Progress in Data Progress in Data Merging/Gridding & CDRMerging/Gridding & CDR

and and Differences Between Global Differences Between Global

SST Averages in Gridded SST Averages in Gridded Data SetsData Sets

Alexey KaplanAlexey KaplanLamont-Doherty Earth Observatory of Columbia Lamont-Doherty Earth Observatory of Columbia

University, Palisades, NY 10964, USAUniversity, Palisades, NY 10964, USA

Recommendations from Recommendations from WHITE PAPER WHITE PAPER

Data Merging, Gridding, and Analysis:Data Merging, Gridding, and Analysis:

1. Begin accounting for correlated errors in 1. Begin accounting for correlated errors in implementations of data gridding implementations of data gridding procedures. procedures.

2. Research into misspecification of a priori OI 2. Research into misspecification of a priori OI parameters on OA fields. parameters on OA fields.

3. Research into non-stationary structures. 3. Research into non-stationary structures.

4. Research into optimizing inter-sensor bias 4. Research into optimizing inter-sensor bias correction.correction.

Other relevant activities:Other relevant activities:

•Dataset Dataset development/improvemendevelopment/improvement/productiont/production

•User guidance (BAMS User guidance (BAMS paper)paper)

•CDRCDR

GHRSST and SST productsGHRSST and SST productsGlobal Multi-Global Multi-Product Product Ensemble (GMPE):Ensemble (GMPE):Met Office Met Office OSTIA SST analysis

NCEP NCEP RTG_SST_HR SST analysis

NAVOCEANO NAVO K10 SST observationsNAVOCEANO NAVO K10 SST observations JMA MGDSST SST analysisJMA MGDSST SST analysis

RSS RSS MW Fusion SST analysisRSS RSS MW Fusion SST analysisRSS RSS MW+IR Fusion SST analysisRSS RSS MW+IR Fusion SST analysis

FNMOC GHRSST SST and sea Ice analysisFNMOC GHRSST SST and sea Ice analysis

MERSEA ODYSSEA SST analysis MERSEA ODYSSEA SST analysis

NOAA AVHRR OI (Reynolds)NOAA AVHRR OI (Reynolds)

Meterological Service of Canada (CMC) Meterological Service of Canada (CMC)

BMRC GAMSSA SST analysisBMRC GAMSSA SST analysis

Data Merging, Gridding, and Analysis:Data Merging, Gridding, and Analysis:

1.1.Begin accounting for correlated errors in Begin accounting for correlated errors in implementations of data gridding implementations of data gridding procedures. procedures.

OSTIA developmentOSTIA development

3. Research into non-stationary structures:3. Research into non-stationary structures:

MUR (1km resolution)MUR (1km resolution)

4. Research into optimizing inter-sensor 4. Research into optimizing inter-sensor bias correction: bias correction: OSTIA, ODYSSEAOSTIA, ODYSSEANew operational products in the New operational products in the U.S.:U.S.:5km geostationary analysis, 5km geostationary analysis, Reynolds 2-step HR OI Reynolds 2-step HR OI

Two seminal papers published, documenting the utility of GMPE and SQUAM:Two seminal papers published, documenting the utility of GMPE and SQUAM:

