NDVI anomaly using atmospheric correction

Hello ,

We are computing NDVI anomaly using the script here : https://raw.githubusercontent.com/sentinel-hub/custom-scripts/master/sentinel-2/ndvi_anomaly_detection/ndvi_anomaly_detection_script.js

We tried to use Sentinel 2 L2A but this do not have enough historical data to compute a proper NDVI anomaly.

So we used atmospheric correction (ATMCOR) over Sentinel 2 L1C.

But this give completely different result from without atmospheric correction.
Here is an example without :

And with correction :

Here is the link for the last image :

Do we use atmospheric correction correctly ?

Thank you,

As described here, the ATMCOR is not a “full scale” atmospheric correction but rather a statistical approach, which works OK in some cases and not that OK in other cases. We would not recommend it to be used for such analysis.