Recruitment limitation is known to influence species abundances and distributions. Recognition of how and why it occurs both in natural and in designed environments could improve restoration. Aquatic insects, for instance, rarely re-establish in restored streams to levels comparable to reference streams even years after post-restoration. We experimentally increased oviposition habitat in five out of ten restored streams in western North Carolina to test whether insect egg-laying habitat was limiting insect populations in restored streams. A main goal was to test whether adding oviposition habitat in the form of rocks that partially protrude above the water surface could be used to increase the abundance and richness of stream insect eggs and larval insects in restored streams. Adding egg-laying habitat enhanced several response variables (e.g., protruding rocks, number of eggs, egg masses, egg morphotype richness, and oviposition habitat stability) to levels similar to those found in reference streams. Following the addition of protruding rocks, egg mass abundance increased by 186 % and richness increased by 77 % respectively in restored-treated streams. Densities of larval insects that attached their eggs to protruding rocks showed an overall pattern consistent with treatment effects due to the combination of non-significant and significant increases of several taxa and not just one taxon. Our results indicate that these stream insect populations are limited by oviposition habitat and that adding egg-laying habitat alleviated this component of recruitment limitation. However, the weaker larval response indicates that additional post-recruitment factors, such as egg or larval mortality, may still be limiting a full recovery of larval insect abundances in these restored streams. This study shows the importance of integrating information from animal life histories, ecology, and geomorphology into restoration practices to improve the recovery of aquatic insects, which are commonly used to assess water quality and the biological efficacy of stream restoration.

Site selection

We obtained a list of 378 North Carolina Department of Environmental Quality projects that included a stream component. To reduce variability, we selected streams with similar restoration priorities, land-use, and stream bed composition. We selected projects that utilized restoration priority type one or two (RP1 or RP2) of the four Rosgen (1997) developed options for the restoration of incised channels. These project types increased channel sinuosity, stream bed heterogeneity, and either elevated channels to reconnect them to floodplains (RP1) or created floodplains at the current stream elevation (RP2). We also selected projects 7–15 years post-restoration to allow sufficient time for riparian planting to mature, fine sediments to stabilize, detritus to accumulate, biofilms to establish, and some insects to recolonize. Further, we selected streams with similar catchment land-use composition (e.g., impervious surface less than 5 % and developed land less than 20 %). We identified large cobble- and boulder-sized rocks (128–256 and >256 mm) as the oviposition habitat for manipulation. Consequently, we excluded streams located in the coastal plain and with a median particle size (D50) < 2 mm (i.e., primarily sand-bed streams) because, given their hydrogeomorphic context, these streams often naturally lack this type of oviposition habitat and we wanted to keep oviposition habitat type similar across streams. Although stream insects use rocks, wood, and live vegetation as well as other materials for egg-laying (Smith and Storey 2018), we focused on rocks in this study because they were the most abundant habitat, more permanent than organic materials such as wood, and are usually available or added during stream restoration.

Ultimately, 12 restoration projects fit all requirements including access. Following physical visits, we selected 10 restored streams. The 10 streams were part of four restoration projects located in western North Carolina (Appendix S1: Figure S1). We grouped streams primarily by proximity to one another (Appendix S1: Table S1; Figure S1) and secondarily by stream characteristics, such as channel size and presence or absence of a culvert above the restored reach (Appendix S1: Table S1). We randomly assigned one stream from each of five groups to receive the treatment (addition of suitable rocks for oviposition) using a coin toss, resulting in five restored-untreated and five restored-treated streams (Appendix S1: Table S1). Two projects included multiple streams (at least one treated and untreated). For example, the two streams at Warrior Creek were located on different tributaries. Morgan Creek had four streams all located on different tributaries and the main channel included an upstream untreated and downstream treated section of stream. We were limited to three reference streams because of a lack of reference streams with comparable catchment size, forested watershed in reasonable proximity to the restored streams and limited access to private property (Appendix S1: Table S1; Figure S1). Reference streams were on large areas of state or federal government land that have provided watershed-scale protection from the land use changes affecting restored streams (e.g., cattle grazing and forest clearing for pasture and crops). The Indian Creek reference site was ~10 m upstream of a State of North Carolina Division of Environmental Quality (NCDEQ) reference site near Hanging Rock State Park and was near both Pinch Gut and Candiff Creek. Boone Fork Creek was nearest to Warrior Creek but at a higher elevation in the northeastern portion of Pisgah National Forest, Lenoir, NC. Beaverdam Creek was located east of Morgan Creek on the southern end of the South Mountains State Game Lands. One reference stream was included in each of groups four and five with restored-untreated and restored-treated streams, but the reference stream Beaverdam Creek was included in its own group, six (Appendix S1: Table S1). Groups 1–3 did not include a reference stream, resulting in a partially balanced incomplete block design (see Statistical Analysis). 

