Effects of Dams on Vertebrate Diversity: A Global Analysis

Dams are crucial for water supply in human populations and are becoming more common globally for hydroelectric power generation. Dams alter natural habitats and their biodiversity; however, studies are inconclusive about their effects on them. This study aimed to examine the effects of dams on vertebrates and the determinants of changes in global biodiversity and their relationship with critical areas for conservation. We evaluated the effects of dams on vertebrate richness and abundance. We performed a meta-analysis based on 120 case studies. We evaluated the overall effect on richness and abundance and examined these effects regarding taxa, disturbance type, latitudinal zone, zoogeographic zone, biodiversity hotspots, dam size and purpose, and species extinction risk. We conducted an overall analysis that included all species, and then we conducted separate analyses for terrestrial and aquatic species. Dams had a negative effect on vertebrate richness but not on vertebrate abundance. These effects were influenced by larger dams with fragmentation and were more pronounced within hotspots and in countries with a low species extinction risk. Such negative effects were explained by terrestrial vertebrates (particularly birds and mammals) because species richness and the abundance of aquatic vertebrates (fish) were not affected by dams in any case. Our results showed that habitat fragmentation created by large dams changes vertebrate communities, negatively affecting species richness in some areas of conservation concern. We propose implementing reservoirs in areas where they would have a lower impact on biodiversity and avoiding large dams in priority areas for conservation and where endangered species inhabit.


Introduction
Dams are closely linked to human welfare; they supply basic needs for modern society: aqueducts and irrigation are used for renewable energy generation [1]. The number of dams in the world has increased dramatically; in 1950, there were approximately 5000 dams, while by 2020 that number was around 58,713 [2,3]. Currently, dams are found in 167 countries and occupy 335,000 km 2 ; 76% of these dams have flood areas over 300 km 2 [4,5]. Dams used for hydropower generation currently contribute a quarter of the world's overall electricity production [6] and support a small percentage of the world's food production through irrigation [3,7]. Due to the steady increase in demand for water resources, water storage in dams is expected to increase by up to 3% between 2040 and 2050. This phenomenon will mainly occur in developing countries, based on their expected growth in energy production using dams [8].
From an environmental perspective, dams have implied changes in terrestrial and aquatic habitats. It has been suggested that disturbances caused by flooded areas are associated with changes in local and regional biodiversity [9][10][11][12]. For instance, local biodiversity changes are related to local extinctions and range shifts, as was found in the Himalayas by Pandit and Grumbine [13]. Moreover, regional changes are related to population declines and modifications in the trophic webs, as found by Wu et al. [14] and New and Xie [12] in China. Dams cause habitat fragmentation (e.g., island formation), habitat loss (i.e., vegetation cover replaced by the flooded area), and river fragmentation and alter water dynamics [14][15][16][17][18]. However, the effects of dams on some vertebrate groups (e.g., birds, mammals, fish, amphibians, or reptiles) are inconclusive due to the large heterogeneity of the responses (ranging from negative to positive) reported in the literature [13,[19][20][21][22]. For example, dams can positively affect several fish species, favoring those species capable of rapidly adapting to the novel conditions, as shown, e.g., by Cooper et al. [23], but have negative effects on mammals and birds that lose their breeding territories due to flooding [24]. Therefore, the effect of dams on vertebrate diversity is a topic of ongoing debate, which has no global consensus so far. Those heterogeneous responses may be determined by many confounding factors such as the type of disturbance caused by establishing dams (i.e., habitat loss or fragmentations, or stream alteration), the geographic context (tropical and temperate species may have contrasting responses), or dam size (large dams may have stronger effects than small ones). In addition, dam establishment may be conducted within critical areas for global conservation [25,26], which is likely to disproportionately affect threatened species. Further, these confounding effects may impact species richness and abundance in different ways.
To examine the effects of dams on vertebrate fauna and establish the possible determinants of changes in biodiversity, we conducted a global meta-analysis and contrasted the effects on terrestrial and aquatic species. In this analysis, we also addressed the potential relationships of dams with biodiversity conservation. We hypothesized that vertebrate species richness and abundance would decrease in response to habitat alteration produced by dams [27], with those effects being stronger for terrestrial (birds, mammals, and herpetofauna) than for aquatic vertebrates (fish). Additionally, as dams lead to habitat loss and fragmentation, we expected larger dams to have more negative impacts on vertebrate richness and abundances irrespective of their purpose or geographic location.

