Seasonally distinct taxonomic and functional shifts in macroinvertebrate communities following dam removal

Study system and experimental design

The 5th Avenue Dam (∼2.4-m high, and 143-m wide) was removed in September 2012 to improve water quality and aquatic habitat in the 5th-order Olentangy River, Ohio, USA (39°59′N, 83°01′W; US Army Corps of Engineers, 2004; Fig. 1A). We studied effects of its removal upstream and downstream using three 360-m study reaches: one reach ∼400 m downstream of previous dam location, and two reaches upstream of the previous dam (i.e., the former reservoir) (Fig. 1B). One of the upstream reaches (∼1,300 m from previous dam) was restored after dam removal (hereafter: upstream restored; Figs. 1C–1E), including river-channel engineering to redevelop and reconnect floodplain wetlands (Ohio Environmental Protection Agency, 2011). The second upstream reach (∼2,300 m from previous dam) was allowed to adjust naturally (hereafter: upstream unrestored). Prior to dam removal, the downstream reach was a shallow riffle whereas the two upstream reaches were characterised by deeper, slower moving-to-impounded water, and wide channel widths (e.g., Fig. 1C). We collected samples from three sub-sites located at the upstream, middle, and downstream sections of each study reach for sub-site level replication (see Methods: Benthic Macroinvertebrates).

Common logistical constraints associated with studying dam removal were associated with our study; for example, we were unable to collect comparable benthic macroinvertebrate samples given the depth and benthic conditions in the study reaches prior to dam removal (see Methods: Benthic Macroinvertebrates; Fig. 1C). Nevertheless, our study encompassing three impact reaches with multiple samples in space and time is consistent with dam-removal study designs in the literature (Kibler, Tullos & Kondolf, 2011). Although we had no true control reach, fish-community data from a reach above an intact dam in the same river generally showed no differences before and during the three years following dam removal (Dorobek, Sullivan & Kautza, 2015), supporting dam removal as a major driver of ecological shifts in our study reaches. We also considered benthic-macroinvertebrate community data collected from a companion study in a reach directly below an intact dam in the Olentangy River (Fig. 1A; Vent, 2015). We used these data to represent the reference state of benthic macroinvertebrates in the study system to evaluate the practical significance (Kibler, Tullos & Kondolf, 2011) of dam removal effects on benthic macroinvertebrates relative to observations below an intact dam during the same time as our study. These reference data were collected in June, August, and November 2014, and March and June 2015 using comparable methods to those outlined below (i.e., Surber samplers; see Methods: Benthic macroinvertebrates, Vent, 2015).

Benthic macroinvertebrates

We established three transects (upstream, middle, downstream) extending laterally across the river channel (i.e., from left to right bank) at each study reach. Benthic macroinvertebrates were collected at random points along each transect using a 0.30-m² Surber sampler (500-µm mesh) for 90 s following Sullivan & Watzin (2008) (n = 3 per study reach). Starting in June 2013, we consecutively sampled each reach 10 times over three years in four seasons: late autumn (December 2013 and November 2014), early spring (April 2014 and 2015), late spring (June 2013 and 2014), and summer (August 2013, 2014, 2015) capturing both seasonal high (spring) and low (summer, late autumn) flows (see also Fig. S1). Note that we also opportunistically sampled in July 2014. Samples were preserved in 70% ethanol, and identified to the lowest taxonomic resolution possible by Rhithron Associates, Inc. (Missoula, Montana, USA). The most common taxonomic resolution was genus, including genus-level identification for Chironomidae in 98% of cases. Invertebrates were collected under Ohio Division of Wildlife Wild Animal Permit 15-49.

Numerical and statistical analysis

We calculated macroinvertebrate generic richness (R), Shannon–Wiener Diversity Index (H′), and evenness (E) for the three transects at each study reach and time interval (except for in Jun–Dec 2013 when subsamples were mistakenly combined), from which we computed a reach-specific mean and standard error. The Shannon–Wiener Diversity Index is indicative of both taxonomic richness and evenness, such that, (1)${\displaystyle H^{\prime }=-\sum _{i=1}^g{p}_i\ast ln{p}_i}$ where p_i is the proportion of the total sample represented by genus (g) i. We computed generic evenness as (2)${\displaystyle E=\frac{H^{\prime }}{H_{max}}}$ where H′ is the Shannon-Wiener Diversity Index, and H_max is the natural log of generic richness, such that evenness ranges from 0 to 1, with values approaching 1 indicating greater generic evenness in the community.

We also quantified the functional structure of the macroinvertebrate communities after dam removal (Tullos, Finn & Walter, 2014). We assigned each taxon to eight life-history (e.g., voltinism), morphological (e.g., attachment, armoring), and ecological (e.g., functional feeding groups, rheophily) traits, using genus-level classifications according to Poff et al. (2006) (Tables S1 and S2). We selected traits based on their expected ability to indicate physical or ecological changes associated with dam removal (e.g., predominant attachment traits could indicate altered streamflow velocities).

