OPS 3-11
Citizen science facilitates analysis of network-level data sets: PhenoCam as a case study

Tuesday, August 11, 2015
Exhibit Hall, Baltimore Convention Center
Margaret Kosmala, Organismic and Evolutionary Biology, Harvard University
Rebecca Cheng, NEON, Inc., Boulder, CO
Sandra Henderson, NEON, Inc., Boulder, CO
Andrew D. Richardson, Organismic and Evolutionary Biology, Harvard University

The PhenoCam network uses more than 200 elevated digital cameras for automated monitoring of vegetation phenology in ecosystems across North America. The cameras take photos of vegetated landscapes every 30 minutes throughout the day, year-round, and automatically send them back to a central database. These photos are used to infer phenological change of forests, grasslands, shrublands, cropland, and tundra. While extracting measures of canopy “greenness” from these images is straightforward, inferring specific phonological events or phases, such as flowering or the presence of snow, is not as simple. Because there are hundreds of thousands of images, it is not practical to manually classify them. We have developed an online citizen science project called Season Spotter (www.seasonspotter.org) that asks volunteers to provide information about the images. A unique aspect of this project is the advancement of both science and education based goals.  Outcomes of the collaboration are to 1) develop a dataset of human observations that can be used to validate the PhenoCam algorithms and 2) develop a complementary dataset of human observations of phenology events that cannot easily be automated with algorithms. Because phenology is such an accessible concept to non-scientists, we have also designed a significant education component that will not only help engage the volunteers longer, but also enrich their understanding of phenology specifically and science more generally.


Preliminary results of PhenoCam citizen science show that volunteers are accurate in identifying snow in images – something that is hard to do algorithmically. For example, identifying images with snow is important for automated processing because snowy images bias the calculation of “greenness”. With as few as three people looking at each of 60,000 images, we achieved 97% accuracy in snow classifications in just three months.  We demonstrate the value of a citizen science approach to the categorization and annotation of remotely sensed plant phenology imagery.  This poster will also address best practices associated with the engagement and retention of individuals in ecology-based citizen science projects.