Last modified on Nov 08, 2023 18:02

Contributors: L. Carrea (University Of Reading), C. Merchant (University Of Reading), O. Embury (University Of Reading), A. Dostalova (EODC)

Issued by: L. Carrea, C.J. Merchant

Date: 15/02/2023

Ref: C3S2_312a_Lot4.WP2-FDDP-LK-v1_202212_LSWT_ATBD-v4_i1.1

Official reference number service contract: 2021/C3S2_312a_Lot4_EODC/SC1

History of modifications

List of datasets covered by this document

Related documents

Acronyms

General definitions

L2P – Geophysical variables derived from Level 1 source data on the Level 1 grid (typically the satellite swath projection). Ancillary data and metadata added following GHRSST Data Specification.
L3U – Level 3 Un-collated data are L2 data remapped to a regular latitude/longitude grid without combining observations from multiple source files. L3U files will typically be "sparse", corresponding to a single satellite orbit.
L3C – Level 3 Collated data are observations from a single instrument combined into a space-time grid. A typical L3C file may contain all the observations from a single instrument in a 24-hour period.
L3S – Level 3 Super-collated data are observations from more than one satellite that have been gridded together into a single grid-cell estimate, for those periods where more than one satellite data stream delivering the geophysical quantity has been available.
Brokered dataset – 'Dataset provided by another institution/initiative and not produced within this service.

Scope of the document

This document is the Algorithm Theoretical Basis Document (ATBD) for the Lake Surface Water Temperature (LSWT) v4.5 C3S product. ("v4.5" is the scientific versioning of the climate data record familiar to users, whereas under C3S numbering this document refers to "CDR v4"). The brokered CDR product was produced within the European Space Agency (ESA) Climate Change Initiative (CCI) Lakes project from the Along Track Scanning Radiometer (ATSR), Terra Moderate Resolution Imaging Spectroradiometer (MODIS), MetOp-A/B Advanced Very-High Resolution Radiometer (AVHRR) and Sentinel-3A/3B Sea and Land Surface Temperature Radiometer (SLSTR) instruments.

Executive summary

This document summarises the algorithms and auxiliary data used in the production of the Level 2 and Level 3 Lake Surface Water Temperature brokered and extended CDR products. The C3S extensions of the brokered CDR are intended to be used in combination with the corresponding ESA CCI-LSWT-4.5 CDR. All CDR datasets are generated using the same software and algorithms originally developed within the ESA CCI Lakes project.

The LSWT version 4.5 ESA CCI Lakes CDR provides a baseline record from 1995 through end-2020.

Input satellite brightness temperatures (BTs) are screened using the water detection method developed within the GloboLakes/ ESA CCI Lakes project and LSWTs are retrieved using an Optimal Estimation technique. Quality level and retrieval uncertainty estimates are calculated, and the LSWTs are remapped to a regular latitude-longitude grid to produce a global daily product. Since the LSWTs have been generated using different instruments, a cross-sensor adjustment is estimated and applied in order to obtain a harmonized product.

This ATBD describes the algorithms used to generate the LSWT products, including the (1) identification of water-only pixels for valid retrieval, (2) the LSWT retrieval itself, (3) estimating the daily average LSWT from the instantaneous skin observation, (4) assigning a pixel quality level, (5) remapping the data to a regular global grid, (6) cross-sensor LSWT harmonization.

1. Instruments

1.1. AVHRR

The Advanced Very High Resolution Radiometers (AVHRRs) are a series of multipurpose imaging instruments carried onboard the National Oceanic and Atmospheric Administration (NOAA) Polar Operational Environmental Satellites (POES) and EUMETSAT Polar System (EPS) MetOp satellites. The first AVHRR instrument was carried onboard the TIROS-N satellite launched in October 1978; as of 2022 there are four AVHRR instruments still in operation with the most recent AVHRR launched onboard MetOp-C in November 2018. For the C3S CDR v4, only the data from AVHRR on the MetOp-A and MetOp-B satellites are processed.

