Methane is invisible, but its impact on the climate is not. Now, Google and NASA’s Jet Propulsion Laboratory, known as JPL, have introduced MAPL-EMIT, a deep learning model that identifies methane emissions on a global scale using images captured from space.
The research was published in PNAS and aims to address one of the major challenges in environmental monitoring: quickly finding methane sources across vast territories, with clouds, vegetation, cities, and other signals that can confuse detection systems.
An AI Trained on Millions of Scenarios
MAPL-EMIT was trained with 3.6 million methane plumes simulated using physical models. A plume is the cloud of gas released into the atmosphere from a source such as a landfill, industrial facility, or energy operation.
This approach combines machine learning with knowledge of how methane behaves in the atmosphere. Why does that matter? Because the model does not rely only on recognizing visual patterns: it also learns to distinguish a real emission within complex and noisy data.
In testing, MAPL-EMIT detected 50% more plumes than human experts and identified more than 23,000 additional emissions worldwide. One result is especially significant: it found 24 of the planet’s 25 largest methane-emitting landfills.
Why Detecting Methane Matters
Methane has a shorter atmospheric lifetime than carbon dioxide, but it causes much more warming during that period. Over a 100-year timeframe, its global warming potential is approximately 30 times greater than that of CO2.
That makes methane leaks a concrete opportunity to reduce global warming. If a source can be located precisely, it is also possible to prioritize inspections, repair equipment, or improve waste management without having to manually review enormous areas.
Detecting an emission does not eliminate the problem, but it shows us where to act first.
From Satellite to a Global Database
MAPL-EMIT uses information from NASA’s EMIT instrument, installed on the International Space Station. Its original purpose is related to studying minerals on Earth’s surface, but its data can also help identify concentrations of gases such as methane.
Google published the global plume database in Google Earth Engine, along with an application for exploring and visualizing the results. This makes it easier for researchers, authorities, and operators to examine the distribution of emissions without having to build the entire analysis infrastructure from scratch.
In addition, the open-source models are available on Kaggle and the inference tools were published on GitHub. In practical terms, inference is the process through which an already-trained model analyzes new data and generates its predictions.
A Technical Model with Practical Impact
The value of MAPL-EMIT is not limited to the precision of its deep learning architecture. It also lies in its ability to turn observations from space into actionable information.
A researcher can use the database to study trends. A public agency can identify facilities that require inspection. A company can locate leaks and prioritize maintenance. And a climate team can assess whether the measures taken are reducing emissions.
In this case, AI does not replace specialists in the field. It helps them focus their work more effectively, especially when the volume of images exceeds what one person can reasonably review.
A Tool for Acting Faster
Global methane detection still depends on data quality, observation frequency, and ground-based verification. An algorithm can flag a plume, but additional analysis is usually needed to confirm its source, measure its magnitude, and decide how to respond.
Even so, MAPL-EMIT shows how AI models can connect planetary observation with concrete environmental decisions. The change is not simply that we are looking at Earth from space, something we already do, but that we can process this information quickly and at enough scale to find emissions that might previously have gone unnoticed.