Experimental addition of oviposition habitats

To experimentally test whether adding protruding rocks, or potentially suitable oviposition habitat, increased the abundance and richness of insect egg masses and larval abundance in restored streams, we used a before and after repeated measures ANOVA incomplete block experimental design with three treatment types: five replicate restored streams in which we manipulated oviposition habitats by adding suitable rocks (restored-treated), five replicate restored streams that were unmanipulated (restored-untreated), and three reference streams (reference). We planned to treat the five restored streams in March 2020 before adult insect emergence, and census egg inputs from May to October, similar to 2019. However, due to the COVID-19 pandemic, treatment was delayed until May 2020 and egg mass censuses started in June 2020 (Appendix S1: Fig S2). Because of this delay and the seasonality of oviposition that includes a May peak, we omitted 2019 May data from the analyses and used egg censuses from June – October for both 2019 (before) and 2020 (after).

In May 2020, we added or repositioned rocks within the five restored-treated streams (Table 1; Appendix S1: Figure S3) to increase the availability of suitable oviposition habitat. Two people added rocks by hand to all five 50-meter-long sections of restored-treated streams in a single visit (Table 1). Rocks added were collected from nearby riparian areas or were formerly submerged within the channel; hence, protruding rocks were weathered and included rounded native stone but most were angular quarry stones added during restoration. Rock size (measured as the intermediate axis width) varied between 181–512 mm depending on availability within or near treatment streams and whether they were large enough to protrude above the water surface. We positioned rocks, with 25–30 % of the surface protruding above the water, in deep portions of riffles or other areas with fast to moderate water velocity (e.g., runs) where fine sediment deposition was low and thus the likelihood of rocks becoming embedded by sediment was low. Rocks were not placed in locations with low water velocity and high fine sediment deposition, such as pools, or low water velocity and shallow areas along the streambank where attached eggs are more likely to dry as water levels recede. We adjusted rocks throughout the study period (June-October 2020) to ensure the supplemental habitats remained suitable (i.e., protruding and unembedded) for egg-laying females. Our rationale for repositioning potential egg-laying habitat was to ensure the treatment was maintained. This decision reflects our aim of investigating the outcomes of maximizing availability of potential oviposition habitat to provide a “proof of concept” test of our hypotheses. The availability of these habitats is a dynamic naturally occurring process in streams (Ciotti et al. 2021), but this process may be delayed or impaired in many restored streams, as evidenced by the lack of egg-laying habitat in these streams (Jordt and Taylor 2021a). We used the mean number of suitable rocks as the response variable for protruding rock abundance, which was calculated as the sum of the number of suitable rocks per m2 stream area for each stream across all censuses for each year (omitting May 2019 census data), divided by the number of censuses for each year (2019 and 2020).

Protruding rocks absent on subsequent sampling dates were recorded as missing (i.e., completely buried or rolled out of the study area). Missing rocks were removed from the count of protruding rocks present but were included in counts if they were rediscovered on subsequent sampling dates. We noted whether rocks rolled (as evidence by the position of the painted number), were underwater, out of water and dry, and/or embedded in sediments. With the exception of treatment streams (during 2020), we did not unembed rocks to census them. In treatment streams in 2020, rocks that became unsuitable (i.e., they became embedded or submerged) were repositioned or unembedded throughout the 2020 egg-laying season in order to maintain a high level of suitable rock availability. Suitable rocks that became embedded or submerged were not considered as unstable, or included in the analysis of rock movement. Lastly, in restored-treated streams, rocks that were within the study area and became suitable for egg laying (became unembedded and protruding) or rolled into the study area were repositioned and counted as protruding rocks during the 2020 census. Rocks that became suitable (i.e., protruding) because water depth declined or because they rolled into the study area were not considered unstable unless subsequently buried, rolled, or missing. 

