Butterfly trends – an integrated analytical framework for disparate data

Abstract

In the midst of an environmental crisis, incomplete knowledge of long-term species trends limits our understanding of the effects of climate change, land-use change and other pressures on biodiversity. Current knowledge is largely based on data since the 1970s, whilst many pressures known to affect biodiversity began much earlier. Historical specimens offer a wealth of data from the period of accelerating anthropogenic pressures, but are rarely used to estimate trends. Specimen data were not systematically sampled, invalidating conventional modelling approaches and presenting substantial challenges for trend estimation. In this thesis, I address these challenges using specimen data. I formulate three integrated occupancy models that explicitly account for inconsistent sampling among and between occurrence datasets of British butterflies (natural history collections, a standardised monitoring scheme and opportunistic observational records). The different formulations represent alternative hypotheses about the data generation process underpinning each dataset. I use the three formulations to estimate trends for British butterflies since 1900 and employ a variety of statistical model-validation techniques to assess the goodness-of-fit for each formulation. Surprisingly, these techniques show limited ability to distinguish between formulations or species, despite some modelled trends being highly implausible. As an alternative approach to model validation, I conduct an expert elicitation to assess the plausibility of trends based on specimen data. The results demonstrate that expert knowledge is a useful addition to the model validation toolkit. Based on models experts deemed plausible, I investigated the timings of change in British butterflies using a functional approach. I show that substantial change occurred prior to the 1970s - indicating that we may indeed be suffering from “shifting baseline syndrome” In conclusion, my research demonstrates the enormous potential of historical specimens in biodiversity science; however, it highlights that substantive methodological challenges remain, including the inadequacy of existing diagnostic tools for evaluating complex hierarchical models.