Do novel ecosystems follow predictable trajectories? Testing the trophic surge hypothesis in reservoirs using fish

Abstract

Hydroelectric reservoirs are novel ecosystems that provide a variety of important services. To manage these ecosystems and their fish populations effectively, we need to develop conceptual frameworks for predicting their short‐ and long‐term responses. To advance this goal, we revisited and tested the “trophic surge hypothesis, TSH.” The TSH has been widely cited in the literature, but has not been empirically tested across numerous reservoirs. The TSH suggests that fish populations should show a hump‐shaped pattern (i.e., the non‐equilibrium phase) after river impoundment. As such, we assembled 40 recruitment and 109 adult fish abundance time series from 19 species distributed across seven reservoirs from temperate and boreal regions, and applied curve fitting analyses and model selection. We found that the hump‐shaped pattern was the predominant pattern across individual time series, providing moderate support for the TSH. Fish recruitment increased substantially during reservoir filling and was followed by an increase in adult fish lagging 3–4 years behind. The non‐equilibrium phase was transient and lasted roughly eight years for recruits, whereas it could be much longer for adults. When time series were combined across regions and sites, the support for the TSH was weaker. However, we observed significant variability in the duration, timing, and magnitude of the surge across individual time series and found that the total flooded area was the most influential predictor to explain this variability. In conclusion, the TSH and related metrics can be a useful and general predictive framework to understand how fish populations may respond to impoundment. In particular, long‐term management recommendations could be short‐sighted if formulated before convincing evidence has emerged to show that the reservoir reached its new trophic equilibrium.

Introduction

One of the clearest examples of novel, or emerging, ecosystems is large hydroelectric reservoirs that are created by damming rivers (Milton , Hobbs et al. , ). Around the world, we have transformed many large rivers into reservoirs (Rosenberg et al. , Renöfalt et al. ), with a current boom in the construction of hydropower dams in emerging economies (Winemiller et al. ). The presence of dams and their associated management bring about pronounced water level dynamics in upstream reservoirs (Kroger , Friedman et al. , Graf , Zohary and Ostrovsky ), as well as decreased and homogenized downstream flow dynamics (Straškraba et al. , Poff et al. ). The shift from a lotic to a lentic ecosystem upstream of the dam can change the direction of key physical, chemical, and biological processes (Friedl and Wüest , Haxton and Findlay ). For example, the water temperature regime, sedimentation processes, nutrient cycling, and lake productivity are all factors that have been observed to change (Baxter , Ostrofsky , Benson ); these alterations can, in turn, impact the biodiversity and ecosystem functions (Nilsson and Berggren , Furey et al. ).

Conceptual frameworks predicting how novel ecosystems, such as large reservoirs, will age and go through non‐equilibrium and equilibrium phases can assist managers and scientists in developing expectations about the costs, benefits, and trade‐offs of balancing ecosystem services and ecological integrity (Hobbs et al. , ). In reservoirs, the trophic surge hypothesis (TSH or upsurge hypothesis; Baranov ) represents an interesting concept to underpin such a framework, but is one that has not yet been rigorously tested empirically. The TSH makes predictions about how the increase in phosphorus loading from the leaching and decomposition of organic matter from newly inundated terrestrial areas would affect productivity in reservoirs (Ostrofsky , Ostrofsky and Duthie , Grimard and Jones , Straškraba et al. ). Following these predictions, newly created reservoirs should undergo three main phases: a non‐equilibrium trophic surge, a non‐equilibrium trophic depression, and then a new equilibrium phase (Fig. ).

A schematic representation of the trophic surge hypothesis, inspired by Kimmel and Groeger (). Three phases should be observed during reservoir filling and after impoundment. The first phase is a non‐equilibrium trophic surge (i.e., increase in general ecosystem productivity), the second phase is a non‐equilibrium trophic depression (i.e., decrease in productivity), and a third phase is a trophic equilibrium where general ecosystem productivity can stabilize to levels similar to higher or lower than pre‐impoundment levels. On the schema, water level is represented by the black line. The orange line represents the trophic surge per se, referring to the increase in nutrients and microorganisms in the water following impoundment. The change in available habitat for aquatic organisms (green dashed line), the abundance of zooplankton and macroinvertebrates (red dotted line), fish recruitment (purple dashed line), and adult fish abundance (blue line) are also shown as relative units for simplicity.

More specifically, during the surge, the large influx of allochthonous inorganic nutrients and organic detritus should translate quickly into elevated autochthonous primary production (Baxter , Ostrofsky and Duthie , Fig. ). The rapid expansion of littoral habitat from the inundated terrestrial area would create new ecological niches and, coupled with high primary production, should result in high zooplankton and benthos productivity. Increases in primary production and secondary consumers should lead to a peak in fish recruitment (i.e., the number of fish surviving and entering the settlement stage), followed by an increase in adults a few years later (Balon and Coche , Ploskey , Kimmel and Groeger ). After the surge, the reservoir should experience a trophic depression, when available nutrients and detritus stocks are exhausted (Grimard and Jones , Dettmers and Stein , Garvey and Stein ), and suitable habitats can decline due to increased sedimentation and shoreline modification (Benson ). The non‐equilibrium phase should be then followed by a new trophic equilibrium where the reservoir stabilizes toward a steady “lake‐type” ecosystem (Grimard and Jones ).

