Contributors: D. Ghent (UoL), K. Veal (UoL), G.Kirches (BC)

Issued by: University of Leicester/ Darren Ghent

Date: 10/12/2025

Ref: C3S2_313e_BC_WP3-DR-LST-LST_CCI-v3.00-1995-2024_202506_PQAR

Official reference number service contract: 2024/C3S2_313e_BC/SC1

<style>
.special_indent ul > li > ul {
    padding-left: 0;
}

.special_indent ul > li > ul > li > ul {
    padding-left: 1.5em;
}
</style>

History of modifications

List of datasets covered by this document

Acronyms

General definitions

Accuracy: is the degree of “closeness of the agreement between the result of a measurement and a true value of the measurand” (JCGM, 2008). Commonly, accuracy is represented as a description of systematic errors and a measure of statistical bias.

Bias: Defined as “the closeness of agreement between the result of a measurement and a true value of a measurand” (JCGM, 2008) and is the inaccuracy by virtue of a systematic error. In LST_CCI documents bias refers to the estimated magnitude of the systematic error.

Brightness temperature (BT): the Brightness Temperature is the temperature of a black body that would have the same radiance as the radiance actually observed with the radiometer. Specifically indicates the directional temperature obtained by equating the measured radiance with the integral over wavelength of the Planck’s Black Body function times the sensor response. (Norman and Becker, 1995).

Brokered Climate Data Record: Climate Data Record processed by a partner agency and then integrated into C3S. 

Calibration: Calibration is the process of quantitatively defining the system response to known, controlled system inputs (Guillevic et al., 2018).

Difference / Discrepancy: The difference between an observation and the reference value.

Error: Result of a measurement minus a true value of the measurand. Note that in practice a true value cannot be determined and therefore a conventional true value is used instead (JCGM, 2008).

Land Surface emissivity (LSE): Surface emissivity of an isothermal, homogeneous body is defined as the ratio of the actual emitted radiance to the radiance which would be emitted from a perfectly emitting surface (i.e. ‘blackbody’) at the same thermodynamic temperature” (Norman and Becker, 1995).

Land surface temperature (LST): Land Surface Temperature (LST) is the aggregated radiative skin temperature derived from thermal radiation of all objects comprising the surface, measured in situ and estimated from satellite. It is a basic determinant of the terrestrial thermal behaviour, as it controls the effective radiating temperature of the Earth’s surface (Norman and Becker, 1995; Guillevic et al., 2018).

Level-1 (L1b): radiometrically calibrated and geometrically corrected radiances or brightness temperatures presented on the orbit swath at native resolution and geolocated to latitude and longitude of centres (and/or corners) of pixels or to tie-points. Should the geolocation information be given at tie-points only, then the user is required to perform an interpolation in order to geolocate pixels in between the tie-points.

Level-2 (L2): geophysical variables, e.g. LST, derived from L1 data at same resolution as L1 data i.e. not spatially or temporally manipulated. Geolocation information may be included with the L2 product, but not necessarily at full resolution, or in other cases the user must source the geolocation information from a L1 product.

Level 3 (L3): geophysical product that has been temporally or spatially manipulated and in a gridded map projection format e.g. daily LST on a 0.01° longitude by 0.01° latitude global grid.

Level-3 uncollated (L3U): L2 data regridded to a spatial grid without combining data from different orbits.

Level-3 collated (L3C): L2 data from a single instrument regridded to a space-time grid, data from several orbits may be combined.

Level-3 super-collated (L3S): L3 data from multiple instruments combined in a space-time grid.

Precision: Defined as the “closeness of agreement between indications or measured quantity values obtained by replicate measurements on the same or similar objects under specified conditions” (JCGM, 2008). In this case the precision will be expressed as the standard deviation of the measured bias, which is the estimated magnitude of the systematic error.

Reference standard: Measurement standard designated for the calibration of other measurements standards for quantities of a given kind in a given organization or at a given location (Guillevic et al., 2018).

Skin temperature: The temperature of a layer of a medium of depth equal to the penetration depth of the electromagnetic radiation at the given wavelengths (Norman and Becker, 1995). Surface brightness and radiometric temperatures are the effective temperature that a radiometer would measure near the surface, including emissivity effects and reflected downwelling radiance.

Uncertainty: A parameter associated with the result of a measurement, that characterizes the dispersion of the values that could reasonably be attributed to the measurand, that is the value of the particular quantity to be measured (JCGM, 2008).

Validation: Validation is the process of assessing, by independent means, the quality of the data products derived from those system outputs (Guillevic et al., 2018).

Executive summary

The Copernicus Climate Change Service (C3S) delivers a Climate Data Record (CDR) for LST for the time period 1995 – 2024 inclusive, which will be continued forward in time as an Interim Climate Data Record (ICDR). The C3S LST CDR provides the global-scale information for climate applications. The output Level-3S global monthly datafiles are provided in NetCDF CF-compliant format and include LST as the main variable, with accompanying variables on LST uncertainty and quality flags. Two datafiles are provided for each month, one for daytime LST and one for night-time LST. Data are produced and output at 0.01˚ spatial resolution.

The C3S LST is brokered from both the Climate Change Initiative Land Surface Temperature (LST_cci) project and the UK Earth Observation Climate Information Service (EOCIS). LST_cci provides the CDR covering the years 1995 – 2023, and EOCIS provides the Interim Climate Data Record (ICDR) which covers the full 12-months of 2024. The data record is a multi-sensor product which is called the Infrared Climate Data Record (IRCDR) with a version number of 3.00. All data across the multi-sensor CDR and ICDR are consistent in terms of intercalibration, calibration database, retrieval algorithm, uncertainty model, and cloud detection approach. The contribution of each of the individual sensors to the global monthly multi-sensor IRCDR can be summarised as:

• ERS-2 Along Track Scanning Radiometer (ATSR-2): 1995-06-01 to 2002-05-31
• Envisat Advanced Along Track Scanning Radiometer (AATSR): 2002-06-01 to 2012-03-31
• EOS Terra Moderate Resolution Imaging Spectroradiometer (MODIS): 2012-04-01 to 2018-11-30
• Sentinel-3B Sea and Land Surface Temperature Radiometer (SLSTR): 2018-12-01 to 2024-12-31.

