Cited by 520Oceanographic and Atmospheric ProcessesClimate variability and modelsOcean Waves and Remote Sensing

The DOI matches its Crossref record; 57 of the 57 references it cites were found in a registry or an index; 23 of the 38 sources its claims cite were checked for support.

Satellite-based time-series of sea-surface temperature since 1981 for climate applications

Open access
Merchant, Christopher J., Embury, Owen, Bulgin, Claire E., Block, Thomas, Corlett, Gary K., Fiedler, Emma, Good, Simon A., Mittaz, Jonathan, Rayner, Nick A., Berry, David, Eastwood, Steinar, Taylor, Michael, Tsushima, Yoko, Waterfall, Alison, Wilson, Ruth, Donlon, Craig·2019·Scientific Data

Abstract

A climate data record of global sea surface temperature (SST) spanning 1981–2016 has been developed from 4 × 10 12 satellite measurements of thermal infra-red radiance. The spatial area represented by pixel SST estimates is between 1 km 2 and 45 km 2. The mean density of good-quality observations is 13 km−2 yr−1. SST uncertainty is evaluated per datum, the median uncertainty for pixel SSTs being 0.18 K. Multi-annual observational stability relative to drifting buoy measurements is within 0.003 K yr−1 of zero with high confidence, despite maximal independence from in situ SSTs over the latter two decades of the record. Data are provided at native resolution, gridded at 0.05° latitude-longitude resolution (individual sensors), and aggregated and gap-filled on a daily 0.05° grid. Skin SSTs, depth-adjusted SSTs de-aliased with respect to the diurnal cycle, and SST anomalies are provided. Target applications of the dataset include: climate and ocean model evaluation; quantification of marine change and variability (including marine heatwaves); climate and ocean-atmosphere processes; and specific applications in ocean ecology, oceanography and geophysics.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This paper describes ESA SST CCI v2.1, a global satellite sea-surface temperature climate data record spanning 1981–2016, built from 4 × 10¹² thermal infrared radiance measurements and provided at native, 0.05° gridded, and gap-filled daily 0.05° levels. It reports median per-pixel uncertainty of 0.18 K and multi-annual stability relative to drifting buoys within 0.003 K yr⁻¹ of zero.

Key findings

  • The CDR spans 1981–2016 from 11 AVHRRs and three ATSRs; the mean density of good-quality observations (QL 4 and 5) is 13 km⁻² yr⁻¹, with pixel SST footprint areas between 1 km² and 45 km².
  • Median evaluated uncertainty for L2P pixel SSTs is 0.18 K; total uncertainty is less than 0.8 K for almost all SSTs, and median sensitivity for QL 4 and 5 data is 101%.
  • Multi-annual global observational stability relative to drifting buoy SSTs is, with 95% confidence, in the range −0.0026 to 0.0004 K yr⁻¹.
  • Products include skin SST, depth-adjusted SST at 20 cm de-aliased with respect to the diurnal cycle, and SST anomalies; the L4 analysis is a gap-filled daily-mean 20 cm SST field produced without assimilating in situ measurements.

Methods

  • Inputs are the full ATSR-series v3/v2.1 level-1b archive and AVHRR GAC archives (AVHRR 7, 9, 11, 12, 14, 15, 16, 17, 18, 19 and Metop-A), with auxiliary ERA-Interim NWP fields, OSI-SAF sea-ice concentration, land-sea masks, and pre-calculated look-up tables.
  • Processing includes Bayesian cloud detection, radiance harmonisation referenced to AATSR/ATSR-2, radiative transfer modelling (LBLRTM and RTTOV), skin SST retrieval (coefficient-based for ATSRs; optimal estimation for AVHRRs), per-pixel uncertainty decomposition, depth/time adjustment to 20 cm and 10:30 or 22:30 local mean solar time, and NEMOVAR variational assimilation for the L4 analysis.
  • Validation includes automated inspection of all 465,302 L2P and L3U products (8 TB L2P, 3.4 TB L3U), visual inspection of random subsets, manual metadata checks, and comparison against drifting buoys in skin-raw, skin-skin, and depth-depth categories.

