COS 63-2 - CANCELLED - Identifying the Identifiers: How iNaturalist facilitates collaborative, research-relevant data generation and why it matters for biodiversity science

Abstract: Public generation of research-quality ecological data relies on proper taxonomic identification. One of the most widely used platforms, iNaturalist, leverages public engagement to identify organisms by community consensus. These identifications are responsible for the annotation and verification of millions of observational records per year. Despite the exponential growth in the use of the iNaturalist platform, little is known about biases in identifier effort across multiple dimensions. Additionally, iNaturalist records have not been wholly embraced as reliable data by some researchers, in large part because they rely on effort by platform members in a non-professional context. Here, we characterize identifier effort from its inception in 2008 through 2021 by aggregating a broad dataset of all observations, identifications, and taxonomic information from the platform during this interval.

We first explored questions related to patterns of identifications by users and how their rates and allocations play out across time, user activity, and in the context of the introduction of computer vision / machine learning algorithms (CV) that assist in identifications. The bulk of identifications on the site are made by a small group of users: 50% of all identifications were made by the top 0.3% of identifiers. We show that identifications are scaling with the rapid growth of observations uploaded to the site, and time to a vetted research-grade (RG) status has continued to fall since the site’s inception, even as iNaturalist’s activity and user base grows rapidly. The most active identifiers are much more likely to be taxonomic specialists than the average user, but are about equally likely to be geographic specialists or generalists. Next, we applied a hierarchical modeling framework to test key characteristics that determine whether an observation reaches RG and how long it takes to get there. We found that the key determinant of whether records hit RG, and the time it takes to do so is identifier expertise. The use of iNaturalist’s computer vision feature itself does not impact the likelihood of an observation reaching RG, but likely reduces the time to reach RG by lowering the initial taxonomic rank of the first identification– a key factor in hastening RG status. We close by discussing methods to decrease gaps and biases by incentivizing identification activity, best practices for enhanced identification quality, and continuing steps to further grow trust across the ecosystem of observers, identifiers, and biodiversity researchers.