Description:

The volume of open Earth and atmospheric science data has substantially increased over the past decade and projected to exponentially increase over the next several years. Over this same period, artificial intelligence and machine learning have emerged as a means to analyze and extract scientific knowledge from these expansive datasets. Proper application of machine learning methods and extraction of meaningful knowledge is dependent on the input training and testing data accurately representing the scientific problem being addressed. These machine learning ready datasets, most often must be large which requires significant curation and management overhead for subject matter experts. Making these datasets open to the community maximizes return on this effort and promotes scientific advancement. The purpose of this session is to provide a forum to showcase open datasets for machine learning in Earth and atmospheric sciences as well as the tools and methods utilized to develop these datasets. This session seeks submissions that describe open datasets that are leveraged in machine learning applications, including phenomenon detection, change detection, and forecasting; methods and implementations that contributed to the development and curation of these comprehensive datasets; description of tools developed to automate the dataset development process; and detailed descriptions and evaluations of these datasets to highlight their capabilities and limitations.  Submissions with unique techniques to address the development of high-quality training and testing datasets, but also address challenging science questions, will be prioritized.