What explains the variation in dam impacts on riverine macroinvertebrates? A global quantitative synthesis

Dams have fragmented rivers and threatened aquatic biodiversity globally. However, the findings regarding dam impacts on riverine macroinvertebrates vary across regions and taxa. We conducted a global meta-analysis to quantify the effects of dams on macroinvertebrate assemblages (i.e. species richness and abundance) based on 3849 data points extracted from 54 publications. Responses of macroinvertebrates to dams varied among climatic zones, dam altitudes, dam sizes (i.e. height), downstream distances from the dams, and taxonomic groups. The overall effect size of dams on macroinvertebrate richness was negative, while that of dams on abundance was positive but varied among different dam types. Richness reductions were most pronounced in cold regions and high-altitude regions and were least pronounced in tropical regions and low-altitude regions, while abundance increases were more obvious in tropical regions and low-altitude regions. Macroinvertebrate richness reduction and abundance increase were coupled (i.e. when the richness slightly decreased, the increase in abundance was more significant, and vice versa) under the influence of dams across different climatic zones, altitudes, dam heights, and downstream distances from the dams. Furthermore, different taxonomic groups responded variably to dams, with stoneflies (Plecoptera), caddisflies (Trichoptera) and true bugs (Hemiptera) being the most sensitive groups (i.e. significant reduction in richness) among the taxa examined. Macroinvertebrate richness reductions were primarily attributed to changes in downstream substrate composition (i.e. from coarse to fine substrates), while abundance increases were potentially caused by replacements among taxa at downstream sites. Collectively, our results contribute to improving the prediction of the effects of dams on riverine macroinvertebrate assemblages and are valuable for guiding assessment and monitoring of river ecosystems, as well as sustainable dam development, planning and restoration.


Introduction
Dams have fragmented rivers across nearly every continent excluding Antarctica (Nilsson et al 2005, Poff and Matthews 2013, Zarfl et al 2015, Couto and Olden 2018, Grill et al 2019. Although dams often deliver economic services such as hydropower, flood risk alleviation, water supply, recreation and more, there are concerns that dams have impaired the key functions of rivers in providing diverse habitats and maintaining ecosystem integrity (Baxter 1977, Bunn and Arthington 2002, Carlisle et al 2011, Ellis and Jones 2013, Tonkin et al 2018a, Reid et al 2019. Numerous of aquatic organisms living in rivers are widely acknowledged to be vulnerable to daminduced thermal and flow alterations, reduced river connectivity and altered hydrochemistry (Bunn and Arthington 2002, Reid et al 2019, Mellado-Díaz et al 2019. Biodiversity loss, e.g. species diversity decline, has pervasively occurred in both large dammed rivers (Cheng et al 2015, Castello and Macedo 2016, Winemiller et al 2016, Zhang et al 2018, as well as small dammed rivers and streams (Lessard and Hayes 2003, Couto and Olden 2018. In addition, many terrestrial organisms, such as plants, birds, and mammals, that dwell within riparian zones and wider catchments are threatened by dam-induced environmental changes (Wu et al 2003, Meijaard 2019, Tang et al 2019. Therefore, dams pose a major threat to global biodiversity (Dudgeon et al 2006, Reid et al 2019. The prospects for freshwater biodiversity around dams (Rolls et al 2018, Turgeon et al 2019, require a more organized assessment for the prediction, restoration and management of the resulting changes in river ecosystems. Macroinvertebrates, which act as key organisms in aquatic food webs, are ideal candidates for studying how aquatic biota and the whole food webs respond to dam construction (Allan and Castillo 2007, Mbaka and Mwaniki 2015, Mor et al 2018, although also other taxonomic groups are considered to show relevant capacity in the detection of anthropogenic stresses (Hering et al 2006). First, because macroinvertebrates are an important linkages in the food web, their responses are influenced by changes in primary productivity (e.g. the productivity of algae and macrophytes, and the input of terrestrial organic matters) and further affect the composition, abundance and dynamics of higher-level consumers (e.g. fishes; Malmgvist and Englund 1996, Wallace and Webster 1996, Mcneely and Power 2007. Second, owing to their relatively small lifetime movement in a river segment, macroinvertebrates are better in portraying local environmental changes than other active consumers (e.g. fishes and waterbirds; Rosenberg and Resh 1993, Ormerod and Tyler 1993, Morse et al 2007. Third, many studies assessing river condition have shown that macroinvertebrates are sensitive ecological indicators for reflecting and monitoring multiple anthropogenic disturbances, including flow alteration, pollution, eutrophication, climate change and biological invasions (Bonada et al 2006, Fornaroli et al 2018, Mellado-Díaz et al 2019, Engels et al 2019, Guareschi and Wood 2019. In addition, macroinvertebrates are a taxonomically and functionally diverse group (Hauer and Resh 2007), constituting a large proportion of global freshwater biodiversity (Balian et al 2008).
Responses of macroinvertebrates to dams occur at multiple organization levels from individuals to communities (Brittain and Saltveit 1989, Krajenbrink et al 2019. For example, thermal and flow alterations can change macroinvertebrate life histories (Brittain and Saltveit 1989), drift and dispersal processes (Kennedy et al 2014, Holt et al 2015, Brooks et al 2018, and interspecific relationships (Mor et al 2018), thus further altering their community structure (Lessard and Hayes 2003, Phillips et al 2015, Mor et al 2018 and ecosystem functions (Martínez et al 2013, White et al 2016. Although numerous studies have evaluated the effects of dams on macroinvertebrates (Mbaka andMwaniki 2015, Wang et al 2019), the conclusions of single case studies diverge and even contradict one another across different regions and taxonomic groups. These discrepancies could be attributed to differences in climate (Turgeon et al 2019, Carr et al 2019, dam size Hart 2002, Mor et al 2018), downstream distance from the dam Stanford 1983, Ruhi et al 2018), life-history traits (Brittain and Saltveit 1989, Petts et al 1993, Petrin et al 2013, and phylogenetic position of the macroinvertebrate group under investigation (Campbell and Novelo-Gutiérrez 2007). Moreover, a recent study has shown that the influence of dams on macroinvertebrates is scale dependent, suggesting that understanding of river impoundment effects on downstream biota should be extended from individual rivers to larger regions (Krajenbrink et al 2019).
Accurate predictions and effective management strategies addressing the negative effects of dams on aquatic organisms and riverine ecosystems rely on the synthesis and quantification of the general response patterns of key taxa (e.g. linkage organisms, such as macroinvertebrates; higher-level consumers, such as fishes; Turgeon et al 2019) across different geographical and environmental settings. Therefore, we conducted a global meta-analysis to examine the various responses of macroinvertebrate assemblages to multiple factors to increase our understanding of dam impacts on macroinvertebrates in a general context. Rather than focussing on single case studies one-by-one, we provided generalizations across studies, which is the main aim of the metaanalytical approach (Gurevitch et al 2018). We aimed to answer the following questions: 1) What are the overall effects of dam construction on macroinvertebrate assemblages (i.e. species richness and abundance)? 2) Do the responses of macroinvertebrate assemblages differ according to the climate conditions (i.e. climatic zones, dam altitudes), main characteristics of the dam (i.e. dam height and downstream distance from the dam), and taxonomic groups? 3) What are the major environmental factors (or local habitat features) that affect macroinvertebrate richness and abundance at regulated sites?