Limitations

  • Stated in the paper: independence from in situ observations is high for 1995 onward, but is significantly lost for 1981–1995 because in situ SSTs were used as a large-scale calibration reference.
  • Stated in the paper: known v2.1 artefacts include unscreened and unadjusted desert dust causing intermittent negative biases of magnitude 1 K in CCI AVHRR SSTs in the north east tropical Atlantic, Red Sea and Gulf of Arabia, with an exaggerated positive trend of about 0.01 to 0.02 K yr⁻¹ in those regions; 1980s single-AVHRR periods with degraded calibration introduce global-mean analysis artefacts of 0.1 to 0.5 K in May 1982, October–December 1982, early August 1983 and late September 1983.
  • Stated in the paper: some error sources are not estimated (cloud-affected pixels passing cloud detection, volcanic aerosol effects, mineral dust, undetected solar contamination), correlation scales for synoptically correlated uncertainty are not fully quantified, and L4 interpolation degrades feature resolution relative to lower-level data.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41597-019-0236-x

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Characteristics of full resolution (L2P) CCI SSTs (derived from ATSR and AVHRR imagery). (a) Number of SST observations per unit area of ocean (“observation density”), per sensor over time. Each dot represents the observations of a single day. The upper panel shows data for the quality levels recommended for climate applications (QL 4 & 5), and the lower panel shows QL 3. (b) Zonal SST observation density per unit area, for all sensors combined and averaged over the time series. (c) Histogram of evaluated uncertainty in SST observations, for all sensors combined over the time series. (d) Histogram of evaluated sensitivity of SST observations, for all sensors combined over the time series.
Fig. 1 Characteristics of full resolution (L2P) CCI SSTs (derived from ATSR and AVHRR imagery). (a) Number of SST observations per unit area of ocean (“observation density”), per sensor over time. Each dot represents the observations of a single day. The upper panel shows data for the quality levels recommended for climate applications (QL 4 & 5), and the lower panel shows QL 3. (b) Zonal SST observation density per unit area, for all sensors combined and averaged over the time series. (c) Histogram of evaluated uncertainty in SST observations, for all sensors combined over the time series. (d) Histogram of evaluated sensitivity of SST observations, for all sensors combined over the time series.CC BY · PubMed Central
Fig. 2 Overview of logic of the production of climate data records (CDRs) for sea surface temperature (SST). Calibrated, geo-located satellite data comprising radiance (or brightness temperature) imagery form the fundamental source data. Many processing parameters have been defined by physics-based algorithm development, addressing the harmonisation of satellite radiances between sensors (i.e., consistency of calibration), probabilistic (Bayesian) pixel classification to determine for which pixels valid SSTs can be obtained, retrieval of the skin SST (to which thermal radiances are directly related), adjustment of skin SST to estimate daily mean SST at a nominal depth comparable to in situ measurements, and estimation of SST retrieval and sampling uncertainty. The retrieval and skin-to-depth adjustment steps are informed by estimates of the past atmospheric conditions at the observation locations from numerical weather prediction re-analysis outputs. The low-level SST CDR is at the full pixel resolution of the input imagery (L2P), is gappy because of clouds, and comprises SST and its per-pixel uncertainty. L2P SSTs are averaged on a grid of 0.05° in latitude and longitude to obtain L3U products, sensor by sensor, with propagated cell-mean uncertainty. L3U SSTs are combined across many sensors to generate an interpolated, gap-filled L4 analysis, with uncertainty estimates.
