DeepEmSat: Deep Emulation for Satellite Data Mining.

Kate Duffy,Thomas Vandal,Sangram Ganguly,Ramakrishna Nemani,Shuang Li,Auroop R Ganguly

doi:10.3389/fdata.2019.00042

Abstract

The growing volume of Earth science data available from climate simulations and satellite remote sensing offers unprecedented opportunity for scientific insight, while also presenting computational challenges. One potential area of impact is atmospheric correction, where physics-based numerical models retrieve surface reflectance information from top of atmosphere observations, and are too computationally intensive to be run in real time. Machine learning methods have demonstrated potential as fast statistical models for expensive simulations and for extracting credible insights from complex datasets. Here, we develop DeepEmSat: a deep learning emulator approach for atmospheric correction, and offer comparison against physics-based models to support the hypothesis that deep learning can make a contribution to the efficient processing of satellite images.

Highlights

Contemporary satellite remote sensing is responsible for contributing Earth science data to public repositories at an unprecedented volume (Overpeck et al, 2011)
We develop an experiment in atmospheric correction and present results to suggest that a deep learning model can be trained to emulate a complex physical process
Non-learning approaches to atmospheric correction (AC) use physical modeling and empirical relationships to retrieve surface reflectance from observations contaminated by atmospheric scattering and absorption processes that occur in the paths between the sun, the Earth’s surface, and the satellite sensor

Summary

Introduction

Contemporary satellite remote sensing is responsible for contributing Earth science data to public repositories at an unprecedented volume (Overpeck et al, 2011) This abundant data has drawn interest to applying machine learning (ML) for data mining (Castelluccio et al, 2015; Xie et al, 2016; Mou et al, 2017), climate data downscaling (Vandal et al, 2017), and to advance process understanding in Earth sciences (Reichstein et al, 2019). SR is a characteristic of the Earth’s surface and is produced from raw, top of atmosphere (TOA) observations by removing the effects of atmospheric scattering and absorption This process, termed atmospheric correction (AC) allows greater comparability between observations across space and time. Atmospheric correction models must be tuned for new sensors, which may have short operational lifespans

Objectives

Methods

Results

Conclusion