Based on earlier studies, it has been suggested that the duration, magnitude of the surge, and the time to reach the trophic equilibrium will vary as a function of several key attributes. Specifically, the following predictors have been proposed: (1) reservoir characteristics such as area flooded, depth, water residency time, geographic location (Dillon , Ostrofsky , Straškraba et al. ), (2) the degree of alteration in the hydrological regime and reservoir management, including the magnitude and timing of drawdown, reservoir filling options, mitigation measures (Grimard and Jones , Pegg et al. ), (3) the land use and the amount of nutrient loading in the watershed (Ostrofsky , Grimard and Jones ), (4) the species life history traits and the strength of trophic interactions (Cherry and Guthrie ), and (5) other external factors such as stocking, species introductions, and fishing (Balon and Coche , Kimmel and Groeger , Pivnička , Costa‐Pierce , Tessier et al. ). However, we have very little direct empirical evidence of the TSH, and a limited temporal perspective on the ecological consequences of creating a reservoir.

Herein, we test the support for the TSH quantitatively using fish population time series from numerous boreal and temperate reservoirs with the goal of advancing a predictive framework for reservoirs as novel ecosystems. Most of the initial ideas and information about reservoir dynamics and responses of fish to impoundment are qualitative and/or are based on studies that typically examined a single reservoir, or described the trophic surge in relation to the trophic depression phase, rather than to pre‐impoundment conditions. In our analyses, we evaluated the prevalence of a hump‐shaped pattern (i.e., surge and then depression during the trophic non‐equilibrium phase) in 40 recruitment and 109 adult fish time series that came from seven different reservoirs. To extract some quantitative generalities, we also isolated TSH metrics from the time series (i.e., duration, magnitude, and timing of the surge) and then related these metrics to some reservoirs characteristics suggested to have an effect on the TS.

Methods

Literature search process

The studies presented in this synthesis were compiled from journals indexed in Thomson ISI's Web of Knowledge (mostly peer‐reviewed articles) and from Google Scholar (i.e., peer‐reviewed articles and textbooks, as well as government and industry reports, non‐peer‐reviewed journals, and conference proceedings). Extensive searches were performed between October 2014 and June 2015 on the references available at that time. We searched for references including the following keywords, individually or in combination: “reservoir*,” “trophic *surge,” “aging,” “time series,” and “impoundment,” but the search included “fish*” at all times. In addition, the reference lists and bibliographies of relevant sources were also examined to find literature that was not identified through Thomson ISI's Web of Knowledge and Google Scholar. We then screened our database to refine our selection criteria, and we included only studies that had strong quantitative evidence for the effects of impoundment on fish population dynamics (data from figures, tables, or databases in Appendix S1). We excluded modeling and simulation exercises. Studies with only a few years of data, lacking data before impoundment or during the first years of filling, or having only 2–3 years of data after impoundment cannot be used to test the TSH and therefore were excluded from our quantitative synthesis.

Data extraction and description of the time series

We were able to use time series data from six studies (Nelson and Walburg , Walburg , Benson , Strange et al. , DesLandes et al. , Doyon and Belzile ) to evaluate the effects of impoundment on fish population dynamics (Appendix S1). Within these six studies, we identified and extracted 40 continuous time series examining fish recruitment or a surrogate of fish recruitment for 18 species, and 109 time series examining adult fish abundance for 19 species. Time series data were mostly extracted from tables presented in the peer‐reviewed articles and gray literature reports. When data were only presented in figures, they were extracted using the GetData Graph digitizer software (v. 2.26).

Time series were collected from seven North American large reservoirs (Table ) used mainly for hydroelectricity production (19 sites; some reservoirs had more than one sampling station) and lasted between 8 and 23 years. The average length of recruitment time series was 11 yr (mean ± SD = 11.45 ± 2.7 yr; range = 5–23 yr) and was of 14 yr for adult fish (mean ± SD = 13.7 ± 2.8 yr; range = 8–22 yr). The vast majority of the recruitment time series came from temperate reservoirs (75%; 30 of 40), and the remaining 25% were from boreal reservoirs. For adult fish time series, 21% of the time series came from temperate reservoirs (23 of 109) and the remaining from boreal reservoirs (79%; Appendix S1). Most of the time series, especially the longer ones, were collected for species of recreational fisheries interest in America (e.g., Esox lucius [northern pike], Perca flavescens [yellow perch], Sander vitreus [walleye], and Sander canadensis [sauger]).