Literature search strategy
We conducted an extensive literature search for primary research papers that tested the effects of dams on riverine macroinvertebrates. The literature search was conducted between October 2018 and April 2019. Relevant studies published from 1900 to 2019 were obtained from two databases (i.e. Web of Science and Scopus) using the following search term combinations: (dam OR dams OR weir OR weirs OR 'bypass tunnel * ' OR dike OR dikes OR dyke OR dykes OR hydropower * OR hydropeaking OR 'hydroelectric project * ' OR 'hydroelectric plant * ' OR 'hydroelectric power') AND (macroinvertebrate * OR invertebrate * OR zoobenthos OR macrozoobenthos OR macrofauna OR 'EPT' OR 'aquatic insect * ' OR Ephemeroptera OR Plecoptera OR Trichoptera OR mayfly OR stonefly OR caddisfly OR Chironomid * OR Oligochaeta OR Limnodrilus OR Mollus * OR Bivalvia OR Gastropods OR Snails OR 'functional feeding group' OR (Crustacea NOT (zooplankton OR Cladocera OR Copepoda))). These terms (e.g. dam, weir and dyke) were chosen due to these physical structures might affect the environmental conditions and the connectivity of a river network, thereby having further effects on the aquatic biota. We refined the retrieved publications strictly followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (appendix figure S2.1 (available online at stacks.iop.org/ERL/15/124028/mmedia); Moher et al 2009).

Inclusion and exclusion criteria
We further refined publications based on the following criteria by scanning the title, abstract and full text: 1) field studies evaluating the effects of dams on macroinvertebrates; 2) studies including comparisons of upstream reference (i.e. upstream sites are assumed to represent unregulated conditions) sites and downstream impaired (which are regulated) sites of dams; 3) studies reporting data for species richness or abundance (or both) of macroinvertebrates; 4) sampling sites located at the same river order as the dam, since rivers of different orders inherently may have completely different species compositions (Vannote et al 1980) and are not directly comparable; and 5) studies reporting extreme hydrological events (e.g. large flow release, flooding) were excluded.

