Compressive Sensing and Deep Learning framework for Multiple Satellite Sensor Data Fusion

Abstract:

We propose Compressive Sensing and Deep Learning framework (CS-DL) for multiple satellite sensor based data fusion. It’s aims to improve spatial and temporal resolution for long term analysis. Compressive Sensing is used as an initial guess to combine data from multiple sources. Deep Learning model, using Long Short Term Memory Neural Network (LSTM/RNN) refines and further improves the resulting data fusion output from Compressive Sensing. Our CS-DL framework has been tested to fuse CO2 from the NASA Orbiting Carbon Observatory-2 (OCO-2) and the JAXA Greenhouse gases from Orbiting Satellites (GOSAT). It achieves lower errors and high correlation compared with the original data. This work demonstrates the use of CS-DL for fusing CO2 from NASA Orbiting Carbon Observatory-3 and GOSAT-2 at higher resolution.