Reservoir characteristics used as predictors to model the occurrence, duration, timing, and magnitude of the trophic surge

Variables	Reservoirs
Opinaca	Robert‐Bourassa	Caniapiscau	Boyd‐Sakami D.	Southern Indian Lake	Oahe	Francis Case
Region	Boreal	Boreal	Boreal	Boreal	Boreal	Temperate	Temperate
Latitude	52°38′58″ N	53°45′00″ N	54°31′46″ N	53°13′14″ N	57°20′11″ N	44°42′40″ N	43°25′01″ N
Longitude	76°19′54″ W	77°00′00″ W	69°51′18″ W	76°45′36″ W	98°20′31″ W	100°37′44″ W	99°09′24″ W
Trophic status	Oligotrophic	Oligotrophic	Oligotrophic	Oligotrophic	Mesotrophic	Mesotrophic	Mesotrophic
Area (km2)	1040	2835	4275	738	2391	1268	412
Area flooded (km2)	740 (71%)	2630 (92%)	3430 (80%)	–	414 (17%)	833 (66%)	280 (68%)
Volume (km3)	8.4	61.7	53.8	–	23.4	23.7	4.1
Filling time (yr)	0.5	1	2	1	1	9	5
Residency time (days)	124	183	803	80	263	526	103
Mean depth (m)	8	22	12	8	10	19	15
Max depth (m)	51	137	49	110	30	55	43
Annual drawdown (m)	3.6	3.3	2.1	1	<1	3.8	6.9
Watershed area (km2)	30,000	97,643	36,800	49,835	242,000	19,253	11,472

Notes

Reservoir area and reservoir volume represent the surface covered with water and volume at maximum pool. The area flooded represents the surface of terrestrial land flooded following impoundment and, in brackets, the percentage of the reservoir that was terrestrial land before impoundment.

We considered two different baselines (t₀) to base our analyses on comparisons with pre‐impoundment data because filling time varied substantially among reservoirs. In the first case, we used t_0BF as the time when the reservoir filling began. Our second baseline was t_0EF, which we defined as the time when filling of the reservoir was completed and the dam was operational. When using t_0BF (t₀ = reservoir filling started), 75% of the time series for recruits and 21% of the time series for adult fish (all in temperate reservoirs) did not have pre‐impoundment data (Appendix S1). For the time series having pre‐impoundment data, the number of years during this period was 3.4 ± 1.9 yr (mean ± 1 SD) for recruits and 1.8 ± 0.6 for adult fish, respectively; the number of years post‐impoundment was 12.9 ± 4.1 yr for recruits and 13.2 ± 2.3 yr for adult fish. When using t_0EF, the number of years before impoundment was 3.4 ± 1.9 yr for recruits and 2.4 ± 1.6 yr for adult fish; the number of years post‐impoundment was 7.9 ± 4.1 yr for recruits and 11.9 ± 3.6 yr for adult fish.

Data analysis

Support for TSH (detecting a hump‐shaped pattern)

To test whether the TSH was supported by each individual time series, we compared the fit of the data to alternative functions (or scenarios) corresponding to plausible general abundance patterns that could be observed for the period covering before impoundment, during reservoir filling, and after impoundment (Appendix S1: Table S1, Fig. S1). Specifically, we tested for the presence of one of the following trends: (1) no pattern over time (flat line function), (2) a linear increasing pattern, (3) a linear decreasing pattern, (4) a non‐linear decreasing trend (negative exponential function), (5–6) two non‐linear increasing patterns (negative exponential and quadratic polynomial functions), (7–8) two hump‐shaped patterns suggested to capture the whole trophic surge hypothesis (using either Ricker and negative quadratic polynomial functions; Appendix S1: Table S1, Fig. S1). Parameters used to fit each function to the observed data were estimated by maximizing the negative binomial, the Poisson, or the Gamma likelihood depending on the data distribution and residuals analysis in each time series, and using the mle2 function with the Nelder‐Mead method available on the bbmle package version 1.0.18 (Bolker and R Development Core Team ). To evaluate which function fit best the observed data, we used Akaike's information criterion modified for small sample sizes (AIC_c). We calculated the difference between AIC_c for each model i and the lowest observed AIC_c (∆AIC_c) and compiled normalized Akaike weights (w_i) for each function to estimate the probability that function i is the best function, given the data and set of candidate functions (Burnham et al. ). All analyses were performed in R (v. 3.2.0). For completeness, we also visually inspected the time series to detect patterns that were not captured statistically by the curve fitting analysis.

To evaluate whether there was support for TSH across datasets, we conducted analyses using generalized additive mixed effect models (gamm; Wood , ). We considered six different combined datasets: (1) all recruitment time series (n = 40), (2) recruitment time series in boreal reservoirs (n = 10), (3) recruitment time series in temperate reservoirs (n = 30), (4) all adults time series (n = 109), (5) adult time series in boreal reservoirs (n = 89), and (6) adult time series in temperate reservoirs (n = 20). We used gamm to control for nested random factors (reservoirs and sites within reservoirs) and to model non‐linear relationships (“smooth terms”). Models were fit by penalized likelihood for each smooth function added and selected by generalized cross‐validation (i.e., GCV, an internal model selection procedure; Wood ). Because of potential temporal autocorrelation between successive years, we used an autoregressive correlation structure (corAR1) in the gamm. Gamm were fit using the mgcv package v. 1.7–28 in R (Wood ). For completeness, we also compared the fit of each of the six combined datasets to the eight alternative scenarios using the curve fitting method mentioned above.

Analyzing TSH metrics to develop a predictive framework

We extracted six TSH metrics that could be used in a predictive framework. These TSH metrics were as follows: (1) the occurrence of the TSH (i.e., detection of a hump‐shaped pattern or not), (2) the duration of the trophic non‐equilibrium phase in years (i.e., the time needed for abundance values to either come back to values comparable to pre‐impoundment or to reach a stable state), (3) the duration of the surge in years (i.e., time needed to reach the peak in abundance from t₀), (4–5) the magnitude of the peak in abundance in relation to t_0BF and t_0EF, and (5–6) the timing of the peak (i.e., year at which the peak occurred) in relation to t_0BF and t_0EF. We defined the duration of the trophic non‐equilibrium and surge phases based on visual inspection of the time series, whereas the magnitude of the peak was extracted by dividing the actual observed value at the peak for a given time series by the mean value of abundance before impoundment.

