Continuous riverine biodiversity changes in a 10‐years‐post‐restoration‐study—Impacts and pitfalls

Evaluating river restoration effects over several years is the exception rather than the rule. The benthic invertebrate fauna of three small mountain streams was investigated yearly from 2010 to 2019 following remeandering measures. Additionally, upstream near‐natural reaches were studied following a Before‐After‐Control‐Impact (BACI) design. Species richness and Ephemeroptera, Plecoptera and Trichoptera richness decreased strongly immediately after restoration but had positive effect sizes in the following 6 years. Abundances increased in all sites after restoration. These patterns were consistent also in the upstream near‐natural reaches, except for the decrease in richness in the second year, indicating that other factors beside the restoration affected the sites. A large flood event coincided with the implementation of the restoration measures depleting also the near‐natural sites. The similarity between paired reaches showed a sharp decline in the first year after restoration, followed by a direct increase, which indicates fast recolonization from the upstream reaches. Community composition analysis showed a shift of all communities with the time axis, underlining a substantial effect of external factors. Generalized linear mixed effects models exhibited that the percentage of tree cover and riparian vegetation had significant effects on changes in richness and abundance. Clear‐cut logging in the floodplains to restore natural floodplain forest supposedly increased water temperatures because many cold‐adapted upstream species were replaced by species naturally inhabiting more downstream reaches. The results emphasize that multiannual samples and a BACI design are necessary to understand restored systems. Furthermore, floodplain restoration and natural hydrology often shape benthic invertebrate communities more than pure instream restoration measures.


| INTRODUCTION
River restoration is a hot topic in the recent two decades (Beechie, Pess, & Roni, 2008;Palmer et al., 2005) and the number of studies on the effects of river restoration measures is strongly increasing (Feld et al., 2011). There is a large diversity in effect studies covering time scales, biotic groups or morphological changes. Most frequently, short-term effects are investigated, targeting effects within the first 2-5 years after restoration (e.g., Howson, Robson, & Mitchell, 2009;Nilsson et al., 2015;Tikkanen, Laasonen, Muotka, Huhta, & Kuusela, 1994). Longer time scales of 10-20 years after restoration are rare (but see Muotka, Paavola, Haapala, Novikmec, & Laasonen, 2002;Louhi et al., 2011). This problem is inherent to the relatively new research topic. Furthermore, besides the relatively short time between restoration and investigation, most studies investigate only 1 or 2 years or seasons. Thus, long-term data sets are largely missing, that is, studies of restoration effects for a time span of more than 5 or 10 years (but see Friberg, Kronvang, Hansen, &Svendsen, 1998 andLouhi et al., 2011). To overcome this shortcoming, several authors focused on meta-analyses with data sets of different restoration measures and different time frames after restoration (Kail, Brabec, Poppe, & Januschke, 2015;Lorenz, Haase, Januschke, Sundermann, & Hering, 2018;Miller, Budy, & Schmidt, 2010). Nonetheless, these do not account for interannual variation, the stability of biotic communities and the inherent succession (Dyste & Valett, 2019).
Fish and benthic invertebrates are the most frequently studied response organism groups. For both taxonomic groups the overall impression is that the list of investigations is long but the list of really observed biotic improvement is very short (Feld et al., 2011;Louhi et al., 2011;Nilsson et al., 2015). Nonetheless, there is a bias in the underlying datasets. Mostly, restoration measures are implemented in short sections of vastly degraded river systems. Thus, the expectation of an improving community might be unrealistic (Parkyn & Smith, 2011), as the chance of sensitive species reaching restored sections is low, not to mention their establishment. Furthermore they often need continuously good water quality (Sundermann, Gerhardt, Kappes, & Haase, 2013) and specific habitats (Pilotto, Bertoncin, Harvey, Wharton, & Pusch, 2014). Thus, the question remains, how does biology react if a restoration measure is implemented in a river system, which is only under minor anthropogenic pressure or in a system where near-natural morphology is the standard and degraded sections are the exception? Dispersal constraints or missing species pools (Sundermann, Stoll, & Haase, 2011) are not a limiting factor, which might probably lead to a fast recolonization by sensitive species from upstream and downstream sections.
Morphological restoration comprises many different targets and techniques. The removal of a large dam can be seen as the one end of the gradient, and abandoning management of the river banks, a passive restoration measure, as the other end. Effects of those measures are for sure different, manifest within greatly different time intervals and act differently on biological groups. Besides the longitudinal aspects of, for example, dam removal, remeandering measures or improvement of instream habitats, the lateral aspect, that is, floodplain restoration, is often overlooked. Floodplain restoration clearly affects riparian vegetation composition and structure (Göthe, Timmermann, Januschke, & Baattrup-Pedersen, 2016;Modrak, Brunzel, & Lorenz, 2017) and might also affect aquatic communities, as differing forms of landuse have substantial effects on aquatic communities (Quinn, Cooper, Davies-Colley, Rutherford, & Williamson, 1997). Not only the agricultural or urban impact needs to be considered here, but also clear-cutting in forested catchments (Noel, Martin, & Federer, 1986). If those general landuse changes detrimentally influence benthic invertebrate communities then floodplain restoration which is a form of landuse change might also have a substantial impact.
A key factor influencing the distribution of benthic invertebrates in river systems is water temperature (Caissie, 2006). Illies and Botosaneanu (1963) developed the Rhithral-Potamal concept mainly based on longitudinal changes of water temperature parameters. For central European benthic invertebrate species the longitudinal preferences are well known and summarized in an online platform (Schmidt-Kloiber & Hering, 2015). But water temperatures increase unnaturally when floodplain vegetation is altered, for example, by changes to agricultural area or by clear-cutting (Brown & Krygier, 1970). Thus, the instream communities might be affected by restoration measures conducted in the floodplain.
This double influence of longitudinal restoration, that is, morphological restoration in the river bed and lateral restoration, that is, floodplain restoration is rarely considered.
The Arnsberg forest is a large forested area in Western Germany.
Morphology of most of the streams is near-natural and only short sections have been previously straightened. Pre-restoration evaluation of the invertebrate communities displayed an overall good ecological status of the catchments. Nonetheless, funded by an EU-life project (www.life-bachtaeler.de), the remaining straightened sections were restored and a general change in the floodplains was initiated from non-native coniferous trees to indigenous deciduous trees. This gave the chance to investigate restoration measures in a more natural surrounding. Benthic invertebrate community patterns were investigated continuously at three sites starting before the restoration measures in the year 2010 until 2019 (1 year before; 10 years of sampling in total).
Besides the restored sites upstream near-natural sites were sampled allowing for a Before-After-Control-Impact (BACI) design in the analysis. The objectives of the study are to shed light on short-and longterm effects of restoration measures in a more natural environment comparing the influence of instream and floodplain restoration. The following questions will be answered: How do the benthic invertebrate communities respond to the morphological changes over the time span of 10 years? And, how do the benthic invertebrate communities respond to the landuse changes in the floodplain?

