I’m tryingt to create a time series of vegetation growth from one farm site, cultivated with winter wheat. As proxy for the plant growth I choose the NDVI (certainly there exist other vegetation indices, but just as a first apporach). As data I dowloaded L2A-data from the copernicus.hub during one vegetation period. As I choose L2A- and not L1C-data I assumed that no further correction would be needed and started working with the data in QGIS.I loaded the red- and the near infrared-band (B04 and B07) of the different days in QGIS, calculated NDVI-values and extracted the mean NDVI-values of each pixel in a shapefile-grid (10 mx10m, fitted to Sentinel-grid).
Now, looking at my time series (one for each pixel in the area), the graphs aren’t very stable, with some unrealistic large decreases and increases just in a few days. I already checked for cloud coverage. Is there still an atmospheric influence or an sun angle influence in the L2A-data causing a date-specific shift?
Additionaly it seems as the satellite data are kind of smoothed - as when I compare it with ground measurements, we meet perhaps the crop growth average at the different days, but the degree by which we can describe the real crop variation differs strongly between the dates. Might this be an atmospheric effect too or is this a problem of the resolution? Do you know how the values of each pixel are determined? Is it the average value of everything in the pixel-area or is it a smoothed values, based on more than one pixel?
No atmospheric correction is needed for L2A- products because it’s already BOA, but you could apply sen2three , Spatio-Temporal Synthesis of bottom of atmosphere corrected Sentinel-2 level 2a images, as they are generated by the Sen2Cor application.
The NDVI equation is (b8-b4)/(b8+b4) , (NIR-R)/(NIR+R), so in your case you used band 0 7, and this might be the first unaccepted values of NDVI!
In this case you could download Level-1C and apply sen2cor, to be able comparing the two results, already L2A and the one you corrected by applying sen2cor,
I think the following post could answer this demand,
Your comment on how to calculate NDVI was of course absolutely right, I just mixed up 7 and 8.
Concerning the sen2thress-application, I am not completly sure if this is the right thing to solve my problem as they describe it as: “Sen2Three takes time series of level 2a images of certain geographical areas (tiles) as input and generates a synthetic output image by replacing step by step all “bad” pixels of previous input images with the collocated “good” pixels of scenes following in time.”
If I understand it in the right way, they are mixing pictures of different dates? In this case we would loose the oppotunity to look at the Time Series, wouldn’t we?
I think, your proposal of comparing L1C with L2A-data might help to search for the effects of the atmospheric correction via the Sen2Processor. But my question is rather if this data processing really corrects every problem with atmosphere and sun angle or if further steps are required?
Finally, your last Link wasn’t really helpfull as I work with Sentinel- and not with Landsat-data and the problem are not to high but not stable or/and not sentitive enough data.
I don’t think so, since the goal of the sen2three is more AC, for the BOA,
of course there are many equations of AC, but these are used, experienced and accredited in SNAP,
I gave this answer, because I got two concepts of your previous question.
I’d like to comeback to your previous comment,
I think this point is absolutely true, how did you take your ground measurement, what is the angle, time of the day, the altitude of your radiometer, the area covers by single measurement? so all of these questions could normalized the results of the S-2.
Thank you for your reponse. Actually, I was trying to do a sort of bottom-up-approach: In the recent years, me and my collegues established by destructive sampling a very stable relationship between plant growth and measurements of one droe-based multispectral camera. Now I had the idea that I could establish a connection between the already evaluated dronal maps and the satellite data (might later save time or close data gaps) but this seems to be a very difficult problem: besides a relatively low R² at the single dates, the correlation between satellite-data and drone-based maps seems sometimes to shift between different dates. Hence, I just got to the question how reliable the satellite data are - especially concerning atmospheric corrections, as this would explain both, the low explanatory value at some dates and the shift of the correlation with the drone data!