Improving the robustness and reliability of population-based global biodiversity indicators

The current global biodiversity crisis is complicated by a data crisis. Reliable tools are needed to guide scientific research and conservation policy decisions, but the data underlying those tools is incomplete and biased. For example, the Living Planet Index (LPI) tracks the changing status of global vertebrate biodiversity, but gaps, biases and quality issues plague the aggregated data used to calculate trends. Unfortunately, we have little understanding of how reliable biodiversity indicators are. In this thesis I develop a suite of tools to assess and improve the reliability of trends in the LPI and similar indicators. First, I explore distance measures as a flexible toolset for comparing time series and trends. I test distance measures for properties related to time series comparisons and rate their relative sensitivities, then expand the results into a framework for choosing an appropriate distance measure for any time series comparison task in ecology. I use the framework to select an appropriate metric for determining trend accuracy. Second, I construct a model of trend reliability from accuracy measurements of sampled trend replicates calculated from artificially generated time series datasets. I apply the model to the LPI to reveal that the majority of trends need more data to be considered reliable, particularly across the global south, and for reptiles and amphibians everywhere. Finally, I develop a method to account for sampling error and serial correlation in confidence intervals of indicators that use aggregated abundance data from different sources. I show that the new method results in more robust and accurate confidence intervals across a wide range of dataset parameters, without reducing trend accuracy. I also apply the method to the LPI to reveal that the current method used by the LPI results in inaccurate and overly wide confidence intervals.