Literature Search and Inclusion Criteria
We conducted a literature search using the Scopus database (January 1960 to June 2021) with the following search terms: "dam OR hydropower OR hydroelectric AND *diversity OR change". The terms "habitat loss OR fragment* OR river flow" and "bird*-mammal*reptile*-amphibian* and fish*" were also included in the search. The keywords used tried to integrate information about dams, biological diversity, issues, and taxa. To ensure a standard of quality in the results and the study's replicability, we limited the search to peer-reviewed articles in English [28].
The initial search yielded a total of 1879 results, which, after eliminating duplicates, resulted in a total of 1662 articles. Then, we narrowed them to 348 after discarding information not relevant to the study. The remaining articles were read in full to determine if they met the following inclusion criteria: (1) comparison of a before (control) and after (treatment) establishment of the dam; (2) having richness and an abundance of data for the taxa assessed before and after the establishment of the dam; (3) having data on mean, sample size, and a measure of data dispersion (standard error or standard deviation). The inclusion of articles was done by one person (MBM) to avoid possible bias among reviewers. As a result, 298 articles that did not meet the inclusion criteria were excluded, resulting in a total of 50 usable articles yielding a total of 120 case studies ( Figure 1). The literature search followed the procedures established by the PRISMA statement [29].

Calculation of Effect Size and Moderators
We used Hedges' unbiased standardized mean to measure effect size in each case study [30]. This measure is used for estimating the magnitude of a particular effect, comparing control and treatment groups [31], and standardizing data from different studies to make them comparable. Separate analyses were performed for both species richness and abundance, then we estimated the overall effects of both, combining all taxonomic groups. Next, we analyzed aquatic (i.e., fish) and terrestrial (i.e., birds, mammals (including semi-aquatic mammals), and herpetofauna) vertebrates separately. Eight moderating variables were defined: (1) type of habitat disturbance (habitat loss, habitat fragmentation, or water alteration) as reported in each article; (2) taxa (mammals, birds, herpetofauna, and fish); (3) latitudinal zone (tropical or temperate); (4) zoogeographic region according to Holt et al. [32]; (5) location of the dam within or outside a biodiversity hotspot according to Myers et al. [33]; (6) Red Index, showing the overall extinction risk of species by country, values close to zero indicate all species extinct, while values close to one indicate all species of least concern [34] (Red Index values were subdivided into five categories of 0.2 increments (0.00-0.19 very high extinction risk, 0.20-0.39 high extinction risk, 0.40-0.59 moderate extinction risk, 0.60-0.79 low extinction risk, and 0.80-1.00 very low extinction risk). The countries included in our data presented values that ranged between 0.6 and 0.9; therefore, they were categorized as low risk (Red Index values between 0.6 and 0.79)