Data extraction
Most data were extracted from tables in the main text or from datasets available in appendices and supplementary materials associated with the research papers. When data were presented in figures, they were extracted using ImageJ software (version 1.57, Schneider et al 2012). Since the direct comparison of pre-damming and post-damming macroinvertebrate communities is often not possible (due to the absence of pre-impoundment baseline monitoring data), most studies have compared communities at regulated sites with those at reference sites. Thus, we adopted this common approach in this meta-analytical study. We extracted the mean richness/abundance, sample size and standard deviation (SD) of macroinvertebrates or specific taxonomic groups for both reference sites and regulated sites, since species richness and abundance are the most widely-used metrics to portray the ecological effects of dams (Martínez et al 2013). For publications including two or more regulated sites/reference sites, we extracted the above data for all possible comparisons, i.e. each regulated site was compared to one or more reference sites. Thus, one publication may contain one or more data points depending on the number of sampling sites and specific taxonomic groups involved.
We recorded data concerning study characteristics as explanatory factors, including the following: 1) geographic location of the dam (i.e. climatic zones, according to the Köppen-Geiger climate classification system: tropical, arid, temperate, cold and polar; Köppen 1936, Beck et al 2018; 2) altitude of the dam location (m); 3) dam types; 4) dam height (m); 5) downstream distance from the dam (km); 6) taxonomic groups; 7) downstream flow discharge (m 3 s −1 ); and 8) changes in downstream environmental factors (the values at the downstream sites minus those at the upstream sites) with high occurrence rates (i.e. water temperature (WT, • C), current velocity (m s −1 ), conductivity (µs cm −1 ), total nitrogen (TN, mg l −1 ), total phosphorus (TP, mg l −1 ), dissolved oxygen (DO, mg l −1 ), pH and substrate types: boulder (>256 mm), cobble (64-256 mm), pebble (16-64 mm), gravel (2-16 mm), and silt (<2 mm)) were used to characterize changes of habitat features (appendix figure S2.2). Although the effects of dams on macroinvertebrates were also dependent on operation (water release) type, for example, surface or deep-water release, only 6 out of 54 publications clearly stated the water release type (5 were deep release, and one was surface release). Therefore, this study could not evaluate the contributions of different dam operation types to changes in macroinvertebrate richness and abundance. When the above data on study characteristics were not available in the paper, we searched for relevant information in other publications and website resources in the Internet (i.e. The Australian National Committee on Large Dams Incorporated, 2010, www.fao.org/nr/water/aquastat/dams/, http://npdp.stanford.edu/data_access/international_ dams_list.php, http://damnet.or.jp/Dambinran/binra n/TopIndex_en.html). Dam coordinates and impoundment areas were obtained from Google Earth. The downstream discharge data were estimated according to the methods of Wanders et al (2019), if not available in the original literature (appendix table S1.1).

Quantification of effect sizes
Differences in the overall effect size (e.g. if the mean of any metrics differed between the regulated sites and the reference sites) were assessed via Hedges' d metric (which is one of the expressions of the standardized mean difference; Hedges and Olkin 1985). Hedges' d is often used for ecological meta-analytical studies because it adjusts for differences among studies in terms of sampling effort, corrects for small sample sizes, and can handle zero values for control or treatment groups (Rosenberg et al 2013). Hedges' d was calculated as where S is the pooled SD and was calculated as J is a weighting factor based on the number of replicates (N) per data, and was calculated as The variance of Hedges' d was calculated as Hedges' d for any given metric is negative (<0) if the estimates (e.g. species richness, abundance) are lower for downstream sites than for upstream sites because of a dam and positive (>0) if the estimates are higher.
Hedges' d values that are close to zero (≈0) indicate little or no effects of a dam. In all the subsequent meta-analyses and metaregressions, the observed effect sizes (Hedges' d) were weighted by the inverse of the sampling variances (i.e. the SD). For studies that do not report estimates of the SD, we used Bracken (1992) approach to impute the SD using the coefficient of variation from all complete cases (Bracken 1992). These analyses were performed using the package 'metagear' in R (Lajeunesse 2016, R Core Team 2016).

