Detecting population trends with historical data: Contributions of volatility, low detectability, and metapopulation turnover to potential sampling bias

In retrospective studies, discrete population units such as ponds may be resurveyed at a later time using only the set of initially occupied sites. There are possible confoundings that affect estimates of occupancy change under these conditions. For most possible parameter values for a metapopulation, simulations and analytical results show that turnover leads to a tendency to observe a decline in the proportion of initially occupied sites that are occupied at a later time even when the overall metapopulation is stable or increasing. For a given time interval, the spurious decline will be greater when metapopulation turnover is higher. If site-level detectability d is <1, a single resurvey of only the initially occupied sites will show a decline of 1-d even if no change has taken place. Finally, volatile populations can be difficult to resurvey, especially if sample units are chosen based on having an abundance of the species at the earlier survey. All three issues can exist simultaneously and their influence on trend estimates can be difficult to distinguish based on samples at only two points in time. A sample of literature illustrates clear cases where these biases could exist, even though a variety of survey methods were used. Suggestions are made for improved sampling, including resampling the entire original set of sites and conducting multi-year resurveys.