An Evaluation of Sentinel-3 SYN VGT Products in Comparison to the SPOT/VEGETATION and PROBA-V Archives

Abstract

Sentinel-3 synergy (SYN) VEGETATION (VGT) products were designed to provide continuity to the SPOT/VEGETATION (SPOT VGT) base products archive. Since the PROBA-V mission acted as a gap filler between SPOT VGT and Sentinel-3, and in principle, a continuous series of data products from the combined data archives of SPOT VGT (1998–2014), PROBA-V (2013–2020) and Sentinel-3 SYN VGT (from 2018 onwards) are available to users, the consistency of Sentinel-3 SYN VGT with both the latest SPOT VGT (VGT-C3) and PROBA-V (PV-C2) archives is highly relevant. In past years, important changes have been implemented in the SYN VGT processing baseline. The archive of SYN VGT products is therefore intrinsically inconsistent, leading to different consistency levels with SPOT VGT and PROBA-V throughout the years. A spatio-temporal intercomparison of the combined time series of VGT-C3, PV-C2 and Sentinel-3 SYN VGT 10-day NDVI composite products with an external reference from LSA-SAF, and an intercomparison of Sentinel-3 SYN V10 products with a climatology of VGT-C3 resp. PV-C2 for three distinct periods with different levels of product quality have shown that the subsequent processing baseline updates have indeed resulted in better-quality products. It is therefore essential to reprocess the entire Sentinel-3 SYN VGT archive; a uniform data record of standard SPOT VGT, PROBA-V and Sentinel-3 SYN VGT products, spanning over 25 years, would provide valuable input for a wide range of applications.

1. Introduction

Sentinel-3A was launched on 16 February 2016 and accompanied by Sentinel-3B on 25 April 2018, with onboard multiple sensing instruments focusing on Earth observation to support Copernicus ocean, land, atmosphere, emergency, security and cryosphere applications. Co-located observations of two of these instruments—the Ocean and Land Colour Instrument (OLCI) and the Sea and Land Surface Temperature Radiometer (SLSTR)—are used as synergistic inputs for the so-called ‘synergy’ (SYN) processing chain [1].

The Sentinel-3 SYN processing chain combines observations in the visible and near-infrared (VNIR) and short-wave infrared (SWIR) channels to provide several higher-level products. One particular series of output products of the SYN processing chain are the SYN VEGETATION (SYN VGT) products, which are designed to provide surface reflectance products similar to those obtained from the VEGETATION1 and VEGETATION2 (VGT) instruments onboard SPOT4 (launched 24 March 1998) and SPOT5 (launched 4 May 2002), respectively, to meet Copernicus—previously known as Global Monitoring for Environment and Security (GMES)—user needs [2].

SPOT5 was, however, decommissioned in early 2015, i.e., before the launch of the first Sentinel-3 satellite. From 2013 to 2020, PROBA-V (Project for On-Board Autonomy—Vegetation), which was launched on 7 May 2013, provided data continuity to the SPOT VGT payloads and acted as a gap filler to the Sentinel-3 mission [3]. The PROBA-V mission was designed, and products were defined to closely match SPOT VGT products [4], and after the latest reprocessing campaigns, both the SPOT VGT Collection 3 and PROBA-V Collection 2 archives were shown to be highly consistent [5,6].

Similar to the standard products of SPOT VGT and PROBA-V, Sentinel-3 SYN VGT products provide complete Earth coverage every 1 to 2 days in four spectral bands (BLUE, with a central wavelength around 460 nm; RED, 655 nm; NIR, 835 nm; and SWIR, 1600 nm) (see also

Table 1. The spectral range (the center wavelength and Full Width at Half Maximum (FWHM) (between parentheses) for SPOT4/VGT1, SPOT5/VGT2 [7], PROBA-V [8] and the relevant Sentinel-3 OLCI (Oa) and SLSTR (S) bands [9].