We then correlated four TSH metrics (see Tables  and ) with reservoirs characteristics that can influence fish population dynamics based on existing literature (see Table ). Values were mainly extracted from the six studies from which we extracted the time series, but some values were also extracted from other complementary articles and reports (Newbury et al. , Schetagne et al. , Dickerson et al. ). We used generalized linear mixed effect model (glmm) to correlate TSH metrics to reservoirs characteristics. We controlled for reservoir and sampling stations identity by using nested random factors in the models. For the occurrence of the trophic surge, we used a binomial error structure, and for the duration of the surge, we used a Gaussian error structure (glmer function in R; lme4 package v1.1‐12). We used the glmulti function (glmulti package 1.0.7) to perform model selection using AIC_c. We excluded reservoir area as a predictor because of multicollinearity problems with other variable and based on the observation that it explained a lower proportion of the variation. For each TSH metric, we evaluated the fit of candidate models, including all plausible additive effects. We considered as plausible models all candidate models that were within two AIC_c from the best model (ΔAIC_c < 2). If more than one model had support, we performed model averaging. To determine the reliability of the predictor estimates from averaging, we calculated the weighted unconditional variance with its associated confidence intervals (95% CI). To assess the relative importance of each predictor, we also use the normalized Akaike weights (w_ip). To calculate w_ip, the Akaike weights calculated for each model that contains the parameter of interest are summed.

Results

Is the TSH supported by individual recruitment time series?

A hump‐shaped trend was the predominant pattern identified across individual recruitment time series based on curve fitting and model selection. Given that 43% of the time series displayed the hump‐shaped pattern, these data suggest that the TSH has moderate support in explaining recruits' abundance after impoundment (Fig. a; Appendix S1: Table S2). Among those time series, the Ricker hump‐shaped pattern had equal support to the quadratic one (1.05 time more likely based on evidence ratio; w_i; 0.2505/0.245; Appendix S1: Table S2). We detected an increasing or decreasing trend in 20% and 17% of the time series, respectively, and were not able to detect any trend in 20% of the time series (Fig. a; Appendix S1: Table S2).

Summary of the number of species‐by‐reservoir time series for which (a) recruitment (young of year fish abundance; YOY) and (b) adult fish abundance showed increasing, hump‐shaped, decreasing, or no trend across the period of record based on statistical evidence from the curve fitting analysis and model selection with AICc.

When a hump‐shaped pattern was detected, the trophic non‐equilibrium phase lasted about nine years after filling, when abundance values came back to those comparable to pre‐impoundment period (Fig. a). The surge (i.e., time from t₀ to reach the peak in abundance) lasted about four years (Fig. b). When using t_0EF, the magnitude of recruitment in boreal and temperate reservoirs was, on average, three to four times (respectively) higher after impoundment, relative to recruitment values during and after filling (Fig. c). When using t_0BF, the peak in recruitment was still three times higher after impoundment than before in boreal reservoirs; no temperate reservoirs had pre‐impoundment data, so this test could not be performed (Fig. d). The peak in recruitment occurred on average just before the dam was operational or during filling in boreal (t_0EF; Fig. e), and eight years after the reservoir filling started in temperate reservoirs (t_0BF; Fig. f).

Frequency distribution (number of time series) of six trophic surge hypothesis metrics using kernel curves: (a) duration of the trophic non‐equilibrium (TNE; in years), (b) duration of the surge (in years), (c, d) magnitude of the peak in abundance for t0EF and t0BF, and (e, f) timing of the peak in abundance for t0EF and t0BF. The white dots on each panel represent the mean of the metrics for adults, and the black dots, the mean for recruitment in boreal and temperate reservoirs.

Based on the data available, there was no clear support for a trophic equilibrium state. We relied on visual inspection of the time series to evaluate this phase as we were now zooming in on only a fraction of the time series. Using this approach, 50% of the time series showed a directional change over time, equally balanced between increasing and decreasing trends and <23% of them convincingly reached the trophic equilibrium state.

Is the TSH supported by individual adult fish time series?

Broad support for the TSH was also apparent with the individual adult fish time series as we found that the hump‐shaped was the predominant pattern (Fig. b; Appendix S1: Table S3). For adult time series, the Ricker hump‐shaped pattern was only mildly more likely than the polynomial pattern (1.1 times; Appendix S1: Table S3). We detected an increasing or decreasing trend in 7% and 23% of the time series, respectively, and were not able to detect any trend in 25% of the time series (Fig. b; Appendix S1: Table S3).

