Student Research Reports

View Research Report

To complement satellite land cover data, researchers today rely on citizen science to record and analyze ground-level land cover because of its ability to efficiently facilitate research at a large scale, quickly gather data across a variety of regions, and ethically engage communities in pertinent matters. The GLOBE Observer app, part of The GLOBE Program, is a citizen science tool used by thousands to document the planet. To make a land cover observation, a user logs their date and location, records the surface conditions, takes six directional photos, and classifies various land cover features in the photos, including reporting visible percentages for each feature. However, in the time-consuming step of classifying photos, our surveys have found that user subjectivity, limited knowledge, and availability of time can cause citizen scientists to classify data inconsistently or leave images unclassified altogether, greatly decreasing data usability for scientists. In this study, we collected unique AOI land cover data uploaded to GLOBE from around the United States and selected 5896 directional photos gathered by the SEES 2023 cohort. These contain information about 3 key land cover classes: sky, land, and water. We received assistance from citizen scientists to identify whether the three classes are present in the land cover photos. This process was facilitated by Zooniverse, a citizen science data annotation tool. Finally, 4717 land cover classifications were used to train and validate a deep learning model, while 1179 photos were used to evaluate the model. To quantitatively assess the model, we compared its outputs for each class to the ground truth citizen science classifications gathered from Zooniverse. Our model begins to relieve the burden on citizen scientists to classify photos, improving citizen science data for better climate modeling.