SENSOR	BLUE [nm]	RED [nm]	NIR [nm]	SWIR [µm]
SPOT4/VGT1	459 (43)	658 (85)	834 (121)	1.649 (0.092)
SPOT5/VGT2	458 (37)	653 (74)	838 (109)	1.635 (0.101)
PROBA-V	464 (47)	655 (82)	837 (130)	1.603 (0.065)
Sentinel-3/OLCI	Oa03: 442 (10) Oa04: 491 (10)	Oa07: 621 (10) Oa08: 666 (10) Oa09: 674 (7.5) Oa10: 682 (7.5)	Oa16: 779 (15) Oa17: 866 (20) Oa18: 886 (10)
Sentinel-3/SLSTR				S5: 1.610 (0.065)

) projected on the same regular 1 km ‘Plate carrée’ grid [1]. Output products include (i) top-of-atmosphere (TOA) reflectance (VGP), (ii) 1-day synthesis surface reflectance and Normalized Difference Vegetation Index (NDVI) (VG1) and (iii) 10-day synthesis surface reflectance and NDVI (V10).

In principle, users thus have access to a continuous series of data products from the combined data archives of SPOT VGT (1998–2014), PROBA-V (2013–2020) and Sentinel-3 SYN VGT (from 2018 onwards). There are, however, several aspects related to the satellite/sensor and processing definitions that, to a greater or lesser extent, impact the consistency between these product series, such as differences in the acquisition scheme, sensor design, spectral response, atmospheric correction and sensor calibration. Another crucial aspect is that since the release of the first Sentinel-3 SYN VGT products in October 2018, several important changes in the SYN VGT processing baseline have been implemented, thereby gradually improving the quality of SYN VGT products. However, to date, reprocessing of the SYN VGT archive has not taken place. The archive of SYN VGT products is therefore intrinsically inconsistent, and consequently, different consistency levels with SPOT VGT and PROBA-V will be reached throughout the years.

Numerous studies have addressed the challenge of transitioning between consecutive wide swath land monitoring sensors while attempting to provide consistent product archives. These include, for example, the assessment of the uncertainties related to the transition from the Moderate Resolution Imaging Spectroradiometer (MODIS) to Visible Infrared Imaging Radiometer Suite (VIIRS) [10], between different AVHRR sensors [11,12,13], from the Advanced Very-High-Resolution Radiometer (AVHRR) to MODIS [14,15,16] or from the AVHRR to SPOT VGT [15,17]. In many cases, spatial aggregation, spectral adjustments, radiometric intercalibration factors or adjustments for the Bidirectional Reflectance Distribution Function (BRDF) are applied in order to reduce differences and to minimize the effect of differences in viewing and illumination geometries. This study focuses on the transition from SPOT VGT (and PROBA-V) to Sentinel-3 SYN VGT. However, in order to conduct this evaluation from a user perspective, we decided to not apply any adjustments or corrections and to evaluate the operational products in the form they are made available to the user.

The goal of this paper is to evaluate how the consecutive improvements in Sentinel-3 SYN processing have (positively) affected the level of consistency of Sentinel-3 SYN VGT V10 products with SPOT VGT Collection 3 and PROBA-V Collection 2 products, and to determine what level of consistency is reached in the SYN VGT products that are currently made available shortly after new Sentinel-3 OLCI and SLSTR acquisitions. Furthermore, we aim to inform users about the quality concerns present in the current SYN VGT data set, particularly when these products are used alongside the SPOT VGT and PROBA-V product archives.

To reach these goals, we first compare the combined time series of 10-day NDVI synthesis products from SPOT VGT, PROBA-V and Sentinel-3 SYN VGT with an external NDVI time series derived from the AVHRR onboard the METOP satellites. Secondly, Sentinel-3 SYN VGT V10 products for three distinct periods with different levels of product quality are compared to 5-year climatology products derived from SPOT VGT and PROBA-V, respectively. Analyses are performed on a systematic global spatial subsample.

2. Materials and Methods

2.1. Data

2.1.1. Sentinel-3 SYN VGT V10

Sentinel-3 SYN VGT products and processing steps are described in [1]. In summary, the SYN VGT processing module consists of (i) the Level-1 module dedicated to the co-registration of OLCI and SLSTR acquisitions and the production of internal SYN Level-1 products, (ii) the SYN Level-2 module dedicated to aerosol retrieval and atmospheric correction and (iii) the SYN VGT module that performs spectral mapping to simulate the SPOT VGT or PROBA-V spectral bands (see Table 1) and projection and compositing to generate simulated VGP, VG1 and V10 products [18]. In this study, we focus on V10 products, i.e., 10-day maximum NDVI value composites of surface reflectance measurements and the NDVI at 1 km spatial resolution.