This Product Quality Assessment Report (PQAR) provides the quality assessment results for the C3S LST product (CDR LST_CCI LST v3.00 1995-2024). This is achieved by validation of the datasets. Specifically, this document reports the results and interpretations of the validations carried out within LST_cci on the datasets brokered by C3S. The methodology was first documented in the LST_cci Product Validation Plan (PVP) (Perez-Planells and Martin, 2022), with the results taken from the LST_cci Product Validation and Intercomparison Report (PVIR) (Perez-Planells and Martin, 2025).

The validation is obtained in two steps:  the satellite datasets are firstly validated against in situ point measurements, and then against reference third party datasets. The first step, known as Category A validation in the CEOS-WGCV Land Surface Temperature Product Validation Best Practice Protocol (Guillevic et al., 2018), is the traditional and most straightforward validation type. The second step, known as Category C validation in this CEOS-WGCV protocol, gives additional insights on the quality of the data and is complementary to the Category A validation. The results are presented in terms of bias and robust standard deviation (RSTD) for the in situ validation (Section 2.1), and in terms of differences for the intercomparisons against the third party datasets (Section 2.2).

For the Category A in situ validation, thermal infrared ground observations from well established stations are used. Specifically, stations managed by the Karlsruhe Institute of Technology (KIT), Atmospheric Radiation Measurement (ARM) network, Surface Radiation Budget Network (SURFRAD), Heihe River Basin network, and the Copernicus Space Component Validation for Land Surface Temperature, Aerosol Optical Depth and Water Vapor Sentinel-3 Products (LAW) network are all sources of these ground observations. The validation time period ranged from 1995 to 2023. The validation is carried out on the individual datasets comprising the IRCDR, since the process operates on Level-2 data. Intercomparisons of satellite - satellite data pairs are a valuable asset to in situ validation results, as not all land covers types and regions are covered by in situ stations. Here intercomparisons are carried out at Level-3U (uncollated) with respect to the Spinning Enhanced Visible Infra-Red Imager (SEVIRI) onboard the Meteosat Second Generation (MSG) satellites.

The results of the in situ validation show that in general the accuracy of the LST_cci datasets comprising the IRCDR are better at night than during the day, which is expected, and generally have absolute biases less than 1 K. The latest version of the IRCDR datasets (v3.00) show a marked improvement on previous versions. Significant negative biases found for different datasets in previous versions have been reduced, with only few outliers remaining. This improvement is mainly due to the improvement in the cloud masking. The intercomparison analysis revealed that all IRCDR LST_cci products are comparable with the LST_cci SEVIRI product. AATSR and Terra-MODIS are well correlated. The variability, estimated considering data pairs across the different continents, is larger during night over Africa and during day over Europe.

For the C3S LST products (CDR LST_CCI LST v3.00 1995-2024) the linear trend of the comparison between satellite estimates and homogenised T2M reference data is 0.11 K/decade has been estimated which meets the GCOS Breakthrough threshold for LST stability. However, although the long-term trend meets the GCOS requirement, the early part of the record (1995-1999) shows fluctuations in the mean bias. When these early years are excluded, and the trend is recalculated for the period from 2000 to the end of the record yields, the resulting trend is 0.03 K/decade (Figure 7b), which is well within the GCOS Goal threshold. Consequently, climate data users are advised to use data starting from the year 2000 onward.

Product validation methodology

An appropriate way to validate the brokered IRCDR dataset in LST_cci adheres to the validation Categories A and C described in the CEOS-WGCV Land Surface Temperature Product Validation Best Practice Protocol (Guillevic et al., 2018), which also follows the LST Validation Protocol (Schneider et al., 2012).

Category A is the traditional and most straightforward validation type, where a satellite LST product is directly compared with in situ LST data from ground-based radiometers. The in situ data needs to be collocated and simultaneously acquired with satellite overpasses. This is the most accurate validation category if the in situ data is taken with well maintained high quality infrared (IR) radiometers, and are representative for the satellite sensor footprint (Göttsche et al., 2017, Guillevic et al., 2018).

Within the Category C validation approach, satellite products are compared with other, generally already validated, LST products (Schneider et al., 2012). For the analysis, these satellite vs. satellite intercomparisons of LST products are temporarily and spatially matched. This method is particularly valuable for identifying disagreements between LST products over large areas and different land cover types (Guillevic et al., 2018). Although these intercomparisons cannot provide an absolute validation, such as the Category A validation with in situ stations, they are a good supplement to it.

Multi-sensor Matchup Database (MMDB)

The Multi-sensor Matchup Database (MMDB) is the tool used for carrying out all validation on the brokered IRCDR dataset. It consists of both the data and the core software for performing the extractions, matchups and analysis of all validation and intercomparison exercises. The MMDB is located on a University of Leicester managed workspace on the UK JASMIN server, which is hosted by the Science and Technology Facilities Council in the UK, and will be used for the continuing validation of the IRCDR dataset under C3S.

The MMDB consists of four different types of data:

• Satellite Extraction files (SE)
• Satellite – Satellite match ups (SS)
• In situ files (IS)
• Satellite – In situ match up files (SI)

All these data files are produced in a common harmonized NetCDF data format that confirms to the ESA Climate Change Initiative (CCI) data format (ESA Climate Office, 2018). The complete time series of matched-up data in the MMDB facilitates a complete assessment and validation of the IRCDR LST product for a broad range of atmospheric conditions over several years.

Product in situ Validation Methodology

Data Preparation and Matching

For selection of the in situ stations for Category A validation, the temporal and spatial coverage of the in situ datasets are taken into account, as well as their suitability for validation. The latter is mainly dependent on the particular measurement devices used, the temporal resolution of the measurements and their availability, as well as on additional geographical information such as surface homogeneity or orography around the station. Furthermore, the in situ measurements should be fully traceable to a calibrated reference radiance blackbody to ensure they are suitable for climate data record validation (Göttsche et al., 2017). Based on these considerations, in situ LST datasets are identified as well-suited for validation of the brokered IRCDR from LST_cci (Table 1). These include the stations run by the Karlsruhe Institute of Technology (KIT) in Africa and Europe, the Surface Radiation Budget Network (SURFRAD) and Atmospheric Radiation Measurement (ARM) stations in the USA, and the Copernicus Space Component Validation for Land Surface Temperature, Aerosol Optical Depth and Water Vapor Sentinel-3 Products (LAW) stations (Table 1). Several of the datasets mentioned above were already used successfully in the GlobTemperature project (Martin and Göttsche, 2016).

