A whole-ecosystem experiment reveals flow-induced shifts in a stream community
Study areaThe study was conducted in the headwaters of the Chalpi Grande River watershed, 95 km2, located inside the Cayambe-Coca National Park in the northern Andes of Ecuador at an elevation range of 3789 to 3835 m (S 0°16′ 45″, W 78° 4′49″). This watershed harbors the primary water supply system for Quito. The system includes two reservoirs and 10 water intakes placed on first and second-order streams that, altogether, provide 39% of Quito’s water supply28. We monitored the Chalpi Norte stream for ~1.5 years prior to conducting our experiment for ~0.5 years (176 days), and ~0.4 years after the manipulation. Further, in the nearby area, we monitored 21 stream sites distributed upstream and downstream water intakes from the supply system (Fig. S4).Experiment for flow manipulation and monitoring flow reduction and recoveryWe conducted our experimental flow manipulation between October 2018 and April 2019 in a mainly rain-fed stream45. The experiment manipulated natural flows encompassing stable low flows and sporadic spates characterizing the high temporal variability of headwaters45,28 (Figs. 2a, b and S1). We set up a full Before-After/Control- Impact (BACI) experiment29 to evaluate ecosystem variables under natural and manipulated flow conditions. We identified a free-flowing stream reach on the Chalpi Norte that was above any water intakes that allowed us to divert flow with an ecohydraulic structure31. The structure was located above a meander, which we used to divert flow and return it to the stream below the meander (Fig. S4). The experimental site was comprised of an upstream/free-flowing reach (L = 25 m) (reference conditions), located ~32 m above the ecohydraulic structure and a downstream/regulated reach (L = 97 m) located immediately below the flow manipulation structure (Fig. 1b–d)31. The control site was located in a free-flowing stream, a tributary of the Chalpi Norte stream, with an upstream reach separated from a downstream reach by a distance of 16 m. We manipulated the instantaneous flow of the Chalpi Norte stream through a series of fixed percentages using different v-notch weir pairs31. We started diversions to maintain in the meander 100, 80, 60, 50, 40, 30, and 20% of the incoming flow for 7-day periods (based on local observations of benthic algal colonization); then we maintained 10% of the upstream flow for 36 days. We started to return flow gradually to recover 20, 30, 40, 50, 60, 80, and 100% of the upstream flow. In response to a natural spate while we maintained the 10% of upstream flow, the manipulated flow briefly (during ~9 h) increased above the targeted reduction (i.e., 54% instead of 10%) (Fig. 2a). We registered the spate of flow on the upstream reach of the experimental site (Figs. 2b and S1).Stream monitoring in adjacent streamsWe monitored 21 stream sites between July 2017 and July 2019. We selected seven streams with water intakes placed on the main channel (Chalpi Norte, Gonzalito, Quillugsha 1, 2, 3, Venado, and Guaytaloma). We sampled one site upstream of the water intake and two sites (i.e., 10 m and 500 m) downstream to obtain a wide range of flow reduction levels (Fig. S4) (see, 30 for further details on stream sites).Global literature surveyWe performed a systematic literature review to explore benthic algae responses to flow alterations (increase or decrease), focusing on cyanobacteria in streams. We used ISI Web of Science, Google Scholar, and Google Search for the entries: “benthic cyanobacteria” + “stream”, and “river”, “benthic algal bloom” + “flow” and all available combinations (Table S1). We selected papers containing information on benthic cyanobacteria and algae biomass and flow or water level measurements; specifically, we explored detailed information regarding experiments, spatial studies with upstream and downstream sites, and temporal replicates, as well seasonal associated benthic cyanobacteria blooms. We used published and/or publicly available data to calculate the percent of flow alteration in streams and calculated a factor on cyanobacteria biomass increase or decrease (quantitative studies) according to reported baseline conditions (either temporal or spatial). Only three out of 53 study sites reported a qualitative decrease in benthic cyanobacteria biomass attributable to flow reduction (Fig. 1d). Most studies (94%, n = 50) reported biomass increases with flow reductions. Among these studies sites, 44% reported qualitative observations where low flows were proposed as one of the environmental drivers responsible for benthic cyanobacteria blooms. While 66% of study sites (n = 33) related cyanobacterium biomass increase in time or space due to flow reductions caused by droughts, extreme low flow events, water abstractions, and experimental flumes manipulations.Abiotic and biotic variables sampling and analysesWater level sensors recording every 30 min (HOBO U40L, Onset USA) were installed at both upstream and downstream sites of water intakes, and on the experimental and control stream reaches (BACI desing), where we conducted multiple wading-rod flow measurements to convert water level into discharge via stage-discharge relationships (ADC current meter, OTT