Calculation of Effect Size and Moderators
We used Hedges' unbiased standardized mean to measure effect size in each case study [30]. This measure is used for estimating the magnitude of a particular effect, comparing control and treatment groups [31], and standardizing data from different studies to make them comparable. Separate analyses were performed for both species richness and abundance, then we estimated the overall effects of both, combining all taxonomic groups. Next, we analyzed aquatic (i.e., fish) and terrestrial (i.e., birds, mammals (including semi-aquatic mammals), and herpetofauna) vertebrates separately. Eight moderating variables were defined: (1) type of habitat disturbance (habitat loss, habitat fragmentation, or water alteration) as reported in each article; (2) taxa (mammals, birds, herpetofauna, and fish); (3) latitudinal zone (tropical or temperate); (4) zoogeographic region according to Holt et al. [32]; (5) location of the dam within or outside a biodiversity hotspot according to Myers et al. [33]; (6) Red Index, showing the overall extinction risk of species by country, values close to zero indicate all species extinct, while values close to one indicate all species of least concern [34] (Red Index values were subdivided into five categories of 0.2 increments (0.00-0.19 very high extinction risk, 0.20-0.39 high extinction risk, 0.40-0.59 moderate extinction risk, 0.60-0.79 low extinction risk, and 0.80-1.00 very low extinction risk). The countries included in our data presented values that ranged between 0.6 and 0.9; therefore, they were categorized as low risk (Red Index values between 0.6 and 0.79) or very low risk (0.80 to 0.90). Red Index values were obtained from the Global SDG Indicators Database [35]); (7) dam size-dams with an area less than 576 km 2 (this threshold value corresponds to the median value of dam sizes obtained for each case study) were categorized as small, dams with an area greater than this value were categorized as large; (8) purpose of the dam, hydroelectric generation or other uses (irrigation, water supply, or recreation), according to FAO [36]. Since there were asymmetries in the sample size among some moderators, we excluded those levels with N < 4 as we considered them uninformative. In addition, moderator levels with N ≤ 10 were interpreted with caution. When appropriate, we also excluded those case studies with no information (e.g., not reporting dam size) from the moderator analyses. To examine heterogeneity among moderator levels, we estimated between-group homogeneity Q between statistics (which follows a Chi 2 distribution statistic with k-1 degrees of freedom that quantify heterogeneity among case studies) [37,38]. When significant heterogeneity (i.e., Q total estimate) is detected, random effects models are advised.

Model and Publication Bias
Because this study included articles reporting on different taxonomic groups, habitats, and geographic locations, a random effects model was appropriate to analyze biological data, as we did not expect all case studies to share a common effect. As many articles provided more than one case study, we included article ID as a random factor in the models to account for the variation among articles [37]. Before we examined the results, we assessed the robustness of the data, using the following tests: First, we used Egger's test to statistically test the skewness of the funnel plot [39]. We then used the Baujat plot to visually explore the individual contribution of each case study to heterogeneity in our meta-analysis. We calculated Kendall's tau correlation with continuity correction to assess potential publication bias to determine whether the effects and sample size were correlated. We then performed a trim and fill analysis [40,41], which accounts for any potential bias related to asymmetries in the distribution of positive and negative cases, and recalculated the mean effect and confidence intervals to check the robustness of the results [42]. All analyses were performed using the "meta" package [43] in R version 4.1.0 [44].

Results
We obtained 120 case studies in the literature search; 83 case studies were related to species richness, while 37 were related to species abundance. Those cases came from 16 countries on different continents ( Figure 2) and covered a wide taxonomic range (Table 1, detailed information is available in Supplementary Table S1). categorized as small, dams with an area greater than this value were categorized as large; (8) purpose of the dam, hydroelectric generation or other uses (irrigation, water supply, or recreation), according to FAO [36]. Since there were asymmetries in the sample size among some moderators, we excluded those levels with N < 4 as we considered them uninformative. In addition, moderator levels with N ≤ 10 were interpreted with caution. When appropriate, we also excluded those case studies with no information (e.g., not reporting dam size) from the moderator analyses. To examine heterogeneity among moderator levels, we estimated between-group homogeneity Qbetween statistics (which follows a Chi 2 distribution statistic with k-1 degrees of freedom that quantify heterogeneity among case studies) [37,38]. When significant heterogeneity (i.e., Qtotal estimate) is detected, random effects models are advised.

Model and Publication Bias
Because this study included articles reporting on different taxonomic groups, habitats, and geographic locations, a random effects model was appropriate to analyze biological data, as we did not expect all case studies to share a common effect. As many articles provided more than one case study, we included article ID as a random factor in the models to account for the variation among articles [37]. Before we examined the results, we assessed the robustness of the data, using the following tests: First, we used Egger's test to statistically test the skewness of the funnel plot [39]. We then used the Baujat plot to visually explore the individual contribution of each case study to heterogeneity in our meta-analysis. We calculated Kendall's tau correlation with continuity correction to assess potential publication bias to determine whether the effects and sample size were correlated. We then performed a trim and fill analysis [40,41], which accounts for any potential bias related to asymmetries in the distribution of positive and negative cases, and recalculated the mean effect and confidence intervals to check the robustness of the results [42]. All analyses were performed using the "meta" package [43] in R version 4.1.0 [44].