The trophic non‐equilibrium phase, when detected in the adult fish time series, was longer in duration than what was observed for recruitment. Adult fish abundance conservatively came back to values comparable to pre‐impoundment after 14 yr in boreal and after nine years in temperate reservoirs (Fig. a). This is a conservative estimate because several time series were still in the depression phase at the end of the time series. The surge lasted about four years in both ecosystems (Fig. b). When using t_0EF, adult fish abundance was on average seven or 16 times higher after impoundment relative to values observed during and after filling in temperate and boreal reservoirs, respectively (Fig. c). When using t_0BF, the peak was about 18 times higher after impoundment than before in boreal reservoirs. No temperate reservoirs had pre‐impoundment data (Fig. d). The peak in recruitment occurred in boreal reservoirs, on average, just before the dam was operational (t_0EF; Fig. e) or four years after reservoir filling (t_0BF; Fig. f); these peaks were slightly later in the temperate reservoirs. Based on visual inspection of the time series, <23% convincingly reached the equilibrium state. Overall, 72% of the time series showed a directional change over time. On average, 28% of the time series demonstrated an increase in abundance (29% of the time series in boreal, and 20% in temperate reservoirs) and 44% demonstrated a decrease in abundance over time (42% of the time series in boreal, and 50% in temperate reservoirs). The other 28% of the time series showed no directional trend over time.

Is the TSH supported by combined datasets?

The support we found for the TSH with the combined datasets varied between regions. We found no support for the TSH when the 40 recruitment time series were combined or examined for temperate reservoirs (n = 30) using both the gamm (P‐value of smooth terms >0.07; Fig. a, b) and curve fitting analysis (Table ). However, we did detect support for the TSH when we examined boreal time series alone (n = 10; gamm, P‐value of smooth terms <0.001 for t_0BF and t_0EF; Fig. a, b, Table ). Patterns were comparable for t_0BF (t₀ = reservoir filling started; Fig. a) and t_0EF (t₀ = time at which the dam is operational and the filling is completed; Fig. b). In boreal reservoirs, data showed a clear hump‐shaped pattern, where recruitment peaked shortly after reservoir filling (i.e., the surge), and decreased after impoundment (i.e., the depression; Fig. a, b). The support for a hump‐shaped pattern was more than two times more likely when using t_0BF than t_0EF (Table ).

General patterns in the time series for combined species and reservoirs, for recruits (a, b) and adult fish (c, d) and with two different t0 (t0BF = at the beginning of reservoir filling, and t0EF = end of reservoir filling, dam is operational). In each panel, the general trend for combined species and reservoirs (black line), for boreal reservoirs only (blue dots and blue line), and for temperate reservoirs only (green dots and green line) is also presented and modeled using gamm.

Support for the alternative functions (see Appendix S1: Table S1 and Fig. S1 for the description of the functions) to examine the support for the trophic surge hypothesis in recruitment and adult fish time series for all species combined using AIC_c and w_i

Time series	No trend	Linear	Negative exponential	Polynomial	Ricker
AICc	w i	AICc	w i	AICc	w i	AICc	w i	AICc	w i
Recruits—All reservoirs (no. of time series = 40)
t 0BF	1388.26	0.402	1390.31	0.145	1390.25	0.149	1390.11	0.160	1390.31	0.145
t 0EF	1388.26	0.335	1390.29	0.122	1388.51	0.297	1390.25	0.124	1390.29	0.122
Recruits—Boreal (no. of time series = 10)
t 0BF	261.13	0.027	263.25	0.009	263.17	0.010	254.41	0.776	257.35	0.178
t 0EF	261.13	0.307	263.25	0.107	263.27	0.105	260.73	0.375	263.27	0.105
Recruits—Temperate (no. of time series = 30)
t 0BF	1137.75	0.328	1139.74	0.121	1138.00	0.290	1139.74	0.121	1139.42	0.142
t 0EF	1137.75	0.335	1139.70	0.126	1138.00	0.296	1139.78	0.121	1139.78	0.121
Adults—All reservoirs (no. of time series = 109)
t 0BF	4302.12	0.000	4282.82	0.764	4285.72	0.179	4288.02	0.057	4304.19	0.000
t 0EF	4304.19	0.000	4270.38	0.999	4285.72	0.001	4288.02	0.000	4302.12	0.000
Adults—Boreal (no. of time series = 89)
t 0BF	3517.61	0.000	3499.40	0.006	3519.74	0.000	3489.34	0.994	3519.62	0.000
t 0EF	3517.61	0.000	3499.40	0.006	3519.62	0.000	3489.34	0.994	3519.74	0.000
Adults—Temperate (no. of time series = 20)
t 0BF	790.60	0.000	776.72	0.062	792.64	0.000	771.28	0.938	788.44	0.000
t 0EF	790.60	0.000	776.72	0.062	771.28	0.938	788.90	0.000	788.44	0.000

Notes

We examined the support for boreal, temperate reservoirs, and for both regions combined using t_0BF = beginning of the filling phase and t_0EF = end of the filling phase. The function with the highest support has AIC_c and w_i values in bold.