Data analysis
To control for non-independence within the dataset owing to multiple effect sizes per study (Nakagawa and Santos 2012), we performed multilevel mixedeffects meta-analyses with restricted maximum likelihood via the R package 'metafor' (version 1.9-8, function rma.mv, Viechtbauer 2010, R Core Team 2016). All analyses were performed separately for the richness and abundance data. We treated study identity and taxon identity as random effects in our models. Random effects were retained or discarded based on the models' Bayesian information criterion, which is more restrictive than Akaike's information corrected criterion (Burnham and Anderson 2003). The final retained random-effects structure was (1|Study + 1|Taxon) for both richness and abundance (appendix table S1.2).
Multilevel random-effects meta-analyses were conducted and single mixed-effects meta-regression models were constructed according to different objectives. We assessed the overall reductions in the richness and abundance of macroinvertebrates in the reference vs. dam-impacted sites via multilevel random-effects meta-analysis. Formal Cochran's Q tests (Q) were used to test the residual heterogeneity of effect sizes, i.e. whether the variability in the observed effect sizes was larger than would be expected based on sampling variability alone. We introduced a number of moderators (appendix table S1.3) to explain the variability in effect sizes if their residual heterogeneity was significant.
We constructed single mixed-effects metaregression models to assess the relationships between Hedges' d and climatic zones, altitude of the dam location, dam height, downstream distance from the dam, and taxonomic groups (appendix table S1.1). Single mixed-effects meta-regression models were also used to estimate Hedges' d of the interactions of pairs of variables, i.e. taxonomic groups × climatic zones, taxonomic groups × dam height/altitude/downstream distance from the dam, and climate × dam height/altitude/downstream distance from the dam. Continuous variables were log-transformed and fitted as quadratic polynomials to account for non-linear relationships. Models with categorical factors were also constructed without an intercept to obtain the mean effect size of each level. The heterogeneity captured by the moderators in each independent meta-regression was assessed with omnibus tests (Q statistics, QM; appendix table S1.3). We also used generalized linear models to determine the relationships between Hedges' d and changes in habitat features, ultimately aiming to explore how environmental factors may affect macroinvertebrate richness and abundance.

Publication bias
Publication bias, i.e. if only studies detecting significant effects were published, was assessed using funnel plots by including precision (1/standard error (SE)) as a covariate with the rma.mv function in the R package 'metafor' and using meta-analytic residuals. Rosenberg's fail-safe numbers were also calculated to assess the robustness of our results against publication bias (Rosenberg 2005). Asymmetric funnel plots and failsafe numbers less than 5n + 10 indicate publication bias, which means that macroinvertebrate richness or abundance significantly affected by dams is easier to publish than that less significantly affected.

Description of the data
In total, 54 publications with publication dates ranging from 1970 to 2019 met our criteria, and 3849 data points (394 for richness and 3455 for abundance) were extracted. Differences in the numbers of data points between richness and abundance were due to the fact that some studies did not report richness data, but only focussed on abundance data. This meta-analysis involved 84 dams spanning 22 countries across five major climatic zones (figure 1). Most studies were from temperate (25) and cold regions (22), whereas studies from arid (5), tropical (4) and polar (1) regions were scarce. The altitude of the dam location ranged from 50 to 3120 m above sea level, and the dam height ranged from 0.35 to 219 m. The downstream distance of regulated sites from dams ranged from 0 to 206 km. This study classified the identified taxa into 14 main taxonomic groups: eight aquatic insect orders (Ephemeroptera, Odonata, Plecoptera, Hemiptera, Diptera, Trichoptera, Coleoptera, and Megaloptera) and six aquatic non-insect groups (Platyhelminthes, Nematomorpha, Oligochaeta, Hirudinea, Crustacea, and Mollusca; appendix table S1.1, appendix Data S3).

Overall effect sizes
The presence of dams resulted in distinct effects on macroinvertebrate richness and abundance along longitudinal river gradients. The overall effect size of dams on macroinvertebrate richness was negative (effect size: −1.46, P = 0.015, figure 2(a)), while that of dams on abundance was positive (effect size: 2.15, P = 0.006, figure 2(b)). There was significant residual heterogeneity in the random-effects meta-analysis for the richness dataset (Q = 23 350.46, P < 0.001) and for the abundance dataset (Q = 202 323.60, P < 0.001).

