Sentinel-2 Anomaly Database

The SENTINEL-2 on-line anomaly database provides a searchable access to the list of known anomalies on SENTINEL-2 products. This database includes both on-board anomalies (usually irrecoverable) and processing anomalies (potentially recoverable), and covers image data as well as metadata and format anomalies. For each anomaly, a short description and illustrations are provided and a list of datatakes, datastrips or products can be retrieved.

In addition to the search interface, this website provides an API (Application Programming Interface) for machine-to-machine access using scripts.

Besides on-board and processing anomalies, some know issues, called product features, can be present in SENTINEL-2 L1C and L2A products. They are listed below.

Features impacting L1C products

Product Features

​​​​​Spectral response non-uniformity

In this section we report on a known feature of Sentinel 2 products created by the spectral response non-uniformity. This feature has been anticipated since the design phase and is compliant with mission specification.

This feature is characterized by along-track soft-edged darker or brighter stripes near the detector boundaries, as shown on the figure below. Indeed, the spectral response is slightly different at the edges of the detectors, especially for bands B03 and B05. When the spectrum of the scene has strong gradient over the spectral bandwidth of the detector, a difference in the measured radiometry can be observed (up to 2% in worst-cases).

Figure 1: Along-track stripes resulting from spectral response non-uniformity (band B03).

Parallax effects

In this section we report on parallax effects created by the staggered configuration of the focal plane. Indeed, the instrument swath is covered by 12 individual detectors assembled in a staggered manner. Because of this configuration, odd and even detectors do not see the ground under the same viewing angles. This can create visible effects on some images, as detailed in the next subsections.

Surface reflectance effects

Because the viewing angles are not the same for even and odd detectors, differences in measured radiometry can be observed on non-Lambertian surfaces. This is especially visible on Sun glint over sea surfaces (see Figure 2 below).

Figure 2: Stripe pattern over sea surface due to the observation parallax effect between odd and even detectors

Misregistration of high-altitude objects

The processing algorithm ensures the co-registration of images acquired by all spectral bands and the detectors for features at ground level. Objects at a higher altitude like planes and clouds cannot be properly co-registered. This effect leads to spectral misregistration (“rainbow” effect) and discontinuities between detectors.

Both effects can be seen in Figure 3 hereafter.

Figure 3: Spectral misregistration and detector misalignment for object at high altitude (plane and contrail). This feature is not an anomaly

Gradient crosstalk

This feature can be seen on contrasted images on band B12 (typically near the coast). It can be explained by a crosstalk signal coming from the along-track gradient of the B11 image. The typical amplitude of the effect is 10 digital counts.

Figure 4: Gradient crosstalk on band B12 (highly enhanced contrast).

Datastrip overlap

Sentinel-2 products are generated by a network of several ground stations around the globe. Data acquired by the satellites are split into processing units called “datastrips” which are processed independently, and subsequently transferred to the Sentinel Data Hub. A given continuous acquisition sequence (or “datatake”) can be split into several datastrips. In that case, two different products are generated for level 1C tiles located at the interface between the datastrips.

The two products can be merged to reconstruct the full image.

However, one should be aware that the geometric refinement may introduce a small shift (a few meters) between the two tiles. This shift is not visible to the naked eye but can be measured by computing the co-registration between the products on their overlap area. This shift is zero if both products are unrefined, as in the figure below.

Figure 5: Example of a pair of products at the overlap between two data-strips. a: product from the first data-strip, processed at Svalbard (SGS) b: product from the second data-strip, processed at Matera (MTI). c: the two products overlap seamlessly to reconstruct the complete acquisition. d: close-up near the transition line

Valid pixels

Users are advised that the pixel validity status may be different for different spectral bands: it is possible to have one band with valid data and one band with No Data (0) at the same location. This happens in particular at the western and eastern edges of the swath. Any multi-spectral processing should be done only on pixels having valid data for all spectral bands.

Features impacting L2A products

Scene classification

The current scene classification algorithm has some known limitations:

  • Over-detection of clouds over bright targets,
  • Under-detection of semi-transparent clouds or cloud edges,
  • Cloud pixels miss-classified as snow (shaded parts of the clouds),
  • Dark areas miss-classified as cloud shadows. This can occur in particular when bright objects are incorrectly classified as clouds,
  • Topographic shadows may be miss-classified as water,
  • Open fires can be miss-classified as cirrus.
  • Degraded pixels from data loss at L1C are currently not supported by the L2A processor. Users are advised to check the TECQUA mask to identify affected pixels.

These problems have been significantly reduced starting with baseline 02.09.