Using the combined adult datasets, reservoirs filling times and regional attributes were related to our detection of the TSH. For example, we found that the delineation of t₀ (i.e., using t_0BF or t_0EF) affected the interpretation of the trends when using adult fish time series. This result is likely because temperate reservoirs took a longer time to fill than boreal reservoirs (Table , Fig. c, d). When using gamm and t_0BF, we found support for the TSH in boreal (gamm; P‐value of smooth term <0.001, n = 89) and temperate reservoirs (gamm; P‐value of smooth term <0.001, n = 20) as well as when both climatic regions were combined (gamm; P‐value of smooth term <0.001, n = 20; Fig. c). Support for the TSH in boreal and temperate reservoirs was corroborated by a curve fitting analysis applied to these same datasets, except that this latter analysis detected a decreasing linear trend when both climatic regions were combined (Table ). When using t_0EF, we detected a hump‐shaped pattern in boreal reservoirs (gamm; P‐value of smooth term <0.001, and curve fitting analysis; Table ), a decreasing non‐linear trend in temperate reservoirs (gamm; P‐value of smooth term <0.001, and curve fitting analysis; Table ), a decreasing non‐linear trend when climatic regions were combined with gamm (P‐value of smooth term <0.001, Fig. d), and a linear decreasing trend with curve fitting analysis (Table ).

Predictors of the TSH metrics

Overall, reservoir characteristics had moderate to mild support in explaining variation in the four TSH metrics extracted from recruitment (Table ) and adult fish time series (Table ). The area of terrestrial land flooded, and the percentage of terrestrial land flooded following impoundment, were important predictors of numerous TSH metrics in both recruitment and adult time series. The probability of detecting a trophic surge tended to increase as a positive function of watershed area for recruitment (Table ) and significantly increased as a positive function of percentage of the reservoir that was land before impoundment for adult fish (Table ). For the duration of the surge, we found a tendency for this metric to decrease as a function of increasing area flooded and increasing percentage of the reservoir that was land before in the recruitment time series (Table ) and to increase with the area flooded and decrease with increasing filling time and water residency in adult fish (Table ). To explain the magnitude of the surge in recruitment time series, no model was better than the null, whereas with the adult fish time series, the magnitude of the surge significantly increased with increasing area flooded (Table ). Finally, both types of time series showed that the timing of the surge (i.e., when the peak in abundance was observed) significantly decreased with increasing area flooded and increased with increasing filling time (Tables  and ).

Predictor estimates ± SE or unconditional variance (UV; if using model averaging), 95% CI, and importance Weight (w_ip) of the predictors used to predict the effect of reservoir characteristics on four TSH metrics for recruitment time series

Predictors	(Intercept)	Watershed area	Area flooded	L. flooded (%)	Filling time	Residency time
Occurrence of the trophic surge (two models within two AICc)
Estimate ± SE	−0.263 ± 0.167	1.303 ± 0.959	−0.198 ± 0.408	–	–	–
95% CI	−0.590 to 0.064	−0.585 to 3.183	−0.998 to 0.602	–	–	–
w ip	1.00	1.00	0.42	–	–	–
Duration of the trophic surge (two models within two AICc)
Estimate ± SE	4.347 ± 0.640	–	−0.312 ± 0.275	−0.240 ± 0.187	–	–
95% CI	3.093 to 5.601	–	−0.851 to 0.226	−0.607 to 0.127	–	–
w ip	1.00	–	0.57	0.43	–	–
Magnitude of the trophic surge (No model had more support that the null model)
Timing of the surge (peak; one model within two AICc)
Estimate ± SE	5.713 ± 0.784	–	−1.676 ± 0.773	–	2.368 ± 0.967	–
95% CI	4.176 to 7.250	–	−3.191 to −0.161	–	0.473 to 4.263	–
w ip	1.00	–	1.00	–	1.00	–

Notes

The predictors that did not include zero within its 95% CI are in bold. The set of best candidate models were selected using candidate GLMMs that were within two AIC_c form the best model.

Predictor estimates ± SE or unconditional variance (UV; if using model averaging), 95% CI, and importance Weight (w_ip) of the predictors used to predict the effect of reservoir characteristics on four TSH metrics for adult fish time series

Predictors	(Intercept)	Watershed area	Area flooded	L. flooded (%)	Filling time	Residency time
Occurrence of the trophic surge (one model within two AICc)
Estimate ± SE	−0.353 ± 0.046	–	–	1.018 ± 0.145	–	–
95% CI	−0.443 to −0.263	–	–	0.734 to 1.302	–	–
w ip	1.00	–	–	1.00	–	–
Duration of the trophic surge (three models within two AICc)
Estimate ± SE	4.730 ± 0.104	–	0.170 ± 0.131	–	−0.322 ± 0.167	−0.085 ± 0.054
95% CI	4.526 to 4.934	–	−0.087 to 0.427	–	−0.649 to 0.005	−0.191 to 0.021
w ip	1.00	–	1.00	–	0.45	0.22
Magnitude of the trophic surge (one model within two AICc)
Estimate ± SE	−0.084 ± 0.035	–	0.413 ± 0.039	–	–	–
95% CI	−0.153 to −0.015	–	0.337 to 0.489	–	–	–
w ip	1.00	–	1.00	–	–	–
Timing of the surge (peak; four models within two AICc)
Estimate ± SE	5.051 ± 0.143	0.307 ± 0.451	−0.869 ± 0.712	–	0.425 ± 0.292	0.163 ± 0.286
95% CI	4.771 to 5.331	−0.577 to 1.191	−2.265 to 0.526	–	−0.147 to 0.997	−0.398 to 0.724
w ip	1.00	0.16	1.00	–	0.68	0.15

Notes

Only the predictors retained in candidate models are represented in the table. The predictors that did not include zero within its 95% CI are in bold. The set of best candidate models were selected using candidate GLMMs that were within two AIC_c form the best model.

