Quantifying uncertainties in biologically-based water quality assessment: A pan-European analysis of lake phytoplankton community metrics

► Phytoplankton ecological quality metrics were calculated for 32 European lakes. ► We modelled sources of variability (within and among lakes) in these metrics. ► Metrics varied more among lakes, than within lakes or due to sampling variation. ► Metrics varied significantly with eutrophication and lake depth. ► Three metrics are considered robust for Water Framework Directive Intercalibration. Lake phytoplankton are adopted world-wide as a sensitive indicator of water quality. European environmental legislation, the EU Water Framework Directive (WFD), formalises this, requiring the use of phytoplankton to assess the ecological status of lakes and coastal waters. Here we provide a rigorous assessment of a number of proposed phytoplankton metrics for assessing the ecological quality of European lakes, specifically in response to nutrient enrichment, or eutrophication, the most widespread pressure affecting lakes. To be useful indicators, metrics must have a small measurement error relative to the eutrophication signal we want them to represent among lakes of different nutrient status. An understanding of variability in metric scores among different locations around a lake, or due to sampling and analytical variability can also identify how best this measurement error is minimised. To quantify metric variability, we analyse data from a multi-scale field campaign of 32 European lakes, resolving the extent to which seven phytoplankton metrics (including chlorophyll a, the most widely used metric of lake quality) vary among lakes, among sampling locations within a lake and through sample replication and processing. We also relate these metrics to environmental variables, including total phosphorus concentration as an indicator of eutrophication. For all seven metrics, 65–96% of the variance in metric scores was among lakes, much higher than variability occurring due to sampling/sample processing. Using multi-model inference, there was strong support for relationships between among-lake variation in three metrics and differences in total phosphorus concentrations. Three of the metrics were also related to mean lake depth. Variability among locations within a lake was minimal (