Satellite remote sensing has revolutionized Earth observing capabilities and plays a major role in the studies of atmosphere, ocean and land systems. Radiative transfer simulations are foundational for satellite remote sensing algorithms, as they connect the physical properties of the atmosphere and the underlying surface, to the satellite measurements of radiation at the top of the atmosphere. Remote sensing techniques have advanced rapidly to provide highly accurate geophysical property retrievals by utilizing the rich information content of observations at multiple spectral bands, viewing angles and polarization states. However, that high dimensionality and accuracy in observations introduces new challenges in terms of the computational efficiency of radiative transfer simulations and inversions. Conventional approaches, such as the use of precomputed lookup tables, quickly become unmanageable when the retrieval of many parameters is desired. Machine learning methods, such as multi-layer perceptron (MLP), convolutional neural networks (CNN), or Gaussian Processes (GPs), etc., can help to resolve this issue, for example, by creating radiative transfer emulators or conducting direct inversion. Several desirable characteristics of deep learning methods include:
• Computational efficiency, due to the use of matrix operations that can be quickly evaluated;
• High accuracy, given sufficient training data volume;
• Minimal file and memory usage compared to lookup tables;
• Flexibility, since the emulator or inversion model can be incorporated into existing remote sensing algorithms; and
• Differentiability, as the Jacobian matrix of NN models can be represented analytically through automatic differentiation.
This Research Topic aims to showcase the advancement of machine learning research and techniques, designed for radiative transfer modeling and remote sensing inversion. The goal is to improve the analysis of observations from existing and future satellite missions such as NASA’s Plankton, Aerosol, Cloud, ocean Ecosystem (PACE), Multi-Angle Imager for Aerosols (MAIA), Atmosphere Observing System (AOS), and ESA’s Tropospheric Monitoring Instrument (TROPOMI), Multi-view Multi-channel Multi-polarization Imager (3MI) and Carbon Dioxide Monitoring (CO2M) missions, whose large data volume poses challenges to current operational retrieval techniques.
We encourage investigators to contribute manuscripts including, but not limited to, the following topics:
• Novel deep learning algorithms to represent complex radiative transfer scenarios over atmosphere, ocean and land system, including scattering and absorption by aerosols, clouds, surfaces, etc.;
• Deep learning frameworks to conduct remote sensing inversions from multiple spectral, angular or polarimetric measurements including spatial or temporal variations;
• Multi-sensor data processing incorporating instrument uncertainties and characteristics;
• Statistical methods to improve understanding and evaluate uncertainties of machine learning methods for remote sensing.
Topic Editor Meng Gao was employed by Science Systems and Applications, Inc (SSAI). All other Topic Editors declare no competing interests with regards to the Research Topic subject.
Satellite remote sensing has revolutionized Earth observing capabilities and plays a major role in the studies of atmosphere, ocean and land systems. Radiative transfer simulations are foundational for satellite remote sensing algorithms, as they connect the physical properties of the atmosphere and the underlying surface, to the satellite measurements of radiation at the top of the atmosphere. Remote sensing techniques have advanced rapidly to provide highly accurate geophysical property retrievals by utilizing the rich information content of observations at multiple spectral bands, viewing angles and polarization states. However, that high dimensionality and accuracy in observations introduces new challenges in terms of the computational efficiency of radiative transfer simulations and inversions. Conventional approaches, such as the use of precomputed lookup tables, quickly become unmanageable when the retrieval of many parameters is desired. Machine learning methods, such as multi-layer perceptron (MLP), convolutional neural networks (CNN), or Gaussian Processes (GPs), etc., can help to resolve this issue, for example, by creating radiative transfer emulators or conducting direct inversion. Several desirable characteristics of deep learning methods include:
• Computational efficiency, due to the use of matrix operations that can be quickly evaluated;
• High accuracy, given sufficient training data volume;
• Minimal file and memory usage compared to lookup tables;
• Flexibility, since the emulator or inversion model can be incorporated into existing remote sensing algorithms; and
• Differentiability, as the Jacobian matrix of NN models can be represented analytically through automatic differentiation.
This Research Topic aims to showcase the advancement of machine learning research and techniques, designed for radiative transfer modeling and remote sensing inversion. The goal is to improve the analysis of observations from existing and future satellite missions such as NASA’s Plankton, Aerosol, Cloud, ocean Ecosystem (PACE), Multi-Angle Imager for Aerosols (MAIA), Atmosphere Observing System (AOS), and ESA’s Tropospheric Monitoring Instrument (TROPOMI), Multi-view Multi-channel Multi-polarization Imager (3MI) and Carbon Dioxide Monitoring (CO2M) missions, whose large data volume poses challenges to current operational retrieval techniques.
We encourage investigators to contribute manuscripts including, but not limited to, the following topics:
• Novel deep learning algorithms to represent complex radiative transfer scenarios over atmosphere, ocean and land system, including scattering and absorption by aerosols, clouds, surfaces, etc.;
• Deep learning frameworks to conduct remote sensing inversions from multiple spectral, angular or polarimetric measurements including spatial or temporal variations;
• Multi-sensor data processing incorporating instrument uncertainties and characteristics;
• Statistical methods to improve understanding and evaluate uncertainties of machine learning methods for remote sensing.
Topic Editor Meng Gao was employed by Science Systems and Applications, Inc (SSAI). All other Topic Editors declare no competing interests with regards to the Research Topic subject.