The AVHRR is an across-track scanning radiometer using six spectral channels (early versions of the instrument had four or five channels), with a spatial resolution of approximately 1.1 km at nadir. There are 2048 pixels in each scan for a swath width of about 2800 km. However, due to hardware limitations when the instruments were originally designed it was not possible to record a complete orbit of full resolution data for transmission to the ground station. Therefore, the onboard processor samples the real-time data to produce reduced resolution Global Area Coverage (GAC) data with a nominal resolution of ~4 km. This is achieved by averaging four pixels along the first scanline and then skipping a pixel before averaging the next four pixels. The next two scan lines are discarded before resuming the sampling on the fourth scanline. Each four-pixel average is then considered to be representative of a 15-pixel cell (5 pixels across track by 3 pixels along track). The more recent MetOp satellites do not have this limitation and record full orbit data at native or full resolution. Note - NOAA distributes the full resolution MetOp data as Full Resolution Area Coverage (FRAC). Table 1 reports the list of platforms carrying an AVHRR instrument where the platforms used for the C3S LSWT v4.5 brokered and extension products have been highlighted, while Table 2 reports the AVHRR channels.

Table 1: List of platforms carrying AVHRR instruments, and ancillary information on their operations.

Table 2: List of AVHRR channels for AVHRR/3 which is used for the C3S dataset. X indicates a channel is used by the instrument.

1.2. ATSR

The Along Track Scanning Radiometer (ATSR) instruments are well calibrated, dual-view radiometers intended to produce long-term, consistent LSWT observations. Three ATSR instruments have flown on board ESA's two ERS (European Remote Sensing) satellites and Envisat (Environmental Satellite) satellite. All three satellites were in stable sun-synchronous orbits with near-constant Local Equator Crossing Times (LECTs) – the ERS-1 and ERS-2 platforms had a LECT of 10:30 and Envisat had a crossing time of 10:00 all of which were maintained within a few minutes.

The ATSR instruments had four key design features making them better suited to climate applications than the AVHRR instruments:

• The instrument spectral response functions were accurately measured during pre-flight calibration and characterisation.
• The ATSRs are exceptionally well calibrated with two accurate on-board calibration targets at temperatures of ~260 K and 300 K which greatly reduces non-linearity errors for ocean observations (for comparison AVHRR instruments have a single on-board target at ~290 K and rely on a space view (2.7 K) to provide the second point.
• The infrared detectors are actively cooled to ~82 K to reduce instrument noise and avoid temperature dependent effects on calibration. In addition, the instrument fore-optics are cooled below ambient temperature to reduce self-emission issues, and the instrument is enclosed to prevent stray-light affecting the detectors.
• The dual-view capability using a single telescope with a conical scanning pattern provides both a nadir-view and an inclined forward view (~55°). Having two views of the Earth's surface allows the instrument to gather more information and more effectively separate surface and atmospheric effects, i.e. the sea surface temperature (SST) retrieval can be made more robust to atmospheric conditions, including water vapour and stratospheric aerosol.

The first ATSR carried onboard ERS-1 was a four channel radiometer with channels centred at 1.6, 3.7, 11, and 12 μm. However, the ATSR-1 instrument would only transmit one of the 1.6 or 3.7 μm channels with the selection based on the 1.6 μm reflectance intended to separate day and night-conditions. There were two major issues affecting the ATSR-1 instrument:

1. The 3.7 μm channel failed in May 1992, less than a year after the satellite was launched
2. In order to preserve mission lifetime, the temperature of the actively cooled detectors was allowed to rise, by-end of mission the detectors were operating at over 110 K.

For these reasons, data from ATSR1 have not been used to generate LSWT. In Table 3, the instruments used for the C3S LSWT v4.5 brokered product have been highlighted.

