Estimating species distributions from spatially biased citizen science data

Abstract

Ecological citizen science data are rapidly growing in availability and use in ecology and conservation. Many citizen science projects have the flexibility for participants to select where they survey, resulting in more participants, but also spatially biased data. It is important to assess the extent to which these spatially biased data can provide reliable estimates of species distributions. Here we quantify the extent of site selection bias in a citizen science project and the implications of this spatial bias in species distribution models. Using data from the BirdTrack citizen science project in Great Britain from 2007 to 2011, we modelled the spatial bias of data submissions. We next produced species occupancy models for 138 bird species, and assessed the impact of accounting for spatial bias. We compared the distributions to those produced using unbiased data from an Atlas survey from the same region and time period. Averaging across 138 species, models with spatially biased data produced accurate and precise estimates of species occupancy for most locations in Great Britain. However, these distributions were both less accurate and less precise in the Scottish Highlands, showing on average a positive bias. Accounting for the spatially biased sampling with weights led to on average greater accuracy in the Scottish Highlands, but did not increase precision. This region is both distinct in environmental characteristics and has a low density of observations, making it difficult to characterise environmental relationships with species occupancy. Accounting for the spatially biased sampling did not affect average accuracy or precision throughout most of the country. Spatially biased citizen science data can be used to estimate species occupancy in regions with stationary environmental relationships and good sampling across environmental space. The reliability of estimated species distributions from spatially biased data should be further validated and tested under a range of different scenarios.