1. Multiple biological invertebrate traits (e.g. body size, body form, dispersal potential) each described through multiple categories (e.g. small, intermediate or large body size) could serve as indicators of particular types of human impacts on large rivers. The trait composition of natural invertebrate communities is scarcely constrained by taxonomic differences among them, i.e. individual trait categories could be used to discriminate various types of human impacts across large geographic areas, which would require the definition of trait patterns for conditions of relatively low human impact.
2. Using large databases to link 14 biological traits (described through 66 categories) of invertebrate genera to their occurrence in running water reaches with known environmental conditions, we examined the accuracy of various approaches to predict expected trait variation across least impacted river reaches (LIRRs) of Europe in a stepwise analytical procedure. This procedure included Monte Carlo simulations and ultimately the assignment of test-LIRRs (reaches not used in previous analyses) to the previously defined LIRR conditions.
3. Distance from the source was an integrative variable capturing some (but not all) landscape features (e.g. altitude) or habitat variables (e.g. reach shear stress). Correspondingly, the relative abundance of many trait categories changed along 13 European running waters, although particularly the intensity of these changes differed among these 13 running waters.
4. ‘Downstream models’ (using only distance from the source as predictor) provided the least accurate predictions of expected invertebrate trait patterns when compared with ‘landscape models’ (using distance from the source in combination with altitude and/or latitude) or ‘habitat models’ (using reach shear stress, mean annual air temperature and/or pH of the water). Landscape models provided more accurate predictions than habitat models, but the improvement of predictions of expected invertebrate traits patterns obtained using landscape models was negligible in comparison with a simpler ‘mean-model’ approach, suggesting that defining LIRR conditions through simple descriptions of frequency distributions would be sufficient for the future biomonitoring of large European rivers.
5. To define these LIRR conditions, we used the average of the relative abundance of each trait category from 68 LIRRs (≥40 m wide) as expected LIRR values, and computed LIRR frequency distributions that described the deviations of the 68 individual LIRRs from these expected values. Computing such deviations from the expected LIRR values for 57 test-LIRRs (also ≥40 m wide), 57 trait categories correctly assigned >90% of the test-LIRRs to LIRR conditions if the latter were defined through the entire range of the LIRR frequency distributions. To the 90%-range enveloped by the LIRR frequency distributions, 42 trait categories correctly assigned >80% and 12 categories >90% of the test-LIRRs.
6. Using a framework that required no regionalisation of a large geographic area, no modelling of expected values using environmental information and no standardised invertebrate sampling, the performance of our trait approach to assign test-LIRRs to LIRR conditions encourages future assessments of deviations from these defined LIRR conditions in large European river reaches with different types of human impacts.