As highlighted above, important updates, improvements and bug fixes were implemented into the SYN VGT processing lines since the first release of SYN VGT products in 2018 (exact dates and processing baseline documents are available at https://sentiwiki.copernicus.eu/web/synergy-processing, accessed on 25 August 2024). The most impactful changes are listed below and schematized in Figure 1:

• September 2019: The correction of a 0.5-pixel displacement in latitude and longitude directions in product gridding. Measurements are provided on the same regular latitude–longitude grid as previous SPOT VGT and 1 km PROBA-V products with an equatorial sampling distance of approximately 1 km (1°/112).
• June 2020: The alignment of the temporal compositing scheme of 1-day and 10-day synthesis products with previous SPOT VGT and PROBA-V products. Per month, three V10 products are provided based on observations in days 1–10, days 11–20 and day 21 to the end of the month, respectively.
• June 2021: Correction in the definitions of VG1 and V10 NDVI to be based on surface reflectance in the RED and NIR bands. Until May 2021, NDVI products were erroneously based on top-of-atmosphere (TOA) reflectances.
• August/September 2022: Improved handling of saturated values.
• July 2023: The last important update includes two aspects, namely (i) correction in the spectral band mapping procedure and (ii) S3A/OLCI and SLSTR calibration adjustments. The first aspect includes a change to the use of PROBA-V spectral response functions (instead of SPOT4/VGT1) in the spectral band mapping procedure. Although differences between the spectral responses between SPOT VGT and PROBA-V are rather small [3], this change has a slight impact on the retrievals in the RED and SWIR bands. A much larger impact is induced by the implementation of important bug fixes, including the correct exclusion of atmospheric absorption bands and the correct handling of wavelength units. The second aspect includes the application of a 2% calibration bias correction on S3A OLCI, as evidenced through the tandem phase study [19], and the application of the Sentinel-3 SLSTR vicarious calibration adjustments [20].

For the statistical evaluation of the consistency between SYN VGT products and the SPOT VGT and PROBA-V product archives, 3 periods of 12 months were selected in between these processing baseline updates as follows(Figure 1):

• Period 1 (P1): June 2020–May 2021;
• Period 2 (P2): August 2021–July 2022;
• Period 3 (P3): August 2023–July 2024.

Sentinel-3 SYN VGT products are made available through the Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/, accessed on 25 August 2024).

2.1.2. SPOT VGT Collection 3 Level 3 S10-TOC

After the end of the SPOT VGT mission in May/2024, the complete archive was reprocessed, resulting in the Collection 3 archive (VGT-C3) [21]. The VGT-C3 Level-3 10-day top-of-canopy (TOC) synthesis products (S10-TOC) containing surface reflectance and NDVI from 2009 to June/2014 were used in this study. For more details on the processing of SPOT VGT data, we refer to the SPOT VGT Products User Manual [7].

From 2009 onwards, SPOT5 experienced orbital drift, causing the satellite overpass time to gradually shift over time. This evolution causes small but systematic changes in illumination conditions and related BRDF effects, e.g., the NDVI tends to increase with higher solar zenith angles [22,23,24].

VGT-C3 data products are available through the Terrascope platform (https://www.terrascope.be, accessed on 25 August 2024).

2.1.3. PROBA-V Collection 2 Level 3 S10-TOC

Designed as a successor for the SPOT VGT, the PROBA-V mission provides continuity products to SPOT VGT, although products at a higher spatial resolution (300 m and 100 m) are also disseminated. Detailed descriptions of the PROBA-V mission and processing chains are provided in [4,25]. Also, the PROBA-V archive was reprocessed after the end of its operational lifetime, aiming at improving the time series and harmonizing its content. The resulting Collection 2 (PV-C2) was released in 2023 [8]. In this study, we use PV-C2 Level-3 S10-TOC surface reflectance and the NDVI for the period of January 2014–June 2020.

It should be noted that at the time of the switch between SPOT VGT and PROBA-V, there was an important difference in the equator local overpass times between SPOT5 and PROBA-V. Since PROBA-V had no onboard propulsion, the satellite experienced a constantly varying overpass time.

