2017 ESA Annual Meeting (August 6 -- 11)

COS 182-9 - Follow that carbon atom! Using machine learning to trace student understanding of the carbon cycle

Friday, August 11, 2017: 10:50 AM
C122, Oregon Convention Center
Bryan Macneill1, Margaurete Romero1, Anne-Marie Hoskinson2 and Luanna Prevost1, (1)Dept. of Integrative Biology, University of South Florida, Tampa, FL, (2)Biology and Microbiology, South Dakota State University
Background/Question/Methods

A fundamental understanding of the processes and pathways of the carbon cycle is essential for grasping complex ecological issues such as climate change, as well as other ecologically-important cycles such as nitrogen and phosphorus. However, biology students of all ages have difficulty with these concepts. We investigated the use of automated computerized scoring to identify undergraduate biology students’ scientific and non-scientific ideas about carbon cycling. We modified a question asking to describe the movement of carbon atoms from a deceased jackrabbit through the carbon cycle into the tissues of a coyote. The question was administered in undergraduate biology courses at three research-intensive institutions. We coded 575 student responses for the presence or absence of correct and incorrect pathways within the carbon cycles as well as processes involved in carbon movement along these pathways. Words and phrases in students’ responses were extracted by the machine learning software LightSide and used to build predictive models. Models were deemed calibrated when they achieved agreement between human and computer generated better than 80% of the time (Cohen’s kappa ≥ 0.8)

Results/Conclusions

We identified correct pathways and incorrect pathways commonly used by students in their responses. Most responses included a pathway for carbon atoms from plants to herbivores (61%), and from herbivores to the coyote (58%). However, only 25% of responses included carbon moving from the atmosphere to plants. The most common incorrect pathway (37% of responses) described the movement of carbon from the soil directly to plants. Additional incorrect pathways included movement from the atmosphere to the coyote via respiration (10%) and direct consumption of plants by the coyote (25%). We also coded for three carbon transfer processes: decomposition (63% of responses), respiration (39%) and photosynthesis (26 %). We developed scoring models that demonstrate high-agreement between human and computer scoring (Cohen’s kappa ≥ 0.8) for four categories: photosynthesis, decomposition, soil to cactus and plant to herbivore. Our results demonstrate that students harbor the misconception that plants obtain carbon from the soil. This misconception can contribute to an incorrect or incomplete understanding of the role of the atmosphere as a carbon pool, thus limiting understanding of atmospheric carbon in climate change. Instruction focused on the movement of carbon from decomposers to the atmosphere may help address this incorrect idea.