MXPT: Masked XCO2 Pre-training Transformer

Implementation code for the paper: "Leveraging Wide Snapshot XCO2 Pre-training to Estimate Urban Fossil Fuel CO2 Emissions from Space", published in Remote Sensing of Environment.

Overview

The core of the approach involves a two-stage process:

1. Pre-training: A Masked XCO2 Pre-training (MXPT) strategy to reconstruct XCO2 observations in snapshots.
2. Downstream Task: Plume intensity regression to detect urban emission signals.

License

This project is licensed under the MIT License.