DEEP NEURAL NETWORK FRAMEWORK FOR INVERTING REMOTELY SENSED CO2 MEASUREMENTS

We propose a deep learning framework for the inversion of CO2 concentrationmeasurements from satellites to estimate the CO2 emissions. Our algorithm startswith informed guess of emission distributions of CO2 and keeps on correcting ittill it is consistent with outcome of transportation model and CO2 measurementsby satellite. We found that our inversion method is capable of identifying emissionsources of CO2 that are not considered in the prior.