| MATERIAL AND METHODS
Three small mountain streams were investigated in the Arnsberg Forest in the western central part of Germany (Table 1). Catchment sizes ranged between 5 km 2 (Kleine Schmalenau) and 47 km 2 (Heve), with catchments being dominated by coniferous forest with urban settlements and agriculture almost completely lacking. Forests are dominated by spruce (Picea abies), which is not native in this region. The geology is schist (siliceous) and the streams' substrate is cobble-dominated with minor percentages of sand, woody debris and macrophyte patches.
Mean annual rainfall for the area is 900 L/m 2 and mean annual temperature is 8.4 C. The streams are naturally perennial though may be partly reduced in extraordinary hot and dry summers to disconnected pools due to ceased run-off. year. A standardized multi-habitat-sampling was conducted (Meier et al., 2006) by taking 20 subsamples (shovel sampler: 25 × 25 cm, 500 μm mesh size) concordant to the distribution of substrates present at each site. The substrate distribution was recorded in 5% steps.
The sites' subsamples were pooled, conserved with 90% ethanol and transported to the lab for sorting. In the lab, a standardized subsampling procedure was applied (Meier et al., 2006). The specimens were identified to the lowest level possible, mainly species and genus with the exception of Chironomidae (tribe level) and Oligochaeta (family level), according to the operational taxa list for Germany (Haase et al., 2004) (Table S1).
On-site pH, conductivity and oxygen were measured for background information.   For analysis, the environmental data, that is, physico-chemical variables, substrate estimation percentages and shading percentages were standardized and z-transformed. Collinearity was explored using Variance Inflation Factor (vif function in R-package usdm). Collinear variables were subsequently removed. The remaining variables (Table S2) were then used as explanatory variables in generalized linear mixed-effects models (GLMM) applying the lmer function of the lme4 package for R (Bates, Mächler, Bolker, & Walker, 2015). Following the cookbook of Feld, Segurado, and Gutiérrez-Cánovas (2016)   Physico-chemical variables showed only neglectable variations between sites and years (Table 1).
Loam and clay represented a considerable percentage of the substrates in the restored watercourses, while gravel had a negative effect size ( Figure S1). Macrophytes and living parts of terrestrial plants increased in percentage particularly in the second part of the 10 years study.
Tree-cover decreased in the floodplains during the investigation time span due to large scale logging of the coniferous forest ( Figure 2). The logging was part of a forest management plan to foster natural floodplain vegetation. In general, 5-10 species more inhabited the near-natural sites than the respective restored sites until 2016, by when the restored sites' species richness matched those of the near-natural sites ( Figure S2). Wilcoxon-test showed no significant differences in taxa richness, EPT richness or abundance comparing the differences of 'after' minus 'before' of restored sites with the 'after' minus 'before' of control sites.

