Sensitivity analyses of exploiting spatial information via deep-learning from radiometer images to identify aerosols and distinct cloud types
Abstract
Current multi-spectral methods use per-pixel spectral and spatial standard-deviation (SD) thresholding techniques to identify aerosols and distinct cloud types from polar orbiting radiometers such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS), which result in extreme aerosol events being misclassified as clouds and the misclassification of distinct cloud types. Clouds and aerosols in radiometer images do have distinct spatial structures and it is not clear whether a simple spatial statistics (e.g. SD) on different spectral images are powerful enough to set clouds apart from aerosol events. Recently new machine learning neural network architectures have been developed to better represent spatial information of images to improve the distinction between different image types. With these recent advances an immediate question is raised: by how much can the identification of aerosols and distinct cloud types be improved upon with the better extraction of spatial information from radiometer images? To address this question 1) the NASA Worldview platform was adapted to create labeled datasets of aerosol and distinct cloud types, 2) a pre-trained convolutional neural network (CNN) was adopted to exploit both spatial and spectral information from multi-spectral images, 3) the CNN labeling accuracy of aerosols and cloud types was quantified via Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) observations. By harnessing CNNs with a unique labeled dataset, we demonstrate the improvement of the identification of aerosols and distinct cloud types from MODIS and VIIRS images compared to a per-pixel spectral and spatial SD thresholding method. In the presentation we show 1) a couple of case studies through which we compare the CNN methodology results with the MODIS cloud and aerosol products, and 2) a sensitivity analyses of where we quantify the identification accuracy of aerosols and cirrus clouds at various optical depths (ODs), where the ODs of the aerosols and cirrus clouds have been retrieved with CALIOP.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.A51U2675M
- Keywords:
-
- 0365 Troposphere: composition and chemistry;
- ATMOSPHERIC COMPOSITION AND STRUCTURE;
- 3336 Numerical approximations and analyses;
- ATMOSPHERIC PROCESSES;
- 0520 Data analysis: algorithms and implementation;
- COMPUTATIONAL GEOPHYSICS;
- 0555 Neural networks;
- fuzzy logic;
- machine learning;
- COMPUTATIONAL GEOPHYSICS