Table 1: Overview of the in situ stations used in the validation of the brokered IRCDR. Surface types are classified based on dominant land cover. Latitude and longitude are given in decimal degrees. Temporal availability reflects the period for which quality controlled data are available. Station names highlighted in bold are those used to label the stations in the corresponding map (Figure 1).

Figure 1: Location on a global map of the in situ validation stations described in Table 1. 

For any given validation site, the in situ data are stored in the IS files. The satellite extractions are stored in the SE files and comprised of 51 x 51 pixels of each overpassing satellite orbit centred on the given station. The temporarily and spatially matched in situ and satellite data are stored in SI files, which are the files utilised in the validation. In order to make the results of different datasets comparable to each other, it is generally attempted to validate IR satellite data on a 0.05° x 0.05° grid against the in situ data. However, during the GlobTemperature project it was found that that it is not always possible due to surface heterogeneity or difficult orography at certain stations (Martin and Göttsche, 2016). Where this is the case, for the Gobabeb and Table Mountain stations for instance, the 0.05° x 0.05° grid box most representative of the land cover surrounding the in situ station is chosen.

The matching itself is performed in the same way as in the GlobTemperature project (Martin et al., 2019), and is described briefly here. For the IRCDR dataset which has a spatial resolution of 0.01° x 0.01° on equal angle global grid, 25 pixels in a square (5 x 5) centred on the pixel overlaying the in situ station are used to derive the desired 0.05° x 0.05° grid box. Furthermore, from these 25 pixels, only those with the same combined land cover class (LCC) as the station pixel are averaged to generate the 0.05° x 0.05° grid box value. This avoids influences from different land covers. The combined land cover classes are described in Caselles et al. (2012), and represent 10 classes grouped from which the original 43 LST_cci land cover classes (which are described in the accompanying C3S LST Algorithm Theoretical Basis Document (C3S, 2025)). The median of the LST of these averaged pixels is taken as the satellite LST for the validation process.

Using 5 x 5 pixels centred on the in situ station is not possible at all sites due to surface heterogeneity around the station or changing orography. In these cases, either 3 x 3 pixels centred on the station, or the single pixel overlaying the station, are used for the validation. Only grid boxes where at least 80 percent of the pixels are not flagged as being cloudy are used for the validation. In fact, grid boxes with less than 80 percent cloud free pixels increased the likelihood of missed cloud. The uncertainty on the averaged LST is calculated using the following formula:

(AveragedUncertainty)^2 = \dfrac{\sum{(ClearPixelUncertainty)^2}}{NumberClearPixels} + \dfrac{NumberCloudyPixels * VarianceClearPixels}{NumberClearPixels + NumberCloudyPixels} \quad (eq. 1)

where ClearPixelUncertainty is the uncertainty on the LST values for the individual pixels that are not flagged as cloudy, NumberClearPixels and NumberCloudyPixels are the number of the clear and cloudy pixels, respectively, and VarianceClearPixels is the LST variance of the clear pixels.

Temporal matching between satellite LST and in situ LST is achieved by linear interpolation between the two in situ measurements that are closest in time to the acquisition time of the overlaying satellite pixel. The temporal resolution of in situ observations varies between the different station networks, with the maximum temporal interval being 3 minutes. If there are gaps in the in situ data, and the closest in situ data is acquired with a time difference of greater than 3 minutes from the acquisition time of the overlaying satellite pixel, then the data point is disregarded.

Validation Process

The in situ validation approach is carried out on the Level 2 LST data from the four contributing satellite missions to the IRCDR (ATSR-2, AATSR, Terra-MODIS and SLSTR-B). The process is the same for all four missions, and efficiently uses the harmonized data format. Differences between satellite and in situ LST are analysed both statistically and visually. Statistical results are presented in terms of bias, which in this case is the median average of the difference of satellite LST minus in situ LST (following Guillevic et al., 2018), and in terms of robust standard deviation (RSTD) (see for example Pearson, 2002).

Daytime and night-time differences, seasonal variations, differences between the individual satellite datasets, and differences between stations are all investigated. Possible influences from satellite observation geometry, cloud contamination, surface elevation and land cover at the station are considered.

Uncertainty validation process

In addition to validation of the satellite LST data, the satellite LST uncertainty is also validated. There are four main components that contribute to the validation of LST uncertainty:

• Satellite LST uncertainty 
• In situ LST uncertainty
• Spatial mismatch uncertainty
• Temporal mismatch uncertainty

Satellite LST uncertainty is a summary term which includes individual components, such as measurement uncertainty, uncertainty in the the retrieval algorithm, uncertainty in the atmospheric correction of the data, and uncertainty in emissivity (Li et al., 2013). In situ LST uncertainty includes uncertainty from the measurement device, and an uncertainty in the land surface emissivity used to calculate the in situ LST. Spatial mismatch uncertainty can be a significant contributor, and is a function of the scale difference between the field of view of the in situ radiometer and footprint of the satellite pixel. The spatial mismatch uncertainty is larger if the land cover surrounding a station is heterogeneous, and is estimated as the standard deviation of the LST of the 5 x 5 pixels surrounding each station (which is the same averaging area as used in the LST validation - see Section 1.2.1). Temporal mismatch uncertainty is a function of the difference in time between in situ observation and the acquisition time of the overlaying satellite pixel. The total uncertainty on the matchup process between satellite and in situ LST (u_diff) is estimated by taking the square root of the sum of the quadrature of its single elements:

u_{diff} = \sqrt{u^2_{sat} + u^2_{insitu} + u^2_{space} + u^2_{time}} \quad (eq. 2)

where u²_sat is the satellite LST uncertainty, u²_insitu is the in situ LST uncertainty, u²_space is the spatial mismatch uncertainty, and u²_time is the temporal mismatch uncertainty.

