Great Work!

[JohnMBrandt]

Hi -- I am an author of one the papers you cited in your study (https://www.sciencedirect.com/science/article/pii/S0034425723001256) which uses multi temporal cloud masking + histogram equalization to construct analysis ready data. Your approach looks very promising! I am going to try it out next week and I'll let you know if it improves our ARD :)

[JohnMBrandt]

Following up on this -- I compared Vint2 to the cloud removal algorithm that we use, which is essentially:

• Identify cloudy areas
• Make a cloud-free composite by histogram equalizing and gapfilling
• Use the cloud-free composite to train a per-image gradient boosting machine to predict values for cloudy pixels

VPint2 looks really good in many areas that I tested that have low band standard deviations (e.g. dense forests), but in areas with high non-cloud values, like on farmland, the reconstruction from VPint2 seems to have really low contrast, e.g. here in Ghana:

Any ideas?