For macroinvertebrate density, taxonomic richness, diversity, and evenness, we used general linear models (GLMs) with time and reach as predictor variables where both were considered fixed effects. Visual inspection of our data revealed potential non-linear responses to dam removal through time. In these cases, we used breakpoint regression (Muggeo, 2003; Dodds et al., 2010) to estimate possible threshold responses (i.e., point at which there is an abrupt ecological change; Dodds et al., 2010). We compared evidence for linear models vs. piece-wise linear models using Davies tests (Davies, 1987; Davies, 2002) to explicitly evaluate non-zero differences in slope through time. We also tested for differences in mean macroinvertebrate density between or among comparable seasons (e.g., late spring 2013 vs. late spring 2014) categorically using orthogonal contrasts (Dowdy, Wearden & Chilko, 2004). Data were ln-transformed (or ln[x + 1] in the case of relative abundance) to meet assumptions of normality and homogeneity of variance when deemed appropriate after visual inspection of residuals versus fitted values.

We used non-metric multidimensional scaling (NMDS; Clarke, 1993) followed by analysis of similarities (ANOSIM) to assess differences in macroinvertebrate communities among the three reaches, and through successive months after dam removal. NMDS incorporated Bray-Curtis dissimilarities between relative densities (no. ind m ⁻²/total no. ind m⁻²) of genera for a given month or reach and was limited to a two-dimensional solution. To evaluate community shifts through time, we used the function ordiellipse in R to compute centroids of ellipses around ordination points grouped by month. We then computed the Euclidian distance traveled by each ellipse from year to year, and compared these distances between years within seasons ad hoc. We also compared each seasonally distinct centroid to the final sampling date centroid (summer 2015).

In an effort to account for patterns attributable to the dominance of specific macroinvertebrate families, we conducted NMDS and ANOSIM with and without Chironomidae, the most abundant family observed throughout the study (see below: Results). Chironomidae is a taxonomically and functionally diverse family (>1,000 species in North America; Ferrington, Berg & Coffman, 2008) that occurs in high densities, suggesting that changes to their relative abundance might contribute disproportionately to observed responses. Lastly, given the potential lack of independence among study reaches, we tested for spatial autocorrelation (Moran’s I) among response variables, and found no evidence for non-random spatial patterns in benthic macroinvertebrate responses including density, richness, and diversity (P > 0.05 in all cases).

All statistical analyses were performed using R statistical software v.3.3.0 (R Core Team, 2016). We used the R package ‘segmented’ to estimate possible breakpoints in the relationship between macroinvertebrate variables and months after dam removal (Muggeo, 2003; Muggeo, 2008). We used the R package ‘vegan’ (Oksanen et al., 2013) to generate and analyse ordination data for macroinvertebrate communities through time, and among our study reaches. In all cases, P < 0.05 was considered evidence of statistical significance, and P < 0.10 was considered evidence of a trend.

Benthic macroinvertebrate density, diversity, and evenness

We identified 7,695 macroinvertebrates across all reaches and years. Mean (±SE) macroinvertebrate density was 4,359 ± 101 ind m⁻² across all reaches and time points. There was a consistent pattern of decreasing macroinvertebrate density between 9 (summer 2013) and 15 (late autumn 2013) months after dam removal in all three reaches. We observed the lowest mean densities ∼19 (early spring 2014), ∼15 (late autumn 2013), and ∼21 (late spring 2014) months after dam removal in the upstream-unrestored, upstream-restored, and downstream reach, respectively. Macroinvertebrate densities subsequently increased through time at both upstream reaches beginning in early spring 2014 but not at the downstream reach (Figs. 2A–2C). The estimated changepoint (i.e., threshold response) occurred 15 months post-dam removal in both the upstream-unrestored (Davies test: P = 0.041) and restored (Davies test: P = 0.065) reaches. There was no significant change from decreasing to increasing macroinvertebrate density in the downstream reach (Davies test: P = 0.114). Although we had no true control data, macroinvertebrate densities generally rebounded to approach values found in a reference reach located directly below an intact dam (3,000 ± 635 ind m⁻²; n = 5; Vent, 2015) (Figs. 2A–2C) by the end of the study.

Macroinvertebrate density was highest in late spring during the first year of the study, but declined during this season in the second year in the downstream reach (orthogonal contrasts: P < 0.05; Fig. 3). Mean macroinvertebrate density was lowest in late autumn compared to the other seasons, but increased marginally in the following year in the upstream-restored reach (orthogonal contrasts: P = 0.089; Fig. 3). Summer densities also tended to increase, but not significantly, particularly in the upstream-unrestored reach (Fig. 3).