For the validation of the brokered IRCDR LST dataset the temporal mismatch uncertainty is deemed negligible, since all the in situ datasets have a very small sampling interval of 3 minutes or less. The first three components are therefore used, and are combined to give a total uncertainty on the matchup process between satellite and in situ LST. This total uncertainty is compared against the robust standard deviation (RSTD) on the differences between satellite and in situ LST. The resulting differences are an indicator of whether the estimated satellite LST uncertainties are realistic. If all components are considered and estimated correctly, the total uncertainty should approximate the empirically determined RSTD. If the total uncertainty is lower than the RSTD, there are either uncertainty components missing or the ones considered are too low. If the total uncertainty is higher than the RSTD, one or more of the uncertainty components are too high.

Product Intercomparison Methodology

Data Preparation and Matching

Category C intercomparison is usually carried out on a spatial grid equal to, or coarser than, the spatial resolution of the coarsest satellite sensor in the intercomparison. The regridding of the data to this matchup resolution mitigates against differences in the fields-of-view of any two satellite instruments, even when they are nominally coincident. For the IRCDR, Level 3U (uncollated) data are used for the intercomparison. These are orbit / granule level data gridded in space but not in time, which preserves the acquisition times of the individual sensors. All data fields (including uncertainties, quality flags, cloud / aerosol information, and other auxiliary data) are propagated to the Level 3U product. In the MMDB the Level 3U data files are internal to the processing, but they represent the baseline for all intercomparison matchups. Moreover, they are also the input for the higher-level Level 3C (collated) and Level 3S (super collated) output products.

There are several methods for the spatial matching of two satellite datasets. The so-called nearest neighbour approach is an effective and relatively straightforward method. The images of the two considered satellites are overlaid with each other by shifting one set of pixels so that they match the other set. The advantage of this method is that data is not averaged or weighted, thus the original data remains unchanged. However, the larger the shift becomes, the further apart spatially the compared LST values are. A second approach is the averaging of the data by polygon weighting, which involves the formation of a polygon tessellation. To account for the fact that the pixel area and the area of interest are often not exactly the same, the data in each polygon is weighted according to the proportion of the area of interest in the polygon to its total area (see Jallego, 2006). A third possibility is a simpler version of the second approach. In this case, each point in space is assigned to the pixel to which its midpoint is closest to, and these are then averaged without weighting. Although this third method is easy to implement, it has disadvantages: i) the original pixels may not be representative of the final sampling grid; and ii) for coarser resolution pixels gaps can occur where the assignment of original pixels means some grid cells can remain unfilled even for a continuous field of original pixels.

The polygon weighting approach is used in LST_cci to regrid the Level 2 data to Level 3U. The reason for this choice is that a high spatial resolution matchup grid is highly sensitive to orbit tracks of the polar orbiters (ATSR-2, AASTR, Terra-MODIS and SLSTR-B) which constitute the IRCDR, and their pixel nearest neighbour binning. This is particularly the case at the edge-of-swath of wide-swath instruments, such as MODIS, where the pixel sizes grow to 5 - 6 km, and can thus be similar in size to a common matchup grid. The brokered LST IRCDR dataset is intercompared against the LST_cci product from the Spinning Enhanced Visible Infra-Red Imager (SEVIRI), which is onboard the geostationary Meteosat 2nd Generation satellite. The Level 2 IRCDR LST data has a spatial resolution of 1 km, and the LST_cci SEVIRI LST has a spatial resolution of 0.05°. Therefore the IRCDR is regridded onto a 0.05° spatial grid, to be common with SEVIRI, by averaging all geo-referenced, cloud free pixels weighted by their respective fractional area overlap with the corresponding common grid cell.

Once the input data are regridded to the Level 3U format they are read in by the Simultaneous Nadir Overpass (SNO) processor. The high temporal variability of LST means that intercomparison of different LST products is a challenging prospect. In order to minimise the impact on the intercomparison results, LST differences due to different observation times have to be minimised. This can be achieved by limiting the data to close temporal matchups. In the case of the intercomparison of the IRCDR with LST_cci SEVIRI, only matchups with a 5 minute temporal threshold between datasets is used. This is consistent with the GSICS criteria (Goldberg et al., 2011), and facilitated increased matchups between the polar orbiters and the nearest SEVIRI acquisition, while limiting the potential for significant temperature change. Larger temporal thresholds would have increased the risk of LST differences representing actual ground temperature changes rather than being attributable to differences between the products. To maintain consistency, no interpolation between adjacent SEVIRI LSTs that temporally bracketed a specific overpass time of the polar orbiter is carried out. Furthermore, interpolation between less frequent SEVIRI observations would have increased the risk that any assumption of a linear relationship between bracketing LST observations would be invalid.

The outputs from the SNO processor are temporally collated into daytime and night-time composites based on their respective solar zenith angles. These are then aggregated to monthly median statistics on the common matchup grid. Specifically, all valid individual SNO matchups are collected and stored as differences between the satellite LST values (ΔLST) along with their associated uncertainty, satellite viewing angles, cloud mask, acquisition time, and geolocation information. These outputs are passed to the seasonal aggregator and analysis processor. The seasonal aggregating processor is identical to the monthly aggregator with the exception that it takes in all the data from the whole period where the two sensor time series overlap. The analysis processor is split into three components. It produces a range of analytical plots and result tables. By design, the complexity of the analysis performed at each stage increases. This allows for the control of the volume of output content. Visualisation of the spatial distribution and time series of variables for each test is also produced. The analysis processor excludes data where the uncertainties in the input data are very high due to very low clear to cloud ratios. The output matched satellite – satellite (SS) datafiles are then stored in the MMDB. Figure 2 illustrates the full intercomparison matchup processing chain.