Hydromet, Germany). Streamwater’s physical and chemical in situ parameters (i.e., pH, temperature, conductivity, dissolved oxygen) were measured three times during biotic sampling on both stream sites and adjacent streams using a portable sonde (YSI, Xylem, USA). We collected water samples (500 ml) during in situ samplings to analyze nutrients (i.e., nitrate and phosphate) at the water supply company’s (EPMAPS) laboratory. We also measured precipitation from a rain gauge (HOBO Onset USA) installed in the Chalpi Norte stream.Our biotic variables included three benthic algae: cyanobacteria, diatoms, and green algae), and aquatic invertebrates biomass (Table 1). To measure Chl-a from cyanobacteria and benthic algae on artificial substrates, we used a BenthoTorch® (bbe Moldaenke GmbH, Germany) on unglazed ceramic plates (200 mm × 400 mm) with a grid of 25 squares of 2500 mm2 to allow algal accrual on a standardized surface. We allowed 21 days for colonization (based on previous observations) and then we placed all substrates5 at the beginning of the experiment. We performed five readings on five squares randomly selected within each plate. To consider the effect of benthic invertebrates to flow variations, we sampled stream sites using a Surber net (mesh size = 250 µm, area = 0.0625 m2). On the experimental and control sites we measured biotic, physical, and chemical in situ parameters every two days (n = 1760), and nutrients and invertebrates every seven days (n = 500) for the duration of the flow manipulation (~0.5 years). On the monitored sites, we measured biotic, physical, and chemical in situ parameters every seven days (n = 1456) and nutrients and invertebrates every 30 days (n = 336). To evaluate differences we calculated mean abiotic and biotic variables during the different phases (BL: baseline, FR: flow reduction, FI: gradual reset to initial flow) in the four-stream reaches to apply the BACI design29: upstream and downstream reaches on the experimental and control sites. We applied a paired one-tail t-test at α = 0.05 to compare FR and FI phases to baseline conditions, based on the expected direction of the response 1,14.Statistics and reproducibilityTo quantify the relationships between environmental variables and cyanobacteria biomass under manipulated and natural flow conditions, including interaction among algae and with invertebrates, we used multivariate autoregressive state-space modeling (MARSS)14,30. We fitted models with Gaussian errors for flow, conductivity, pH, water temperature, nitrate, phosphate, cyanobacteria, benthic algae, and invertebrate biomass time series via maximum likelihood (MARSS R-package)48. The state processes Xt includes state measurements for all four benthic components (cyanobacteria, diatoms, green algae, and invertebrates’ biomasses) considering the interactions between benthic components and environmental covariates (flow, conductivity, pH, water temperature, nitrate, phosphate) evolving through time, as follows:$${X}_{t}={{BX}}_{t-1}+U+{C}_{{Ct}}+{W}_{t}; {W}_{t} sim {MVN}(0,Q)$$
(1)
$${Y}_{t}={{ZX}}_{t}+{V}_{t} ; {V}_{t} sim {MVN}(0,R)$$
(2)
with Xt a matrix of states at time t, Yt a matrix of observations at time t, Wt a matrix of process errors (multivariate normally distributed with mean 0 and variance Q), Vt is a matrix of observation errors (normally distributed with mean 0 and variance R). Z is a matrix linking the observations Yt and the correspondent state Xt. B is an interaction matrix with inter-specific interaction (diatom and green algae) and with invertebrate strengths, Ct is a matrix of environmental variables (flow, conductivity, pH, water temperature, nitrate, phosphate) at time t. C is a matrix of coefficients indicating the effect of Ct to states Xt. U describes the mean trend. We computed a total of 12 models from the most complete to the simplest, the best-fitting model was identified as having the lowest Akaike Information Criterion adjusted for small sample sizes (AICc)14,30. To detect structural breaks in cyanobacteria biomass time series we calculated the differences between the smoothed state estimates at time t and t-1 based on the multivariate models. Sudden changes in the level were detected when the standardized smoothed state residuals exceed the 95% confidence interval for a t-distribution. We estimated the strength of environmental variables on cyanobacteria biomass and fitted models independently for each stream reach.To analyze cyanobacteria biomass across a gradient of flow alterations we compared weekly paired data (n = 1456) from upstream and downstream sites (i.e., at 10 m and 500 m). We thus calculated how much downstream site(s) biomass changed in comparison to upstream site biomass and assigned a factor for the increase or decrease. We determined the relative fraction of the instantaneous upstream flow in the downstream site measured within a 30-min time-step. We applied the same analysis to data from experiments obtained on the web search. We applied the Ramer–Douglas–Peucker (RDP) algorithm to find a breakpoint (ε lower distance to breakpoint) and the best line of fit for the local and global survey data distribution, we used the kmlShape-R package 48.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More