| Community descriptors
In all three GLMM models, using taxa richness, EPT richness and abundance as response variables, the percentage of living parts of terrestrial plants and the percentage shading 500 m upstream of the sampling sites were important and often significant descriptors for the changes (Table 2). Furthermore, gravel, loam/clay and cobbles occurred in the models. AIC was low indicating a better fit in the models for taxa numbers and EPT richness (121.8 and 98.5 respectively) but high in the model for abundance (255.0).
In the first year after restoration, the community similarity to the upstream near-natural sites dropped by 31.4% in the Kleine Schmalenau and by 16.1% in the Große Schmalenau (Figure 4). In the second year after restoration, Bray-Curtis similarity directly increased F I G U R E 3 Change expressed as effect size ("after" minus "before") of taxa richness, EPT richness and abundance in the three restored sites and the nearnatural sites between 2011 and 2019. Symbol/colour/linetype codes express: Triangle/blue/dashed = near-natural, circle/red/dot-dashed = restored. Wilcoxon tests showed no significant differences in richness, EPT richness or abundance in the BACI design comparing the differences of before minus after of restored sites with the before minus after of control sites [Colour figure can be viewed at wileyonlinelibrary.com] by 26.4% and 9.7% respectively. In the following years, the similarity levelled out to on average 65%.
Over the whole 10-year period, the NMDS showed that samples of the same year of restored and respective near-natural sites cluster close to each other ( Figure 5). However, samples from the first years cluster far away from samples of the later years. Additionally, when the factor "years after restoration" is overlaid on the NMDS ordination space, it aligns predominantly with axis one. This axis explains more than 47% of the variance in the long-term community composition.
The change of the invertebrate community is also observed from the longitudinal preferences of the species (Figure 6). In 2010, the mean share of epirhithral preferences (preference for small mountain brooks) was 26.2%. This decreased continuously by 10% until 2019. A decrease can also be seen in the average metarhithral preferences (lower trout region).
In contrast, hyporhithral and epipotamal (grayling and barbel region) preferences increased from mean values of 17.7 and 8.6%, respectively, to 21.8 and 14.1%. This change of longitudinal preferences is inherent to all sites irrespective of their restored or near-natural status.
Furthermore, epirhithral and metarhithral preferences are positively correlated to the percentage of tree cover (Table 3) This is supported by Stoll, Breyer, Tonkin, Früh, and Haase (2016) who found that if local degraded stream morphology is imbedded in regional good morphology then community improvement will be most likely successful after restoration.
Overall, the system shows large changes in community composition and a high beta diversity. Interestingly, species numbers and abundances increased also in the near-natural sites, indicating that restoration was not the only factor influencing biodiversity pattern. This is supported by the results of the tests in the BACI Design. No  Note: The year of investigation and the sites within streams were included as random variables; significant correlations are in bold. Gravel (diameter 6-20 cm); LPTP (living parts of terrestrial plants, roots), cobbles (diameter 0.2-2 cm); shading 500 m (% shading 500 m upstream of the sampling site). *p < .05. **p < .01.
have been investigated with a before-after-comparison the results might have led to a fallacy.
Abundance patterns followed species richness patterns in the first years. Abundance increases in restored reaches were also described by Kail et al. (2015), as one of the few benthic invertebrate community patterns that changed significantly due to restoration. A potential cause for the multi-annual accrual is the flood of spring 2011, which has depleted the benthic invertebrate fauna in all sites; in subsequent years, species re-established and increased in abundance. Abundances boosting in the years after large flood events are reported several times in the literature (see Death, 2008 for a review).
Starting from 2016 species richness declined at all sites. This pattern was most probably fuelled by the summer drought of 2018, when a loss of more taxa, particularly many EPT, was observed. Beetles and dipterans were less influenced and had only minor decreases. In contrast, abundances exhibited high increases from 2016 to 2019, which was also seen by Stone and Wallace (1998) in an effective study of forest succession after logging on the invertebrate community structure.
This abundance increase is particularly interesting in light of the drought of 2018. Except for some pools, the stream channels fell dry for several weeks. The majority of Central European taxa are not desiccation tolerant (Schmidt-Kloiber & Hering, 2015), thus many species obviously have not survived the drought. The recolonization was then supplemented by other taxa, particularly fostered by drifting individuals and aerial colonists like Simuliids and beetles (Brittain & Eikeland, 1988), who replaced EPT taxa. Furthermore, the logging in the floodplains reduced substantially the shading on the sampling sites, which enhanced algae growth (Noel et al., 1986). Elmidae beetles feed by grazing ( et al. (1998). They assumed that river maintenance stopping was the main cause and then macrophytes increased habitat heterogeneity fostering richness and abundance. Interestingly, they found these patterns also in their control site which supports the finding of this study that floodplain management and instream biotic habitats are significant drivers of biodiversity. Additionally, Miller et al. (2010) showed that species richness responded positively to restoration in forested catchments like in this study.