PV-C2 data products are available through the Terrascope platform (https://www.terrascope.be).

2.1.4. LSA-SAF METOP/AVHRR ENDVI10 Version 2

The European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Polar System (EPS) consists of a series of polar-orbiting meteorological satellites known as METOP. The AVHRR instruments onboard METOP are used to generate the 10-day Enhanced NDVI (ENDVI10) by the Land Surface Analysis Satellite Application Facility (LSA-SAF). ENDVI10 version 2 is processed in a similar way to the S10-TOC of PROBA-V, with the same water vapor and ozone inputs and similar atmospheric correction and compositing method [26]. The ENDVI10 compositing method is similar to the SYN V10 temporal compositing scheme, and the products are also provided on the same regular latitude/longitude grid as Sentinel-3 SYN VGT products. Although there are differences in spectral response, calibration, cloud detection and overpass time stability that will influence the intercomparison, the ENDVI10 is a relevant external reference since no spatial or temporal remapping is needed, which would introduce additional uncertainties. Global ENDVI10 version 2 products derived from METOP-A (launched 19 October 2006) for the period of January 2009–April 2013 and METOP-B (launched 17 September 2012) for the period of May 2013–July 2024 are used in the evaluation as an external reference.

2.2. Methods

2.2.1. Sampling

A systematic global spatial subsample is taken from each global product by considering one pixel out of every twenty in both the latitude and longitude directions. This (arbitrary) subsample is representative of the global patterns of vegetation and considerably reduces processing time while retaining the original resolution. To generate Hovmöller diagrams (see below), this subsample is grouped for latitude bands of 12°.

Only pixels that are not identified as cloud, cloud shadow or snow in the respective product status maps are considered for the analyses. For PV-C2 data, the aerosol optical thickness (AOT) mask is applied, and observations affected by a high AOT (AOT ≥ 0.8) are discarded [8].

In order to evaluate gradual changes in SYN VGT data quality, temporal sampling of SYN VGT V10 products includes the delineation of three 12-month periods labeled as P1, P2 and P3 (see above).

2.2.2. Long-Term Statistics

Long-term statistics (LTS) are retrieved from the extracted VGT-C3 and PV-C2 spatial subsamples by calculating an average value for each 10-day period in the year and for each sample. For both VGT-C3 and PV-C2, a 5-year period is considered for LTS calculation: the 2009–2013 period for VGT-C3 and the 2014–2018 period for PV-C2, respectively. The resulting climatology for each product series is a measure of the average status of the surface for the respective LTS periods and is used as the reference in statistical intercomparison.

2.2.3. Geometric Mean Regression and Coefficient of Determination (R²)

The geometric mean (GM) regression model is used to identify the relationship between two datasets of remote sensing measurements. Because both data sets (X: reference; Y: product under evaluation) in this case are subject to noise, it is appropriate to use an orthogonal (model II) regression. The GM regression model minimizes the sum of the products of the vertical and horizontal distances (i.e., errors on both X and Y). By applying an eigen decomposition to the covariance metrics of X and Y, two eigenvectors are obtained that describe the principal axes of the point cloud [27].

The coefficient of determination (R²) indicates agreement or covariation between two data sets with respect to the linear regression model. It summarizes the total data variation explained by this linear regression model—higher R² values indicate higher covariation.

2.2.4. APU Metrics

The deviations between two data sets are evaluated through an assessment of the Accuracy, Precision and Uncertainty (APU) metrics [28]. The Accuracy (A) or mean bias measures the average actual difference between two data sets, X and Y:

$$A={\displaystyle \frac{1}{n}}\sum _{i=1}^n(X_i-Y_i)=\overline{X}-\overline{Y}$$

(1)

As such, A retains the sign of the difference between the data sets and is a measure for systematic bias. Precision (P) or repeatability represents the dispersion of product retrievals around their expected value and is estimated by the standard deviation of the bias between retrieved satellite products:

$$P=\sqrt{{\displaystyle \frac{1}{n-1}}\sum _{i=1}^n{(X_i-Y_i-A)}^2}$$

(2)

Finally, Uncertainty (U) is defined as the overall difference, including random and systematic differences, and is measured through the root mean squared difference:

$$U=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n({X_i-Y_i)}^2}$$

(3)

High APU values reflect discrepancies between two compared data sets, whereas low values indicate high consistency [29]. Accuracy is mainly influenced by differences in absolute radiometric calibration and spectral response. Precision is influenced by temporal, spatial and angular sampling; spectral response and the performance of the spectral resampling algorithm applied for SYN VGT; and the atmospheric effects and performance of pixel classification algorithms to exclude satellite observations perturbed by clouds, cloud shadows or snow. Uncertainty provides an overall measure of differences between two datasets.

