Google researchers used ERA to develop a single-pixel, physics-guided neural network that estimates column-averaged CO2 from existing GOES East observations. The approach combines data from 16 wavelength bands with lower-troposphere meteorology, solar angles, and day of the year, then uses sparse observations from OCO-2 and OCO-3 for training.
The work is aimed at a long-standing gap in satellite monitoring of carbon dioxide. Regular CO2 observations began at Hawaii’s Mauna Loa Observatory in the late 1950s and produced the Keeling Curve, but current space-based CO2 sensors only cover a small part of Earth’s surface and typically revisit the same location every 16 days. By contrast, the GOES East satellite can scan an entire hemisphere every 10 minutes, although it was not designed to map CO2.
According to research shared at the International Workshop on Greenhouse Gas Measurements from Space, the model can derive estimates of column-averaged CO2 everywhere and every 10 minutes. The researchers say this gives CO2 measurements unprecedented spatial and temporal resolution by using the high density of GOES East observations.
The results were compared against independent data from additional years of OCO-2 observations and the ground-based total column carbon observing network, which confirmed the model’s ability to capture real CO2 variability.
The project shows how AI can extract more value from existing observational instruments, particularly for resource-intensive satellite research missions. Google researchers said the project is one of several climate and greenhouse gas questions they are exploring using ERA.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