Effect sizes in different climatic zones and along with dam characteristics
The response of macroinvertebrate assemblages to dam revealed obvious and consistent zonation across different climatic zones and altitudinal regions ( figure  3). However, the responses of macroinvertebrate richness and abundance were distinctly different. Specifically, below-dam reductions in richness were most pronounced in the cold region (mean effect size: −3.02, P < 0.001) and high-altitude regions and were least pronounced in the tropical region (−1.44, P < 0.001) and low-altitude regions (figures 3(a) and (c)). In contrast to the pattern of richness reduction, downstream abundance was much higher than upstream abundance in all climatic regions, except the polar region, with downstream abundance increases being noticeably higher in the tropical (7.49, P = 0.008) and arid regions (6.26, P < 0.001) than in the temperate (0.72, P = 0.018) and cold regions (2.28, P = 0.025; figure 3(b)). Meanwhile, abundance increases were most pronounced in lowaltitude regions, and they were less pronounced in high-altitude regions (figure 3(d)). However, compared with dams in other regions, significant relationship could not be found in polar regions (richness: 0.319, P = 0.890; abundance: −1.543, P = 0.842), due to the low number of studies in this climatic zone. In addition, the dams in the tropical regions had different effects on macroinvertebrate assemblages, with the reduction of richness likely decreases and addition of abundance increases along with the altitude of the dam location (appendix figures S2.3(a) and (d)).
Different dam types had distinct influences on macroinvertebrate assemblages (table 1). Macroinvertebrate richness reduction was significant for hydropower and multiple usage dams. While the increase of macroinvertebrate abundance was shown in most cases, it significantly increased especially at sites downstream of water supply dams. In terms of dam size (height as a proxy), the reduction in downstream richness intensified with increasing dam height, while the abundance increase was more pronounced in the sites downstream of small dams than that of large dams (figures 4(a) and (b)). Besides, dams in tropical regions showed different patterns of impact on macroinvertebrate richness, with the effect on richness decreasing with increasing dam height (appendix figure S2.3(b)). Along with the increase of downstream distance from the dam, the richness reduction decreased while the abundance addition increased (figures 4(c) and (d)). Additionally, the abundance increases along the distance were more significant in the tropical regions than in other regions (appendix figure S2.3(f)). Furthermore, the richness reduction and abundance increase of macroinvertebrates under dams were coupled across climatic zones, dam altitudes, dam sizes, and downstream distances from the dams (figures 3(a) vs. vs. (d)), i.e. when the richness slightly decreased, the increase in abundance was more significant (or vice versa).
Downstream WT and TN additions were higher in tropical regions than in other regions, while velocity reduction was low (appendix figure S2.9). The reductions in coarse substrates, pH and velocity, accompanied by the addition of fine substrates, DO, and TN, all intensified along with altitude (appendix  figure S2.10). The reductions in coarse substrates, pH, conductivity and TP, accompanied by the addition of fine substrates, WT and TN, were more significant at sites downstream of large dams than at downstream of small dams (appendix figure S2.11). In addition, the additions of WT, DO, and TN and the reduction in velocity were high at sites in closer proximity to the dams, while pH and TP reductions were high at sites farther downstream (appendix figure S2.12). The variation in downstream richness reduction could be primarily explained by the decrease in the amount of coarse substrate (boulder: 22.4% variability; cobble: 10.8%) and the increase in the amount of fine substrate (silt: 27.9%), pH (17.2%) and conductivity (12.8%). Other factors, such as WT and velocity, were significantly correlated with richness reduction, but they explained very low percentages of the variability ( figure 6(a)). The increase in downstream abundance was significantly correlated with WT, DO, pH, TN and TP, but all of these factors also explained a very low percentage of the variability ( figure 6(b)).

Publication bias
The funnel plots revealed no obvious publication bias (appendix figure S2.13). Rosenberg's failsafe numbers were large enough for both richness (N = 362 047, P < 0.001) and abundance (N = 802 400, P < 0.001) to be confident about the reliability of the estimates, which were larger than 5n + 10 (richness: 1980; abundance: 17 305), implying that publication bias can be ignored in our study.

Discussion
This global meta-analysis revealed the heterogeneity of dam impacts on macroinvertebrate assemblages (i.e. species richness and abundance) among climatic zone, dam altitude, dam height, downstream distance from the dam, and taxonomic groups. Overall, while dams had negative effects on macroinvertebrate richness, they had positive effects on macroinvertebrate abundance. The reduction in richness and the increase in abundance varied consistently across climatic zones and altitude levels. Macroinvertebrate richness reduction was greater in cold and highaltitude regions, while the increase in abundance exhibited the opposite response pattern to dams, i.e. the increase in abundance was greater in tropical and low-altitude regions. The degree of macroinvertebrate abundance increase was generally coupled with the degree of richness reduction, and vice versa. In terms of taxonomic groups, the response of aquatic insect richness and abundance to dams varied, with Plecoptera, Trichoptera, and Hemiptera being the most sensitive groups, but for aquatic noninsects, the overall abundance increased at damimpacted sites. Richness reductions were likely to be explained primarily by changes in downstream substrate composition, while abundance increases can barely be attributed to changes in any of the considered environmental factors characterizing habitat features.