2.2.5. Hovmöller Diagrams

In order to perform a combined assessment of the spatial and temporal variability of the APU metrics of the intercomparison between the combined time series of the 10-day NDVI synthesis products from VGT-C3, PV-C2 and SYN V10 on the one hand with the stable time series of ENDVI10 on the other hand, Hovmöller diagrams are made. The metrics are derived on the global subsample for each time step (10-day period) and for each secondary spatial subset, and they are defined as latitude bands of 12°. The resulting time–latitude Hovmöller diagrams allow for the space–time features of the time series evaluation to be summarized, thereby depicting the temporal evolution of the spatial agreement [30].

3. Results and Discussion

3.1. Spatio-Temporal Intercomparison with LSA-SAF ENDVI10

As the ultimate goal of Sentinel-3 SYN VGT products is to provide continuity to the SPOT VGT and PROBA-V product archives, we focus first on the spatio-temporal intercomparison of the combined series with an external data set. The Hovmöller plots of the intercomparison between LSA-SAF ENDVI and the combined series of 10-day NDVI composites from VGT-C3 (2009–2013), PV-C2 (2014–June 2020) and S3A SYN VGT (July 2020–July 2024) are illustrated in Figure 2. These allow us to obtain a broad overview of the temporal stability of Sentinel-3 SYN V10 NDVI products in relation to the SPOT VGT and PROBA-V archives. The results for S3B SYN V10 are very similar and not shown.

Whereas higher values for P and U are observed around the equator and up to 20° S in the Southern Hemisphere summer period, these (mostly densely vegetated) areas show, in general, a lower A; the average bias between both time series is lower, but with higher dispersion. This could be related to a larger influence of atmospheric effects in tropical areas and more cloud contamination that is still present in the maximum NDVI value composites [31].

The three 12-month periods that are used to evaluate the Sentinel-3 SYN V10 products’ consistency with the SPOT VGT and PROBA-V data archives are delineated in Figure 2. An important discontinuity in A and U is observed at the switch from PROBA-V to S3A (July 2020). This is caused by the fact that before June 2021 (i.e., in P1), Sentinel-3 SYN V10 was (incorrectly) based on TOA reflectances. After the correction, in P2, A stabilizes but at a rather high level. The corrections implemented in July 2023 result in A being closer to zero in P3.

Towards the end of both the SPOT VGT (2013) and PROBA-V (June 2020) periods, a deviation of A is visible. An orbital drift in both instruments (see above) leads to gradually earlier overpass times, resulting in increasing solar zenith angles. Although the effects of the orbital drift have been reported to be mitigated to some extent in the NDVI [24,32], the effect is not negligible and can only be mitigated through anisotropy correction based on the BRDF [33,34].

3.2. Intercomparison with VGT-C3 and PV-C2 LTS for P1, P2 and P3

In order to evaluate the evolution of the consistency between Sentinel-3 SYN V10 and the SPOT VGT and PROBA-V data archives, Figure 3 and Figure 4 show the results of the statistical consistency analysis between VGT-C3 LTS resp. PV-C2 LTS and S3A SYN V10 for P1, P2 and P3 and for the NDVI and four spectral bands. The results for S3B are very similar and are not displayed.