Discussion

By testing the TSH on 149 time series of recruitment and adult fish abundance from seven reservoirs, this study is the most comprehensive work that quantitatively evaluated the short‐ and long‐term responses of fish population to impoundment. It also represents the first attempt to correlate metrics of the TSH with reservoir characteristics to build a general framework to predicting the trajectories of fish over time in reservoir. Three major insights emerged from our analysis: First, we provided broadscale support for the TSH by conservatively demonstrating that the expected hump‐shaped pattern was the predominant pattern in recruitment and adult fish time series. Second, regarding the generality of TSH metrics, when a hump‐shaped pattern was detected, a peak in recruitment was observed quickly after the start of reservoir filling, followed by an increase in adults. The non‐equilibrium phase (i.e., surge and depression) lasted on average seven to eight years for recruits and conservatively more than 10 years in adults, although variation was apparent among reservoirs and climatic regions. Finally, we found that reservoir characteristics provided mild to moderate support for the variation in TSH metrics and identified the total flooded area as being the most influential predictor.

Variation in the trophic surge

Our analyses showed that a trophic surge was observed in many fish time series in newly created temperate or boreal reservoirs (Fig. ). Our estimates are likely to be conservative because visual inspection of the time series suggested that an additional 13% and 17% of the time series from recruits and adult fish could be categorized as being consistent with the TSH (see Appendix S1: Fig. S2). These results strongly suggest that fish productivity levels observed within the first 15 years of impoundment, during the non‐trophic equilibrium, should not be used to formulate long‐term management recommendations. We also found that the area of terrestrial land flooded following impoundment was a recurring predictor explaining several TSH metrics. The probability of occurrence of the surge (i.e., detecting a hump‐shaped), the magnitude of the surge, and the timing of the surge increased with increasing area flooded and with increasing percentage of flooded area that was terrestrial before impoundment. These results make ecological sense as the amount of flooded area represents a reasonable proxy for the amount of nutrient exported in reservoirs and the availability of newly created habitat for fish (Ostrofsky , Grimard and Jones ). We also found convincing relationships between the timing of the surge (when the peak in abundance occurred) in the adult time series and reservoir filling time. Based on phosphorus budgets, a longer filling time should result in a longer trophic surge and trophic non‐equilibrium stage that is of a lower magnitude (Grimard and Jones ). These correlations suggest that phosphorus budgets in reservoirs are a key factor controlling the magnitude and duration of the trophic non‐equilibrium phase, and they also suggest that newly created reservoirs are strongly bottom‐up regulated. We view these correlations as informative and help to advance the development of a conceptual framework, but could be improved upon in the future as our sample size is still relatively small (n = 7 reservoirs).

Interestingly, when time series were considered at the individual or climatic zone level, we detected support for the TSH, but this support was weaker when time series were combined. These findings echo the variation reported in the literature and suggest that there is considerable variability at the level of reservoirs, sampling stations, and species. The duration of the trophic surge for fish reported in the literature ranges between 2 and 25 years (2–3 yr: Baranov ; 5–10 yr with some lasting more than 25 yr: Cooper and Hubbell ; 10–15  yr: Allen ). However, these estimates were mostly based on qualitative analyses of the data, and it is unclear whether authors referred to the trophic surge (i.e., the increasing portion of the curve) or the whole trophic non‐equilibrium phase (i.e., surge and depression).

Other sources of heterogeneity can explain the variability observed. All else being equal, one could expect that as physical, chemical, and biological processes are accelerated under warmer conditions (Grimard and Jones , Straškraba et al. ). As such, one could hypothesize that the trophic surge and trophic non‐equilibrium phase should be shorter in tropical ecosystems, intermediate in temperate, and longer in boreal ones. Consistent with this logic, we found that the trophic non‐equilibrium phase was on average longer in boreal reservoirs compared to temperate ones. Furthermore, indirect evidence from commercial and recreational fisheries landings and knowledge of tropical reservoirs suggest that the duration of the trophic surge can be relatively short (2–10 yr; Henderson et al. , Balon and Coche , Williams et al. , Agostinho et al. , Hallwass et al. , Cottet et al. ).

Fish species characteristics, trophic interactions among species, and management actions can also explain the variability observed in our ability to detect a trophic surge. Fish recruitment, production, and abundance in reservoirs strongly depend on species reproductive success, growth rate, behavior, and trophic position in this novel ecosystem. Some authors observed declines in abundance for species requiring flooded vegetation and littoral zones when drawdown is important, whereas an increase in abundance has been reported for species using open water, tributary streams, the upper riverine zone of the reservoir, or the gravel or cobble shorelines (Benson , Hendrickson and Power ). In this study, some taxa showed a striking consistency in trends. For example, the northern pike showed a clear and consistent hump‐shaped pattern in boreal reservoirs (Fig. a), whereas the white sucker showed a consistent decline over time (Fig. b). Other species, such as the walleye, showed no consistent patterns (Fig. c). Based on visual assessment of the time series, some walleye and whitefish (Coregonus clupeiformis) time series in boreal reservoirs demonstrated an obvious “U”‐shaped pattern, depicted by a trend where abundances were relatively high before impoundment, then reached low values shortly after impoundment, and then increased to reach their highest values roughly 10 years after impoundment. This “U”‐shaped pattern has been suggested to be related to rapid redistribution following new conditions and to the strength of trophic interactions, among species, such as predation and competition (Cook and Bergersen , DesLandes et al. , Nunn et al. , Probst et al. ).

