Colombia's Autoridad Nacional de Licencias Ambientales (ANLA) maintains environmental monitoring databases for all of Colombia's hydropower projects generating > 100 MW (26 power stations). Hydropower installations are distributed throughout the country's three branches of the Andes, but they are concentrated in the Magdalena–Cauca Basin, which is home to roughly 70 % of the total national population of 50 million people. Colombian hydropower installations span a wide range of elevations – from 70 m in the foothills to more than 2200 m in the high Andes, with a correspondingly high range of mean ambient temperatures. Because they are distributed across the many slopes of the Colombian Andes including within relatively dry inter-Andean valleys, they experience highly divergent precipitation regimes, including monomodal and bimodal rainfall patterns. The western slope of the western Andes is one of the Earth's wettest regions, with some municipalities reporting annual rainfall of more than 10 000 mm yr-1; in contrast, the Sogamoso Reservoir near Bucaramanga receives roughly 1/10 this amount of annual rainfall. The climate settings of Colombian dams are enormously diverse. Of the hydropower projects under ANLA jurisdiction, 22 are associated with a dam and reservoir. The remaining four are micro-hydro systems that divert a portion of river discharge into turbine intakes but do not completely impound their associated rivers; thus, we exclude them from analysis.

As part of the environmental licensing process, the companies that operate the hydropower projects are required (by ANLA) to monitor the environmental impacts of their operations and submit annual reports and monitoring data sets. The companies (which may be public or private) contract environmental consulting firms to collect field samples and analyze them in their laboratories. The data are assimilated into a central georeferenced database maintained by ANLA, which screens the data for quality. In a legal sense, these data are public, and ANLA works to guarantee access; however, to date, plans to build an online portal to facilitate direct data acquisition have not yet been implemented. Currently, data are available to the public via direct request to ANLA. For this study, we requested and were granted access to systematized hydropower monitoring data from the years 2017 and 2018, which were the most recent years of fully quality-controlled data (Table 1). ANLA is working to incorporate historical data into its database; however, as our primary goal was to examine changes across the portfolio of hydropower systems, rather than to examine the evolution of dammed rivers over time, we did not request data from previous years. Data are stored in .gdb files, which require graphical information systems software (i.e., Esri ArcMap) to read them. We screened all data for statistical outliers and for chemical and physical plausibility and did not find the need to discard any data from our focal parameters.

Data collection frequency (summarized in Table 1) is highly heterogeneous across sites, with two dams (Urrá and Porce III) being measured monthly, whereas most dams provide just one to three data points per year. Sampling frequency is generally higher for more recently constructed reservoirs, reflecting an update to regulations in 1993. Older reservoirs (pre-2000) are normally monitored only once or twice per year. Each dam's monitoring approach is uniquely tailored to its geographic circumstances, but it generally includes surface water sampling from one or more stations upstream of the reservoir and downstream of the dam as well as samples from various parts of the reservoir itself, often including depth profiles. Adding to the data heterogeneity is a lack of standardization, leaving key fields missing in some entries. For example, Colombia's largest dam by volume, Sogamoso Reservoir, has been extensively profiled, but the depth and time information has been omitted, limiting interpretability.

As stratification is a fundamental driver of thermal and oxygen dynamics within and downstream of hydropower reservoirs (Winton et al., 2019), we first qualitatively assessed depth profiles (where available) for stratification strength. Because the profile data availability is quite variable, including the number of depth points sampled and seasonal coverage, we focus on deepwater oxygen levels (below the thermocline) and the magnitude of the oxygen concentration difference between surface and deep waters to sort each reservoir into coarse categories of strongly stratifying, weakly stratifying or non-stratifying. We classified reservoirs as strongly stratifying if they showed oxygen concentrations of < 2 mg O2 L-1 at depth and a difference between the surface of at least 3 mg O2 L-1. We classified all other reservoirs as weakly stratifying, as all showed differences in temperature and oxygen between the surface and deep waters of at least 2 ∘C and 1 mg O2 L-1 (for a summary, see Table S1 in the Supplement). Secondarily we use the densiometric Froude number (Parker et al., 1975), which compares the inertial force of reservoir water, based on mean flow-through velocity, with the gravitational force tending to maintain densiometric stability (Orlob, 1983; Deas and Lowney, 2000). The Froude number can be approximated using the following simplified formula including reservoir length (L), depth (D), discharge (Q) and volume (V): F= 320(L/D)(Q/V) (Parker et al., 1975). This formula has been applied as a metric for stratification behavior (Ledec and Quintero, 2003).