Results
We obtained 120 case studies in the literature search; 83 case studies were related to species richness, while 37 were related to species abundance. Those cases came from 16 countries on different continents ( Figure 2) and covered a wide taxonomic range ( Table 1, detailed information is available in Supplementary Table S1).

Vertebrate Richness
We identified a negative effect of dams on vertebrate richness (Figure 3a). Regarding differences among vertebrate taxa, birds and mammals were the two taxonomic groups most negatively affected in species richness by dams (Q between = 11.54, p = 0.009; Figure 3b). Habitat fragmentation was the determining factor for negative changes in vertebrate richness (Q between = 27.39, p < 0.001; Figure 3c). From a geographic context, dam location did not affect vertebrate richness differently with respect to latitude (Q between = 2.81, p = 0.093; Figure 3d) or zoogeographic region (Q between = 6.33, p = 0.176). However, we could evidence a negative effect of dams on species richness for the Oriental zoogeographic region (Figure 3e). Dam size had a marginally significant effect on species richness (Q between = 3.25, p = 0.072), with large dams negatively affecting vertebrates the most (Figure 3f). In this sense, dam purpose was not related to a change in species richness (Q between = 1.48, p = 0.224; Figure 3g). Regarding conservation prioritization, dams located within or outside hotspots had a similar effect on vertebrate species richness (Q between = 1.78, p = 0.181; Figure 3h). Finally, countries with low Red Index values experienced the greatest negative change in vertebrate species richness (Figure 3i).

Vertebrate Abundance
Vertebrate abundance was not significantly affected by dams (Figure 4a). Consistent with this, dams did not significantly affect the abundance of birds, mammals, and fish (Q between = 1.47, p = 0.478; Figure 4b). None of the habitat disturbance factors negatively affected vertebrate abundance (Q between = 1.74, p = 0.419; Figure 4c). Geographically, dam location did not differentially affect vertebrate abundance with respect to latitude (Q between = 2.11, p = 0.146; Figure 4d) or zoogeographic region (Q between = 4.23, p = 0.238; Figure 4e). Likewise, dam size had no significant effect on vertebrate abundance (Figure 4f). We found, however, a contrasting result when examining dam purpose, as those used for hydroelectric generation had the greatest negative effect on vertebrate abundance (Q between = 5.83, p = 0.015; Figure 4g). Concerning conservation, vertebrate abundance was negatively affected within biodiversity hotspots (Q between = 4.29, p = 0.038; Figure 4h). Finally, countries with a low or very low Red Index did not differ in the change in vertebrate abundance due to dams (Q between = 0.63, p = 0.429; Figure 4i).

Vertebrate Abundance
Vertebrate abundance was not significantly affected by dams (Figure 4a). Consistent with this, dams did not significantly affect the abundance of birds, mammals, and fish (Qbetween = 1.47, p = 0.478; Figure 4b). None of the habitat disturbance factors negatively affected vertebrate abundance (Qbetween = 1.74, p = 0.419; Figure 4c). Geographically, dam location did not differentially affect vertebrate abundance with respect to latitude (Qbetween = 2.11, p = 0.146; Figure 4d) or zoogeographic region (Qbetween = 4.23, p = 0.238; Figure 4e). Likewise, dam size had no significant effect on vertebrate abundance (Figure 4f). We found, however, a contrasting result when examining dam purpose, as those used for hydroelectric generation had the greatest negative effect on vertebrate abundance (Qbetween = 5.83, p = 0.015; Figure 4g). Concerning conservation, vertebrate abundance was negatively affected within biodiversity hotspots (Qbetween = 4.29, p = 0.038; Figure 4h). Finally, countries with a low or very low Red Index did not differ in the change in vertebrate abundance due to dams (Qbetween = 0.63, p = 0.429; Figure 4i).

