Study Identification and Selection

We focused on vertebrates as this group includes fish which is perhaps the most intensively studied wildlife group in terms of hydropower impacts globally (Algera et al., 2020; Arantes et al., 2019; Turgeon et al., 2021). Additionally, this group also includes “mega-fauna” (vertebrates >30 kg) that have a disproportionately high risk of extinction (Turvey et al., 2010) and extirpation due to hydropower developments (He et al., 2018). As such vertebrates should present a best case scenario for the scientific evidence documenting hydropower impacts on Amazonian wildlife. Searches were conducted for articles published from 1945 to 2020 using four different databases: ISI Web of Science (Core Collection), SCOPUS, PubMed Central, and SciELO Analytics. The databases were searched using the following combination of terms: (Amazon*) and (hydroelectric or hydropower or dam) and (mammal or fish or bird or reptile or amphibian or vertebrate) and (impact* or effect*). The same terms were translated and searches repeated in Portuguese and Spanish (official languages of the five countries in the Amazon basin,  Supplementary Material Appendix 1). Searches were conducted twice, once on March 28, 2020 and again on March 29, 2021 to update publications from 2020.

Studies were selected following guidelines established by the Preferred Reporting Items for a Systematic Review and Meta-analysis (PRISMA [Moher et al., 2015; Shamseer et al., 2015], Figure 1). First, we screened all titles, keywords, and abstracts and excluded duplicates and any studies that were not related to hydropower developments and vertebrates within the Legal Brazilian Amazon. This region (comprising nine Brazilian states covering 5 Mkm² [(IBGE, 2020)]) was selected as it includes a wide range of tropical ecosystems and hydropower developments regulated by the same legal and governmental institutions, which enabled us to reduce confounding external factors that can affect design and implementation of impact assessments. The full-text of all articles that passed initial screening was then read to establish eligibility.

Figure 1.

PRISMA flow chart. Showing process used to assess and select studies.

As our focus was on evaluating impacts, the studies needed to include results from comparisons with at least one of the following: control areas (including space-for-time) and/or the impacted area after the hydropower was operational. Selected articles needed to present basic data/primary studies (Salafsky et al., 2019) from operational hydropower dams. Laboratory experiments, modelling studies, simulations, reviews, and meta-analysis were not included. Studies that used novel reservoir environments to test theories (e.g., species-area relationships on reservoir islands) were also not included. In addition, studies with lists of species compared with other areas in only a qualitative narrative form or where comparisons were only discussed (not included as part of the sampling methodology or analysis) were also excluded at this stage.

Based on the initial results, few studies were returned for some regions of the Legal Brazilian Amazon. To examine if the search terms were generating a strong effect, we updated the term used to establish geographic range of results (Amazon*), with the addition of a state name (Amazon* or Mato Grosso). This state was chosen as it included a relatively large number of dams and few studies. Full searches were repeated with this updated term on August 2, 2021. As no additional studies were added to the final selected list, we decided to retain results from the original search terms and not proceed with any additional search term refinements.

Study Data Extraction

Each study was evaluated by one author (ERS), who compiled: publication year, vertebrate groups, period of data collection, study design, geographic coordinates for the studied dams (obtained by joining dam name with coordinates provided by SIGEL (2021)), evidence type, and whether the study received funding/data from the developer/operator ( Supplementary Material Appendix 2). Study design typology followed definitions in Christie et al. (2019), and evidence types were classified following Burivalova et al. (2019) (Table 1). Although there is no agreed classification across different scientific disciplines, we followed Burivalova et al. (2019) in including Case-Control II and Quasi-Experimental as evidence types that should enable causal inference to be established. Finally, the PRISMA process and data extraction stages were independently reviewed by two authors (DN and FM) and corrections made to ensure reproducibility and consistency.

Table 1.

Study Designs and Evidence Types. Typology used to Classify Selected Studies. Descriptions Summarized From Christie et al. (2019) and Burivalova et al. (2019).

