YIP SSEC aerosol and cloud type machine learning methods developments for GEO and LEO imagers in Support of ONR
Abstract
We propose a three-year period effort which is aligned with ONR s atmospheric science interests. The effort we propose is to continue with the work of ONR-322 grant (no. N00173-19-1-G005) where we developed a machine learning (ML) methodology that has the unique capability to enhance existing operational NASA and NOAA cloud and aerosol products. This previous ML work demonstrated that texturewith spectral information from deep ocean Low Earth Orbit (LEO) imagery provide the ability to more confidently identify severe aerosol events and a wider range of cloud types compared to existing operational NASA and NOAA aerosol and cloud identification algorithms which rely primarily on spectral tests. Specifically, the current limitation of operational aerosol identification methods is that spectral signatures of severe smoke aerosols and optically thick clouds share spectral features which makes it difficult to confidently discriminate between these with an algorithm that is meant to work ona global scale. Also, operational cloud identification methods are limited to identifying cloud phase (i.e., water vs. ice clouds) since spectral tests alone are not indicative of atmospheric physics such as different convection processes and thermodynamically stability. In contrast, ML methods allow quantitative analyses of texture and spectral information to identify severe aerosols, cumuliform (i.e., convection processes), closed stratiform (i.e., thermodynamically stability) clouds, transitional/mixed cloud (transitional between the closed-stratiform and cumuliform) and cirrus/high-altitude clouds.The goals of the proposed effort are to 1) extend the ML method to identify a richer set of cloud types indicative of specific convective atmospheric physics and 2) extend the ML to operate on both geostationary imager data (e.g., ABI and AHI) instead of just low earth orbit imager data (e.g., MODIS and VIIRS). The proposed efforts as based on interests expressed by our MAGPIE collaborators to have the ability to identify specific convective cloud types such as hexagonal cells, cold pools, transverse cloud streets, and clouds formed in vicinity of islands such as island streamers and A von Karman vortex clouds from both GEO andLEO imagery. Part of the research challenge of this effort will be to understand the limits to identifying unique convective cloud types. For example, the texture and temporal features of cold pool clouds are similar to other convective cloud types, hence there will be a limit to which how accurately cold pools clouds can be identified.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 24, 2025
- Source ID
- N000142512165
Entities
People
- Willem Marais
Organizations
- Office of Naval Research
- United States Navy
- University of Wisconsin System