Appendix F: Assessing Citizen Science Labeling to Improve Training Data Quality for Land Cover Protocols within the GLOBE Observer Community

The proposed project aims to use hackathons as the enabling platform to label images collected through the Land Cover tool in the NASA GLOBE Observer app, the citizen science app of the GLOBE Program. The labeled images would be used to train, validate, and test machine learning algorithms in NASA Earth Science missions. GLOBE Observer has cataloged over 15,800 entries (up to 6 photos per entry) but only 40% of entries are labeled, rendering their use and applications to be limited compared to their potential. The work efficiency will be maximized by leveraging the existing Land Cover Type classification systems for labeling scheme design and leveraging already labeled images for ground truth label generation. Labels will be structured in a modularized hierarchy, initially using Carbon Monitoring System as the application but extensible to other applications. Hackathon will run throughout the performance period marked by four meetings. Trainees will be recruited inclusively from diverse groups, and trained via computing platforms centered on Microsoft Teams at the University of Vermont and leveraging Amazon SageMaker services geared for efficient labeling of ground truth images. The quality of image labels generated through hackathon will be assessed using a direct measure (comparing with the ground truth labels) and an indirect measure (comparing the outputs from using multiple image classification models); and, additionally, image labelers will be assessed as well in conjunction with the quality of image labels they produce. Successful completion of this work requires multidisciplinary expertise spanning from data management, machine learning and data science to land-cover and land-use to workforce training and community development. The project team collectively covers all necessary expertise. The project outcome will include not only high-quality image labels but also assessment metrics for the quality of image labels and labelers. A success of this project would enable a longer-term vision to build a comprehensive, compact land-cover labeling system encompassing heterogeneous applications. Another vision would be to let the data deficiency identified during the hackathon engage citizen scientists to contribute data better customized to meet the need – namely "hackathon-in-the-loop".

This is a mature technology (TRL 7+) — the realistic path in is usually NASA's Technology Transfer Program: licensing an existing NASA patent, or a Space Act Agreement to use NASA facilities/expertise directly. NASA also runs a startup licensing program with no upfront fee for companies formed to commercialize a specific NASA technology.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.