The second ATSR carried onboard the ERS-2 added three reflectance channels centred at 0.55, 0.67, and 0.87 μm mainly for vegetation monitoring. The channels had only limited availability over ocean due to telemetry bandwidth limitations; depending on instrument operating mode the visible channels may be transmitted for a reduced narrow swath, reduced 8-bit digitization, or interlaced (record every-other) pixels. The final, Advanced ATSR (AATSR), instrument onboard Envisat was functionally the same as ATSR-2 but without the bandwidth limitations so all seven channels are always available in full resolution.

Table 3: List of ATSR channels for each instrument. X indicates a channel is used by the instrument, X¹ indicates that ATSR1 only transmitted one of 1.6 and 3.7 μm. The 1.6 μm channel helped to separate day and night conditions. However, after the failure of the 3.7 μm channel only 1.6 μm data was transmitted. X² indicates that ATSR2 visible channels had limited availability over the ocean resulting in a reduced width swath and/or reduced 8-bit digitisation.

1.3. SLSTR

The Sea and Land Surface Temperature Radiometer (SLSTR) instruments are well-calibrated, dual-view radiometers intended to produce long-term, consistent water surface temperature observations. The design of the SLSTR instrument builds on the heritage of the earlier ATSR instruments adding more spectral bands and a wider swath. The swath width is about 1400 km at nadir. The first SLSTR instrument is carried on-board the Sentinel-3A satellite launched in February 2016. A second platform, the Sentinel-3B was launched in April 2018. The Sentinel-3A orbit is near-polar, sun-synchronous orbit with a descending node equatorial crossing at 10:00h Mean Local Solar time. The Sentinel-3B orbit is identical to the Sentinel-3A's but Sentinel-3B flies +/-140° out of phase with Sentinel-3A, after an initial 6 month 'tandem phase' where the two satellites observed the same place on Earth within 30 seconds. The SLSTR channels, together with bandwidth and spatial resolution, are reported in Table 4 where the channels used for the LSWT are highlighted.

Table 4: List of SLSTR channels. X indicates the channels used for the C3S product.

1.4. MODIS

The Moderate Resolution Imaging Spectroradiometer (MODIS) instrument that will be used is the sensor on Terra, which is in a stable sun-synchronous orbit with LECT of 10:30 am, which is relatively consistent with the ATSRs and Metop AVHRRs. It has a viewing swath width of 2330 km and views the entire surface of the Earth every one to two days. Its detectors measure 36 spectral bands between 0.405 and 14.385 µm, and it acquires data at three spatial resolutions – 250 m, 500 m, and 1000 m, with the channels relevant to our algorithms being those with 1000 m resolution.

The MODIS data used are the L1b MOD021KM and the corresponding MOD03 geolocation data (collection 6.1)¹ with data from the 24/02/2000 until present.

2. Input and auxiliary data

2.1. AVHRR L1b

AVHRR data is available from NOAA and EUMETSAT; both agencies distribute data in their own “level-1b” formats, but the formats are not equivalent. NOAA L1b contains the instrument counts plus other calibration and telemetry information, the user must apply an appropriate calibration method to obtain instrument radiances or BTs. EUMETSAT L1b (EUMETSAT Polar System) contains the calibrated radiances.

The algorithms described in this document have been applied to the EUMETSAT format data.

AVHRR L1b is available from EUMETSAT².

2.2. ATSR L1b

ATSR data from the two instruments used with GloboLakes are now available in the Envisat format used for all products from the Envisat satellite. The data are calibrated and geo-located onto a common grid for both nadir and forward views.

ATSR L1b available from ESA³.

2.3. SLSTR L1b

SLSTR data is available in ESA Standard Archive for Europe (SAFE) format⁴ which comprises a collection of netCDF4 files and an XML format manifest file. The L1b file contains also Numerical Weather Prediction (NWP) fields from ECMWF operational data. The data are calibrated and geo-located onto a common grid for both nadir and forward views.

SLSTR L1b available is available from the Copernicus Open Access Hub⁵.

2.4. MODIS L1b

The MODIS L1b data on Terra are calibrated and geolocated at-aperture radiances⁶ and they are available from NASA⁷.

