Directions for SST estimation activities Chris Merchant
A typology of estimation techniques Empirical regression to buoys Optimal estimation of SST & atmosph. Regression to RT modelling Joint estimation of SST-WV- cloud-aerosol? Empirical screening thresholds Probabilistic / dynamic RT RT-based screening thresholds Retrieval Cloud detection
Need to make progress Current coefficient-based retrievals have “features” that are not widely appreciated (NLSST)
NLSST 3 channel Same observations, two different retrievals
Optimal estimation Calculation local coefficients “on the fly” Based on –Local atmospheric state –Physics of radiative transfer What should OE give us? –Minimized random retrieval error –No regional biases related to coefficients –Reduced sensitivity to water vapour –New and powerful metric of retrieval quality But, potentially can introduce different biases
Potential of optimal estimation techniques: Metop-A study Showed that potential benefits of OE can largely be obtained in practice
Two main estimators Maximum a posteriori – MAP –Minimizes SST error variance –Explicitly embeds prior information in result –But may attenuate frontal gradients etc –Probably appropriate for NWP, oceanography Maximum likelihood – ML –Minimizes BT residuals, SST is noisier –“Zero” prior influence in result –Preserves frontal gradients, diurnal variability etc –The only type of SST that should be used in climate re-analyses (if you are a purist) or frontal studies
Reduced SST single-pixel error Operational NLSST Error= 0.47 K MAP split window Error= 0.39 K
Cost: powerful quality indicator MAP errors by confidence level CL 5 (44%) / CL 4 (36%) / CL 3 (19%) / MAP errors by cost Lowest 44% / Next 36% / Poorest 19% /- 0.53
Satellite Radiances Satellite Radiances Satellite Radiances SSTs Operational analysis Re-analysis Value-added service Users Exploiting system of sensors: referencing, bias adjustment
Retrieved SST in some region Biasing factor System-adjusted SST Use differential sensitivity and or auxiliary observations to detect & do mutual correction SSTs of three different sensors Exploiting the system of sensors to reduce biases at source: (1) Learn better how to exploit “reference” sensors (2) Move beyond privileging one reference, mutual correction
Conclusion There is plenty for an Estimation WG to do, some at a pretty fundamental level –Didn’t mention Marginal Ice Zone, MW-IR differences, cloud-related biases, aerosols … –Open questions on OE: spatial correlations of water vapour, state vector elements, forward model errors … Exploiting the system is a big new challenge