| Community changes
The communities of all sites, irrespective of restored or near-natural, shifted over the 10 years sampling period. The similarity of near-natural sites and reflects an increase in water temperature because the basis for the longitudinal preferences is water temperature preferences (Illies & Botosaneanu, 1963). This may be attributed to the loss of woody riparian cover by the logging activities in the floodplains and thus water temperatures potentially have increased due to less shading (Noel et al., 1986;Ringler & Hall, 1975). Logging activities were particularly undertaken between 2011 and 2014 and decreased the woody riparian buffer by on average 50%. Interestingly, the invertebrate communities followed the presumed increase of water temperatures with a time lag of 1 year, which is equivalent to the 1-year life cycle of most species. Thus, species preferring colder stream temperatures were replaced by species preferring warmer F I G U R E 5 Joint plot of NMDS results of the invertebrate samples of the three restored sites and the respective near-natural sites and the factor "years after restoration" as overlay. The coding explains the site and the respective sampling year. Symbol/colour codes express the groups: square/blue = year before restoration and the first year after restoration and respective samples in near-natural sites; triangle/ green = second to fifth year after restoration and respective samples in near-natural sites; circle/red = sixth to ninth year after restoration and respective samples in near-natural sites. Samples are abbreviated according to As part of the floodplain restoration not only logging was conducted but also reforestation. Native young alder trees have been planted and start shading the channels and provide coarse particulate organic matter for shredding invertebrates. A small increase in shading is already visible in the 2018 data for the Kleine Schmalenau.
Thus, future changes in the invertebrate community are to be expected, for example, an increase of shredder to grazer percentage (Hernandez, Merritt, & Wipfli, 2005;Stone & Wallace, 1998). Furthermore, Hernandez et al. (2005) showed in Alaskan streams that alderdominated forest succession after clear-cut enhanced densities of invertebrates and led to a richer and more diverse fauna. The question remains if future shading will decrease the stream temperatures again and subsequently cold-adapted invertebrates will recolonize the sites.

| CONCLUSION
This 10-years consecutive data set unveils several messages.

DATA AVAILABILITY STATEMENT
The benthic invertebrate data analysed in this study are provided as supplementary material.

ORCID
Armin W. Lorenz https://orcid.org/0000-0002-3262-6396 F I G U R E 7 Trajectory of the longitudinal preferences (in % of the community) of the invertebrate samples in the three restored sites in relation to the tree cover in the 500 m section of the sampling site between 2010 and 2019