How do you account for cloud contamination and atmospheric correction in Sentinel-2 imagery?

[Tanay-s5]

As Sentinel-2 imagery undergoes processing to remove clouds and haze, it can sometimes distort the true ground reality. If multiple farmers rely on this data, even small inaccuracies could lead to serious losses. To combat this risk, what systems are you planning to ensure reliable and trustworthy analytics?

[Thanas-R]

That is definitely something I have been thinking about seriously while building Virdis.

Right now, the approach is to avoid treating a single satellite snapshot as absolute ground truth. Instead, the system tries to improve reliability through multiple safeguards such as cloud filtering, temporal analysis across multiple dates, vegetation trend comparison, and combining satellite insights with other environmental datasets like weather, soil, rainfall, and land use data.

I also intentionally position the analytics as decision support rather than guaranteed predictions. Agriculture is influenced by many dynamic real world factors, so the goal is to provide informed insights while still being transparent about uncertainty and limitations.

Going forward, I also want to explore confidence scoring, anomaly detection, and eventually integrating additional satellite sources or historical validation pipelines to improve consistency further.