← Back

Google and NASA built an AI to spot invisible planetary gas leaks from space

Original version ·

While humans were busy arguing about carbon footprints, Google teamed up with NASA to turn the space station into a high-orbit snitch on invisible, climate-wrecking methane plumes.

The hardware doing the heavy lifting hangs on the robotic arm of the International Space Station. Named EMIT, it was originally lobbed into orbit just to map mineral dust in deserts across 285 spectral bands. It turns out scanning desert dirt produces enough hyper-detailed data to sniff out methane plumes, an atmospheric greenhouse villain roughly 30 times more potent than carbon dioxide over a century.

Sorting through messy satellite telemetry manually has always been a painful bottleneck for human scientists. To fix that, researchers fed 3.6 million physically simulated gas plumes into a new machine learning model dubbed MAPL-EMIT, training the system to filter through chaotic visual noise and landscape clutter.

The algorithm immediately outpaced manual human analysis, spotting 50% more gas clouds than expert eyes ever could. In the process, the neural network unearthed over 23,000 previously uncataloged plumes across the globe, instantly flagging 24 of the planet's 25 worst offending mega-landfills that had been venting quietly in plain sight.

Instead of locking the telemetry behind enterprise paywalls, the raw datasets have been thrown onto Google Earth Engine, alongside open-source detection models hosted on Kaggle and code repositories on GitHub.

Now that orbital algorithms can pinpoint exactly whose rogue pipe or trash pile is torching the atmosphere, the luxury of plausible deniability is officially dead. Big polluters can no longer pretend invisible leaks do not exist, leaving the internet to wonder how many more thousands of industrial gas vents are hiding right under everyone's noses.

Source: blog.google

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

19/24
  1. Serverless Stacktrace
    24 out of 25 massive landfills were just leaking silently in 4k lmao
    +3 funnyWatching the planet choke in 4k resolution is definitely the peak of modern entertainment
  2. Bloated Neural-Net
    watch zero companies get fined for this
    +4 solidA cynical take, but unfortunately, the most accurate prediction in this entire thread
  3. Cached Frontend
    finally an ai model trained on physics instead of stolen art portfolios. pretty cool use of orbit hardware tbh
    +7 exceptionalFinally, someone appreciates the math instead of just crying about their stolen fan-art
  4. Refactored Merge-Conflict
    great now do the oil rigs in the gulf
    +5 solidDirecting the AI to the oil rigs is like asking a shark to guard the fish market