Explainer • Field Change Alerts
How Farmbit Decides When to Alert You
Farmbit keeps alerts conservative. It throws out weak scenes, waits for enough history, and only surfaces the changes that are strong enough to deserve a real walk.
A good alert should answer one question: is this worth a field check today?
Farmbit stays conservative on purpose. Most satellite scenes never become alerts.
Cloud, thin history, too little affected area, or a short wobble after irrigation can all make a block look noisy for a day. The system is built to throw those scenes out before they create work for the grower.
What Farmbit throws out first
Farmbit starts from regular Sentinel-2 scenes clipped to the field boundary. Behind the scenes it cleans the scene, checks how much of the field is actually usable, and compares the current block against recent history.
Most scenes stop there. That is intentional.
When an alert is worth trusting
Four conditions matter more than the rest.
- Enough clean coverage: At least 70% of the field needs to be usable.
- Enough recent history: The system needs enough earlier observations to know whether this is drift or a real change.
- Enough affected area: A tiny patch should not trigger a field-wide alert.
- Enough strength or persistence: The drop has to either stick around or be big enough to justify immediate attention.
If one of those checks fails, Farmbit records a no-alert reason and stays quiet.
What turns a change into an alert
Medium alerts are for changes that persist. High alerts are for bigger drops across more of the block, with cleaner coverage, where waiting for another pass would slow the response more than it would help.
That is why the system sends fewer alerts than a loose monitoring feed. It is trying to filter out the scenes that create work without helping the decision.
What the grower actually sees
When Farmbit does send an alert, it includes the alert type, severity, confidence, affected area, usable coverage, persistence count, and the baseline window behind the call.
It also makes one thing clear: Farmbit is not naming the cause. It is saying this block deserves a closer look now.
What happens after that
Alerts go out by email or push, based on your settings. They also show up in the dashboard, where the grower can record what happened next: looked normal, irrigation issue, wetness, pest-like signs, and so on.
That feedback does not replace the field check. It helps Farmbit learn which alerts were worth surfacing and which ones were just noise.
Why the system stays conservative
The thresholds are kept conservative and fail safe. If internal settings are wrong, Farmbit falls back to safer defaults instead of sending noise.
That tradeoff is deliberate. Missing a noisy alert is cheaper than training people to ignore the real ones.