To understand dam impacts, we rely on contemporaneous paired measurements of upstream vs. downstream conditions, and we interpret the differences to be attributable to dam effects, which is a widely used approach (Fovet et al., 2020). Colombian regulations stipulate that industrial or commercial activities should not alter water temperatures by more than 5 ∘C, which is a temperature difference that has been associated with acute responses from tropical biota, such as fish mortality from warm water (Cooper et al., 2019) or disruption to fish reproductive cycles from cold water (King et al., 1998). A more conservative thermal change threshold of ±2 ∘C may be warranted given that the community composition of macroinvertebrates is highly sensitive to subtle changes in the thermal regime (Eady et al., 2013; Preece and Jones, 2002) and assessments of global climate change effects on fish delimit a 2 ∘C threshold for impacts (van Vliet et al., 2013). Furthermore, as we are limited (by data availability) to just a few random comparisons for most sites for Colombia, we are unlikely to be capturing the most extreme moments of thermal impact; thus, the precautionary principle would dictate a stricter approach. Therefore, we consider a change of +2 or -2 ∘C to be evidence of warm-water or cold-water pollution, respectively. For changes in dissolved oxygen, we found that a loss of > 2 mg L-1 imposed by dams always corresponded to a downstream concentration of < 5 mg L-1, which is the regulatory minimum concentration for cold freshwaters in Colombia. Oxygen availability imposes a fundamental constraint on many aquatic species (Coble, 1982; Spoor, 1990; Ekau et al., 2010); therefore, we assess impact along a change threshold of 2 mg L-1 to distinguish between “minor” and “severe” oxygen loss. For sediment trapping, although the literature is clear about the potential consequences of dam-induced sediment loss from rivers, choosing an appropriate threshold demarcating what constitutes a severe loss is highly subjective, and there are no regulatory guidelines for total suspended solids. We report the gradient of responses, noting how many dams trapped > 50 % and > 99 % of inbound suspended sediments.

We note that, in some cases, dams can be oriented in a cascade whereby the outflow from one reservoir rapidly (or immediately) enters the reservoir for the next power station. In such a configuration, the inflowing water does not necessarily represent a neutral reference condition, as it has likely already been altered by the previous dam. This may bias us to underestimate the potential of lower-chain dams to alter water quality; however, without access to data on pre-dam river conditions, there is no alternative metric for reference conditions beyond upstream waters. As only 1 year of recent data is available for most projects, we focus our analyses on just the most recent year (2017 or 2018) under the assumption that covering an annual climate cycle is more important for answering our research questions than studying variation between years.

Our analyses of reservoir mixing behavior indicate that most, if not all, Colombian reservoirs stratify strongly. Of the 22 reservoirs evaluated, 12 had available depth profile information; of these 12 reservoirs, 8 exhibited anoxia (dissolved oxygen (DO) concentration < 1.5 mg O2 L-1) in deep waters, indicating that they stratify sufficiently to prevent consistent reoxygenation from the surface (Table S1). The remaining four reservoirs lack evidence of acutely hypoxic deep water; however, with just two or three depth profiles per year, it is possible that this limited sampling did not coincide with periods of stronger stratification, which can develop rapidly in tropical lakes (Lewis, 1996). From the few reservoirs that have been profiled several times per year, it appears that Colombian reservoirs are rarely, if ever, well mixed (Figs. S1, S2 and S3 in the Supplement). This result is supported by our calculations of the densiometric Froude number (Fr) (Parker et al., 1975; Orlob, 1983), which indicate that, of the 12 dams with discharge data available (necessary for calculating Fr), all reservoirs except for Guatapé fall into the strongly stratifying domain of Fr < 0.3, regardless of whether mean depth or maximum depth is used for the calculation (Fig. 2). In Colombia, large reservoirs tend to stratify strongly, and this limnological reality creates the potential for thermal and biogeochemical disruptions to downstream aquatic ecosystems.