2.5. Lake mask

The lake water location input consists of the land water mask (Carrea et al. 2015) derived from maximum water extent observed in ESA CCI Land Cover (v3.0) at 300-m resolution within the GloboLakes project. The land water mask was generated within the ESA CCI Lakes project for 2024 lakes distributed globally (Carrea et al. 2022) and it contains water body identifiers and distance to land for each water pixel.

The ESA CCI Lakes mask [Carrea et al, 2022] is available from zenodo⁸.

2.6. NWP data

NWP (Numerical Weather Prediction) data are used as input to the radiative transfer model (RTM) needed for LSWT retrieval. The required parameters for the various algorithms are shown in Table 5 below. The NWP LSWT currently are not sufficiently accurate for lake processing, therefore LSWT has been developed within the GloboLakes/ESA CCI Lakes project⁹ to be used as a prior LSWT for optimal estimation retrieval.

The ESA CCI Lakes and therefore the C3S LSWT processing used ERA-Interim inputs for ATSRs and AVHRRs instruments, ERA5 for MODIS and the ECMWF operational meteorological fields stored in the L1b files for SLSTRs products (brokered and extension CDR). The lake surface water temperature reported in the table is the prior temperature used for the RTM and the Optimal Estimation (OE) algorithms. A climatology has been created and used for the purpose, as a sufficiently accurate lake surface water temperature is currently unavailable.

Table 5. Inputs to algorithms: RTM – Radiative Transfer Model; OE – Optimal Estimation LSWT retrieval. Note: the OE algorithm uses the output from the RTM. Concerning “Type”, “Analysis” data is opposite to forecast data, “Profile” is a 3D field, and “surface” refers to a 2D field.

3. Algorithms

The algorithm to derive the LSWT product from data of visible and infrared radiometers consists of many components which aim to retrieve the LSWT using the observed reflectance and brightness temperature for water-only pixels. The core of the algorithm is the retrieval part which is based on Optimal Estimation (OE) given simulations and observations. The other components of the algorithm prepare the inputs for the retrieval part, namely simulating the brightness temperatures and classifying a pixel as water or non-water. The classification consists of a water detection algorithm applied to pixels defined as lake water in the ESA CCI Lakes mask (Carrea et al. 2022) derived from the GloboLakes defined in (Carrea et al. 2015). Finally, the observations are gridded into a regular 0.05^o resolution grid. Since the observed LSWTs come from different instruments, an adjustment at level 3 has been implemented, in order to stabilize the temperatures across the instruments. After the validation of the ESA CCI Lakes LSWT dataset using independent in situ measurements, the reference sensor/mission for the adjustment (harmonisation) has been selected to be the observed lake temperatures from the AVHRR on MetOp-A.

The ESA CCI Lakes LSWT algorithm was based on work carried out within the ESA ARCLake, the ESA CCI SST and the GloboLakes projects¹⁰. An important component of the ESA CCI SST algorithm (Figure 1) is the Bayesian cloud detection which depends on how accurate the prior surface temperature is. In particular, it has been found that NWP-based lake surface water temperature values are not (at present) sufficiently accurate for this purpose (MacCallum and Merchant, 2012).

Within the GloboLakes project, Plymouth Marine Laboratory (PML) has derived an indicator of cloudiness using Medium Resolution Imaging Spectrometer (MERIS) data. Since the MERIS instrument was on the same platform as AATSR and had higher resolution, the cloudiness indicator was used to decide if a pixel was cloudy or cloud-free. However, data from other instruments (such as AVHRR, SLSTR, MODIS, amongst others) needed to be processed. Therefore, a classification of water-only pixels, which does not rely on the Bayesian cloud detection and therefore on an accurate prior surface temperature, was implemented. To this aim, the probability of clouds generated by PML using MERIS data for the GloboLakes lakes has been used to

• derive a climatology which was then used as prior surface temperature for the AATSR data,
• tune the water detection algorithm introduced within ARCLake to detect water in presence of clouds.