Fig. 2 Overview of logic of the production of climate data records (CDRs) for sea surface temperature (SST). Calibrated, geo-located satellite data comprising radiance (or brightness temperature) imagery form the fundamental source data. Many processing parameters have been defined by physics-based algorithm development, addressing the harmonisation of satellite radiances between sensors (i.e., consistency of calibration), probabilistic (Bayesian) pixel classification to determine for which pixels valid SSTs can be obtained, retrieval of the skin SST (to which thermal radiances are directly related), adjustment of skin SST to estimate daily mean SST at a nominal depth comparable to in situ measurements, and estimation of SST retrieval and sampling uncertainty. The retrieval and skin-to-depth adjustment steps are informed by estimates of the past atmospheric conditions at the observation locations from numerical weather prediction re-analysis outputs. The low-level SST CDR is at the full pixel resolution of the input imagery (L2P), is gappy because of clouds, and comprises SST and its per-pixel uncertainty. L2P SSTs are averaged on a grid of 0.05° in latitude and longitude to obtain L3U products, sensor by sensor, with propagated cell-mean uncertainty. L3U SSTs are combined across many sensors to generate an interpolated, gap-filled L4 analysis, with uncertainty estimates.CC BY · PubMed Central
Fig. 3 Timeline of contributing missions, geophysical and on-orbit events affecting data quality, and El Nino Southern Oscillation (ENSO) phase.
Fig. 3 Timeline of contributing missions, geophysical and on-orbit events affecting data quality, and El Nino Southern Oscillation (ENSO) phase.CC BY · PubMed Central
Fig. 4 Logic and thresholds for assigning pixel quality level (QL) flags to L2P SSTs. “Invalid data” means input data are flagged as invalid. “Land” means the centre of the pixel is not over ocean. “Bad NWP” means tests on the integrity of the prior fields are failed. Pclear is the posterior probability of the pixel being clear-sky over ice-free ocean. “sens” is the evaluation of sensitivity of the retrieval to true changes in SST. χ2 is the channel-normalised result of a test on the goodness of fit achieved in retrieval (applies to optimally estimated SST only). θsat is the satellite zenith angle of the retrieval. “twilight” means the solar zenith angle at the target pixel is within 5° of 90°. “DDI” refers to an index for presence of desert (mineral) dust aerosol (available for ATSRs only). The Pclear threshold of 0.90 for designation as QL > 3 is for the ATSRs and AVHRRs at night-time, and for AVHRRs observations during daytime is instead 0.99.
Fig. 4 Logic and thresholds for assigning pixel quality level (QL) flags to L2P SSTs. “Invalid data” means input data are flagged as invalid. “Land” means the centre of the pixel is not over ocean. “Bad NWP” means tests on the integrity of the prior fields are failed. Pclear is the posterior probability of the pixel being clear-sky over ice-free ocean. “sens” is the evaluation of sensitivity of the retrieval to true changes in SST. χ2 is the channel-normalised result of a test on the goodness of fit achieved in retrieval (applies to optimally estimated SST only). θsat is the satellite zenith angle of the retrieval. “twilight” means the solar zenith angle at the target pixel is within 5° of 90°. “DDI” refers to an index for presence of desert (mineral) dust aerosol (available for ATSRs only). The Pclear threshold of 0.90 for designation as QL > 3 is for the ATSRs and AVHRRs at night-time, and for AVHRRs observations during daytime is instead 0.99.CC BY · PubMed Central
Fig. 5 SST content in data in different product levels. L2P data are on the original viewing geometry, with gaps in SST from cloud cover. These are gridded per orbit to L3U. A day’s worth of L3U SSTs from a given sensor is collated to form an L3C product. Data from multiple sensors are merged and interpolated to give the daily gap-free SST field of the L4 analysis.
Fig. 5 SST content in data in different product levels. L2P data are on the original viewing geometry, with gaps in SST from cloud cover. These are gridded per orbit to L3U. A day’s worth of L3U SSTs from a given sensor is collated to form an L3C product. Data from multiple sensors are merged and interpolated to give the daily gap-free SST field of the L4 analysis.CC BY · PubMed Central
Fig. 6 Overview of validation results. Robust standard deviation (a) and median discrepancy (b) for comparison of CCI ATSR and AVHRR SSTs with drifting buoys. Depth-depth validation results are shown, in which satellite skin SSTs have been adjusted to 10:30 am/pm local time and in situ records have been temporally interpolated to the same time. (c) Time/latitude variation of analysis minus drifting-buoy SST differences, averaged zonally and monthly.
Fig. 6 Overview of validation results. Robust standard deviation (a) and median discrepancy (b) for comparison of CCI ATSR and AVHRR SSTs with drifting buoys. Depth-depth validation results are shown, in which satellite skin SSTs have been adjusted to 10:30 am/pm local time and in situ records have been temporally interpolated to the same time. (c) Time/latitude variation of analysis minus drifting-buoy SST differences, averaged zonally and monthly.CC BY · PubMed Central

