Fig 1.
Description of the study design.
A) The taxonomic composition of benthic macroinvertebrates at each sampling site was assessed with two methods: Kick-net and eDNA sampling. Subsequently, the focal index on the biological state (IBCH Index) was calculated from kick-net and eDNA data. B) Map of Switzerland showing the spatial setup of the biomonitoring sampling sites. Sampling sites are given as black points overlaid on the main network of rivers and lakes. Different blue shading highlights major catchments of Switzerland.
Fig 2.
Filtering of raw sequencing data.
Distribution of A) mean read number per sampling site and B) mean OTU number per sampling site using thresholds based on detection rate in the four field filter replicates. In the violin plots, the black dots indicate A) the mean read number over all sampling sites and B) the mean number of OTUs over all sites, the black vertical lines span the 95%-quantiles of all values. The detection rate is given with increasing stringency: Detection of an OTU in at least 1, 2, 3, or 4 out of 4 filter replicates per site, respectively. C) Read abundance distribution of OTUs (n = 4599) detected in at least 2 out of 4 replicates per site. Read abundances of OTUs that were taxonomically assigned to indicator taxa are highlighted in green.
Fig 3.
The taxonomic richness of indicator groups in Swiss rivers at each sampling site based on A) kick-net monitoring and B) eDNA monitoring. For the latter, only macroinvertebrates also considered in the traditional biological assessment are included. The color gradient is adjusted to the respective range of indicator richness values.
Fig 4.
Proportions of indicator groups detected by kick-net versus eDNA monitoring. The stacked bars indicate proportions of indicator groups inferred by the two methods (proportion of counts for kick-net and of reads for eDNA). For the most common indicator groups, names are given. The flows between the two stacked bars connect families within indicator groups between the methods. A change in flow width indicates a change in proportion depending on the method used.
Fig 5.
Biotic index based on bioindicators.
Comparison of the index on the biological state (IBCH index) based on kick-net-derived scores (IBCH index observed) versus the predicted index derived from eDNA data. The predictions are the output from a random forest model deriving IBCH index scores using OTU presence-absence as input. A linear regression model gives the relationship between observed and predicted values (adjusted R2 = 0.61, p-value < 0.001). The colored boxes summarize the numerical index scores ranging from 5 to 20 into categories ranging from “unsatisfactory” to “very good”.
Fig 6.
Distribution of predicted classifications of the biotic state.
Comparison of the biological state of sampling sites when comparing classifications based on kick-net or random forest predictions. A) The density distributions for the observed (kick-net-based, grey) and the predicted (eDNA-based, green) IBCH index scores. The x-axis indicates the range of the biological index from 5 to 20. B) Barplot showing the percentage of sites that fell in the same (x = 0) or different (x ≠ 0) category by the random forest predictions based on eDNA data compared to the kick-net-based classifications. The majority of sampling sites were classified in the same category (72%). All other sites (28%) were maximally deviating by one category.