1. Arctic SST retrieval in the CCI project Owen Embury Chris Merchant University of Reading

2. SST CCI Phase 1 Combine ATSR accuracy with AVHRR coverage Optimal Estimation (OE) retrieval – Cross referenced to ARC SST Diurnal variability adjustment – Report SSTs at standard depth and time of day Uncertainty estimation Product specification – NetCDF 4 with classic data model – GDS2.0 compliant

3. SST CCI Phase 1 Long-term (Aug 1991 – Dec 2010) – ATSR OE retrieval Bayesian cloud screening L3U (0.05°) – AVHRR GAC OE retrieval CLAVR-X cloud screening Ice detection L2P – SST skin at time of observation – SST 0.2m at 10:30 local time Adjusted to nearest am/pm

4. Bayesian Cloud Detection Use RTTOV to simulate expected observations from ECMWF NWP Calculate P(obs | clear) from obs-sim differences Get P(obs | cloud) from empirical lookup table Use Bayes to get P(clear | obs, nwp) P(obs | clear) is dominant factor – ECMWF NWP – RTTOV forward model – Prior error assumptions Prior SST error is location dependent (0.5 to 1.8 K)

5. Bayesian Cloud Detection Can consider the current system as Bayesian “clear-sky” detection Problem detecting conditions which look like clear-sky in infrared – Seaice – Fog Potential improvements – Add fog / seaice as extra classes for detection Needs software refactoring – Use visible channels Not done yet due to ARC software heritage (ARC needed method applicable to ATSR1) Daytime only – Review prior SST error assumption in Arctic areas

6. OE SST retrieval MAP formulation with prior SST error of 5 K – Reduces influence of prior on retrieval QC check on χ 2 to remove bad retrievals – Calculation of χ 2 similar to P(obs | clear) in Bayesian cloud detection QC check to remove SSTs < 271.35 K – Not applied in pre-release data

7. Comparison with other datasets Compare 5 day composite images – Pathfinder v5.2 – ARC v1.1.1 – AMSR-E v7 – OSTIA – CCI AVHRR L2P ATSR L3U Show with OSISAF sea ice concentration – 15% and 85% contour lines

8. 2008 Norwegian and Greenland Seas

22. 2010 Norwegian and Greenland Seas

32. 2008 Ice melt in Beaufort Sea and outflow from Mackenzie river

46. 2010 Ice melt in Beaufort Sea and outflow from Mackenzie river

60. Summary AVHRR – CCI generally has better coverage than Pathfinder – Pathfinder less likely to reject extreme warm SSTs in Mackenzie outflow ATSR – CCI and ARC consistent in images shown CCI-ARC differences exist but << 1 K Large temperature anomalies can be problematic for Phase 1 CCI – Both SST retrieval and Bayesian cloud screening used ECMWF-interim SST (OSTIA) as prior

61. SST CCI Phase 1 Demonstration (two 3 month periods) – AATSR L2P (1km) OE retrieval Bayesian cloud screening – Metop-A AVHRR L3C (0.05°) Meteo France retrieval Meteo France cloud screening – SEVIRI L3C (0.05°) Meteo France retrieval Meteo France cloud screening – AMSR-E L2P (0.25°) RSS retrieval – TMI L2P (0.25°) RSS retrieval

62. Known bugs in pre-release data Two biggest issues caused by “minor” bugs No data after midnight – Unsigned int bug in pre-processing code – Last orbit in day cut off at mid-night – Regular gaps in ATSR data Missing data flagged as quality_level 5 – QC based on cloud mask and uncertainty information – OE retrieval can produce SST < 271.15 K – SST-CCI product can not store SST values < 271.15 K