Martin, M., P. Dash, A. Ignatov, V. Banzon, H. Beggs, B. BrasnettMartin, M., P. Dash, A. Ignatov, V. Banzon, H. Beggs, B. Brasnett , J.-F. Cayula, J.-F. Cayula, J. , J. CummingsCummings, C. Donlon, C. Donlon, C. Gentemann, R. Grumbine, S. Ishizaki, E. Maturi, R. W. , C. Gentemann, R. Grumbine, S. Ishizaki, E. Maturi, R. W. Reynolds, J. Roberts-Jones, Reynolds, J. Roberts-Jones, Group for High Resolution Sea Surface temperature Group for High Resolution Sea Surface temperature ((GHRSSTGHRSST) analysis fields inter-comparisons. Part 1: A GHRSST multi-product ) analysis fields inter-comparisons. Part 1: A GHRSST multi-product ensemble (ensemble (GMPEGMPE), ), Deep Sea Research Part II: Topical Studies in OceanographyDeep Sea Research Part II: Topical Studies in Oceanography , , Available online 2 May 2012, ISSN 0967-0645, 10.1016/j.dsr2.2012.04.013. Available online 2 May 2012, ISSN 0967-0645, 10.1016/j.dsr2.2012.04.013.   Dash, P., A. Ignatov, M. Martin, C. Donlon, B. Brasnett, R. W. Reynolds, Dash, P., A. Ignatov, M. Martin, C. Donlon, B. Brasnett, R. W. Reynolds, V. Banzon, H. Beggs,V. Banzon, H. Beggs, J.-F. Cayula, Y. Chao, J.-F. Cayula, Y. Chao, R. Grumbine, E. Maturi,R. Grumbine, E. Maturi, A. Harris, J. A. Harris, J. Mittaz, J. SapperMittaz, J. Sapper, T. M. Chin, J. Vazquez-Cuervo, E. M. Armstrong, , T. M. Chin, J. Vazquez-Cuervo, E. M. Armstrong, C. Gentemann, J. Cummings, J.-F. Piollé, E. Autret, J. Roberts-Jones, C. Gentemann, J. Cummings, J.-F. Piollé, E. Autret, J. Roberts-Jones, S. Ishizaki, J. L. Høyer, D. Poulter, S. Ishizaki, J. L. Høyer, D. Poulter, Group for High Resolution Sea Surface Group for High Resolution Sea Surface Temperature (Temperature (GHRSSTGHRSST) analysis fields inter-comparisons—Part 2: Near real ) analysis fields inter-comparisons—Part 2: Near real time web-based level 4 SST Quality Monitor (time web-based level 4 SST Quality Monitor (L4-SQUAML4-SQUAM), ), Deep Sea Research Part Deep Sea Research Part II: Topical Studies in OceanographyII: Topical Studies in Oceanography, Available online 17 April 2012, ISSN 0967-, Available online 17 April 2012, ISSN 0967-0645, 10.1016/j.dsr2.2012.04.002.0645, 10.1016/j.dsr2.2012.04.002.

Special Thanks to Jorge Vazquez, a Special Thanks to Jorge Vazquez, a Special Editor of this DSR-II Special Special Editor of this DSR-II Special Issue!Issue!

Recommendations from Recommendations from WHITE PAPER WHITE PAPER

Data Merging, Gridding, and Analysis:Data Merging, Gridding, and Analysis:

1. Begin accounting for correlated errors in 1. Begin accounting for correlated errors in implementations of data gridding implementations of data gridding procedures. procedures.

2. Research into misspecification of a priori 2. Research into misspecification of a priori OI parameters on OA fields. OI parameters on OA fields. Reynolds et Reynolds et al. 2013 synthetic data test for L4 al. 2013 synthetic data test for L4 methodologymethodology

GMPE ensemble median GMPE ensemble median isismore more accurateaccurate than than individual members individual members (left); GMPE ensemble (left); GMPE ensemble spread can be used as spread can be used as a a proxy proxy for the error in for the error in its median (up). its median (up). Suppose we’ve reduced Suppose we’ve reduced the number of L4 the number of L4 products to 1 or 2: products to 1 or 2: GMPE ensemble will GMPE ensemble will disappear!disappear!

Impacts of AATSR loss on the L4 Impacts of AATSR loss on the L4 productsproducts

• Sasha noticed the recent increase in the Sasha noticed the recent increase in the spread between some L4 products and spread between some L4 products and independent data in SQUAM. It might be a independent data in SQUAM. It might be a manifestation of the the AATSR data loss.manifestation of the the AATSR data loss.

• Prasanjit agreed to investigate further and Prasanjit agreed to investigate further and to try to document this using SQUAM to try to document this using SQUAM ((action itemaction item). If successful, this might ). If successful, this might become a useful illustration of the become a useful illustration of the importance of the AATSR data source. importance of the AATSR data source.

