COS 82-5 - Turning observations into biodiversity data: Spatial biases in community science

Background/Question/Methods

Community science – community members and scientists collaborating to collect data of the natural world – is rapidly growing in popularity. Community science (CS) projects collect data on many biological phenomena such as species ranges, species interactions, and phenology. However, despite all these potential benefits to science and ecological monitoring, CS data are infrequently incorporated into standard survey monitoring due to the limited information on their biases and errors. With many CS datasets, the extent and influence of biases (e.g. sampling and spatial biases) are unknown.

To address this, we modeled the spatial biases present in the popular CS platform, iNaturalist. iNaturalist uses crowdsourcing to collect georeferenced and time-stamped observations of all taxa worldwide. We focus on the more than 1.75 million observations available from British Columbia, a biodiversity hotpot. With its wealth of biodiversity data, iNaturalist is poised to answer large ecology and conservation management questions, but little is known about the platform’s spatial biases. Using machine learning and species distribution modeling techniques, we examined how landscape features (e.g. roads and parks) influence the distribution of iNaturalist observations.

Results/Conclusions

There are strong road biases for observations in iNaturalist, with over 94% of observations within 1 km of roads in British Columbia. iNaturalist observations are closer to roads (mean = 309 m) than a random point null model (mean = 5276 m; Welch two sample t-test: t = 55695, df = 10422, p-value < 0.0001). Using Maxent for species distribution modeling of iNaturalist observations, we examined which landscape factors (protected areas, roads, travel time, land cover types) were most important in determining where observations are taken and created a probability map of the province revealing how likely different regions are to be sampled by community members. These methods provide tools for modeling the effects of spatial biases, which provides opportunities for species occupancy and biodiversity modeling from this enormous dataset. In addition, we document areas that remain under-sampled on the iNaturalist platform to target future sampling priorities.