The prior information is still needed, because it is an input for the OE retrieval algorithm. However, the OE algorithm is more robust with respect to the prior than the Bayesian cloud detection. Moreover, the OE algorithm outputs precious information beside the retrieved temperature: the uncertainty in the retrieval, the sensitivity to the prior, and the χ² of the retrieval.

The same algorithms are used for the brokered ESA CCI Lakes LSWT CDR and across all instruments. Figure 1 shows how the algorithms are used to generate the different product levels: L2P though L3S.
In this ATBD, the following aspects of the ESA CCI Lakes LSWT algorithm are described:

• Pixel classification: water detection
• Prior temperature: derivation of the temperature prior to be used in the OE algorithm
• LSWT retrieval: the Optimal Estimation algorithm
• Definition of the quality levels
• Remapping (L3U)
• Daily collation (L3C)
• Sensor adjustments (L3S)

Figure 1. Overview of algorithm steps as applied to different processing levels. Input datasets are shown as open blue slanted boxes. Processing steps are shown in black rectangles. Output and intermediary datasets are shown in blue filled slanted boxes. Boxes with rounded edges indicate the final part of the algorithm which require the L3C from all the instruments. Note the following sections describing each step are not arranged in the same progression order as displayed in this workflow.

3.1. LSWT retrieval

Optimal Estimation retrieval has been used for LSWT for all the sensors since it is based on physics and can be applied where no in situ data for retrieval tuning are available. This gives good reasons to expect stable performance across domains in time and space.

For single-view instruments, the LSWT was retrieved using an optimal estimation (OE) scheme (MacCallum and Merchant, 2012):

3.1.1. Retrieval uncertainties

3.1.2. Retrieval χ²

3.1.3. Sensitivity

3.2. Water Detection

3.2.1. Overview

Water detection is a fundamental pre-processing step for LSWT retrieval. It was introduced during the ARCLake project to enable inclusion of targets with variable surface area. Within the GloboLakes and ESA CCI Lakes projects, the water detection is applied to a fixed mask defined in Carrea et al. (2015) for day-time observations. Moreover, the thresholds set within ARCLake have been tuned using the probability of clouds derived by PML on MERIS observations. This enables the algorithm to detect water in the presence of clouds not relying on the Bayesian cloud detection. The ARCLake thresholds are reported in Table 6.

Table 6: ARCLake thresholds test for water detection using radiance and indices thresholds.

A score s has been introduced to be able to evaluate the performance of the test for X especially close to the thresholds:

3.2.2. Tuning the water detection with MERIS probability of clouds

Within the GloboLakes project a probability of cloud for each pixel has been generated by PML using MERIS data. Since both instruments were on the same platform, Envisat, it was possible to verify and adjust the water detection performance on the AATSR data. The algorithm for the detection and classification of clouds over water was proposed by Schiller et al. (2008). It produces a graduated scale, as an indicator of the extent to which a signal is influenced by the presence of clouds. The indicator was evaluated for the 1000 GloboLakes lakes for the whole set of MERIS data, and a cloud indicator quality has been introduced which is related to the number of cloud-free MERIS pixels (300m resolution) within one AATSR pixel (1km resolution).

Figure 2: False colour composite image of reflectance at 1600 nm (R), 870 nm (G) and 670 nm (B), quality of the MERIS cloud indicator (with 0=clear up to 4=full cloud) and water detection score (with 0=no water up to 4=full water) for lake Michigan in USA on the 15-Feb-2011.

Figure 2 shows an example (lake Michigan in the USA) where the water detection misclassifies cloudy pixels as water pixels. In order to use the water detection algorithm in the presence of clouds, we tuned the thresholds of the water detection algorithm applied to the AATSR data according to the quality of the cloud indicator computed on the MERIS data. In order to statistically investigate the relationship between the MERIS cloud indicator and the water detection thresholds, histograms of clear and cloudy pixels according to quality of the MERIS cloud indicator were inspected for the different parameters involved in the water detection threshold test (described in Sec. 3.2.1).