Hydropower Dam Data

To contextualize the systematic literature review, we compiled data on the operational hydropower dams in the Legal Brazilian Amazon. For each hydropower dam, we obtained geographic coordinates, operational start date, and installed capacity from the Brazilian Electric Sector Geographic Information (SIGEL—Sistema de Informações Georreferenciadas do Setor Elétrico”), provided and maintained by the Brazilian National Agency of Electricity (ANEEL—“Agência Nacional de Energia Elétrica”, downloaded from:  https://sigel.aneel.gov.br/Down/, accessed on March 30, 2021). We retained only hydropower dams with an installed capacity greater than 30 MW. We used ArcGIS 10.3 (ESRI, 2015) in order to produce the final distribution map of the hydropower dams and study locations. We chose 30 MW as a representative sample of current operational hydropower dams in the study region as hydropower dams >30 MW (classified as large [30–1000 MW] and mega-dams [>1000 MW]) represented 52% of operational and planned dams across the Amazon basin (Latrubesse et al., 2017). Installed capacity was not included as part of our literature search or selection/inclusion/exclusion criteria, only to obtain an informative sample of the current distribution of hydropower dams.

Data Analysis

All analyses were performed in R (R Development Core Team, 2020) with functions available in base R and “tidyverse” collection of packages (Wickham et al., 2019). Patterns in the geographic and temporal distribution of publications were evaluated using maps and descriptive analysis. As Brazilian states are an important administrative and legislative unit for the management of environmental resources, we compared the distribution of hydropower dams and publications between the nine states of the 5 Mkm² Legal Brazilian Amazon (Acre, Amapá, Amazonas, Mato Grosso, Maranhão, Pará, Rondônia, Roraima, and Tocantins, [(IBGE, 2020)]). The distribution of study designs and evidence types were compared between studies that (i) received funding and/or data from the hydropower developer/operator and (ii) independent research studies without any declared association with the hydropower developer/operator.

Temporal and Spatial Distribution of Studies

A total of 25 peer-reviewed studies were included in our systematic review, most of which (n = 17) were published between 2015 and 2020 (Figure 2). The first article found was published in 1981 (Junk et al., 1981). This was 4 years after the hydropower dam under study (“Curuá-Una”) became operational in 1977 and 6 years after the first hydropower dam became operational in the Legal Brazilian Amazon in 1975 (Figure 2). Although the number of operational hydropower dams increased steadily in the subsequent decades, the number of published articles started to increase only recently (Figure 2). After the first published study, there was a 12-year gap until the next publication, and few studies (n = 4) were published by 2012, despite there being 15 operational hydropower dams in 2010.

Figure 2.

Temporal distribution of published studies and operational hydropower dams. Annual frequency of (a) published articles documenting impacts on vertebrates (n = 25) and (b) newly operational hydropower dams >30 MW (n = 29) across the Legal Brazilian Amazon (area comprising nine states and covering approximately 5 Mkm²). Dashed lines show cumulative totals.

Based on our inclusion criteria, we were able to identify studies assessing impacts on three groups of vertebrates (Figure 2): fish (n = 20), mammals (n = 3), and reptiles (n = 2). There were no integrated studies including species from different groups. The major research interest was related to fish (80.0% of studies) with the four articles published during the first three decades (1981–2013) focusing exclusively on this group (Figure 2). The three mammal studies (Calaça & de Melo, 2017; Calaça et al., 2015; Palmeirim et al., 2014) were published between 2014 and 2017 and all focused on the semi-aquatic giant otter (Pteronura brasiliensis). The two reptile studies assessed impacts on dwarf caimans (Paleosuchus palpebrosus and Paleosuchus trigonatus [Campos et al., 2017]) and yellow-spotted river turtles (Podocnemis unifilis, [Norris et al., 2018]).

The studies assessed impacts caused by 12 of the 29 operational hydropower dams >30 MW in the Legal Brazilian Amazon (Table 2). Impacts of large (30–1000 MW) and mega-dams (>1000 MW) were investigated, but there were no studies of small hydropower dam impacts, with the lowest capacity of a studied dam being 42.8 MW (Curuá-Una dam, Table 2). The size-class distribution of the studied dams (0, 58.3, and 41.7% for small, large, and mega-dams, respectively) was significantly different (Fisher’s Exact Test p < .0001) to the size-class distribution of 288 planned dams (48, 45, and 7% for small, large, and mega-dams, respectively) reported by Latrubesse et al. (2017). The distribution of studies tended to follow the installed capacity of the dams in each state (Figure 3), and we found a positive but not significant correlation between installed capacity and number of studies per hydropower dam (Spearman Correlation rho = 0.41, p = .181).

