Evaluation of the Spatial Biases and Sample Size of a Statewide Citizen Science Project. (16th December 2021)

• Record Type: Journal Article
• Authors: Kays, Roland; Lasky, Monica; Parsons, Arielle W.; Pease, Brent; Pacifici, Krishna
• Is Part Of: Citizen science. Volume 6:Number 1(2021)
• Publication Date: 2021-12-16
• Subjects: sample design -- camera trap -- spatial bias -- mammals -- large scale; Science -- Citizen participation -- Periodicals; Volunteer workers in science -- Periodicals; 507.2
• DOI: 10.5334/cstp.344
• Languages: English
• ISSNs: 2057-4991

Abstract

Although quality control for accuracy is increasingly common in citizen science projects, there is still a risk that spatial biases of opportunistic data could affect results, especially if sample size is low. Here we evaluate how well the sampling locations of North Carolina's Candid Critters citizen science camera trapping project represented available land cover types in the state and whether the sample size (4, 295 sites) was sufficient to estimate ecological parameters (i.e., species occupancy) with low bias and error. Although most sampling was opportunistic, we used a "Plan, Encourage, Supplement" approach to improve our spatial coverage. We assessed potential biases by comparing seven dimensions of habitat (i.e., land cover, elevation, road density, etc.) sampled by camera traps with those available in the state, using a minimum sample threshold approach, and found that the variation of habitat across the state was sufficiently sampled. At the ecoregion level we sampled 99.2% (±0.01) of the variation of potential habitat "adequately" and 96.4% (±0.03) "very adequately." Supplemental sampling by staff helped meet sampling adequacy for 6.8% of ecoregion-habitat classes, especially in less populated parts of the state. Compared with results from the full data set, the relative bias and error with subsets of the data dropped below 10% relatively quickly with increasing sample size for estimates of occupancy, suggesting that results estimated with the full sample are robust, although the precision of particular ecological relationships were more variable. These analyses show that opportunistic sampling can be representative of large areas if sample size is high enough and that a priori sampling goals can help improve coverage by encouraging volunteers to sample in certain places or through supplemental data collection by staff.