Verification

Claims the paper makes

51 No basis given · 26 Own result · 22 Own interpretation · 20 Own method · 10 Supported · 7 No full text of the sources · 6 Partly supported · 4 Sources not found · 2 Supported so far · 1 Couldn’t confirm · 1 Not checked yet · accuracy not yet measured

Show all 81 claims

  • It may state two separate claims as one. 34
  • Probably not a substantial claim. 47
  • The quote only partly states this claim. 4
69 of 69
  • The authors state this as an assumption.

    In the cloud detection calculation, P(x|c) is assumed equal to P(x).

    “This expression simplifies assuming P(x|c) = P(x), since the background state has length scales of ~100 km and does not resolve cloud structures at the finer scales ~1 to ~10 km relevant to the cloudiness of individual pixels.”
  • The authors state this as an assumption.

    The smooth-atmosphere-but-variable-SST assumption: TCWV for the centre pixel and surrounding clear pixels are identical.

    “The smooth-atmosphere-but-variable-SST assumption is imposed by fixing the TCWV for the centre pixel and the surrounding clear pixels to be identical (hence only one TCWV term in z).”
  • The paper relies on this without support.

    Many users seek an estimate of SST at depths of order tens of centimetres.

    “many users seek an estimate of SST at depths of order tens of centimetres, whether because this makes the observations compatible with many historic in situ SSTs from drifting buoys and bucket measurements, or because this depth is more comparable to the upper layer of an ocean model.”
  • The paper relies on this without support.

    Users typically treat the gridded SST as a value representative of the cell as a whole.

    “users typically treat the gridded SST as a value representative of the cell as a whole”
  • The authors state this as an assumption.

    The errors in the new SSTs are uncorrelated within the analysis system.

    “For computational simplicity, the errors in the new SSTs are assumed to be uncorrelated within the analysis system.”
  • The paper relies on this without support.

    The degree of independence between the measurements being compared is important in interpreting validation results.

    “In interpreting validation results, the degree of independence between the measurements being compared is important.”
  • The paper relies on this without support.

    Independence of validation personnel from remote sensing and product generation teams secures the objectivity of validation results.

    “To further secure the objectivity of validation results, the personnel performing the validation analysis were independent of the teams undertaking the remote sensing research and product generation.”
7 of 7

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Cited by

Sorted by citations

See all 520

Metadata

Authors
Merchant, C. J., Embury, O., Bulgin, C. E., Block, T., Corlett, G. K., Fiedler, E., Good, S. A., Mittaz, J., Rayner, N. A., Berry, D., Eastwood, S., Taylor, M., Tsushima, Y., Waterfall, A., Wilson, R., & Donlon, C.
Venue
Scientific Data
Published
2019-10-22
Volume
6
Issue
1
Pages
223-223
ISSN
2052-4463 (ISSN-L)
OpenAlex
W2981879507
Cited by
520
References
56
Citations by year
2026: 27 · 2025: 88 · 2024: 89 · 2023: 94 · 2022: 111 · 2021: 77 · 2020: 27 · 2019: 4
Type
data paper
Language
en
Open access
Locations
  • Scientific Data (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • CentAUR (University of Reading) (acceptedVersion, cc-by)
  • UEA Digital Repository (University of East Anglia) (acceptedVersion, cc-by)
  • PubMed Central (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • ePubs (Science and Technology Facilities Council, Research Councils UK) (submittedVersion)
Funders
European Space Agency (4000109848/13/I-NB)
Topics
  • Physical Sciences › Earth and Planetary Sciences › Oceanography › Oceanographic and Atmospheric Processes
  • Physical Sciences › Environmental Science › Global and Planetary Change › Climate variability and models
  • Physical Sciences › Earth and Planetary Sciences › Oceanography › Ocean Waves and Remote Sensing
Keywords
sea surface temperature, satellite SST, climate data record, thermal infrared radiance, drifting buoys, skin temperature, sea surface temperature anomalies, marine heatwaves, ocean variability, ocean model evaluation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-16, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-16 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review