Table 2.

Studies of Operational Hydropower Dams in the Legal Brazilian Amazon. Data for Hydropower Dams with Installed Capacity >30 MW Obtained From the Online Database Maintained by the Brazilian National Agency of Electricity (ANEEL—“Agência Nacional de Energia Elétrica”, Downloaded From:  https://sigel.aneel.gov.br/Down/, Accessed on March 30, 2021, [SIGEL, 2021]).

Figure 3.

Spatial distribution of published studies and operational hydropower dams. Geographic location of (a) operational hydropower dams >30 MW (circles, n = 29) and (b) studies documenting impacts on vertebrates (triangles, n = 25) across the Legal Brazilian Amazon. The size of the circles showing dams locations is proportional to the power (installed capacity) of each hydropower, and light grey lines represent major rivers. Plots show distribution of power (MW) by (c) state of all 29 operational hydropower and (d) for the 12 hydropower dams included in 25 studies. The sequence of states is ordered by total power output of operational dams in each state (high to low from left to right).

Nearly half of studies (n = 12) investigated impacts of only three hydropower dams, namely Jirau and Santo Antônio (n = 8, with six studies including both) in the state of Rondônia and Peixe Angical (n = 4) in Tocantins state. The two most intensely studied dams (Jirau and Santo Antonio, installed capacity 3750 and 3568 MW, respectively) accounted for 32% of all studies, and nearly half (8 of 17) of recent studies published since 2015. The remaining nine hydropower dams had one or two studies each (Table 2). We also found a weak positive correlation between the number of hydropower dams and number of published studies per state (Spearman Correlation rho = 0.21, p = .686). Mato Grosso was the state with most hydropower dams (n = 13), but was severely under-represented with only two published studies (Figure 3), both of which focused around the recently operational Teles Pires dam (1,819 MW, operational in November 2015, [Calaça & de Melo, 2017; Calaça et al., 2015]).

Geographic, Dam-Type, and Taxonomic Bias in the Literature

We found that studies were not evenly distributed in relation to the geographic location of operational hydropower dams within the states in the Legal Brazilian Amazon. There was a concentration of studies in only three hydropower dams (Jirau, Santo Antônio, and Peixe Angical) located in Rondônia and Tocantins states. However, Mato Grosso state, which had the most large hydropower dams of any of the nine states, was severely under-represented with only two published studies. Thus, a better geographical representation of studies across Brazilian Amazonia is needed as hydropower impacts will vary across different regions that have distinct social, economic, and environmental characteristics (Albert et al., 2021; Almeida, Shi, et al., 2019; Castello et al., 2013; Doria et al., 2018).

Studies did not represent the planned expansion of smaller hydropower dams across the Amazon basin (Latrubesse et al., 2017). Indeed, we found that studies across Brazilian Amazonia were biased by a focus on mega-dams. A major part of the increasing number of studies since 2012 can be attributed to studies of only two dams (Jirau and Santo Antonio). Although the sustainability of both projects was questioned (Fearnside, 2014, 2015), both received certification by Hydropower Sustainability Assessment Protocol ( https://www.hydrosustainability.org/published-assessments/santo-antonio and  https://www.hydrosustainability.org/published-assessments/jirau, accessed June 23, 2021). Our results show that scientific evidence documenting the impacts of both dams was generally weak (i.e., below expected best practice). A finding that supports recent analysis showing a link between superficial EIAs and a lack of social and environmental sustainability of Amazonian hydropower developments (Baird et al., 2021; Doria et al., 2018; Fearnside, 2018; Gerlak et al., 2020; Kahn et al., 2014).

Our findings support those from previous studies that show that there is an urgent need to quantify impacts of small hydropower dams across Amazonia (Athayde, Duarte, et al., 2019). A recent study identified 351 proposed dams across the Amazon basin with installed capacity >1 MW (average 262 MW, [Almeida, Shi, et al., 2019]), with an earlier study suggesting that nearly half (48%) of 288 planned dams were <30 MW (Latrubesse et al., 2017). Although international assessments include measures such as power density (ratio of electricity generation capacity to reservoir flooded area [Almeida, Shi, et al., 2019]) to establish carbon credits for hydropower developments (Clean Development, 2018), such metrics are not applied within Brazilian legislation for hydropower developments (Doria et al., 2018; Fearnside, 2015; Gerlak et al., 2020). For example, the need for and quality of EIAs are simplistically determined by installed capacity, with vastly streamlined requirements for small hydropower developments (less than or equal to 50 MW for future hydropower dams [MME, 2021]). As such there is a clear need for independent and scientifically robust assessment of small hydropower dam impacts across the Legal Brazilian Amazon.