Effects of changing habitat features on macroinvertebrate richness and abundance
The reduction in downstream macroinvertebrate richness detected in this meta-analysis study can be attributed to changes in substrate composition (figure 6(a)), which is consistent with the findings of other studies (Ward and Stanford 1979, Cortes et al 2002, Extence et al 2011. The decrease in coarse substrates and increase in fine substrates were generally concurrent at the sites downstream of dams (Petts 1984). Coarse substrates can provide a wide range of refuges and a high environmental heterogeneity, and thus can support diverse sets of macroinvertebrate taxa (Beisel et al 1998, Mathers and. In contrast, fine substrates can homogenize benthic habitats and decrease the available space among coarse substrates for macroinvertebrates (Harrison et al 2007, Extence et al 2011, Buendia et al 2013. Moreover, fine sediments can particularly damage the gills of macroinvertebrates and thus are detrimental to taxa with a high oxygen demand, e.g. some mayfly species (Jones et al 2012, Descloux et al 2013. Secondarily, changes in physicochemical factors, such as pH, conductivity and velocity, may also reduce richness by affecting life-history processes, food sources and species interactions (Hart and Finelli 1999, Wellnitz et al 2001, Lancaster and Downes 2010. However, we found that these factors explained less variation in richness reduction than did the changes in benthic substrates. Therefore, the richness reduction was very likely driven by loss of taxa and the replacement of coarse substrates by fine substrates. We found that the macroinvertebrate abundance downstream of dams generally increased ( figure 2(b)) and, interestingly, this increase was associated with reduction of richness (figures 3, 4 and 6). However, on the basis of the data and results, it is difficult to infer whether there is a causal relationship between the patterns shown by the response metrics. The first likely explanation for abundance increase would be replacement of taxa, reflecting by the recolonization by taxa with high reproductive output ones, such as chironomids and Oligochaeta (appendix Data S3.1; Cortes et al 2002, Katano et al 2009). These taxa might benefit from the released large amounts of planktonic and sestonic production which were accumulated for a long time in the reservoir, finally resulting in the increase of their density (Lessard andHayes 2003, Tao et al 2020). Second, the increase in abundance could also be explained by competitive release. Once a few species are lost because of a disturbance, the remaining taxa are able to obtain more resources for growth and reproduction, resulting in increased population densities (Kareiva 1982). Third, the increased WT, DO, and nutrient conditions and the reduced flow and velocity may also have contributed to the increase in macroinvertebrate abundance through providing macroinvertebrates with more energy and food sources (Matthaei et al 2010) and causing macroinvertebrate individuals be concentrated into a smaller area of the benthic habitat (Dewson et al 2007). However, due to the very low amount of variation explained by these factors (figure 6(b)), the data and results of this study do not provide adequate support for this explanation.

Effects of climatic zone, dam altitude, dam height and downstream distance from dams on macroinvertebrates
The results of our study demonstrated that the effects of dams on macroinvertebrate richness showed clear and consistent variation across climatic zones and altitudes. The influencing mechanisms associated with climatic zones and altitudes could be the same and may be related to biodiversity dynamics (e.g. stability, resistance and resilience) and the magnitude of environmental change. First, from the perspective of processes, disturbance outcomes may be initially determined by the resistance of ecosystems, which is largely dependent on biodiversity dynamics. It has been demonstrated that high biodiversity can increase ecosystem resistance to disturbances (Naeem andLi 1997, Isbell et al 2015). Generally, broadscale variation in biodiversity is strongly correlated with climate and available energy (Currie et al 2004). In tropical and low-altitude regions, high primary productivity and habitat heterogeneity can support diverse aquatic biota and complex community structures (Dudgeon 2008), thus forming a stable network structure and potentially increasing the resistance of river ecosystems to dam disturbance. In contrast, in high-altitude and high-latitude waterbodies, low primary productivity generally restricts the survival, development, and reproduction of macroinvertebrates, resulting in lower diversity and resistance (Maiolini and Bruno 2007, Scott et al 2011, Culp et al 2019. Second, the outcomes of disturbances are determined by the resilience of ecosystems. High-diversity regions naturally have a large regional species pool, which is an important source for macroinvertebrate recolonization at regulated sites (Sundermann et al 2011, Tonkin et al 2014. In such regions, the diminished richness due to dam disturbance could be supplemented from adjacent river sections, especially by aquatic insects with a strong dispersal ability. Third, the degree of environmental changes under dam disturbance varies from region to region (e.g. Carlisle et al 2011), which could lead to among-region differences in richness reductions. Environmental conditions in cold and highaltitude regions are often challenging (e.g. Heino et al 2020), and the disturbances caused by dams could thus lead to greater changes in the river environment in these regions (Baxter 1977). For example, the negative WT increase in cold regions (appendix figure S2.9) and greater fine sediment addition at high altitudes (appendix figure S2.10) caused a severe reduction in downstream macroinvertebrate richness compared with that in other regions. Moreover, flow regulation has resulted in seasonal interruptions in high-altitude rivers (Gabbud andLane 2016, Bruno et al 2019), and many studies have shown that interruptions (e.g. flow intermittence) have a significant effect on macroinvertebrate community structure and composition (Mcintosh et al 2002. The overall effect of dams on macroinvertebrate richness reduction was found to increase with dam size and decrease with downstream distance, which is in line with previous studies (Mellado-Díaz et al 2019) and conforms to the logic of disturbance intensity and recovery processes. Large dams can cause more substantial environmental changes (e.g. those related to WT, flow regime, sediment dynamics and nutrients; Hart 2002, Mbaka andMwaniki 2015), whereas small dams tend to pass peak flows and are therefore less likely to substantially change the WT and nutrient conditions (Poff and Hart 2002). The influence of large dams on macroinvertebrate assemblages in the present study was also partly indicated by a greater reduction in nutrient levels (e.g. TP) at downstream sites (appendix figure  S2.11). In addition, the effects of dams were shown to be more drastic in the areas just downstream of the dam than in areas farther downstream (Ellis and Jones 2013). Generally, water release has the greatest effects in the areas immediately below the dam through scouring and flow regime alterations (Mellado-Díaz et al 2019). However, it has been suggested that this influence can be attenuated during the environmental recovery process along longitudinal gradients (Song et al 2019). Notably, tributaries of downstream rivers can also contribute to community recovery in disturbed river sections by providing new colonists and by creating transitional habitats (Jones andSchmidt 2018, Milner et al 2019).