Figure 3 shows an example of the Modified Normalised Difference Water Index (MNDWI) histogram of cloudy and clear pixels (according to the MERIS cloud indicator) for lake Michigan in USA.

Figure 3. Histogram for MNDWI for AATSR pixels classified as cloudy with the MERIS data (grey) and as cloud-free pixels with different quality levels for lake Michigan. The legend shows for each colour the quality level of the MERIS data according to the size of the window. For example, 5: 7 pixels x 7 pixels represented in cyan would be considered a higher quality estimate of pixel quality as it is based on more pixels. In contrast, 2: 3 x 3 in yellow would be considered a lower quality assessment of pixel quality. The legend shows also for each colour the fraction of pixels having the correspondent quality level (e.g. for cyan: 0.33, for yellow: 0.01). The solid red line in the plot represents the threshold defined in Table 6 and the red dotted line represents the level where the score reaches the value of 1 (see Sec. 3.2.1). 

In order to tune the thresholds according to the cloud indicator from the MERIS data, the threshold was found in the overlapping area of the two histograms defined as the location of the maximum entropy of the posterior probability that a certain pixel is cloudy or cloud-free. An example of the plot of the entropy together with the posterior probability is shown for MNDWI for lake Michigan in Figure 4.

Figure 4. The upper plot shows the pdf (probability density function) for cloud (grey) for cloud-free (light-blue) and for all pixels (red) for lake Michigan in USA. The lower plot shows the posterior probability (grey line for clouds and light blue line for cloud-free) and the entropy (red line) for MNDWI. A pixel with a MNDWI value above the maximum entropy threshold (the red shaded area) is considered to be water.

This has been done per–lake and then a common threshold for all the lakes has been chosen for each of the water detection parameters. However, note that all the lakes with a spatial coverage of less than 500 pixels (less than 1% of the lakes) were not used in the evaluation. The new thresholds are reported in Table 7.

Table 7: GloboLakes thresholds test for water detection.

A total score (s) is then calculated, which is the sum of the scores for each parameter (see Section 3.2.1)

is introduced to quantify the performance of the water detection algorithm, where the reflectance at 670 nm has been removed. This accounts for the reality that lakes can have different spectral properties, especially in the visible range. The new water detection thresholds improve the water detection for lake Michigan as shown in Figure 5.

Figure 5. ARCLake water detection, GloboLakes water detection and MERIS probability of cloud quality for lake Michigan. The colors in the second and third plots represent the water detection score.

3.3. Prior temperature

The prior temperature is a necessary input for the OE algorithm (Figure 1 and Section 2.6). For lakes examined under ESA CCI Lakes programme, it has been generated in form of a spatially filled climatology by an iterative method. The method starts by processing AATSR data and uses the MERIS cloud indicator together with the ARCLake water detection, using as a prior the 2m T_air with high uncertainty. Then, the first climatology is generated and it is used as a prior LSWT for the next processing of AATSR where now the prior is a LSWT and not the air temperature.

3.4. Determining Quality Level

Quality level is treated as a concept that is distinct from uncertainty. The quality level reflects the degree of confidence in the validity of the uncertainty estimate, not the magnitude of data uncertainty. For example, a highly uncertain LSWT can have the highest quality level if all the conditions for giving a valid LSWT and valid LSWT uncertainty are met.

The quality level assigned to a pixel will be the lowest quality level (row of Table 8) which matches any of the conditions shown in the Table 8. The assignments are compatible with GHRSST conventions [D1]: i.e., a particular level is given if none of the conditions higher up any column of the table are met.

Table 8: Quality Level definitions for pixels in the LSWT data product. The distance d is in km.

For instance, any pixel where s is unavailable (value is less than zero), required input BTs are unavailable, or which is over land will be assigned quality level of 0. Next, any pixels close to land which have s < 0.5, calculated LSWT sensitivity < 0.1 etc. will be assigned quality level of 1 and so on.

