Detecting methane leaks from free satellite imagery not built for it — 5 of 7 SBIR Phase II awards are exactly $1.25M (GeoFinancial Analytics)
A project letting oil and gas producers detect emissions early by combining sub-weekly high-resolution AI satellite scans with continuous onsite monitoring. The central technical risk is a deep learning model detecting and quantifying asset-level emissions in freely available 30-metre satellite imagery that was never designed to detect them.
Grant overview (primary data)
- Award amount$1,250,000
- RecipientGEOFINANCIAL ANALYTICS, INC. (California)
- ProgramSBIR Phase II
- Period2026-08-01 〜 2028-07-31
- FunderU.S. National Science Foundation (NSF) / NSF
Key points
- A deep learning model automatically detecting and quantifying asset-level emissions in freely available 30-metre satellite imagery never designed for the purpose.
- The theoretical detection threshold under traditional methods is 200 to 500 kg per hour under ideal conditions; the project aims below 100 to 150 kg per hour with proprietary training data.
- Sub-weekly satellite scans are combined with continuous 24/7 ground sensor observation at a subset of sites.
- Cited effects include early leak detection and credible certification data enabling premium pricing for produced natural gas.
- Of the 120 NSF awards this site holds as of 2026-08-31, 7 are SBIR Phase II and 5 are exactly $1,250,000, the other two being $1,207,205 and $1,222,567.
- Using the same Landsat and Sentinel-2 imagery, images from the time of known emissions become training data that lowers the detection floor.
1Reading data outside its purpose
At the core of this project sit Landsat and Sentinel-2, satellite imagery freely published. Both were built to observe the land surface, not to measure methane emissions. At 30-metre resolution, traditional handling cannot detect emissions smaller than 200 to 500 kg per hour even under ideal conditions.
The premise is that feeding the model training imagery coincident with known emissions can push that below 100 to 150 kg per hour. Rather than launching a dedicated instrument, capability is sought by changing how existing data is read.
2Detection turning into price
Among the effects the record cites, alongside early discovery of leaks, is generating credible certification data enabling premium pricing for produced natural gas. Demonstrating that emissions were reduced makes the gas worth more. Environmental purpose and commercial interest point the same way, and a reason to keep monitoring arises outside the cost. That route differs from mandating it by regulation.
35 of 7 SBIR Phase II awards are exactly $1.25 million
Of the 120 NSF awards this site holds as of 2026-08-31, 7 belong to SBIR Phase II. Five of them are for exactly $1,250,000, with the remaining two at $1,207,205 and $1,222,567. A ceiling is evidently set and most receive it in full — in contrast with investigator-led awards whose amounts differ case by case, since a small-business program fixes the envelope first. The 120 awards span 67 programs, so how amounts are determined differs by framework.
4Raising sensitivity without launching anything
What this venture uses is free satellite imagery that was never built to observe methane. Rather than improving the instrument, it sets out to gain sensitivity by changing how the images are read.
Launching a dedicated observing satellite raises sensitivity, at a cost in money and years. Changing how already-public data is read is a route open to those without an instrument of their own.
Why it matters
It is an example of new capability arising from reading existing public data for a purpose other than its own — cheaper than commissioning dedicated sensors or satellites, and more frequently updated. When the motivation for environmental monitoring ties to price rather than regulation alone, a reason to continue arises on the business side.
FAQ
Why use free satellite imagery?
How small a leak can be detected?
Why does premium pricing come up?
Sources (primary)
Source: NSF Award Search (U.S. National Science Foundation, public domain). Amounts are the obligated amount. For privacy, we do not handle principal investigator names.
- NSF Award (original, official)
- NSF Award ID: 2537735