Superpower Daily
AIR Raises $50M for Continuous Checks on AI-Agent Add-OnsMeta Introduces Muse Voice Transcribe for Live Speaker Labels i...Anthropic’s EFS Puts Misuse-Monitoring Data in Customer Clouds,...World Labs Launches Atlas for 1440p Camera-Controlled Video, 3D...HUMAIN Picks Compal for Cloud-to-Edge AI Build, With Nvidia and...t54 Puts a Hard Trust Gate on Bedrock Agent Payments, Citing 20...AIR Raises $50M for Continuous Checks on AI-Agent Add-OnsMeta Introduces Muse Voice Transcribe for Live Speaker Labels i...Anthropic’s EFS Puts Misuse-Monitoring Data in Customer Clouds,...World Labs Launches Atlas for 1440p Camera-Controlled Video, 3D...HUMAIN Picks Compal for Cloud-to-Edge AI Build, With Nvidia and...t54 Puts a Hard Trust Gate on Bedrock Agent Payments, Citing 20...
Modelspublished

Google Releases MAPL-EMIT to Map Facility-Scale Methane Plumes From Space

The database, trained model and inference tools give researchers and local monitors new ways to process EMIT scenes. Greater sensitivity still leaves a practical trade-off between catching weak plumes and filtering false alerts.

By 2 min read
Google Releases MAPL-EMIT to Map Facility-Scale Methane Plumes From Space
Google Releases MAPL-EMIT to Map Facility-Scale Methane Plumes From Space
Sign in to save and react

Listen to this story

The audio brief

About 1:27
0:001:27
Read transcript
Google has released MAPL-EMIT, a system that turns satellite measurements into facility-level methane monitoring. Built with NASA JPL, it can estimate how much methane is in each plume pixel, outline overlapping plumes, and trace the gas back toward likely sources in a single pass. The underlying EMIT instrument sees hundreds of separate light bands in every pixel, allowing it to detect methane’s chemical signature. Its view spans 80 kilometers, with 60-meter resolution—detailed enough to focus on places like landfills and industrial sites, rather than only producing broad global maps. The challenge is that some terrain and surface materials can look like methane in hyperspectral data. MAPL-EMIT addresses that with a Swin-S vision transformer, which considers both the spectrum and the surrounding scene. Google says it captured 84 percent of expert-annotated plumes and found roughly 50 percent more plausible detections than its baseline. It also mapped plumes at 24 of the world’s 25 highest-emitting landfills. Since millions of labeled real-world emissions do not exist, the team trained on 3.6 million physics-based synthetic plumes inserted into real EMIT scenes. Lagrangian puff models simulated how those emissions disperse. The key constraint is false positives, particularly in complex terrain. Confidence scores let operators choose between catching weaker plumes and filtering questionable alerts. That trade-off remains the central test as researchers and local monitors put the released tools to work.

Story brief

3 key points

The release turns EMIT’s hyperspectral satellite data into an open facility-monitoring workflow: MAPL-EMIT can estimate plume strength, outline overlapping plumes, and infer source locations in one pass. Google reports that it captured 84% of expert-annotated plumes, found 50% more plausible detections than a baseline, and mapped plumes at 24 of the 25 highest-emitting landfills. Training used 3.6 million synthetic...

  1. 01

    EMIT’s 80-kilometer field of view and 60-meter resolution target facility-scale monitoring rather than coarse global mapping.

  2. 02

    Google released a plume database, trained model, synthetic dataset, and inference library for emissions-monitoring work.

  3. 03

    The Swin-S vision transformer combines spectral data with surrounding-scene context to distinguish plumes from methane-like terrain.

Google has released MAPL-EMIT, a deep-learning system developed with NASA JPL that detects methane plumes in NASA EMIT satellite measurements, estimates methane enhancements and identifies likely emission-source locations. Google also published a global plume database, trained model, synthetic dataset and inference library for emissions-monitoring work.

From a noisy spectrum to a source on the ground

EMIT records hundreds of separate light bands for every pixel, which can capture methane’s chemical signature. Its 80-kilometer field of view and 60-meter spatial resolution are designed for facility-scale measurements, rather than the coarse coverage used by broader global mappers.

Detail alone does not solve the problem. Some landscapes and surface materials resemble methane in hyperspectral data, complicating smaller or diffuse detections. MAPL-EMIT uses a Swin-S vision transformer that reads the spectrum and the surrounding scene together. Google says that context helps the system tell a wind-blown plume from similar-looking terrain.

Three jobs in one model pass

The framework is built to measure methane enhancement in each plume pixel, draw plume boundaries and trace dispersed gas back to a source. That combination is intended to help where emissions from nearby facilities overlap downwind and appear as one cloud.

Synthetic plumes supply the missing labels

A global labeled dataset of millions of real methane emissions does not exist, so the team generated 3.6 million physics-based synthetic plumes and placed them into real EMIT scenes. Lagrangian puff models simulated how particles move and disperse in air. Because the plumes were generated, the training data included source locations and boundaries for overlapping plumes.

Sensitivity does not remove judgment

The reported results suggest the system can surface more candidate plumes, but Google says false positives remain an ongoing problem, especially in complex terrain. Outputs include physics-based confidence information and lower- or higher-confidence tags, so users can filter detections according to their tolerance for missed plumes and questionable alerts.

Sources

  1. research.googleMapping global methane emissions from space with deep learning

Related coverage

Continue reading