Responses of different taxonomic groups
Different taxonomic groups respond differently to dam construction, which could be caused by the differences in assemblage diversity (i.e. taxonomic and functional diversity), life-history characteristics, and niche breadth (e.g. environmental tolerance) among taxonomic groups. For instance, Ephemeroptera comprises numerous species (Barber-James et al 2007) that are affected by different mechanisms and are widely adapted to a range of aquatic environments Saltveit 1989, Malmqvist andEnglund 1996). In addition, the various responses of macroinvertebrates to dams may also be related to differences in dispersal capability (Heino et al 2015, Tonkin et al 2018b because the reduced connectivity caused by dams hinders organisms' dispersal in the river network Stanford 1983, Dynesius andNilsson 1994). The distinct dispersal abilities of aquatic insects (with overland flight capability) and non-insects (passive overland dispersal at best) could leads to differences in their recolonization abilities after dam construction and subsequent continuing disturbance. In addition, macroinvertebrate taxa have different life-history characteristics (i.e. distinct differences in the voltinism of mayflies and stoneflies) in regulated lotic ecosystems, which reflects their resilience to environmental changes and their different responses to dam disturbance (Petrin et al 2013). Lastly, macroinvertebrate taxa with different tolerances to natural and anthropogenically-altered environmental conditions (e.g. sensitive taxa, such as EPT, and tolerant taxa, such as Oligochaeta) also respond distinctly to dams because of changes in the downstream environment.
Generally, EPT are considered to be sensitive taxonomic groups that respond quickly to dam-induced environmental changes (Stanford and Ward 1979, Mihalicz et al 2019, Krajenbrink et al 2019. However, the results of our meta-analytical study demonstrated that Ephemeroptera, as a whole, are not significantly affected by dams ( figure 5). This may be related to species turnover within Ephemeroptera in relation to environmental change; for example, sensitive species could have been replaced by tolerant species Saltveit 1989, Buendia et al 2013). This kind of a turnover could also be attributed to changes in species with different life-history characteristics (Petrin et al 2013). However, because EPT taxa are commonly used as indicators of water quality and anthropogenic disturbance (Carlson et al 2018, Krajenbrink et al 2019, Mihalicz et al 2019, the importance of Ephemeroptera in further dam impact assessments and monitoring should be given due attention. The richness of Hemiptera notably decreased under dam disturbance, while their abundance significantly increased. This finding contrasts with decreases in richness being dependent on sensitive taxa, but is consistent with increases in abundance relying on tolerant taxa (Camargo et al 2005). Aquatic Hemiptera are abundant and occur on most continents (Polhemus and Polhemus 2008), with various species displaying different tolerances to environmental change (Jansson 1977). Hemiptera have a strong dispersal capability and are early colonizers if newly-flooded habitats (Turic et al 2015). Dam-induced flow alterations can cause fluctuation events in the habitats of Hemiptera species, which could result in a rapid decline in richness, at least to some extent. However, some tolerant species can also recolonize and show increases in population abundance. Owing to the diverse responses of Hemiptera species to heterogeneous environmental conditions (e.g. Jansson 1977, Skern et al 2010 and the substantial decrease in the richness of Hemiptera in response to dam disturbance, our study suggests the use of this taxonomic group (via its richness) as a potential indicator of dam disturbance in river ecosystems.