Figure 2: Intercomparison matchup and analysis processing chain

Intercomparison Process

The median and RSTD of the differences of the non-reference LST product (IRCDR) to the reference LST product (LST_cci SEVIRI) per pixel are analysed. Statistically the matchups are analysed in terms of their differences, i.e. the median of IRCDR LST minus SEVIRI LST and their corresponding RSTDs (see e.g. Pearson, 2002). Data are composited over each month and over all four seasons. The influence of satellite viewing geometry, sun angle, orography, elevation classes, and land cover classes on the resulting differences are also investigated. In the assessment by satellite viewing geometry, differences are binned and analysed with respect to the product of the satellite zenith angle (satze) multiplied by the sign of the satellite azimuth angle (sataz):

satellite \; viewing \; angle \; and \; direction = satze * \dfrac{\vert{sataz}\vert}{sataz} \quad (eq. 3)

The number of averaged pixels are also considered, as this can influence the statistical significance of the results. A similar approach had already been used successfully in the GlobTemperature project (Martin et al., 2016). While no quantitative assessment could be made based on differences in cloud contamination, since “true” manual masks have only been produced for a few orbit scenes, interpretation of the results included discussion on the likely impact of cloud contamination.

 Stability Assessment methodology

 Data

Climate Research Unit temperature version 5 (CRUTEM5, Osborn et al., 2021) is a gridded data set of monthly near-surface air temperature anomalies over the land surfaces of the world, running from 1850 to the present. It is a collaborative product of the Climatic Research Unit at the University of East Anglia, the Met Office Hadley Centre and the National Centre for Atmospheric Science. CRUTEM5 is the fifth major version of the dataset, covering the time period from 1850, with a spatial resolution of 5° latitude by 5° longitude and a monthly-mean time resolution. Hemispheric and global mean time series of land air temperature anomalies are also provided. The underlying station database, the gridded temperature anomalies, the global and hemispheric timeseries and their uncertainty intervals will be available e.g. from the Met Office website and the CRU website.

Matchup & Stability Assessment Process

The reference for the stability analysis is the Climate Research Unit temperature version 5 (CRUTEM5, Osborn et al., 2021) station air temperature anomaly dataset. CRUTEM5 is a long-term homogenized station record of T2M which covers the whole of the CDR period and is regularly extended in time. Therefore, CRUTEM5 is a suitable reference for the stability analysis of the CDR and ICDR. The LST CDR was compared with CRUTEM5 (version 5.1.0.0) using the method of Good et al. (2017, 2022). Monthly mean LST anomalies are calculated from the LST CDR at a spatial resolution of 5° latitude-longitude to match the resolution of CRUTEM. The CRUTEM anomalies are adjusted to the same climatological period as the LST anomalies. Anomaly differences (LST - T2M) are calculated and spatially averaged to give global (60° N to 60° S) monthly mean differences. The resulting time series for the whole period of the CDR are investigated for the estimation of the stability.

Validation results

For the brokered IRCDR dataset from LST_cci covering the years 1995 – 2023 the validation results are presented directly from the LST_cci Product Validation and Intercomparison Report (PVIR) (Perez-Planells and Martin, 2025). All validation results are presented with respect to the individual datasets which constitute the IRCDR (ATSR-2, AATSR, Terra-MODIS, and SLSTR-B).

Analysis of In Situ Results

In Situ LST Validation

First, an overview of the validation results for all satellite datasets over all in situ stations is given. The bias and RSTD are given in Table 2 for night-time and in Table 3 for daytime data. When interpreting the results, it should be kept in mind that the time period over which the data is averaged varies between stations and satellite datasets due to their temporal and spatial availability. The matched time periods can be seen in Table 4.

Table 2: Median night-time bias and RSTD in K for the analysed time period

Table 3: Median daytime bias and RSTD in K for the analysed time period

Table 4: Analysed years for the validation of satellite LST datasets over in situ stations

In general, the night-time biases are smaller than the daytime biases and also have lower RSTDs. This is expected and mainly caused by the influence of solar radiation during daytime, which results in different surface temperatures observed by the satellite sensors over sunlit and shadow areas. These differences can also be present in the point measurements of the in situ sensors, although to a lower extent than for the satellite pixels. Previous versions of the datasets had larger negative biases over certain stations. This improved for the current version (3.00) of the IRCDR datasets, where an improved cloud mask is introduced and only few negative outliers remain. Overall the biases are generally within ± 1.0 K.

Tables 2 and 3 indicate that the magnitude of the biases varies significantly between different stations and between satellite datasets. These differences can be caused by the amount of heterogeneity in the area surrounding the in situ stations at the scale of the satellite pixel, as larger surface heterogeneities are often not represented in the field of view of the in situ radiometers. The distinction between day and night is based on the solar zenith angle at the time of the satellite observation. The RSTDs are usually larger during daytime, such as for EVO___, which is located in an area covered with cork-oak trees, where the influence of shadows is significant during day. RSTDs are also larger for stations with more heterogeneous land covers, such as DAH_T_ and TBL___.

The number of matchups varies between satellite datasets and stations due to the specific time span for each dataset and the spatial coverage of the datasets, which are dependent on the width of the respective satellite swathes. Terra-MODIS has more matchups as it has the longest period of operation.

In Situ Uncertainty Validation

Figure 3 illustrates the validation of LST uncertainty for ATSR-2 and AATSR for two example stations. Here the total uncertainty on the matchup process is displayed on the x-axis divided into bins of 0.1 K. On the left y-axis, the RSTD of the bias for each bin is shown as bars. On the right y-axis the number of data points per bin is displayed, represented by dots. For the validation of LST uncertainty, where the tips of the bars meet the cone shape then we can interpret the LST uncertainty as being realistic (Ghent et al., 2019). For some bins the number of data points is very low, which can lead to a larger influence of outliers and more fluctuating RSTD values. Thus, the number of data points needs to be considered when interpreting the results. We therefore focus on the bars where there are a large number of data points.

Figure 3: RSTD of the bias (satellite LST – in situ LST) vs. total matchup uncertainty for ATSR_2 and ATSR_3 datasets over GCM___stations (top-left(a)) and TBL___stations (top-right (b)). The total matchup uncertainty (x-axis) is divided into bins of 0.1 K. The y-axis on the right displays the number of data points per bin.

