The STAR common metrics approach to the WFD intercalibration process: Full application for small, lowland rivers in three European countries

Class boundaries of three European assessment systems based on macroinvertebrates were compared and harmonized. Three different approaches to comparison, one based on regression analysis and the other two on statistical testing, were described and used, however only one was considered useful for the harmonization of boundaries. In all cases, the calculations were based on a set of six Intercalibration Common Metrics, combined into a simple multimetric index (ICMi). The ICMi was calculated for three test datasets from Italy, Poland and the UK, all belonging to the same stream type (small lowland siliceous sand rivers). For comparison, a regression model was employed to convert national assessment boundary values into ICMi values. The ICMi was also calculated on samples included in a strictly WFD-compliant benchmark dataset. The values of the ICMi obtained for the quality classes Good and High for the test and benchmark datasets were statistically compared. When significant differences were observed in the harmonization phase, the boundaries of the national method were refined until no further differences were observed. For the test datasets and assessment systems of Italy (IBE index) and Poland (Polish BMWP index) small refinements of the boundaries between High/Good and Good/Moderate classes were sufficient to remove the differences from the benchmark dataset. After harmonization, in the studied stream type, the percentage of samples requiring restoration to Good quality increased by 22 and 6% for Italy and Poland, respectively. For the UK dataset (EQI ASPT) the comparison to benchmark dataset showed no significant differences, thus no harmonization was proposed. A general discussion of the options used to compare boundaries based on the ICMi and their potential for harmonization is provided. Lastly, the option of harmonizing class boundaries through comparison to an external, benchmarking dataset and then re-setting them until no differences are found is supported.