Similar to macroinvertebrate density, generic richness generally decreased between ∼9 (summer 2013) and ∼15 (late autumn 2013) months after dam removal, and richness subsequently increased through time (Figs. 2D–2F). In contrast to macroinvertebrate density, we detected no threshold response of generic richness in the upstream unrestored reach (Davies test: P = 0.141, Fig. 2D), while there was a significant changepoint in the richness versus time relationship 15 months following dam removal in the upstream-restored reach (Davies test: P = 7.365 × 10⁻⁷, Fig. 2E). The threshold for the change from increasing to decreasing generic richness was delayed (∼19 months after dam removal) in the downstream reach (Davies test: P = 0.056, Fig. 2F).

We found evidence for seasonal differences in generic richness, but not Shannon–Wiener diversity. Generic richness was lowest in late autumn during the first year of the study, but increased significantly during this season in the second year (orthogonal contrasts: P = 0.047; Figs. 2D–2F). Generic richness tended to be highest in late spring to summer, but was comparable across years (orthogonal contrasts; all P > 0.1; Figs. 2D–2F). Shannon-Wiener diversity (H′) also generally decreased between ∼9 and ∼15 months after dam removal, and subsequently increased through time (Figs. S2A–S2C). Similar to generic richness, we detected no threshold in the trajectory of H′ in the upstream unrestored reach (Davies test: P = 0.277, Fig. S2A), while there was a significant threshold response in H′ 15 months following dam removal in the upstream-restored reach (Davies test: P = 0.008, Fig. S2B). The change from decreasing to increasing H′ was delayed (∼19 months after dam removal) and significant in the downstream reach (Davies test: P = 0.017, Fig. S2C). In contrast to both density and generic richness, we found no evidence for seasonally distinct changes in Shannon–Wiener diversity (Figs. S2A–S2C; orthogonal contrasts: all P > 0.10). Generic evenness was relatively stable through time after dam removal, and this pattern was consistent among the three reaches (Fig. S3A–S3C). Generic evenness increased slightly in spring and summer in the upstream unrestored and downstream reaches (Figs. S3A–S3C).

Macroinvertebrate community composition

The four most abundant families surveyed were Chironomidae (3,793 individuals), Hydropsychidae (1,736 individuals), Elmidae (625 individuals), and Baetidae (513 individuals), which represented 47, 23, 8, and, 7% of all macroinvertebrates, respectively. The taxonomic composition of the macroinvertebrate community differed across months after dam removal (NMDS: stress = 0.186; ANOSIM: R = 0.54, P = 0.001; Fig. 4A), but showed no differences among all reaches (ANOSIM: R = 0.0005, P = 0.452; Fig. 4B). Patterns of community structure were consistent between the chironomid assemblage and the complete macroinvertebrate community (see Fig. S4A and S4B). We found no evidence for changes in the relative abundance of the four most prevalent families through time (Figs. S5A–S5D; GLMs: all P > 0.1).

Macroinvertebrate community similarity responded differently depending on season and year (Fig. 4A). The largest distances between the same season but different years were for early spring (distance = 0.99), followed by late autumn (distance = 0.94), summer (distance = 0.77), and late spring (distance = 0.46). The smallest distance between years occurred between summers 2014 and 2015 (distance = 0.07). Within each season, macroinvertebrate community structure tended to become more similar to our final sampling date (summer 2015; Fig. 4A), where the distance between summer 2015 and late autumn 2014 was the smallest (0.32), followed by the distance between summer 2015 and early spring 2015 (0.47), then summer 2015 compared to late spring 2014 (0.75).

Functional traits

We found no differences in the trajectory of relative functional-trait abundance among the three reaches; therefore we pooled the data across reaches for the functional analyses. In general, macroinvertebrate community functional traits shifted toward greater representation by traits such as semi- or univoltine life history, greater affinity for attaching to substrates, stronger armoring, and preferential use of erosional habitat (Table S1) based on declining relative abundance of multivoltine, free-ranging, poorly armored, depositional taxa (Figs. S6A–S6D; GLM: all P < 0.05) most strongly expressed by Chironomidae (Fig. S6E–S6H). The three most abundant FFGs were collector-gatherers (5,016 total individuals), collector-filterers (1,841 individuals), and herbivores (scrapers, piercers; 773 individuals), which represented 65, 24, and 10% of all macroinvertebrates, respectively. Among the functional feeding groups, relative abundance of collector-gatherers declined by ∼3.8% per month after dam removal (GLM: P = 2.16 × 10⁻¹²), while all others increased by a similar magnitude (i.e., collector-filterers, herbivores, predators, and shredders (detritivores); all P < 0.05; Table S1). There were also seasonal differences in the relative abundance of several functional traits (Table S2). Among our focal traits, there were significant declines in multivoltinism during early spring and late autumn (orthogonal contrasts: P = 0.0002, 0.019, respectively; Table S2). We also found decreased relative abundance of taxa that commonly occur in the drift, free-ranging taxa, poorly armored taxa, and those that prefer depositional habitat during all seasons except late spring (orthogonal contrasts: all P < 0.05; Table S2).