Just-in-Time Training Improves Accuracy of Citizen Scientist Wildlife Identifications from Camera Trap Photos

Abstract

Citizen scientists can help professional scientists amass much larger datasets than would be possible without their input, but the quality of these data may impact their utility. Therefore, it is imperative to develop standard practices that maximize the accuracy of data produced by citizen scientists. One method increasingly used to improve data accuracy in citizen science-based projects is just-in-time training (JITT), in which volunteers are given on-demand resources to train them on the spot or in conjunction with the research they are performing. In this article, we examine whether JITT improves citizen scientist accuracy of subject identification, specifically wildlife identification from camera trap photos. Ninety-four participants with varying degrees of experience in biology were asked to identify photos from camera traps in Los Angeles, California set to capture photos of wildlife in an urban habitat. Without access to JITT, citizen scientists with no background in biology had lower accuracy than professional biologists (no background: mean = 51.8%, standard error [SE] = 6.0%; professional biologist: mean = 77.6%, SE = 2.1%). However, when participants with no background in biology received JITT, they were able to identify wildlife with a similar level of accuracy as professional biologists (no background: mean = 81.9%, SE = 3.6%; professional biologist: mean = 85.1%, SE = 2.5%). There was a significant interaction between biology background and training treatment (F-ratio = 7.61, 'p' = 0.0009). The increase in accuracy of novice citizen scientists who received JITT was due primarily to fewer misidentifications of species overall but also to increased confidence in classification of species (participants selected the “Don’t Know” option less frequently). From these results, we conclude that the use of JITT can significantly improve subject identification accuracy for citizen scientists with no background in biology.

Keywords

• citizen science
• wildlife images
• trail camera
• urban wildlife
• human computation
• zooniverse