Accounting for spatial autocorrelation and environment are important to derive robust bat population trends from citizen science data.

• Uneven sampling is a common problem in citizen science data impacting inference. • Controlling for spatial biases results in robust bat population trend estimates. • National bat population trend differences are masked by the overall British trend. Monitoring wildlife populations is essential if global targets to reverse biodiversity declines are to be met. Recent analysis of data from the UK's long-term National Bat Monitoring Programme (NBMP) suggests stable or increasing population trends for many bat species, and these statistics help inform progress towards national biodiversity targets. However, although based on robust citizen science survey designs, it is unknown how sensitive these trends are to spatial and environmental biases. Here we use Bayesian hierarchical modelling with integrated nested Laplace approximation (INLA), to examine the impact of these types of biases on the population trends using relative occupancy of four species monitored by the NBMP Field Survey in Great Britain (GB): Pipistrellus pipistrellus, P. pygmaeus, Nyctalus noctula and Eptesicus serotinus. Where possible, we also disaggregated trends to national levels using the best model per species to determine if national differences in trends remain once sampling biases are accounted for. Although we found evidence of spatial clustering in the NBMP Field Survey locations, the previously reported GB-wide population trends are broadly robust to spatial autocorrelation. In most species, accounting for spatial autocorrelation and species-environment relationships improved model fit. The nationally disaggregated models highlighted that GB-wide trends mask differences between England and Scotland, consistent with previous analysis of these data, as well as illustrating large gaps in survey effort, especially in Wales. We suggest that although bat population trends were found to be broadly robust to sampling biases present in these data, small differences could propagate over time and this impact is likely to be more severe in less structured citizen science data. Therefore, ensuring trends are robust to sampling biases present in citizen science datasets is critical to effective monitoring of progress towards biodiversity targets, managing populations sustainably, and ultimately a reversal of global declines. [ABSTRACT FROM AUTHOR]