Consistent with the expectation that stratifying reservoirs will alter thermal regimes, we find that 9 of the 12 (75 %) Colombian reservoirs assessed generated temperature anomalies in the river of at least 2 ∘C (Fig. 3). Four reservoirs create cold-water pollution of at least -2 ∘C downstream, and seven reservoirs create warm-water pollution of at least +2 ∘C downstream. Two reservoirs – Urrá and El Quimbo – generate both cold- and warm-water pollution at different times of the year, illustrating the importance of temporal dynamics. Seasonal climate cycles that govern stratification, fluctuations in inflowing discharge associated with droughts/floods and dam operation (i.e., hydropeaking) may all influence the direction and magnitude of downstream thermal effects over timescales of months to minutes. Our ability to more broadly assess thermal changes is severely limited by the lack of frequent (seasonal or monthly) and detailed (with depth profiles) monitoring for most projects.

We find that minor loss of oxygen (change in DO < 2 mg L-1) in waters downstream of Colombian dams was common (evident in 9 of 15 reservoirs), but severe oxygen loss relative to upstream (change in DO > 2 mg L-1) was associated with only 4 reservoirs (Fig. 4). These hydropower schemes (Urrá, El Quimbo, Sogamoso and Prado) are the same reservoirs that exhibited cold-water pollution (Fig. 3) – a coincidence that supports our assumption that both effects are driven by thermal stratification (Hutchinson and Loffler, 1956; Lewis, 1987). As with temperature, we observe high seasonal variation in upstream–downstream oxygen dynamics, which, again, supports the notion that hypoxic effects are sensitive to the seasonality of climate, stratification and dam operations.

Loss of total suspended sediments (TSS) associated with Colombian dams is pervasive; however, for several reservoirs, loss of TSS is observed to be extreme (Fig. 5). A total of 6 out of the 10 reservoirs that we assessed showed TSS decreases of more than 50 %, and two dams – El Quimbo and Sogamoso – logged losses of more than 99 %. Some reservoirs exhibited an increase in TSS downstream of the dam relative to upstream conditions, which may reflect local erosion processes or activities, while other reservoirs exhibited upstream sediment losses. For example, the Porce River has been dammed by several dams in a cascade system and, as a result, carries a relatively low sediment load; however, the river reach immediately below Porce III dam is turbid because of local illegal mining activity. Beyond such local artifacts, turbid inflow and clear outflow are a typical modality of Colombian dams, especially during rainy periods.

In contrast to sediment loss and warm-water discharge, which appear to be ubiquitous, cold-water pollution and hypoxic effects appear to both afflict the same subset of four Colombian reservoirs (Urrá, El Quimbo, Prado and Sogamoso). Examining the characteristics of these reservoirs reveals some important commonalities: all are very deep (at least 70 m) and have long mean residence times, sufficient for anoxia to develop in the hypolimnion (as is evident at Urrá, which has depth profile data; Fig. S1). As all of the power plants have fixed depth intakes to run their turbines, there is no opportunity for operators to avoid discharging cold and hypoxic water when the thermocline/oxycline is shallower than the depth of intakes. In the case of El Quimbo, hydropower authorities inject liquid oxygen into discharge waters to meet the minimum dissolved oxygen requirement of 4 mg L-1, which is an expensive measure that they hope will be phased out by 2023 as oxygen demand in the reservoir lessens, as typically happens as reservoirs age. For Urrá, authorities have found a policy solution, passing a rule relaxing the minimum oxygen requirement to 2 mg L-1 for the first 5 km downstream of the dam, which is good for hydroelectric generators but does not reduce impacts on aquatic biodiversity and its associated ecosystems services. Both Prado and Sogamoso are out of compliance with the 4 mg L-1 threshold, and authorities are evaluating management options (unpublished documents from Autoridad Nacional de Licencias Ambientales).

