Taking Climate Change Into Account: Non-Stationarity In Climate Drivers Of Ecological Response

Changes in the global climate system are creating increasingly non-analogue climate conditions with expectations of non-stationarity among climate drivers. Decoupling among climate drivers complicates the assessment of ecological response to the changing climate as characteristics that could be once treated as a suite of conditions now need to be treated as potentially independent with possible synergistic effects.Ecologists commonly use ordination techniques (often principal component analysis; PCA) on large climate and environmental datasets to reduce a range of variables to a few axes that are uncorrelated with each other and often explain large proportion of the variation in the original data. However, non-stationarity, with correlations among variables changing over time, can affect this approach. Here, we use a 37-year climate dataset from the Niwot Ridge Long Term Ecological Research site (Colorado, USA) to present the use of both moving window principal component analysis and moving window correlation analysis to determine non-stationarity in climate data.Relationships among climate variables and between input variables and PCA axes changed over time; this obscured interpretation of relationships between PCA axes and an ecological response (plant biomass), suggesting that one-time PCA for environmental variables may lead to inappropriate inferences.Synthesis. Care must be taken in analysing climate-ecological relationships when predictor variables exhibit non-stationarity. We present a conceptual decision-making tree to help ecologists consider when to use PCA and extract axis scores or use alternative approaches for incorporating non-stationarity into subsequent analysis, including testing variables individually to aid in interpretation, break point analyses and averaging PC scores.