• Quality level 0 pixels should contain no other data.
• Quality level 1 pixels contain data, but the data is not suitable for use (bad_data). For instance, the LSWT retrieval may have been attempted, but rejected as bad_data due to low sensitivity etc.
• Quality level 2-5 pixels should always contain valid data.
• We recommend using quality level 4 and 5, and consideration of use of quality level 3 with caution, depending on the user's application.

3.5. Remapping (L3U)

The remapping from the L2 data in swath projection to the fixed L3 grid (lat/lon WGS1984 datum) proceeds as follows:

• Identify L2 pixels contributing to a L3 cell
• Select highest quality level pixel(s) in the L3 cell
• Calculate average LSWT from the pixels that share the highest quality level and propagate uncertainties to the uncertainty in this average (as in Bulgin et al. 2016a)

When averaging from the pixel scale to L3 grid scale, the component or uncertainty from uncorrelated errors reduces (uncertainty in the mean is scaled by the familiar "1/√n"). Uncertainty in the form of correlated error components are not reduced by averaging, since over these small scales the degree of correlation will be very high and is assumed to be perfect ("r = 1"). The total uncertainty in the average is found by the propagation of uncertainty law.

If the L3 grid cell contains L2 pixels which were not included in the averaging (e.g. due to the presence of cloud etc.), then there is an additional uncertainty due to incomplete sampling. This is calculated following Bulgin et al. (2016b) (derived for application to sea surface temperature uncertainty estimation) and is added to the uncorrelated component.

3.6. Daily Collation (L3C)

The polar orbiting satellites carrying the MODIS/SLSTR/AVHRR/ATSR sensors typically complete 14-15 orbits each day resulting in the same number of L2P or L3U products. While L3U files are on a global grid, they are very sparse as the sensor will only observe a small fraction of the Earth’s surface in each orbit, with each file only containing a single orbit. For ease of use, the LSWT outputs are collated to produce one file for each 24-hour period, corresponding to day-time observations.

Following the GHRSST conventions [D1], when collating observations from overlapping orbits in the same day the L3C will contain the highest quality observation available in the 24-hour period. The selection of best observation is done as follows:

• Choose input cell values with the highest associated quality level flag.
• If multiple observations have the same quality level, then calculate the average of the high quality level values.

3.7. Inter-sensor adjustment (L3S)

The adjustment due to different sensors has been implemented using the AVHRR sensor on MetOp-A as the reference (except for MODIS and SLSTR). This follows from a validation study within the GloboLakes project, which indicated a better agreement throughout the lakes between estimates derived from AVHRR measurements, and in-situ data. The inter-sensor adjustment has been calculated per lake averaging per month. It has been applied only if:

• Enough observations were available to estimate the adjustment for the lake (more than 3 months of data), and
• The uncertainty of the adjustment was < 0.049 – which was valid for about 800 of the lakes.

For MODIS, only LSWTs of quality level 4 and 5 have been used for the final dataset. Overall, adjustments of 0.19 K and 0.11 K for quality level 4 and 5 respectively have been applied.

A flag indicating whether the adjustment has been applied is present in the files and the uncertainty of the bias correction is included in the total uncertainty. For lakes where the flag is not set, the impact of changes in sensor on the long-term trends in LSWT is less well constrained, and trends should be treated with caution.

The final L3S product has been created using the available observations. In case of more than one observation occurring per pixel, the best quality has been chosen in a manner similar to L3C, i.e.:

• Input cell values with the highest associated quality level are chosen, and
• Averaging of values is implemented, if multiple observations have the same quality level.

4. Output data

The following sections present a brief overview of the data file format, naming convention and structure. Further information can be found in the Product User Guide and Specification for these products [D2].

4.1. File format and naming scheme

The output data are in netCDF4-classic format¹¹, and are guided by the GHRSST data specifications [D1]. The file names have the format:

<Date><Time>-<RDAC>-<Level>-LSWT-<Dataset>-fv01.0.nc

Note that fv01.0 refers to the file version. This increments upwards if an amended file version (containing the same data) needs to be distributed. Such file updates could occur in the event of issues being identified by quality control reviews, or by users of the files who choose to submit feedback. The incrementation will appear in the filename, taking the form v01.1, v01.2 and so on.