Egg mass and egg abundance, egg mass morphotype richness, and aggregation

In 2019, the year prior to adding rocks to streams, we censused a 50 m length of the ten restored and three reference streams for egg masses throughout the egg-laying period (May to October). Suitable egg-laying habitats were identified as protruding rocks with submerged spaces that adult insects could access along the sides and underneath (unembedded), as several previous studies have shown these specific rocks to be important stream insect egg-laying habitat (Peckarsky et al. 2000, Reich and Downes 2003b, Encalada and Peckarsky 2006, 2007, Miller et al. 2020). Protruding rocks were numbered using a nontoxic paint pen and removed from the water to identify egg masses to family or morphotype, count egg masses, photograph, and make sketches of mass type and locations on each rock. New masses on rocks were identified by comparing sketches to those made previously. Rocks were placed back into the stream in their original location and position. Individual egg masses were distinguished from one another by the gelatinous matrix, or spumaline surrounding each egg mass, and by size, shape, color, and spatial ordering of eggs within each mass. We returned every one to two weeks to census existing and newly protruding rocks for egg masses 10 times each in 2019 and 2020. We were not able to quantify or identify the eggs of some insects that lay individual eggs, egg masses in rock crevasses, or in moss growing on rocks (e.g., some Trichoptera, Tipulidae, and Elmidae).

Egg masses were identified to order using published descriptions and photographs (Murvosh 1971, Williams 1982, Nolte 1993, Peckarsky et al. 2000, Hoffmann and Resh 2003, Reich 2004, Reich and Downes 2004, Lancaster et al. 2010a, Lancaster et al. 2010b, Lancaster and Downes 2014, Smith and Storey 2018, Miller et al. 2020, Jordt and Taylor 2021a) or by identifying females actively ovipositing. Although family-level identification of eggs was possible for some taxa (e.g., Baetidae, Psephenidae), we could not identify the majority of egg masses beyond order, however, we were able to identify distinct morphotypes within orders based on color, shape (e.g., mat, globular, dendritic, strings, tombstone), texture (e.g., smooth, rough, firm) and relative size of the mass (e.g., globular tiny, globular strings, mat white, mat yellow). Distinguishing egg masses by morphotype provides finer taxonomic resolution when testing for changes in richness in restored-treated streams.

As the response variable for egg inputs, we used the mean number of egg masses (either total masses or by morphotype) per stream area, which was calculated as the sum of the number of egg masses divided by the stream area censused for each stream, divided by the number of censuses for each year (2019 and 2020). Egg masses per stream area censused was used rather than total stream area of each 50 m study section because, on some occasions (8 of 222), the entire 50 m study section could not be censused due to lightning storms. We also calculated the mean number of egg masses per individual rock area, as previous research revealed when rocks were scarce the density of egg masses on individual rocks is high (Jordt and Taylor 2021b) and more aggregated (Lancaster et al. 2021); therefore, we tested whether the density of egg masses on individual rocks in the restored-treated streams changed from before to after treatment. Mean egg masses per individual rock area was calculated as the mean of the sum of masses on an individual rock divided by the underside two-dimensional area of the rock for each stream, census date, and year.

Larval insect abundance

We sampled the abundance of benthic larval invertebrates during May 2020 (before rock addition) and March 2021 (9 months after adding rocks) using a 0.09 m² Surber sampler with 363 µm mesh size net. We selected springtime for larval sampling because by this time all eggs oviposited the previous year should have hatched (Appendix S1: Figure S2), and also because we were primarily interested in whether the additional egg-laying habitat caused an increase in larvae that survived until the next egg-laying period and could potentially emerge into egg-laying adults. Hence, we were not interested in quantifying transient increases in larval abundance. Invertebrates were preserved in ethanol, stained with Rose Bengal stain, picked under 10X magnification, and counted and identified to order and family because this was the lowest taxonomic resolution we could identify eggs, as there are currently no taxonomic keys for eggs.

