Using ecosystem monitoring data to detect impacts

The purpose of ecosystem monitoring programs is to indicate the state of ecosystems and whether they have been impacted by activities such as fishing. This paper discusses a range of methods for inferring such impacts using monitoring data with no control sites. These methods assess either (i) the expected probability of an observed value in an unimpacted system, or (ii) the frequency of values below a fixed reference point. The second approach allows inference criteria based on changes in this frequency rather than by reference to a critical probability. All methods would have provided a sustained indication of a significant decline in Antarctic fur seal (Arctocephalus gazella) pup production at South Georgia from the early 1990s within a few years of its onset, but a fixed reference point method could have provided this sustained indication from the onset. Furthermore, simulation of all methods suggests that the total probability of error (false positives and false negatives combined) is lowest with fixed reference point methods. The probabilities of Type I and Type II error can be evaluated analytically for these methods, which facilitates decision-making based on attitudes to risk.