Most of the studies found in our systematic review focused on fishes. This finding follows global patterns where fishes were one of the most frequently studied groups used to evaluate effects of hydropower dams in both temperate (Algera et al., 2020) and tropical regions (Arantes et al., 2019). But, impacts of run-of-river dams are poorly studied even for fish (Almeida, Hamilton, et al., 2019; Turgeon et al., 2021). The lack of studies documenting impacts to aquatic mammals (freshwater dolphins and manatees) was particularly alarming. There is an urgent need for more robust studies to be developed that could enable effective mitigation actions including bypass canals to be implemented (at both existing and planned dams) to avoid predicted range-wide population declines (Araújo & Wang, 2014; Arraut et al., 2017) or even extinctions as experienced in other basins (Turvey et al., 2010). Such measures would also likely benefit other species including large bodied economically and culturally important fish species (Baird et al., 2021; Doria et al., 2018; Duponchelle et al., 2021; Sant’Anna et al., 2020) that are also heavily impacted by dams that are physical barriers to movements and migrations (He et al., 2018). Yet, there was a lack of studies on multiple vertebrate groups, which is essential to understand hydropower dam effects on complex and biodiverse tropical ecosystems such as the Amazon (Park & Latrubesse, 2017).

Our systematic review showed a lack of studies assessing multiple hydropower dams and/or multiple vertebrate groups along the same river. In Brazil, several hydropower dams are commonly arranged in the same river, creating “cascades” (Athayde, Duarte, et al., 2019; Mendes et al., 2017). Although many studies focus on mega-dams, the combined effect of multiple dams, which can cause cumulative impacts (Athayde, Duarte, et al., 2019), remains poorly documented (but see dos Santos et al., 2017). For example, Coaracy Nunes was the first dam installed in the Legal Brazilian Amazon in 1975; since then two additional dams have become operational along the same river, providing a total of three dams with a combined installed capacity of 549 MW (78, 252, and 219 MW) within a 18 km stretch of river. The impact of these multiple dams is thought to have drastically altered both upstream and downstream flow rates, and following the installation of the second dam (Ferreira Gomes) in 2014, the downstream river course changed, draining predominantly to the Amazon river, not the Atlantic Ocean (Silva dos Santos, 2017). While individual studies focused on fish (Sá-Oliveira et al., 2015, 2016) and turtles (Norris et al., 2018, 2020) along the impacted river, these studies focused on different dams and adopted different sampling designs, which limits the ability to integrate results for important basin-wide analysis necessary to inform mitigation actions.

Sampling Design and Evidence Types

The lack of robust evidence was surprising considering hydropower dam development impacts are so strong and well known at a global scale (Grill et al., 2019; Liermann et al., 2012; Maavara et al., 2020). We found that studies in the Legal Brazilian Amazon generally adopted weak sampling designs (e.g., lacking controls) and lacked evidence necessary to generate reliable causal inference (Burivalova et al., 2019; Christie et al., 2019, 2021; Salafsky et al., 2019). Although randomized-control is widely recognized as the most robust study design, logistically simpler designs such as before-after control-impact can be equally effective in generating robust evidence for impact assessments of abrupt changes induced by large-scale development projects including dam construction. Additionally, dams are so widespread across Amazonia (Anderson et al., 2018; Athayde, Duarte, et al., 2019; Grill et al., 2019) that there are few remaining free flowing river sections that could be included within a randomized-control design.

As impacts were so poorly documented, it is unsurprising that there is limited evidence documenting the effectiveness of mitigation actions for vertebrates impacted by hydropower developments across Amazonia. For example, from a total of 48 actions identified in the Conservation Evidence database ( https://www.conservationevidence.com/data/index?pp=50&terms=dam&country%5B%5D=&result_type=interventions&sort=relevance.desc#searchcontainer, accessed July 14, 2021), there were no studies from the Amazon basin. Although it is possible to suggest some general actions based on documented global experiences, to our knowledge no studies have evaluated effects of installing bypass channels for aquatic mammals (Berthinussen et al., 2021), and only three short-term studies (10–18 months) evaluated translocations, two in French Guiana, both for primates (Richard-Hansen et al., 2000; Vié et al., 2001) and one in central Brazil for lesser anteater (Rodrigues et al., 2009). Indeed, we could not find any studies that have implemented or evaluated mitigation actions that are likely to generate multiple conservation benefits such as post-reservoir filling habitat creation for vertebrate conservation in the Legal Brazilian Amazon.

