Phenomena Portal for Machine Learning Applications in Earth Science

Satellite remote sensing is essential in advancing the state of knowledge of the Earth’s surface and atmospheric processes. Scientists commonly apply classification schemes to data and imagery retrieved from these platforms to gain a conceptual understanding of these processes. However, classification schemes currently employed by the science community lack efficiency and are often not scalable to large datasets. Advancements in computational resources over the last decade have allowed for the development of machine learning (ML) techniques that serve as a potential solution to this problem. We will present an attempt to consolidate multiple Machine Learning models, applied to Earth and atmospheric remote sensing datasets, into a single visualization and analysis platform called the Phenomena Portal. Independent architecture and integration of machine learning models into the portal for transverse cirrus bands, high latitude dust, and smoke are shown. This tool serves as a mechanism for visualizing the spatiotemporal frequency of these phenomena while performing basic statistical analysis. Furthermore, it serves as an event database resource for scientists seeking to further scientific discovery. The framework supports API access to promote searchability, discoverability, and integration of these event catalogs with Earth science data systems and existing web services. This presentation will provide a high level overview of the workflow developed by the NASA IMPACT team and give some practical examples of successful deployments of an ML lifecycle from inception to operational deployment.