|SREL Reprint #2264|
Measuring Trends in Ecological Resources
Questions about the existence or magnitude of trends in ecological resources are an important element of many current ecological and environmental issues. Some examples include: whether the number of lakes showing signs of stress from acidification is increasing, decreasing, or showing no change; whether amphibian or songbird populations are declining over large regions; and whether ecological communities are changing, as might be expected from global climate change. Good answers to these questions depend on using appropriate statistical designs to sample the environment and appropriate statistical techniques to estimate the trend. Although the concepts of trend and change appear simple, the ecological and statistical issues associated with good design and analysis can be quite complex.
Part of the complexity is that "trend" is difficult to define precisely. Trend is often defined as long-term change in the mean, but what constitutes long-term depends on the temporal scale of the study and the relevant ecological dynamics. It is also difficult to separate long-term trends from other components of temporal variation, including multi-year cyclical variation, within-year seasonal variation, and erratic fluctuations. These components cannot be uniquely identified without using a model that precisely defines the temporal scale of each component. In practice, all four components may not be separated; instead, two or more (e.g., long-term trend and cyclical variation) are combined into a composite concept of annual temporal change.
Another complexity is defining an appropriate spatial coverage. Most ecological and environmental monitoring programs are intensive studies of single (sentinel) sites. If the spatial variation among sites is small, trends at a single site may be representative of much larger areas. Unfortunately, large amounts of spatial variation are the norm, not the exception, especially for ecological quantities like population abundance and species composition. Making generalizations about regional trends requires more spatially complete data, either from networks of sites or complete spatial coverage. This requires that the concept of change be expanded to include spatiotemporal trends summarized across the ecological resource within the region. The ecologically interesting trend may be the change in spatial distribution across a region, not a change in the average amount.
Statistical methods are used to make conclusions about unsampled places from observed data. A third complexity is that these conclusions may be justified in two different ways, design-based or model-based inference. In design-based inference, conclusions about unobserved sites are justified by the random choice of observed sites. This requires a probability-based sampling scheme, such as a simple random sample or more complex sampling design. However, sites for many long-term studies are not randomly chosen, and any random sampling scheme is difficult to apply for many ecological and environmental problems. In model-based inference, a statistical model is used to make conclusions from the observed data. Justifiable conclusions about unsampled sites depend on the correctness of the model. Although it is unfair to generalize to all ecological studies, we suspect that most have either assumed that regional trends were similar to those in study sites or not worried about generalizing to unstudied sites.
SREL Reprint #2264
Dixon, P.M., A.R. Olsen, and B.M. Kahn. 1998. Measuring trends in ecological resources. Ecological Applications 8:225-227.