DeepEmSat: Deep Emulation for Satellite Data Mining.

Area of Science:

• Earth Science
• Climate Science
• Remote Sensing

Background:

• Increasing volumes of Earth science data from climate simulations and satellite remote sensing present computational challenges.
• Atmospheric correction, crucial for retrieving surface reflectance, relies on physics-based models that are computationally intensive and not real-time.
• Machine learning (ML) shows promise for accelerating complex simulations and extracting insights from large datasets.

Purpose of the Study:

• To develop and evaluate DeepEmSat, a deep learning emulator for atmospheric correction.
• To compare the performance of DeepEmSat against traditional physics-based models.
• To support the hypothesis that deep learning can enhance the efficiency of satellite image processing.

Main Methods:

• Development of DeepEmSat, a novel deep learning emulator architecture.
• Implementation of atmospheric correction algorithms within the deep learning framework.
• Comparative analysis of DeepEmSat against established physics-based atmospheric correction models.

Main Results:

• DeepEmSat demonstrates potential for efficient processing of satellite images.
• The deep learning emulator provides a faster alternative to computationally intensive physics-based models.
• Results support the hypothesis of deep learning's contribution to efficient Earth observation data processing.

Conclusions:

• Deep learning, through approaches like DeepEmSat, can significantly improve the efficiency of atmospheric correction.
• This advancement facilitates faster and more accessible analysis of Earth science data from satellite remote sensing.
• DeepEmSat offers a viable solution for real-time or near-real-time processing of satellite imagery.