The large scatter in the scatter density plots overall (leading to high values for P) is partly related to the fact that two different periods are compared: the 5-year period on which the respective LTS are based versus the 12-month periods in the more recent SYN V10 product archive. Land cover transitions, e.g., vegetation loss, urban expansion, reforestation or agricultural expansion, will have taken place on the surface in a minority of the pixels sampled. Also, natural fluctuations in vegetation development occur, plus human-induced variations, such as, e.g., crop rotation. In addition, and more importantly, the satellite observations are influenced by angular and atmospheric effects, although these effects are attenuated to some extent through the maximum NDVI value composite algorithm [35] that is used to generate the 10-day composite products. Both SPOT VGT and PROBA-V have a very wide swath coverage, with swath widths above 2200 km, whereas both OLCI and SLSTR have more narrow swath widths (around 1400 km). The OLCI swath is not centered at nadir but tilted 12.6° westward to minimize the impact of sun glint contamination [2]. In addition, whereas Sentinel-3 flies in a sun-synchronous orbit with descending node equatorial crossing at 10 h mean local time, both SPOT VGT and PROBA-V experienced orbital drift (see above). Different sampling in both viewing and illumination angles contributes to unsystematic differences observed in the comparison. Secondly, incomplete cloud and cloud shadow masking can influence the intercomparison, especially in areas with persistent cloud cover where the probability of not having clear sky observations in the 10-day compositing period is higher. In this respect, it should noted that currently, no cloud shadow masking is carried out in Sentinel-3 SYN VGT products. Finally, differences in atmospheric correction lead to unsystematic differences in TOC reflectances and the NDVI. One aspect is the estimation of the aerosol optical thickness (AOT). For VGT-C3 atmospheric correction, the AOT is estimated from the BLUE band and the NDVI through an optimization process [36]. PV-C2 uses an external AOT data set, namely the MERRA-2 (Modern-Era Retrospective analysis for Research and Applications, version 2), as an input for atmospheric correction [37]. In Sentinel-3 SYN Level-2 processing, the AOT is retrieved using collocated OLCI and SLSTR TOA radiances based on methods developed for the predecessors of both sensors [38], and previous studies have shown that this AOT tends to be overestimated [39].

Apart from the large scatter, discussed above, overall high linear correlations are found between the VGT-C3 resp. PV-C2 LTS and the SYN V10 products, with R² values around or well above 0.7. The largest discrepancies are found for the NDVI in P1, with a GMR slope well below 1, and for the SWIR band in P1 and P2, with GMR intercepts well below 0. Figure 5 shows the gradual improvements in APU statistics from P1 to P2 and to P3. The largest improvement for the NDVI, leading to drops in A from 0.1 to 0.05 (in comparison to SPOT VGT) and 0.04 (in comparison to PROBA-V), is related to the correction in the definition of the product in June 2021: in P1, the product was incorrectly based on TOA reflectances. Finally, corrections and adaptations in the spectral band mapping and calibration, applied in July 2023 (P3), resulted in A around 0.02 and 0.01 in comparison to SPOT VGT and PROBA-V, respectively. This effect is obtained by small improvements in the consistency for the RED band. The largest effects of the changes applied in July 2023 are visible in the SWIR band, where A drops from around 0.09 in P1 and P2 to values around −0.02 in P3.

The overview of the APU statistics in Figure 5 also shows that there is very little difference between the results based on Sentinel-3A and Sentinel-3B SYN V10 products. The application of the OLCI-A calibration adjustments (implemented between P2 and P3) seem to have only minor impacts. Values for P remain roughly constant over the three periods because the reasons for high scatter in the comparison (see above) have not been influenced by the subsequent processing baseline updates.

3.3. The Current Consistency Level between Sentinel-3 SYN V10 Products and the SPOT VGT and PROBA-V Archives

Finally,

Table 2. APU metrics for the intercomparison between the VGT-C3 and PV-C2 LTS and Sentinel-3 SYN V10 products for P3 (August 2023–July 2024). The statistics give a measure for the current consistency levels between Sentinel-3 SYN V10 and the SPOT VGT resp. PROBA-V product archives.

Intercomparison	Metric	BLUE	RED	NIR	SWIR	NDVI
VGT-C3 LTS vs. S3A SYN V10	A	−0.003	−0.024	−0.035	−0.022	0.022
	P	0.023	0.037	0.057	0.059	0.083
	U	0.023	0.044	0.067	0.063	0.086
VGT-C3 LTS vs. S3B SYN V10	A	−0.002	−0.024	−0.037	−0.020	0.018
	P	0.023	0.037	0.057	0.057	0.083
	U	0.023	0.045	0.068	0.060	0.085
PV-C2 LTS vs. S3A SYN V10	A	0.012	−0.014	−0.019	−0.025	0.013
	P	0.025	0.036	0.057	0.056	0.08
	U	0.028	0.039	0.06	0.061	0.081
PV-C2 LTS vs. S3B SYN V10	A	0.012	−0.014	−0.021	−0.023	0.009
	P	0.026	0.036	0.056	0.055	0.080
	U	0.029	0.039	0.06	0.059	0.080