For the ATSR-2 and AATSR datasets the total matchup uncertainty frequently fits well to the RSTD of the bias (see Figure 3a (top-left)), such as at the Goodwin Creek (GWM___) site. The total uncertainty of the ATSR datasets is higher (i.e. total matchup uncertainty shifted to the right) over stations located in a more heterogeneous landscape (see Figure 3b (top-right)), such as at the Table Mountain (TBL___) site. This is primarily due to an increase in the spatial mismatch uncertainty. In the Table Mountain example, where data points are high then the LST uncertainty still validates well. For Terra-MODIS the results vary between stations. For some stations the fit is good, whereas for other stations the total matchup uncertainty overestimates the RSTD of the satellite LST - in situ LST. The reason for this is likely to be an overestimation of the emissivity uncertainty in the specific regions. Similarly to the ATSRs, where the number of data points is high the fit is much better. For SLSTR-B, the total matchup uncertainty fits well with the RSTD of the satellite LST - in situ LST over most stations. Only for a few stations does the total matchup uncertainty overestimate the RSTD of the satellite LST - in situ LST. This overestimation is likely to be related to an overestimation of the spatial mismatch uncertainty. The number of data points for SLSTR-B is in general higher for the total matchup uncertainty range between 1 and 2 K at most of the sites. This is the uncertainty range which is in better agreement with the RSTD of the satellite LST - in situ LST.

Analysis of Intercomparison Results

Three satellite vs. satellite data pairs are intercompared against LST_cci SEVIRI LST over Africa and Europe (which corresponds to the main regions covered by the geostationary disk) for the years 2008 – 2010 or for 2018 – 2020, depending on data availability. For the comparison over Africa we used a continental mask which included all continental Africa as well as Madagascar. For Europe we used a continental mask with an northern limit of 60°N and an eastern limit of 30°E. The data pairs are AATSR vs. SEVIRI, Terra-MODIS vs. SEVIRI and SLSTR-B vs. SEVIRI. The first two data pairs are temporally analysed for the time period of 2008 – 2010 when AATSR was operational, with SLSTR-B vs. SEVIRI analysed for 2018 – 2020 when SLSTR-B has been operational. ATSR-2 is not included in the intercomparison due to the lack of a complementary LST_cci geostationary dataset (SEVIRI was not operational during the operational lifespan of ATSR-2).

All analysis is divided into daytime and night-time data, based on solar zenith angle, and assessed on a seasonal basis. The seasons are named “DJF” (December, January, February) for Northen Hemisphere (NH) winter, “MAM” (March, April, May) for NH spring, “JJA” (June, July, August) for NH summer, and “SON” (September, October, November) for NH autumn. The LST differences between the IRCDR datasets (AATSR, Terra-MODIS and SLSTR-B) and the reference dataset (SEVIRI) are evaluated for the following metrics:

• Time series of monthly differences
• Differences as a function of satellite zenith angle
• Differences by land cover classes
• Differences as a function of elevation

An overview of the seasonal differences are shown in Table 5. The differences over Africa for all sensors are within the ± 1 K range, and for the most part over Europe also, with just a few outside of this range. All these data are averaged for the full regions over a large time window. Thus, postive and negative grid box differences may average out at these regional scales. Nonetheless, these results provide a good first indication about the overall differences between the datasets.

Table 5: Overview of seasonal results showing the median differences of the analysed satellite vs. satellite pairs over the continents Africa and Europe, where data from all satellite vs. satellite pairs are available. Results are presented in terms of difference (diff) and RSTD.

Results over Africa

Monthly differences over the investigated time windows are shown in Figure 4. The number of averaged data points in the month is also shown. The highest number of data points is found for the Terra-MODIS dataset with respect to SEVIRI. Overall, the seasonal variability is small. There is a strong positive outlier for night-time data in 2010/09, most likely caused by the SEVIRI data since it is visible in the comparison both against AATSR and Terra-MODIS. Indeed, other comparisons between polar-orbiting LST_cci datasets and SEVIRI showed this same strong positive outlier. As yet no explanation has been found for this anomaly. SLSTR-B vs. SEVIRI shows very small differences with no noticeable seasonal cycle, for daytime or for night-time data.

Figure 4: Monthly time series for AATSR vs. SEVIRI (top-left(a)) and Terra-MODIS vs. SEVIRI (top-right (b)) over Africa from 2008 - 2010, and for SLSTR-B vs. SEVIRI (bottom(c)) over Africa from 2018 - 2020. Red stars represent the daytime data, blue dots the night-time data, the green stars and dots on the right axis display the averaged pixel numbers for daytime and night-time, respectively.

The analysis of the differences by elevation showed that most pixels have an elevation between 50 – 200 m and between 200 – 500 m. No significant differences between the different elevation classes are found. In terms of land cover classes, most pixels have a land cover of either tree_broadleaved_deciduous_open, shrubland, or Bare_areas_of_soil_type_Entisols_Orthents. No significant differences between them could be seen.

The influence of satellite viewing angle on the results is investigated by calculating the differences for different ranges of satellite zenith angle (satze) for the IRCDR dataset, multiplied by the sign of the satellite azimuth angle (sataz) as shown in Equation 3. When the sataz is negative, it means that the satellite is viewing the scene from west to east, and vice-versa when the sataz is positive. As the local overpass time is in the morning for Terra-MODIS, the sun is shining from the east and casting more shadows to the west. These results provided no significant differences for AATSR (Figure 5a - left panels). The reason for this is that AATSR has a narrow swath, with satze ranging only between -22° to +22°. For the intercomparison of Terra-MODIS against SEVIRI (Figure 5b - right panels), the LST differences increase towards the edge of the MODIS swath. The daytime differences in particular increase substantially for larger MODIS satze values, due to the increased influence of sunlit and shadow areas.

Figure 5: Seasonal differences between AATSR vs. SEVIRI LST data (left panels (a)) and Terra-MODIS vs. SEVIRI LST data (right panels(b)) for different satellite zenith angles (labelled as satze*sign(sataz) on the x-axis) of AATSR and Terra-MODIS respectively. Red stars represent the daytime data, blue dots the night-time data, and the green stars and dots on the right axis represent the averaged pixel numbers for daytime and night-time data, respectively.