Example of time series of normalized adult fish abundance for (a) Esox lucius (northern pike), (b) Catostomus commersoni (white sucker), and (c) Sander vitreus (walleye) before impoundment, during filling (represented by the gray area with open symbols), and after impoundment. References: (1) Doyon and Belzile , (2) DesLandes et al. , (3) Strange et al. , (4) Nelson and Walburg , (5) Benson .

Reservoir management actions such as fish stocking, fishing, and mitigation measures could also alter population trajectories and affect our ability to detect the TS. For example, one mitigation measure applied in some reservoirs is to add nutrients in the reservoir to delay the depression phase (Agostinho et al. , Pegg et al. ). In these particular circumstances, the detection of a hump‐shaped pattern will be very unlikely because the depression phase will be replaced by a plateau or a slow decreasing trend in abundance. However, accounting for reservoir management actions on fish population trajectories will be difficult because some actions were successful, but some have failed, and most available data on management actions are qualitative.

Opportunities for future research

Although this is the most comprehensive compilation of fish time series to address the TSH quantitatively, the dataset nonetheless is limited and prevented us notably from drawing strong conclusions regarding the equilibrium phase. Below, we have distilled a number of key attributes regarding the time series, which are key considerations that should be addressed in future work.

First, the length of the time series before filling can limit the detection of a hump‐shaped pattern and may influence the interpretation of the trends. Unfortunately, <25% of the available time series had pre‐impoundment data. Even with this more select group of time series, our ability to appropriately characterize the pre‐impoundment conditions was limited because we had, at best, 1–2 years of data before filling. It is well recognized that fish recruitment is highly stochastic and fluctuates among years (Karjalainen et al. , Myers ), but the consistent trends reported with the adult time series give us confidence that the trophic surge patterns detected are robust.

Second, the paucity of long time series after impoundment impeded our ability to develop strong conclusions regarding the abundances that should be observed at equilibrium, and the time needed to reach such an equilibrium. The longest time series in this study covered 17 years after impoundment in adult abundance (Doyon and Belzile ), and more than 72% the time series for adult fish were still in the depression phase at the end of the series.

Finally, in the future, it would be ideal to have time series from multiple stations in reservoirs as well as comparable data on adjacent reference ecosystems. It is well known that reservoirs combine features of both river and lake environments, creating a very diverse set of habitats along the reservoir (lacustrine, transitional zone, and river; Kimmel and Groeger , Siler et al. , McKenna et al. , Okada et al. , de Paiva Affonso et al. ). As such, it would be very useful to have multiple sampling stations at different distances from the dam so that variation in time series trends could be attributed to site‐specific difference. Likewise, reference sites would allow investigators to control for changes in climate or environmental stochasticity unrelated to the reservoir. Despite these caveats, we still found that the predominant pattern across the recruitment and adult time series collected at different times from two contrasting climatic zones was supportive of the trophic surge hypothesis.

Concluding remarks and recommendations

Our analysis of boreal and temperate reservoirs highlights a predominant, but transient positive response in fish recruitment to impoundment, followed by similar response of adult abundance. We have shown that the TSH and related metrics are useful in developing a general conceptual framework to predict and understand the response of fish populations to reservoir creation. Testing the TSH on fish was a first step. The next logical step would be to test the generality of the TSH on biogeochemical processes (e.g., phosphorus budget, sedimentation, GHG emissions) and other organisms from different trophic levels (e.g., primary producers, zooplankton, zoobenthos) across geographic latitudes. Then, we could develop a general framework exploring the ecological interactions between the different components (e.g., distance and lag between the peaks in abundance) from a bottom‐up perspective. Interestingly, Ostrofsky (), Straškraba et al. (), and Kimmel and Groeger () graphically explored the trajectory of several factors relevant to reservoir aging and dam closure (e.g., water conductivity, oxygen, total phosphorus, phytoplankton biomass, fish landings) and noted that many relationships were clearly hump‐shaped. Based on these trajectories, long‐term management recommendations should be not formulated before convincing evidence that the reservoir reached its new trophic equilibrium.

To grow this theory further, more sites with monitoring of fish populations before, during, and after reservoir filling are needed as well data on adjacent reference sites (lakes and rivers/streams; BACI design). A strong predictive framework is a key management tool that serves to reduce the risk of inappropriate actions, where a transient surge might be falsely interpreted as very productive fishery (Ploskey , Quinn and Kwak ). As suggested by Kimmel and Groeger (), an integrated and transdisciplinary research effort of the effects of impoundment and aquatic ecosystems regulation is desirable for the future of reservoir water quality, fisheries, aquatic ecosystem services, functions, and integrity.

Acknowledgments

This work was supported by a MITACS Accelerate grant to IGE, CS, and CN, and a MITACS Elevate scholarship to KT with Hydro‐Québec as an industrial partner. We thank C. Turpin (Hydro‐Québec), A. Tremblay (Hydro‐Québec), and P. Johnston (Hydro‐Québec) for helpful suggestions and comments on an early draft of this manuscript.