winds in data assimilation leads to a loss of mesoscale spatial information. Are there better ways to use the scatterometer winds than thinning? Moreover, what scales are traceable in the analysis over the ocean? We experimented with superobbing and supermodding to target those spatial scales that are effectively initialized in the analysis. Some results will be shown.