The development of hypoxia in lakes and reservoirs can be exacerbated by long hydraulic residence times and elevated oxygen demand from organic matter inputs. Colombian dam data illustrate the importance of these reservoir and catchment characteristics for determining the risk of hypoxia developing in downstream rivers. Patángoras (Miel), in some important respects, appears to be like El Quimbo – they have similar depths, hydrologic residence times and fixed-depth intakes – but, in contrast to El Quimbo, Miel shows no sign of downstream hypoxia. Miel also relies on a reoxygenation system – an air bubbler in its oscillation cavern – but this is a much less intensive and less costly intervention than El Quimbo's need for liquid oxygen injection. Part of Miel's behavior may be attributable to the fact that its inflowing water is near saturation and has a very low biochemical oxygen demand (BOD; Table S2), meaning that severe hypoxia only appears rarely and typically well below 50 m depth (Fig. S2). This is likely a function of the well-preserved state of the Miel catchment, which includes robust riparian buffers that buffer the reservoir against extreme hydrologic fluctuations. In contrast, Porce II, which lies downstream of the Medellín metropolis (population of 4 million people), receives inflows that are already hypoxic (< 5 mg L-1) and have an elevated BOD (Table S2). Therefore, anoxia downstream of Porce II is not surprising. Further downstream, the Porce III Reservoir receives river water with substantial oxygen deficits, but its short residence time of just over 8 d prevents the dam from discharging water with significantly less oxygen than its inflows. Although the evidence from Colombian dams is anecdotal, it is consistent with the conventional logic that reservoirs loaded with high levels of organic matter from inflows (or left behind by inundated terrestrial ecosystems during reservoir filling) and with long residence times have greater potential to develop hypolimnetic anoxia for discharge downstream.

Dam design features, rather than environmental factors such as oxygen demand, offer an alternative explanation for why some dams avoid discharging hypoxic water downstream. A tower system with multiple intakes spanning 46 vertical meters in the Chivor Reservoir allows for discharged waters to be sourced from different depths as water levels fluctuate seasonally. Although we see no evidence of hypolimnetic hypoxia in this reservoir (Table S2), its selective withdrawal system could ensure oxic surface waters are passed downstream, as has been proposed as a solution for tailwater hypoxia in other tropical dams (Kunz et al., 2013). As previously stated, an air blower in the oscillation cavern at Miel dam provides reaeration of seasonally hypoxic (2 to 5 mg L-1) turbinated waters, such that discharged waters typically maintain DO concentrations of at least 6 mg L-1 (Table S2). A low BOD of inflows, as described above, may be an important mitigating factor, but evidence of the effectiveness of an engineered reoxygenation system for avoiding downstream hypoxia cannot be ignored.

Many tropical freshwater aquatic species are highly susceptible to thermal regime changes (Olden and Naiman, 2010), and a temperature change of a few degrees (Fig. 3) may very well cause disruptions to the life cycles of sensitive species (King et al., 1998; Clarkson and Childs, 2000). In the Andes, surface-releasing reservoirs that discharge warm waters might hypothetically favor lowland species over the cooler-water species that we would normally expect to be present at a given altitude. Several potamodromous fish in Colombia's Magdalena River, such as bocachico (Prochilodus magdalenae) and pimelodids (Pseudoplatystoma magdaleniatum and Pimelodus yuma) migrate seasonally from lowlands to up to 1200 or 500 m elevation, respectively, to spawn, historically transiting river reaches that have been dammed in recent decades (López-casas et al., 2014, 2016). Therefore, these species may avoid spawning in rivers altered by upstream dams, effectively reducing the available reproductive habitat (López-Casas, 2015). Alternatively, if the fish do spawn, thermal changes may disrupt the timing of embryo development; this can be lethal, as has been documented in Colombian fish farms (Harvey and Hoar, 1980) and the Mekong River of tropical Southeast Asia (Li et al., 2021). Researchers have found that hydropower generation is associated with changes in the production of a hormone driving oocyte maturation in P. magdalenae, thus disrupting its spawning cycle in dammed rivers (de Fex-wolf et al., 2019). For Andean fish species, the ranges of thermal tolerance are poorly known; therefore, more research would be needed to test such a hypothesis.

