Authors: Valdez, Jose; Callaghan, Corey;Junker, Jessica;Purvis, Andy;Hill, Samantha;Pereira, Henrique;

Quantifying biodiversity trends at the global scale remains largely unresolved. This is because global analysis of biodiversity change is typically based on a collection of locally monitored sites which are affected by certain sampling biases and measurement errors. Here, we assess the ability to detect global trends using local species richness and the effect of time series length, site-selection biases, measurement error, and taxa mobility (represented by spatial resolution). We find that even with error-free data thousands of sites are needed for decadal times series and detecting trends within a year was practically impossible. Measurement errors in local species richness trends drastically reduced the ability to detect a change, with ten times more sampling sites required with up to 5% measurement error. The ability to detect change was lower at higher resolution. Additionally, spatial sampling biases, such as failing to sample areas in the top 1% of largest losses will not only reduce the ability to detect negative change but can even result in a positive trend. We conclude that detecting global trends using estimates from locally monitored sites is currently unreliable, and requires implementing refined observation networks combined with models to account for biases, measurement error, and environmental drivers.