Differential Effects on Aquatic and Terrestrial Vertebrates
Fish were the only aquatic vertebrates in our database. The analyses conducted for this group showed no effects on species richness (Table 2a) or abundance (Table 2b). Conversely, terrestrial vertebrates showed negative effects on their species richness (Table 3a), which affected birds and mammals. They are explained by habitat fragmentation as well as dam size and purpose and are more acute in the tropical zone and outside biodiversity hotspots.
Terrestrial vertebrate abundance (Table 3b), however, showed no overall effect, but we found a negative effect on those animals within biodiversity hotspots and a positive effect of those dams with purposes other than hydroelectricity generation. Table 2. Effects of dams on aquatic vertebrate species richness and abundance. Means and 95% confidence intervals are shown for: overall effect, disturbance type, geographic location, zoogeographic zone, dam size, dam purpose, location within or outside biodiversity hotspots, and Red Index. The number of case studies is indicated in parenthesis (we excluded moderator levels with N < 4 or with no information). Q values correspond to Q total for the overall effect and Q between for the moderator variables (N/A indicates that there is no heterogeneity estimator as the moderator has only one level). Significance codes: NS p ≥ 0.05, *** p < 0.001.

Moderator
Level  Table 3. Effects of dams on terrestrial vertebrate species richness and abundance. Means and 95% confidence intervals are shown for: overall effect, vertebrate group, disturbance type, geographic location, zoogeographic zone, dam size, dam purpose, location within or outside biodiversity hotspots, and Red Index. The number of case studies is indicated in parenthesis (we excluded moderator levels with N < 4 or with no information). Q values correspond to Q total for the overall effect and Q between for the moderator variables (N/A indicates that there is no heterogeneity estimator as the moderator has only one level). Significance codes: NS p ≥ 0.05, * p < 0.05, ** p < 0.01, *** p < 0.001. Bold type indicates significant effects.

Publication Bias
We found no correlation between effect and sample size for species richness (Kendall's tau = −0.03, p = 0.637) or abundance (Kendall's tau = −0.17, p = 0.126) data. Funnel plots showed an even distribution of positive and negative cases for richness and abundance data (Figures S1 and S2 available online as Supplementary Material). Egger's test showed that there is no significant skewness for richness (intercept = 0.56, df = 81, p = 0.575) and abundance (intercept = −0.72, df = 35, p = 0.476). In addition, the Baujat plots showed a consistent pattern of heterogeneity among studies ( Figures S3 and S4 available online as Supplementary Material). The direction and significance of our results did not change after performing the trim and fill procedure (Table S2 available online as Supplementary Material), indicating that they are robust and not influenced by a possible skewness bias.