Hypoxic conditions in the tailwaters of dams impose even more dramatic ecological constraints than temperature. Dissolved oxygen concentrations from below 3.5 to 5 mg L-1 trigger escape behavior in most macroscopic organisms (Spoor, 1990), and monitoring data reveal concentrations of less than 5 mg L-1 below seven Colombian hydropower dams. Because the data come from sparse grab samples, it is not clear how persistent hypoxia is in these river reaches, but, at minimum, they indicate a loss of viable habitat for hypoxia-sensitive species for at least some parts of the year. The fish communities downstream of Colombia's Porce III dam – one of the projects with hypoxic tailwater – have shifted, with a loss of some native species and replacement by invasive exotic species (Valencia-Rodríguez et al., 2022). The authors of this study attribute these shifts to habitat fragmentation, but anoxia/hypoxia could be an important factor, as many other studies have commented that the multiple changes and stressors imposed by dams are difficult to disentangle (van Puijenbroek et al., 2021; Young et al., 1976). A review from Brazil indicates that poor oxygen management associated with dams is likely to be a major driver of fish mortality in South American rivers, accounting for roughly 40 % of fish kills covered by the media (Agostinho et al., 2021). As with temperature, a limited understanding of species-specific fish tolerance thresholds makes it impossible to fully grasp the impact of hypoxia/anoxia on fish communities.

Sediment trapping in dams, which can reach 99 % efficiency in some Colombian cases, exerts ecological impacts at different spatial scales. Unnaturally clear river water below dams may favor a different pool of top predators (those adapted for visual hunting), which may displace species adapted to turbid conditions, setting off trophic cascades rippling down the food chain. This effect has been documented at a large dam in Brazil (Granzotti et al., 2018), but it has not yet been reported in the Tropical Andes. If the sediment load is not replenished through additional erosion in excess of deposition downstream of the dam, the floodplain and delta ecosystems, which depend on riverine sediment delivery, will ultimately be starved, disrupting the cycles of nutrient retention and transport (Kondolf et al., 2014). A modeling exercise estimates that up to 40 % of sediments in the heavily dammed Magdalena River basin are currently being trapped behind dams and that this figure could increase to up to 68 %, threatening the ecological functioning of the Colombia's Mompós Depression, one of South America's largest wetland complexes (Angarita et al., 2018). Scientists have sounded the alarm that, globally, sediment trapping at dams in concert with sea level rise may lead to a massive loss of coastal deltaic wetlands (Giosan et al., 2014; Dunn et al., 2019). Colombian river sediments support mangroves on the Pacific and Caribbean coasts, although the extent to which dams may be impacting floodplain lakes, coastal mangroves and deltaic processes within Colombia's coastal zones is unclear.

It is a challenge to pinpoint the exact mechanism by which dams alter their associated aquatic ecosystems, as they impose so many changes, spanning physical, hydrological, chemical and biological dimensions, simultaneously (Young et al., 1976). Many studies of dammed river ecology focus on hydromorphological changes in habitat structure or availability, loss of connectivity, and alterations to flow regimes (Bratrich et al., 2004; García et al., 2011). In this study, we focus on oxygen, temperature and suspended sediments and find that these are just as plausible mechanistic pathways for reductions in habitat quality and availability, thereby driving shifts in ecological communities in dam-adjacent ecosystems. Environmental assessments of dams should take care not to ignore the changes dams impose on the physicochemical condition of downstream waters nor to underestimate temperature, oxygen and suspended sediments as modes of ecological change.