Statistical analysis

We used a linear mixed-model, repeated-measures ANOVA (proc mixed in SAS) partially balanced incomplete block design to analyze whether adding rocks increased 1) the abundance of suitable oviposition habitat, 2) the density of eggs, and density and richness of egg masses, 3) the density of larval insects, and 4) the stability of suitable oviposition habitat (i.e., decreased the percent of rocks rolled, buried, or missing). This was a partially balanced incomplete block design because references did not occur in all blocks; therefore, the precision of differences in restored-treated versus restored-untreated was greater than the precision of differences in restored-treated versus reference streams (Oehlert 2000). We used pre-planned comparisons, or a priori contrasts, because we were interested in testing specific and sometimes directional hypotheses associated with the stream type x time interaction. For example, a before-after increase of a response in restored-treated streams compared to restored-untreated streams. For the number of rocks rolled, buried, or missing, we hypothesized a decrease in restored-treated streams compared to restored-untreated streams and reference streams. For all tests, including a priori contrasts, we report two-tailed test p values. We omitted a priori contrast between restored-untreated and reference streams because this was already tested by Jordt and Taylor (2021b). For unplanned comparisons, such as comparing the means of restored-treated streams versus reference streams after adding rocks, we reported Tukey-Kramer two-tailed test statistics only if the type III ANOVA main effects p value was < 0.05, an adjustment to reduce type I error. Stream type (i.e., restored-untreated, restored-treated, and reference), time (2019 and 2020), and their interaction were fixed factors. Group (i.e., block) and group x stream type were random factors. For larval density analyses, we fit generalized linear mixed models (proc glimmix in SAS) with the same fixed and random effects as for linear mixed models above and assessed the fit of either gamma or lognormal distributions using Akaike’s information criterion (AIC). We did not adjust for multiple tests in the analyses of changes in multiple egg and larval taxonomic groups because in some cases the model structure did not support it and such adjustments with limited replicates (n=5) increase the likelihood of type II errors. Instead, we report unadjusted probabilities and the magnitude of effects and do not use any one significant result to make broad overall conclusions (Voelkl 2019). We used linear regression to test whether annual egg mass densities per stream area of all taxa combined increased as the stability of protruding rocks (i.e., percentage of suitable rocks rolled or missing) decreased in each stream, regardless of type. We also used linear regression to assess the relation between changes in protruding rock densities and changes in egg mass density and egg mass morphotype richness from 2019 to 2020, as this could be used to predict the effects of the number of rocks added on egg masses density and richness. We tested relationships between rock size and cumulative number of egg masses per rock per year using a negative binomial regression and the relationship between rock size and probability of egg masses per rock per year using logistic regression. We transformed response variables to meet assumptions of normality and homoscedasticity. For graphical display of transformed variables, we report the back-transformed means (an estimate similar to the median in the untransformed scale) and back-transformed 95 % confidence intervals.

Egg masses were not distributed randomly among rocks, so the mean density of egg masses per protruding rock area was not a suitable measure of aggregation or crowding. The distribution of number of egg masses per protruding rock and the number per individual rock area were highly right-skewed. Therefore, we analyzed the mean number of egg masses per individual rock area for rocks with > 0 egg masses. Additionally, to assess aggregation, or crowding, of egg masses on individual protruding rocks among streams and whether aggregation changed following the addition of protruding rocks, we used a density-independent metric developed by Lang et al. (2017, eq. 23 with q=1), which is more appropriate than Lloyd’s (1967) patchiness index when mean abundances are small. As a descriptive measure of the distribution of egg masses on individual rocks, we calculated egg mass prevalence as the number of protruding rocks with one or more egg masses divided by the number of all protruding rocks, including those with zero egg masses. All statistical analyses were performed using SAS OnDemand for Academics (SAS Institute Inc. 2014).