Limitations, predictions and restoration implications
Notably, most data points of macroinvertebrate richness and abundance in this study were from sites downstream of hydropower plant-related and multiple-used dam. However, our results showed that the responses of richness and abundance of macroinvertebrate are similar in most cases among dam types (table 1). This finding indicates a general response pattern of macroinvertebrate assemblages to dams, although different type of dams may have distinct operation modes that could result in varied response patterns shown by organisms downstream from dams. Meanwhile, the alterations of downstream macroinvertebrate richness linked to environmental conditions (e.g. thermal regime and seston availability) are suggested to be affected by the type of water release of dams (e.g. where, how and when; Ward and Short 1978, Stanford and Ward 1979, Haxton and Findlay 2008, Olden and Naiman 2010. Unfortunately, owing to limitations of the literature data, we were unable to examine the effects of dam release type on macroinvertebrate richness and abundance. In addition, dams with different age (i.e. the interval time from dam construction to sampling) may have distinct influence on downstream biota. However, the heterogeneity of effect sizes caused by the interval time were smaller than those of other parameters, even for richness, the Q statistics (QM) were not significant (appendix table S1.3), suggesting a slight effect of interval time on the heterogeneity of effect sizes for macroinvertebrate richness in the present dataset. Thus, we did not detect the effect of interval time on macroinvertebrate, but which should be highlighted in future case studies and global syntheses. Furthermore, although taxonomic resolution often varies among different regions, it would have little effect on the main results because previous studies have suggested that the responses of richness or composition of biotic communities to ecological gradients may not necessarily be strongly affected by taxonomic resolution (e.g., Jones 2008, Vijapure andSukumaran 2019).
By untangling the roles of different factors in dam impacts on macroinvertebrates, our study provides evidence-based knowledge to help aquatic environmental managers to make defensible decisions. These include predicting dam effects on riverine biota, assessing and monitoring of river ecosystems, as well as guiding sustainable dam development, planning and restoration. First, for example, the finding showing the coupled relationship between reduced richness and increased abundance increases the ability to predict macroinvertebrate responses to dam construction. The clear variation in macroinvertebrate responses across climatic zones and altitudes highlights the vulnerability of river ecosystems to dam disturbance specifically in cold regions. Additionally, further studies in polar regions are recommend to be conducted due to the paucity of data to predict the effects of dam construction in the present study (n = 4 for richness and abundance, respectively). Second, the different sensitivities of the different taxonomic groups in their response to dams could be informative for the further use of macroinvertebrates in the evaluation and monitoring of the ecological effects of dams. Particular attention should be paid to the traditionally-used sensitive taxa (e.g. EPT) and overlooked taxa (e.g. Hemiptera) identified in this study. Additionally, due to the large variation in the responses among taxonomic groups, traitbased analyses would aid the understanding of the ecological responses of macroinvertebrates to dam disturbance (Martínez et al 2013, Alahuhta et al 2019. Third, findings concerning climatic zones and dam characteristics (i.e. the altitude of the dam location and dam height) also have implications for future dam development, including site selection and the scale of new dams. Lastly, considering the strong influence of substrate composition on macroinvertebrate assemblages, future restoration and management programs should devote more attention to this aspect of the river environment (Cortes et al 2002).

Conclusions
Collectively, this study quantitatively synthesized general response patterns of macroinvertebrates (i.e. richness and abundance) to dams across different environmental settings based on data derived from case studies covering a wide range of geographical areas. Factors contributing to the variation in dam impacts on macroinvertebrates were comprehensively considered. The ecological consequences of dams on macroinvertebrates depend largely on the ecological context. Changes in downstream substrate composition likely play a vital role in driving richness reduction, replacement of taxa, and abundance increase of macroinvertebrates in response to dams. According to this global quantitative synthesis, it is obvious that dams negatively affect macroinvertebrates to various degrees, but the responses of macroinvertebrate assemblages become more predictable when abiotic factors related to dam-caused changes can be quantified. In addition, the findings of our study also have broad implications for the assessment and monitoring of dam impacts, providing constructive suggestions for future dam development in rivers over the world. First-Class' Construction Project (Nos. C176210215, C176240208002). JH was partly supported by the project 'Regoverning the existing hydropower system: integrating ecological, economic and societal aspects of sustainability' funded by the Academy of Finland.

Statement of authorship
C D, J T, D H and J W conceived and designed the research, J W and W S extracted the data; J W, L D and M H analyzed the data and made the figures. J W wrote the manuscript; C D, J H, J T, X, J and D H contributed to the discussion and, subsequently, various versions of the manuscript; all authors reviewed the manuscript before submission.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary information files).