Effects of temporal bias on the assessment of an ecological perturbation: a case study of the Prestige oil spill

Abstract

The impacts of unpredictable ecological perturbations are often assessed via measurements of environmental change only after the event has occurred. Temporal series of satellite images provide a cost-effective way to gather information before ecological perturbations occur. However, in previous studies, the disturbances have neither been always centred in time in the series of the focal environmental variable nor has the relevance of the temporal coverage been explicitly tested through factorial designs. In this study, we manipulated the temporal coverage and the position of the disturbance event in the temporal series to examine whether and how the assessment is affected. Specifically, we tested the effect of the Prestige oil spill on monthly sea chlorophyll concentration and net primary productivity along the north-western Spanish coast. We designed planned comparisons through factorial analyses to test two alternative hypotheses: (1) the spill has negative consequences on phytoplankton activity and/or abundance due to physiological constraints or (2) it has positive consequences on phytoplankton abundance as a result of changes in biotic interactions. The relevance of the statistical effects was critically dependent on the temporal coverage and the position of the spill event in the temporal series. Short periods (three years) were insufficient to cover the range of variability even if the disturbance was centred in the time series. Similarly, results from longer time series (up to eight years) in which the event was temporally biased (at the beginning of the time series) also differed from those that were centred in the entire time window. Temporal series for the study of ecological impacts should be as long as necessary to encompass the temporal variability of the study systems (up to nine years in our study case), and the disturbance event should be centred in the time series to reduce potential spurious effects of temporal autocorrelation. However, our results revealed that each one of these requirements alone was not sufficient to encompass all of the natural variability, and thus both requirements should be met. For impact assessments we encourage the use of unbiased satellite data series to complement in situ measurements.