Deep learning-based spatio-temporal estimate of greenhouse gas emissions using satellite data

Summary

Sandia National Laboratories is developing deep learning approaches to estimate anthropogenic greenhouse gas (GHG) emissions using satellite data, with NO2 concentration as a case study. The project, LDRD Project Number 227215, leverages data from Sentinel-2 (10-60m resolution, 12 channels) and Sentinel-5P (3-5km resolution, 1 channel) for the period 2018-2020. A Deep Neural Operator (DNO) architecture, Model 3, demonstrated superior performance, achieving an R2 of 0.64 and a Mean Absolute Error (MAE) of 5.45 µg/m³ for monthly NO2 predictions, significantly outperforming prior work (R2 = 0.53, MAE = 6.31 µg/m³). This research lays new groundwork for tracking climate change causes and developing mitigation strategies by enhancing the capability to estimate GHG concentrations using remote sensing data, particularly by efficiently integrating multiscale and multimodal datasets without requiring zero-padding for missing data points.

Key Facts

Source Document

https://example-government.gov/policy-document-link