Data for: Revealing hidden sources of uncertainty in biodiversity trend assessments

Idiosyncratic decisions during the biodiversity trend assessment process
may limit reproducibility, whilst “hidden” uncertainty due to collection
bias, taxonomic incompleteness, and variable taxonomic resolution may
limit the reliability of reported trends. We model alternative decisions
made during assessment of taxon-level abundance and distribution trends
using an 18-year time series covering freshwater fish, invertebrates, and
primary producers in England. Through three case studies, we test for
collection bias and quantify uncertainty stemming from data preparation
and model specification decisions, assess the risk of conflating trends
for individual species when aggregating data to higher taxonomic ranks,
and evaluate the potential uncertainty stemming from taxonomic
incompleteness. Choice of optimizer algorithm and data filtering to obtain
more complete time series explained 52.5% of the variation in trend
estimates, obscuring the signal from taxon-specific trends. The use of
Penalized Iteratively Reweighted Least Squares, a simplified approach to
model optimization, was the most important source of uncertainty.
Application of increasingly harsh data filters exacerbated collection bias
in the modelled dataset. Aggregation to higher taxonomic ranks was a
significant source of uncertainty, leading to conflation of trends among
protected and invasive species. We also found potential for substantial
positive bias in trend estimation across six fish populations which were
not consistently recorded in all operational areas. We complement analyses
of observational data with in silico experiments in which monitoring and
trend assessment processes were simulated to enable comparison of trend
estimates with known underlying trends, confirming that collection bias,
data filtering and taxonomic incompleteness have significant negative
impacts on the accuracy of trend estimates. Identifying and managing
uncertainty in biodiversity trend assessment is crucial for informing
effective conservation policy and practice. We highlight several serious
sources of uncertainty affecting biodiversity trend analyses and present
tools to improve the transparency of decisions made during the trend
assessment process.