Discussion
Our results highlight the negative effects of dams on vertebrate species richness but not on its abundance. Furthermore, dams had contrasting effects on aquatic and terrestrial vertebrates, as fish were not affected in terms of species richness or abundance, but terrestrial vertebrates and particularly birds and mammals were negatively affected. These changes in biodiversity appear to be related to habitat fragmentation caused by dams, which are exacerbated by dam size. Moreover, the effects are more pronounced in the tropical zone, but we found no consistent trends when comparing zoogeographic regions. These changes in vertebrate communities often occur in critical areas for conservation and in regions that have experienced drastic biodiversity losses in recent decades. Despite how common dams are worldwide, comparative studies assessing their impact on vertebrate fauna are rather scarce. Consequently, our synthesis is based upon a relatively small sample size and a limited set of species. Therefore, we urge for more studies to be conducted on this subject to improve our understanding of dam effects on biodiversity.
On the one hand, fish seem to be the vertebrate group least affected by dam establishment. Previous work on this subject reported positive and negative effects on fish groups [23], in which most fish species benefited from these novel habitats, while some other species were more sensitive to stream alteration [45]. Such large response heterogeneity may explain why we did not find any significant effects when examining fish data separately. On the other hand, dams negatively affect the richness of birds and mammals; several studies have suggested that these taxa are highly susceptible to habitat loss and fragmentation within dams, leading to reductions in species richness [46][47][48]. The information available for herpetofauna (i.e., amphibians and reptiles) is very limited and precludes making further inferences. Habitat loss and fragmentation caused by dam establishment reduce the amount of suitable habitat for many species and generate and isolate habitat patches of different sizes [47,49]. Such a fragmentation process causes alterations in vertebrate population dynamics, dispersal capabilities, and persistence in the landscape [50]. In particular, it has been suggested that birds, specifically forest specialist birds, are affected by the size and distance between remnant wooded patches within the dam landscape or by features that limit their dispersal [51]. Similarly, large mammals are particularly affected due to poor connectivity among forest fragments within dams [52]. Additionally, these areas that remain flooded after dam establishment may limit bird and mammal breeding sites [24]. Although our results do not evidence a significant effect of dams on herpetofauna and fishes, habitat loss and fragmentation, and hydric alteration appear to be responsible for changes in their community composition [53]. These changes in fishes are due to constraints to their migration and transformation from lotic to lentic ecosystems [54].
Consistent with the above, it appears that large dams (usually used for hydroelectricity generation) are associated with greater changes in terrestrial vertebrate richness (and particularly in birds and mammals). This negative effect of larger dams may be associated with the formation of islands (fragmentation) that have been recorded within this dam type [55], along with changes in water regimes [56]. Lees et al. [57] have documented abrupt changes in vertebrate species richness and local extinctions due to insularization and variations in the water resource in the Brazilian Amazon. In this regard, hydroelectric dams were large dams that had the greatest negative effects on terrestrial vertebrate abundance [57]. Our results are consistent with these findings, and we also found negative effects of terrestrial vertebrate abundance on dams located within biodiversity hotspots.
Regarding biodiversity conservation, our analyses show that dams affected vertebrate species richness in priority regions for conservation (e.g., the Atlantic Forest) and in regions such as the Neotropical or the Oriental zoogeographic regions that have experienced drastic changes in biodiversity due to other factors such as the expansion of the agricultural frontier [28,58]. All vertebrate species included in this meta-analysis were classified as having low or very low extinction risks using the IUCN's Red Index [34]. This could mean that either no threatened species inhabit the sites where dams are located or studies are biased toward common and abundant species (which usually have a low extinction risk and are easier to sample), resulting in an underrepresentation of threatened vertebrate species in these kinds of studies. This may be more common than we can imagine, as large infrastructure projects (such as dams) require a formal environmental assessment report from the local authorities in many countries. The presence of threatened species means it is usually difficult to obtain permits to disturb large areas [59]. Our results are in addition to those expressed by other researchers, who have stressed the importance of avoiding implementing dams in these regions [10]. In the same vein, we consider it a priority for countries such as China, Malaysia, Thailand, and India (with a low Red Index and many dams), and other developing countries within the tropics where the number of dams is expected to continue to increase, to seriously consider their effects on biodiversity [60,61].

Conclusions
The results of this research synthesize and integrate research on the effects of dams on terrestrial and aquatic vertebrates. The work allowed us to identify the taxa that show the greatest negative effects on community structure and the possible determinants of these changes within the dams. In the future, the need for hydroelectric power generation worldwide will promote the construction of dams in many nations or priority regions for biodiversity conservation. Therefore, it is essential to consider the relevance of the implementation of dams in these sites, where other forms of energy generation could be regarded as perhaps having less of an impact on biodiversity.

Acknowledgments:
We are grateful to Vanessa Velásquez, Estefani Martínez, and Belén Murillo for their help and support. We are grateful to two anonymous reviewers for their comments, which helped us to improve the manuscript. FEF acknowledges the support of the project ANID/PIA/ ACT192027.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study, in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Abbreviation
IUCN: International Union for Conservation of Nature.