A new deep learning model maps global methane emissions.

A new deep learning model maps global methane emissions from space.

In a new study published in PNAS, Google and NASA’s Jet Propulsion Laboratory (JPL) introduced MAPL-EMIT, an AI model that tracks methane emissions globally from space using NASA’s EMIT instrument.

Methane is a potent greenhouse gas. Over a 100-year timeframe, its warming potential is 30 times greater than that of carbon dioxide. MAPL-EMIT tackles a critical bottleneck in methane detection. Trained on 3.6 million physics-simulated methane plumes (clouds of methane gas released into the atmosphere), it cuts through complex, noisy terrain to detect 50% more plumes than human experts and identifies more than 23,000 additional plumes globally, including 24 out of 25 of the world’s largest-emitting landfills. By making methane sources easier to find at scale, MAPL-EMIT enables faster, more targeted climate mitigation.

Google has released the global plume database on Earth Engine alongside an Earth Engine app to visualize the data. Open-source models are available on Kaggle and inference tools are on GitHub to support researchers, policymakers, and operators. Read more on the Google Research blog.

A six-panel satellite graphic showing detected methane plumes and emission heatmaps in California (USA), Turkmenistan, Delhi (India), Sao Paulo (Brazil), Katowice (Poland) and Shanxi (China).

Google Research and NASA’s Jet Propulsion Laboratory (JPL) have developed a groundbreaking deep learning model called MAPL-EMIT (Methane Analysis and Plume Localization with EMIT) to track and quantify global methane emissions from space. Published in the Proceedings of the National Academy of Sciences (PNAS) in September 2026, the model significantly advances climate mitigation efforts by making methane point sources much easier to locate at scale. [1, 2, 3, 4]

How MAPL-EMIT Works

Data Source: The model processes hyperspectral radiance data collected by NASA’s Earth Surface Mineral Dust Source Investigation (EMIT) instrument, an advanced imaging spectrometer mounted on the exterior of the International Space Station (ISS). It resolves imagery at 60 meters per pixel. [2, 3, 5, 6]
Architecture: Built as an end-to-end Swin vision transformer, the model analyzes full radiance spectra across 285 bands, combining spectral signatures with spatial context to separate actual wind-blown methane gas from ground features. [4, 7, 8]
Simultaneous Tasks: The framework performs three critical functions at the same time: quantifying gas enhancements, delineating plume boundaries, and localizing the exact source. [2, 7, 8]
Synthetic Training: To overcome the lack of real-world labeled datasets, the team used physics-based Lagrangian puff models to simulate 3.6 million synthetic methane plumes, which they injected into real EMIT background scenes to robustly train the AI. [2, 9]

Performance and Environmental Impact

Metric / Capability	Performance Details
Detection Increase	Detects ~50% more plumes than human analysts and prior NASA tools.
Recall Rate	Captures 84% of expert-annotated plumes.
New Sources Found	Identified over 23,000 additional global plumes previously unmapped.
Landfill Detection	Successfully mapped emissions at 24 of the 25 largest-emitting landfills globally.
Weak Plumes	Significantly lowers detection thresholds for weaker plumes (250–500 kg/h).

Open-Source Accessibility

To accelerate global climate action, the creators have made the technologies fully public:

The global plume database and visualization tool are hosted on Google Earth Engine.
The trained model architecture is open-sourced on Kaggle.
Inference tools and source code are available on GitHub. [1, 5, 10]