provides the APU metrics that are based on the latest 12 months of available Sentinel-3 SYN V10 data (P3), as these provide a measure of the current consistency level between SYN V10 and the SPOT VGT resp. PROBA-V archives. Overall, the consistency is higher with the PV-C2 LTS, with absolute A values around 0.01 (i.e., ±1% surface reflectance) for BLUE and RED, around −0.02 (−2%) for NIR and SWIR, and around 0.01 for the NDVI. Consistency with VGT-C3 is slightly higher for BLUE (A below 1%), similar for SWIR, and lower for RED and NIR (−2% and −4%, respectively) and the NDVI (0.02).

P (and consequently U) values are relatively high, indicating a high dispersion of product retrievals around their expected values. This is caused by many factors: the intercomparison between different periods (recent products vs. product archive); differences in temporal, spatial and angular sampling; the performance of the spectral resampling algorithm; atmospheric effects and the performance of atmospheric correction; and the performance of pixel classification algorithms. These consistency levels are, however, in line with the initial expectations, based on the simulated Medium Resolution Imaging Spectrometer (MERIS) and Advanced Along-Track Scanning Radiometer (AATSR) data, which state that, depending on the spectral band, an accuracy of 3 to 5% could generally be expected [18].

4. Conclusions

Sentinel-3 SYN VGT products were designed to provide continuity to the SPOT VGT standard products, but since SPOT5 was decommissioned before the launch of the first Sentinel-3 satellite, PROBA-V acted as a gap filler, and thus, consistency with PROBA-V is as (if not more) relevant. In principle, a continuous data series of standard TOA reflectance, daily and 10-day composite TOC reflectance and NDVI products should be available to users; for applications related to long-term, large-scale vegetation monitoring; climate change studies; agricultural monitoring; etc. However, some flaws were identified in the Sentinel-3 SYN VGT processing baselines that were only gradually resolved in the course of recent years. These changes have progressively increased the quality of Sentinel-3 SYN VGT products, as is evidenced through the spatio-temporal intercomparison of the combined time series of SPOT VGT, PROBA-V and Sentinel-3 SYN VGT 10-day NDVI composite products with an external reference from LSA-SAF, and the intercomparison of Sentinel-3 SYN V10 products with the LTS of VGT-C3 resp. PV-C2 for three distinct periods with different levels of product quality.

For the past 12 months regarding SYN V10 products, mean absolute bias values compared to the PV-C2 climatology of ~1% (for BLUE and RED surface reflectance), ~2% (for NIR and SWIR) and 0.01 (for NDVI) are reached. Consistency with VGT-C3 is slightly higher for BLUE, similar for SWIR, and slightly lower for RED and NIR and the NDVI. Overall, a high dispersion of product retrievals around their expected values, i.e., a high standard deviation of the bias, is observed because of the many factors that influence the intercomparison.

The subsequent Sentinel-3 SYN VGT processing baseline updates have thus resulted in better-quality products. Nevertheless, there are a few possible improvements that could still lead to better consistency with the SPOT VGT and PROBA-V product archives: (i) an enhancement in the atmospheric correction scheme with more reliable AOT estimation; (ii) improvements in cloud and snow screening and the inclusion of cloud shadow masking; (iii) a combination of S3A and S3B observations in single composite products to increase daily global coverage and to reduce the effect of residual cloud contamination; and (iv) an upgrade of the spatial resolution of SYN VGT products to 300 m (1°/336) to align with the higher resolution PROBA-V global data products archive.

Finally, from users’ perspective, it is essential to reprocess the entire Sentinel-3 SYN VGT archive using the latest and most advanced processing baseline. The current archive is inconsistent, preventing users from accessing a uniform data set of standard SPOT VGT, PROBA-V and Sentinel-3 SYN VGT products. A consistent data record, spanning over 25 years from 1998 onwards, would provide valuable input for a wide range of applications.