Catching the satellite errors that human operators miss.

NOAA's GOES and VIIRS feeds generate terabytes of imagery per day. A specific class of degradation was slipping past the legacy QC pipeline. We built a model that catches it in under a second.

Client
NOAA
Category
Federal · Earth observation
Accuracy rate
99.2%
Detection time
<1s

NOAA's GOES and VIIRS feeds generate terabytes of imagery per day. A specific class of degradation was slipping past the legacy QC pipeline. We built a model that catches it in under a second.

01The problem, stated plainly

A small but operationally significant fraction of GOES-R and VIIRS frames contained artifacts that made downstream products unreliable — but passed every existing QC check.

02Why not just tune the rules

The artifacts we were chasing were structural, not signal-level. A single pixel looked fine. You had to see a whole band, laid out spatially, to recognize the artifact. That is what a CNN is good at.

03What changed

Analyst retrospective review hours dropped to about a fifth of their previous volume. One of the artifact classes we flagged was traced to a specific demodulator firmware issue.

Measured outcomes

99.2%
Accuracy rate
<1s
Detection time
2
Satellite fleets
Work like this

Think we could help you ship the thing that's been stuck?

Our engagements typically start with a two-week fixed-fee diagnostic. Fast, narrow, and ends with a plan you can execute with or without us.