Synthetic data tests for L4 Synthetic data tests for L4 products products ((Chelton and ReynoldsChelton and Reynolds method)method)• HR ECCO model SST is sub-sampled/corrupted HR ECCO model SST is sub-sampled/corrupted

as if it came as the satellite data stream. Obj. as if it came as the satellite data stream. Obj. An. procedure for a given L4 product is applied, An. procedure for a given L4 product is applied, the results are compared with the true model the results are compared with the true model values. First (by G12) was done for the NCDC values. First (by G12) was done for the NCDC OIOI

• Now has been done for OSTIA too (J.R.-J & M.M.)Now has been done for OSTIA too (J.R.-J & M.M.)

• L4 producers at GHRSST were enthusiastic to L4 producers at GHRSST were enthusiastic to do this too (Helen, Jacob, Viva, Mike, Jean-do this too (Helen, Jacob, Viva, Mike, Jean-Francois) and have a joint paper about it.Francois) and have a joint paper about it.

• We’ll offer this to all L4 producersWe’ll offer this to all L4 producers

Transition to the modern Ocean Observing SystemTransition to the modern Ocean Observing System

From From Woodruff et al.Woodruff et al. [2008],[2008],In In Climate Variability and Climate Variability and Extremes during the PastExtremes during the Past100 Years100 Years, Bronniman , Bronniman et al.et al.(eds.) (eds.)

Merchant et al.Merchant et al., JGR, , JGR, 20122012

Despite the efforts to correct inter-Despite the efforts to correct inter-platform biases in the SST data platform biases in the SST data used for producing gridded data used for producing gridded data sets, the remaining biases are sets, the remaining biases are significant enough to create easily significant enough to create easily discernible differences between discernible differences between global means estimated from such global means estimated from such gridded data sets. gridded data sets.

• Global means from annually averaged Global means from annually averaged OSTIA SST is systematically colder than OSTIA SST is systematically colder than that from the NCDC Daily 0.25that from the NCDC Daily 0.25oo AVHRR- AVHRR-only OI data set by about 0.1only OI data set by about 0.1ooC, while the C, while the latter is colder than the same estimated latter is colder than the same estimated from the (older) NCEP monthly 1from the (older) NCEP monthly 1oo OI by OI by approximately the same amount. approximately the same amount.

• While historical SST data sets that make While historical SST data sets that make use of the AVHRR data (HadISST1 and use of the AVHRR data (HadISST1 and COBE SST) show very good consistency COBE SST) show very good consistency with the NCEP monthly 1with the NCEP monthly 1oo OI, they are OI, they are colder than the products that use only colder than the products that use only in in situsitu data (ERSST v3b, HadSST2, HadSST3, data (ERSST v3b, HadSST2, HadSST3, ICOADS). ICOADS).

• The global mean difference between The global mean difference between these two groups of gridded historical these two groups of gridded historical data sets becomes especially prominent data sets becomes especially prominent after 2000, exceeding 0.1after 2000, exceeding 0.1ooC in some C in some years. All these differences are years. All these differences are not not due due to differences in the domains of the data to differences in the domains of the data sets (they appear in co-located sets (they appear in co-located calculations as well) or can be reasonably calculations as well) or can be reasonably explained by random error effects on explained by random error effects on global annual SST averages. global annual SST averages.

• Systematic differences between ship and Systematic differences between ship and buoy data and remaining cold biases in the buoy data and remaining cold biases in the AVHRR data seem responsible for the global AVHRR data seem responsible for the global mean differences between historical data mean differences between historical data sets during the satellite period. Global mean sets during the satellite period. Global mean differences between individual L4 products differences between individual L4 products have to be traced to their input data sets have to be traced to their input data sets and their inter-platform bias removal and their inter-platform bias removal procedures.procedures.

• Homogenization of historical data sets in Homogenization of historical data sets in terms of a common reference across terms of a common reference across satellite and pre-satellite periods is yet to satellite and pre-satellite periods is yet to be satisfactorily resolved in the community, be satisfactorily resolved in the community, even with regards to the annual global SST even with regards to the annual global SST means means