Starting with baseline 02.10, terrain correction is no longer applied for pixels identified as cloudy. This can lead to visual artefacts at the edges of semi-transparent clouds, see figure below.

Figure 6: Visual artefacts at the edges of semi-transparent clouds

Another known issue concerns the occurrence of blocky patterns on the Scene Classification mask, as illustrated in the figure below. This issue is due to the coarser resolution of the CCI auxiliary data used to improve the scene classification. In some cases (as on Figure 7 – left) it can lead to a local over-detection of clouds.

Figure 7: Blocky patterns on the scene classification layer (SCL). Left: near the coastline. Right: near city boundaries.

Overlap between tiles

The L2A products are processed at tile level and some differences can occur in the overlap area between adjacent tiles:

  • The scene classification may be different for a few pixels
  • The AOD and surface reflectances are generally different, although the difference should be small.

Terrain over-correction on shaded areas

Due to inaccuracies of the Digital Elevation Model, a strong terrain correction may be applied in totally or partially shaded areas. This results in a bluish colour in colour composite and inaccuracy in the surface reflectance.

Figure 8: Terrain over-correction on shaded areas. Left: Level 1C true colour image, right: L2A true colour image.

Maximal Sun-Zenith Angle

Users are advised that products with a Sun-Zenith Angle (SZA) higher than 70° are processed with a clipped SZA value of 70°. This results in an under-correction of the atmospheric signal, which results in a bluish colour on the L2A products. The surface reflectance of products with SZA > 70° should not be used for quantitative/scientific analysis. The value of the SZA can be obtained from the GRANULE metadata (MTD_TL.xml, field Mean_Sun_Angle/ZENITH_ANGLE). In the coming period, a warning message will also be introduced in the GENERAL_QUALITY report to identify these products.

Figure 9: L2A True Colour Image of tile 30VVH. Left: 10/10/2018, SZA = 62°. Right: 24/12/208, SZA = 80°. The radiometric quality for surface reflectance is not ensured for SZA > 70°.

Missing packets

In L2A products generated with previous PB, corrupted pixels affected by missing or degraded instrument source packets are not reported in the Scene Classification Layer. Users are advised to check the TECQUA mask to identify affected areas where the SCL is not reliable.

Figure 10: Left: TCI image of a product affected by missing packets. Right: Scene Classification Layer

Note that missing packets in atmospheric correction input bands (B10 and B09) can affect surface reflectance of other spectral bands.

Since the deployment of the Processing Baseline 05.09 on December 6th, 2022, pixels affected by missing or degraded instrument source packets are identified as defective pixels (SCL=1) in the Scene Classification Layer (see figure below).

Figure 11: Left: TCI image of a product affected by missing packets. Right: Scene Classification Layer, PB = 05.09

Discontinuities visible in Terrain Corrections on very flat areas

This apparent contouring in L2A products arises on images over very flat area with high Sun-Zenith angles. This artifact comes from the impact of the vertical quantization of the Digital Elevation Map (1 meter steps in the current production) on topographic correction.

Figure 12: Contour-like line features (red box – right) visible in L2A products over Antarctica

Artefacts at the edge of the swath due to L2A NoData mask

The nodata mask at L2A is common to all the bands and computed at 20 m resolution, while each band has its own nodata mask at L1C. Indeed, the nodata masks of the 10 m bands are downsampled at 20 m, merged with the nodata masks of all the other bands, and the resulting common mask at 20 m is then oversampled at 10 m during the processing of 10 m bands.

In some rare cases, this process of generating a common nodata mask to all bands from heterogenous spatial resolution, by nearest downsampling to 20 m followed by nearest oversampling to 10 m can induce some artefacts at the edge of the swath. That means that it could happen that some artificially created 10 m bands pixels are filled with interpolated data, resulting in very low reflectance values at the edge of the swath border area.

This behaviour can also be observed in the 10 m bands that have been downsampled at 20 m (but also at 60 m, even if less noticeable), and in the 20 m (60 m) TCI as it is created by a downsampling of the 10 m TCI.

Users are advised to pay particular attention to the swath border area for downstream applications (e.g. mosaicking, temporal synthesis). It is recommended to use the MSK_DETFOO per band (available in L2A Granule QI_DATA folder) to filter out those additional artificial pixels and keep only pixels for which data has been acquired.

Figure 13 : Artefacts observed at the swath border of the S2B_MSIL2A_20220228T102849_N0400_R108_T31UGR_20220228T134712.SAFE tile highlighted in the red rectangles on a True Color Image at 10 m resolution (left) and on the B04 image (right)

Menu Display