Results over Europe

Monthly differences over the investigated time spans are shown in Figure 6. The number of averaged data points in the month is also shown. A seasonal cycle for daytime data is found for all data pairs intercompared against SEVIRI, with positive differences in the summer and negative differences in the winter. The main reason for this is the satellite zenith angle of the reference sensor (SEVIRI). Since SEVIRI is onboard a geostationary satellite centred at 0° latitude and 0° longitude, the satellite zenith angle increases as you move away from the centre of the disk. For Europe, the zenith angles are large. This is in contrast to the results over Africa, where the SEVIRI zenith angles are smaller and no strong seasonal cycle for the daytime data is evident. For these satellite intercomparisons against SEVIRI, there is a strong positive outlier for night-time data in 2010/09, which is also seen over Africa (Section 2.2.1).

Figure 6: Monthly time series for AATSR vs. SEVIRI (top-left (a)) and Terra-MODIS vs. SEVIRI (top-right (b)) over Europe from 2008 - 2010, and for SLSTR-B vs. SEVIRI (bottom (c)) over Africa from 2018 - 2020. Red stars represent the daytime data, blue dots the night-time data, the green stars and dots on the right axis display the averaged pixel numbers for daytime and night-time, respectively.

Analysis of the differences by elevation classes (not shown) indicate that most pixels fall into an elevation class of 50 m – 200 m. The differences and RSTD increased for elevations > 1500 m, but no other significant differences between single elevation classes are found. The main land cover classes found over Europe are “tree_needleleaved_evergreen_closed_to_open”, “cropland_rainfed_herbaceous_cover ” and “cropland_rainfed”. Only small differences between the different elevation classes are found. The analysis of the differences with respect to satellite zenith angle of the IRCDR datasets showed an asymmetric distribution of data points for Terra-MODIS against SEVIRI, which is less pronounced for autumn and winter when the influence of solar radiation is less.

Stability assessment

The stability of the dataset is the change in bias over time and, for LST, is assessed as the long-term monotonic trend with respect to a related ECV, in this case, surface air temperature at a height of 2 m (T2M). Coincident T2M and LST observations can differ by 2-20 °C (depending on location and surface and atmospheric conditions, Good 2016) however the long-term signals in T2M and LST are very similar (Good, 2017, 2022). As described before, the reference for the stability analysis is the Climate Research Unit temperature version 5 (CRUTEM5, Osborn et al., 2021) station air temperature anomaly dataset. The applied methodology is described in sec 1.4 and the resulting time series for the whole period of the CDR is plotted in Figure 7a. The linear trend, calculated using the Theil-Sen median of pairwise slopes method (Sen, 1968), is 0.11 K/decade which meets the GCOS Breakthrough threshold for LST stability. However, although the long-term trend meets the GCOS requirement, the early part of the record (1995-1999) shows fluctuations in the mean bias (particularly in the transition from ATSR-2 to AATSR). Removing these years and calculating the trend for the period from the beginning of 2000 to the end of the record yields a trend of 0.03 K/decade (Figure 7b) which is well within the GCOS Goal threshold. Climate users are, therefore, recommended to consider using only the data from the beginning of 2000 onwards.

Figure 7: Monthly mean anomaly difference (LST minus T2M) with linear trendline for: (a) all months (1995-2025) and (b) for 2000-2025. The trend envelope shown as a shaded region around the trendline signifies the 95% confidence interval for the trend. 

Climate Change Assessment

Time series of monthly mean land surface temperature anomaly for 2000-2025

Figure 8 shows time series of mean monthly daytime and nighttime LST anomaly for the period January 2000 to September 2025. LST anomalies are the differences from the mean state for the period 2000 to 2024. The trends indicate that there has been an increase in LST over the period 2000-2025. The increase in nighttime temperature is 0.53 K/decade and is higher than the increase in daytime temperature of 0.21 K/decade. Although linear trends have been calculated, the plot shows that the increase is not linear and that LST anomalies are highly variable.

Figure 8: Time series of monthly mean LST anomaly for daytime (red) and nighttime (blue). Also, shown are the linear trend lines and the 95% confidence intervals.

Application(s) specific assessments 

This section will be periodically updated as consistency studies with other ECVs are carried out. As this is the first version of this document, application specific assessment are still not available.

Compliance with user requirements concerning data quality

In this section, we compare how the validation results for the brokered LST dataset compare with the GCOS-245 requirements for LST (GCOS, 2022). Table 6 details the assessment against the specific requirements. 

Table 6: Compliance with GCOS-245 LST ECV Requirements. Green shading indicates the dataset meets the "goal" requirements, orange indicates the dataset meets the "threshold" requirements, and red indicates the dataset does not meet the threshold requirement.   

For horizontal resolution, only polar orbiting data can meet the requirements, and the IRCDR, which is regularly gridded at 0.01°, approximately meets the GCOS requirements depending on location over the globe. For temporal resolution, only geostationary data can provide data at these resolutions but these are regional datasets. In contrast, polar orbiting satellites cover the whole globe but are restricted to day/night temporal resolution, which do not meet the threshold requirement. Thus the IRCDR does not meet the threshold requirement. The notes related to the GCOS-245 temporal resolution requirement state that the day/night temporal resolution from polar orbiting satellites still satisfies 70% of climate users in the survey of 80 users which was used to derive the LST ECV requirements (GCOS, 2022).

For timeliness, the GCOS requirement was the outcome of a survey of 80 users which revealed a “threshold” need of 30 days for long-term data records, and a “breakthrough” of 48 hours for long-term data records. The C3S LST products (CDR LST_CCI LST v3.00 1995-2024) can be produced with a timeliness of 30 days, thus meeting the threshold requirement.

For required measurement uncertainty, this is the required total uncertainty per pixel combining four groups of uncertainty components: uncorrelated, locally correlated atmospheric, locally correlated surface, and large scale systematic (as described in Algorithm Theoretical Basis Document (ATBD) of the C3S LST Products (E.U. Copernicus Climate Change Service, 2025)). This is assessed by way of the validation of the uncertainties, with these being within 1 

For the C3S LST products (CDR LST_CCI LST v3.00 1995-2024)  a stability value of 0.11 K/decade was estimated, thus almost meeting the goal requirement.

References

Caselles, E., Valora, E., Abadb, F., Caselles, V. (2012). Automatic classification-based generation of thermal infrared land surface emissivity maps using AATSR data over Europe. Remote Sensing of Environment, 124, 321–333.

