R project with data and code to reproduce the analysis on Uribe-Rivera et al.

This R project contains everything to reproduce the results found in the paper titled "Tackling spatial sampling bias in biodiversity records “blindly”: A simple solution to improve species distribution models and conservation decisions". Authors' list and affiliations David E. Uribe-Rivera1,*, Scott D. Foster2, Wen-Hsi Yang3, Caley, Peter4, Andrew J. Hoskins5,6. 1 CSIRO Environment, Brisbane, QLD, Australia2 CSIRO Data61, Hobart, TAS, Australia3 CSIRO Data61, Brisbane, QLD, Australia4 CSIRO Data61, Canberra, ACT, Australia5 North Australian Indigenous Land and Sea Management Alliance, Darwin, NT, Australia6 The University of Queensland, Brisbane, QLD, Australia * Corresponding author: [e-mail] ; [e-mail] Data contained is publicly available from diverse sources, all of which are detailed both in the main text of the manuscript and within the code documentation.The paper's abstract is reproduced below:Georeferenced occurrence records (also known as presence-only data) are widely available biodiversity data that provide information on species’ distributions. These data are available in semi-structured compilatory databases, which compile data arising from a diverse range of sampling processes and intensities/rates. Failing to account for variability in sampling rate, meaning the likelihood of a species being observed AND reported if present, may bias our understanding of ecological patterns. This bias is typically referred to as spatial sampling bias. Various methods have been developed to account for spatial sampling bias. Robust strategies work, for example, by jointly modelling the species distribution and the sampling rate as separate processes. These approaches assume the sampling rate patterns in a dataset of records are relatively consistent. But that implicit assumption is typically violated when records are a mix of opportunistically collected and systematically surveyed data with standardised sampling rate. We highlight the importance of being able to accurately and consistently discriminate opportunistic from systematically collected records using Australian marsupial records in the Atlas of Living Australia (ALA). We found that the spatial distribution of opportunistic records more closely resembled the spatial patterns of well-known spatial sampling bias covariates after filtering out records derived from systematic surveys. Thus, the sampling rate is more effectively explained in the model and a less biased estimate of the distribution is obtained.Metadata from compilatory databases such as the Global Biodiversity Information Facility or the ALA currently contain insufficient information on the observation process, making manual filtering time-consuming and likely imperfect. To ensure methods to minimise spatial sampling bias are more accurately applied, we suggest that georeferenced biodiversity record databases should incorporate a binary metadata column on whether a record has been opportunistically or systematically collected. KeywordsSampling bias, sampling effort, citizen science, observation process, species distribution models, niche modelling, habitat suitability models.