Table 9: Filenaming convention components used for the LSWT products.

4.1.1. Date

The identifying date for this file, using the ISO8601 basic format: YYYYMMDD.

4.1.2. Time

The identifying time for this file in UTC, using the ISO8601 basic format: HHMMSS. The time used depends on the processing level of the dataset:
L3S:centre time of collation window (120000 for daily files)

4.1.3. RDAC

GHRSST Regional Data Assembly Centre (RDAC) where the dataset was generated. Three codes are used for C3S products:

C3S : Copernicus Climate Change Service
GloboLakes : NERC GloboLakes
ESACCILakes: European Space Agency Climate Change Initiative Lakes

4.1.4. Level

The GHRSST processing level for this product will be L3S.

4.1.5. Dataset

Indicates the scientific version number of the product (this corresponds to the versioning used on the CDS). Current string in use is:

v4.5

4.2. File contents

L3S files are supplied on a regular latitude/longitude grid and variables have 3 dimensions, namely:

• time:1 (defined as unlimited)
• lat: Number of latitude points (3600)
• lon: Number of longitude points (7200)

Table 10. Output fields for data in the downloadable LSWT product files (see the Product User Guide and Specification document [D2] for details).

References

Bulgin, C. E., Embury, O., Corlett, G. and Merchant, C. J. (2016a) Independent uncertainty estimates for coefficient based sea surface temperature retrieval from the Along-Track Scanning Radiometer instruments. Remote Sensing of Environment, 178. pp. 213-222. ISSN 0034-4257 doi:10.1016/j.rse.2016.02.022

Bulgin, C. E., Embury, O. and Merchant, C. J. (2016b) Sampling uncertainty in gridded sea surface temperature products and Advanced Very High Resolution Radiometer (AVHRR) Global Area Coverage (GAC) data. Remote Sensing of Environment, 117. pp. 287-294. ISSN 0034-4257 doi:10.1016/j.rse.2016.02.021

Carrea, L., Embury, O. and Merchant, C. J. (2015) Datasets related to in-land water for limnology and remote sensing applications: distance-to-land, distance-to-water, water-body identifiers and lake-centre co-ordinates. Geoscience Data Journal, 2(2). pp. 83-97. ISSN 2049-6060 doi:10.1002/gdj3.32

Carrea, L., Merchant, C. and Simis, S. (2022) Lake mask and distance to land dataset of 2024 lakes for the European Space Agency Climate Change Initiative Lakes v2 (Version 2.0.1). zenodo https://doi.org/10.5281/zenodo.6699376.

MacCallum, S.N. and Merchant, C. J. (2012) Surface water temperature observations for large lakes by optimal estimation. Canadian Journal of Remote Sensing, 38(1). pp. 25-45. ISSN 1712-7971 doi:10.5589/m12-010

Rodgers, C.D. (2000) Inverse methods for atmospheric sounding: Theory and Practice. Edited by RODGERS CLIVE D. Published by World Scientific Co. Pte. Ltd

Schiller H., Brockmann C., Krasemann H. and Schoenfeld W. (2008) A method for detection and classification of clouds over water. In Proc. of the 2^nd MERIS (A)ATSR User Workshop. Sept 2008.

Related articles

• Page:
  ECV - Glaciers, Ice Sheets, Lakes, Soil Moisture
• Page:
  IV Version 1.2; SEC Version 2.0; GMB Version 2.0: Product User Guide Specification (PUGS)
• Page:
  IV Version 1.2; SEC Version 2.0; GMB Version 2.0: System Quality Assurance Document (SQAD)
• Page:
  IV Version 1.2; SEC Version 2.0; GMB Version 2.0: Product Quality Assurance Document (PQAD)
• Page:
  IV Version 1.2; SEC Version 2.0; GMB Version 2.0: Product Quality Assessment Report (PQAR)