Uncertainty (workpackage 6.1)
Within the WISER project, many data analyses are being undertaken. There are equally many potential sources of uncertainty: sampling methodologies, sorting and identification errors, and uncertainties related to the statistical analysis, all of which will affect both the metrics for the various Biological Quality Elements (BQEs) and the relationships of these elements with anthropogenic pressures. Workpackage 6.1 is providing support to the other WISER workpackages in their assessment of this uncertainty. In particular, help is being provided to the workpackages in the lake and transitional/coastal waters modules 3 and 4, where we analyse the implications of sampling uncertainty on their BQE metrics.
The WISER field campaign has produced new hierarchical datasets on all BQEs. For example, water bodies have been sampled within countries at different stations, and in some cases, different replicates at those stations. Sampling variation at each of these levels will contribute to overall sampling uncertainty in metric calculation at the water body level. Such variation can be analysed using mixed-effects (multilevel) statistical models. Calculation of variance by averaging at successively higher levels in the hierarchy does not give correct results.
Outcome and products
A first workpackage 6.1 workshop was already run during the Kick-off Meeting in March 2009 (see Deliverable 6.1-1), which brought together scientists from all relevant workpackages to compare approaches for uncertainty estimation and to raise the awareness of uncertainty issues. Based on this workshop, each work package in Module 3 and 4 developed a sampling programme which could address the various potential sources of sampling uncertainty. A second workshop was recently run during the WISER mid-term meeting in Debe, Poland (6–9 Sep 2010). The workpackage collaborators re-iterated the importance of quantifying variation in metrics due to sampling variability, and presented examples of variance components analysis using mixed-effects models.
Consistency of the datasets and coding systems is vital, and is aided by the planning that has gone into the field campaign and the presence of an over-arching data management in workpackage 2.1. An example has been elaborated for lake chlorophyll (ISO method) data in workpackage 3.1 at five hierarchical levels: country, lake, station, replicate and sub-sample. Preliminary results demonstrated that approximately 90% of the variation was between lakes, with the remaining 10% being partitioned between stations, replicates and sub-samples.
The example for lake macrophytes compared sampling variation in Lake Trophic Rank using two hypothetical sampling strategies, firstly sampling in the nearest zone to the shore (0–1m), and secondly sampling from shore to 4m depth. The example for lake fish demonstrated how sampling variation in fish metrics within and between lakes can be analysed simultaneously with environmental and pressure data for lakes.
Uncertainty analysis in WISER is implemented using a specific software: WISERBUGS.
The software allows the effects of sampling uncertainty on status classification to be quantified. This includes the calculation of multimetric indices and rules for combination of different BQEs.