The recent increase in deforestation and expansion of agricultural frontiers across northern and central Brazil (including states of the Legal Brazilian Amazon [Schneider et al., 2021]) combined with planned expansion of small and large hydropower dams is likely to generate synergistic impacts on vertebrates. We failed to find studies including important cofounding impacts such as deforestation (Stickler et al., 2013). Although deforestation and tree mortality have been widely documented as important impacts of Amazonian dams (Athayde, Mathews, et al., 2019; Resende et al., 2019; Stickler et al., 2013), no studies included these important cofounding variables in the assessments of vertebrates. For example, the lack of studies in Mato Grosso was particularly surprising considering previous studies on effects of forest fragmentation on vertebrates in this state (Michalski & Peres, 2007; Norris & Michalski, 2009). Studies were developed in isolation from basin-wide changes. Hydropower dams do not occur in isolation, but together with development impacts that can have irreversible changes to river basins and freshwater biodiversity. For example, in Colombia, a reduction in 99% of wetlands was associated mainly with agriculture (sugarcane), dam construction (for both hydropower and irrigation), and urban expansion (Ocampo-Marulanda et al., 2021).

We found few studies considering the overall number and investment in hydropower projects across the Legal Brazilian Amazonia. Even fewer studies were found when considering only those with a robust design and able to establish causal inference. It could be suggested that weak evidence is a reflection of a lack of investment in science and technology, together with a reduction in investment in the Brazilian Ministry of the Environment over the past 20 years (de Area Leão Pereira et al., 2019). Although there is undoubtedly support for such considerations, the lack of robust survey designs can also perhaps be attributed more simply to a failure of researchers to adopt robust designs (Christie et al., 2019, 2021).

One of the biggest challenges to implementing Before-After Control-Impact study designs comes from the need to know when a dam project is going to occur and then acting to acquire baseline data at relevant temporal and spatial scales before the new development starts. Currently, data on planned hydropower dams across Brazil are freely available online (SIGEL, 2021), and previous studies have compiled this information to establish research priorities (Almeida, Shi, et al., 2019; Latrubesse et al., 2017). What appears to be lacking is a coordinated multi-disciplinary effort to establish a network dedicated to monitoring and mitigating hydropower dam impacts across the Amazon basin. The adoption of more robust study designs that can be used by multiple disciplines to develop integrated studies and robust conservation evidence could then be used to establish and strengthen collaborations among researchers, policymakers, and stakeholders (Albert et al., 2021; Christie et al., 2021). This coupled with longer-term support and training of local research groups as well as local communities could help increase the robustness and relevance of studies evaluating hydropower dam impacts on the environment, fauna, and flora (Baird et al., 2021). Such actions would also likely support community-based conservation management that has been proven to be efficient across Amazonia (Campos-Silva et al., 2018; Norris et al., 2018), providing improved long-term conservation outcomes and engaging a far wider diversity of local people (Baird et al., 2021; Doria et al., 2018; Fearnside, 2018).

We need to highlight that our systematic review has some limitations, as we did not include “grey literature” in our searches. Thus, it is important to recognize the potential for gaps or missing studies that were not published in peer-reviewed journals. However, due to the superficial content of EIAs (Baird et al., 2021; Doria et al., 2018; Gerlak et al., 2020), we would expect peer-reviewed published studies to have more robust designs and analysis compared with grey literature or reports. Therefore, our systematic review, performed in searches across four different databases and in three languages, is likely to be a best-case representation of the scientific evidence base documenting hydropower impacts on vertebrates in the Brazilian Amazonia.

Implications for Conservation

There is an urgent need to take advantage of freely available data to organize and plan effective surveys and sampling strategies to evaluate sustainability of current and future hydropower dams across the Brazilian Amazon. Below we provide recommendations to help develop a more robust evidence base.