ESA Climate Office (2018) CCI data standards 2.3 (CCI-PRGM-EOPS-TN-13-0009), available from https://climate.esa.int/media/documents/CCI_DataStandards_v2-3.pdf

E.U. Copernicus Climate Change Service (C3S) (2025), Algorithm Theoretical Basis Document (ATBD) v1.0 - CDR and ICDR Land Surface Temperature v3.00, C3S2_313e_BC_WP3-DR-LST-LST_CCI-v3.00-1995-2024_202506_ATBD_v1.0 (not yet published)

Goldberg, M., Ohring, G., Butler, J., Cao, C., Datla, R., Doelling, D., Gärtner, V., Hewison, T., Iacovazzi, B., Kim, D., Kurino, T., Lafeuille, J., Minnis, P., Renaut, D., Schmetz, J., Tobin, D., Wang, L., Weng, F., Wu, X., Yu, F., Zhang, P., Zhu, T. (2011). The Global Space-Based Inter-Calibration System. Bull. Amer. Meteor. Soc., 92, 467 – 475.

Li, Z.L., Tang, B.H., Wu, H., Ren, H., Yan, G., Wan, Z., Trigo, I.F., Sobrino, J.A. (2013). Satellite-derived land surface temperature: Current status and perspectives. Remote Sensing of Environment, 131, 14–37.

Perez-Planells, L., and Martin, M. (2022). CCI LST Product Validation Plan (LST-CCI-D2.5-PVP) V2.0

Perez-Planells, L., and Martin, M. (2025). CCI LST Product Validation and Intercomparison Report (LST-CCI-D4.1-PVIR) V4.0

Schneider, P., Ghent, D., Corlett, G., Prata, F., Remedios, J. (2012). AATSR Validation: LST Validation Protocol. ESA Report, Contract No.: 9054/05/NL/FF, European Space Agency (ESA). UL-NILU-ESA-LST-LVP Issue 1 Revision 0. 

Ghent, D., Veal, K., Trent, T., Dodd, E., Sembhi, H., and Remedios, J. (2019). "A New Approach to Defining Uncertainties for MODIS Land Surface Temperature". Remote Sensing, 11, 1021. doi: 10.3390/rs11091021

Good, E. J. (2016). An in situ–based analysis of the relationship between land surface “skin” and screen-level air tempera­tures. J. Geophys. Res. Atmos., 121, 8801–8819, https://doi.org/10.1002/2016JD025318.

Good, E. J., Ghent, D. J., Bulgin, C. E., and J. J. Remedios (2017). A spatio­temporal analysis of the relationship between near-surface air temperature and satellite land surface temperatures using 17 years of data from the ATSR series. J. Geophys. Res. Atmos., 122, 9185–9210, https://doi.org/10.1002/2017JD026880.

Good, E. J., Aldred, F. M., Ghent, D. J., Veal, K. L., and C. Jimenez (2022). An analysis of the stability and trends in the LST_cci land sur­face temperature datasets over Europe. Earth Space Sci., 9, e2022EA002317, https://doi.org/10.1029/2022EA002317.

Göttsche, F., Olesen, F.S., Høyer, J.L., Wimmer, W., Nightingale, T. (2017). Fiducial Reference Measurements for Validation of Surface Temperature from Satellites (FRM4STS), Technical Report 3 - A Framework to Verify the Field Performance of TIR FRM. Reference: OFE- D120-V1-Iss-3-Ver-1-ISSUED, ESA Contract No. 4000113848 15I-LG. Techreport, ESA. http://www.frm4sts.org/wp-content/uploads/sites/3/2018/04/FRM4STS_D120-TR3_24Apr17_v3-signed.pdf

Guillevic, P., Göttsche, F., Nickeson, J., Hulley, G., Ghent, D., Yu, Y., Trigo, I., Hook, S., Sobrino, J., Remedios, J., Román, M., Camacho, F. (2018). Land Surface Temperature Product Validation Best Practice Protocol, Version 1.1. https://lpvs.gsfc.nasa.gov/PDF/CEOS_LST_PROTOCOL_Feb2018_v1.1.0_light.pdf

GCOS. (2022). The 2022 Global Observing System for Climate ECVs Requirements, GCOS-245. https://library.wmo.int/records/item/58111-the-2022-gcos-ecvs-requirements

Jallego, F.J. (2006). Stratified sampling of satellite images with a systematic grid of points. ISPRS Journal of Photogrammetry and Remote Sensing, 59, 369–376.

Joint Commitee for Guides in Metrology (JCGM). (2008). Evaluation of Measurement Data - Guide to the Expression of Uncertainty in Measurement.

Martin, M.A., Ghent, D., Pires, A.C., Göttsche, F.M., Cermak, J., Remedios, J.J. (2019). Comprehensive In Situ Validation of Five Satellite Land Surface Temperature Data Sets over Multiple Stations and Years. Remote Sensing, 11.

Martin, M., Göttsche, F.M. (2016). ESA DUE GlobTemperature Project: Satellite LST Validation report. Technical report, ESA.

Martin, M., Göttsche, F.M., Ghent, D., Trent, T., Dodd, E., Pires, A., Trigo, I., Prigent, C., Jimenez, C. (2016). ESA DUE GlobTemperature Project: Satellite LST Intercomparison Report. Technical report, ESA.

Norman, J.M. and F. Becker. (1995). Terminology in thermal infrared remote sensing of natural surfaces. Agricultural and Forest Meteorology, 77: p. 153-166.

Osborn, J., Jones, P. D., Lister, D. H., Morice, C. P., Simpson, I. R., Winn, J. P., Hogan, E., Harris, I. C. (2021). Land Surface Air Temperature Variations Across the Globe Updated to 2019: The CRUTEM5 Data Set, Journal of Geophysical Research: Atmospheres, 126, 2, e2019JD032352, DOI: https://doi.org/10.1029/2019JD032352, https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2019JD032352.

Pearson, R.K. (2002). Outliers in Process Modeling and Identification. IEEE Transactions On Control Systems Technology, 10, 55–63

Sen, P. K., (1968). Estimates of the regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 63, 1379–1389, https://doi.org/10.1080/01621459.1968.10480934.

Related articles