This post in the third and likely last part of a report on the recently released Sentinel-3 level 1 data from the OLCI and SLSTR instruments. The first part provided general context, the second part discussed the data distribution form and how this data can be processed and this third part will look more in depth into the quality of the data itself by comparing it to other open data satellite imagery available. Due to the problems with the SLSTR geolocation information explained in the second part this will only cover the OLCI instrument.
On data quality – OLCI
It is generally difficult to judge radiometric accuracy without reliable reference data. Noise characteristics of the visible bands appear to be pretty good – since recordings are limited to relatively high sun angles there is not really a good possibility to properly assess the dynamic range of course.
We remember the spectral bands chart from the first part – in the OLCI instrument there are essentially 10 spectral bands in the visible range, all of them very narrow. The usual convention will likely be to interpret band 3, 6 and 8 as blue, green and red. Bands 1 and 2 are near the blue edge of the visible range, band 7 is also well in the red range and can be used as a red band as well or in combination with band 8. Band 9 and 10 are near the long wavelength end of the visible range. Finally band 4 and 5 are in the blue-green area. The most problematic aspect for visualization purposes is the green band and its location right where red and green sensitivities of the human eye are about the same. Overall such narrow spectral sensitivities are not ideal for accurate color representation but it could be worse.
Resolution wise OLCI offers a 300m resolution at nadir in all bands. For comparison: MODIS and VIIRS provide different spatial resolutions in the red band and the green and blue bands. For MODIS that is 250m for red and 500m for grenn/blue. For VIIRS we get 375m for red and 750m for green/blue. Both also provide the higher resolution for an NIR band (and with VIIRS also for SWIR and thermal bands) so many analytic applications can also make use of higher resolution data.
The nominal resolutions tell only half of the story though. The wider the field of view of the satellite the stronger the dropoff in resolution is towards the edges. This means VIIRS has an additional disadvantage relative to MODIS if you consider the whole image average. The asymmetric view of Sentinel-3 OLCI means despite the more narrow recording swath the maximum viewing angle and therefore the resolution drop-off are comparable to that of MODIS.
Comparing with other satellites
So far this is all theory so here are some comparisons of images from the major open data satellite systems currently operating that provide visible light color images. It is not easy to find image sets where all of these record the same area on the same day, preferably all in the middle of the recording swath. I picked two areas and compromised by using MODIS images from one day earlier than the rest of the images which are all from the same day. The two areas are Lake Tchad and Northern Patagonia:
Images are based on calibrated radiances for all satellites. Normally i would have made this comparison based on reflectance values but apparently the necessary data to calculate reflectance values is currently not available for OLCI. This means differences in recording time and therefore illumination is not compensated for so the images differ also because of that. For reference here the recording times (in UTC) for all images shown.
|Sensor||Time Tchad||Time Patagonia|
As you can see Sentinel-3 is generally the earliest due to the westward view while VIIRS with a noon viewing window is much later than the rest. Here is how the recording footprints look like, you can well see the different image sizes:
And here small area crops for all of these for comparison:
Keep in mind both the differences in recording time and the different date for the MODIS image affect the results. What can still be observed however is:
- that MODIS and Landsat 8 are fairly close in color calibration.
- Sentinel-2 seems off relative to that – which i reported earlier. No sure way to say one of them is correct and the other is wrong though.
- Sentinel-3 OLCI seems somewhere in between those in terms of base brightness but you need to take into account the earlier timeframe here. It does not show the same tint towards blue as the Sentinel-2 data despite the fact that the shortwave blue OLCI band 3 will likely feature significantly stronger atmosphere influence. To me this kind of indicates Sentinel-2 is the outlier here while the rest of the crowd is relatively well synchronized.
- VIIRS is hard to compare due to the much later recording time but is probably also calibrated in combination with MODIS and Landsat by the USGS.
- positional accuracy is generally not that great for the lower resolution images compared to the high resolution (Sentinel-2, Landsat) which are both very close so form a suitable reference.
- resolution of Sentinel-3 OLCI and MODIS is fairly close. This is difficult to compare both due to the variability across the field of view and since MODIS provides higher resolution in the red channel than in the others – there are different possible approaches to use this to produce a high resolution full color image. Which is better also depends on how homogeneous colors are in the location you are looking at. Overall i think OLCI will usually have a slight advantage here overall for color images although for analytics involving mainly red and NIR like NDVI calculation MODIS probably offers measurably higher resolution. In any case the difference is fairly small compared to the difference between MODIS and VIIRS – which is not extremely big either near nadir but overall more significant due to the resolution falloff.
- in the Patagonia sample area you can also see the advantage of the tilted view of Sentinel-3. MODIS, Sentinel-2 and Landsat all show sun glint on the lakes while OLCI does not. VIIRS also lacks sun glint in this area since the area is significantly off-center in the image to the east while the sun is slightly to the west.
Note for VIIRS and OLCI i mixed spectral bands for better approximation of visual colors.
So far we only have the level 1 data. There are plans for higher level products to be made available in 2017 but these are fairly unspecific regarding the underlying methods. It is for example unclear if there will be a surface reflectance product of competitive quality. If we take MODIS as reference here – higher level MODIS products usually come with significant issues and limitations but their easy, uncomplicated and reliable availability makes them an attractive option if you are able to deal with these issues.