Fig 1.
(A) Temperature and predation directly and indirectly affect population density and metabolic rates in aquatic communities. In our experimental communities, predation directly (solid lines) affects the abundance, size, and species composition of prey, and predation by notonectids on grazers leads to an indirect effect (dashed line) called a trophic cascade on algal abundance. Temperature directly affects per capita metabolic rates (solid lines) and indirectly affects algal abundance (dashed lines) by increasing grazing rates, and may have stronger effects on heterotrophic metabolic rates relative to algal metabolic rates (thicker lines represent a stronger direct effect of temperature). Other indirect effects of temperature are possible. (B) Experimental communities varied in their trophic structure. Ten communities included algae only (A), 10 comprised algae + grazers (AG), and 10 included algae + grazers + predators (AGP). We sampled net ecosystem oxygen production (NEP), ecosystem respiration (ER), and total phytoplankton biomass (MB) weekly for 8 weeks. AG, algae + grazers; AGP, algae + grazers + predators; ER, ecosystem respiration; MB, total phytoplankton biomass; NEP, net ecosystem oxygen production.
Fig 2.
(A) Estimated phytoplankton biomass (chlorophyll a concentration) declined with increasing temperature and varied with trophic structure (A, AG, AGP). Lines are estimated effects of temperature on phytoplankton biomass based on LMMs (Eq 12) for Eq 3, with temperature dependence in model terms for the intercept and slope (Table 1). From the best model, the intercept and slope of each line were estimated by pooling terms for the intercept and temperature dependence in Eq 12 (see Methods, Eq 14). All observations for phytoplankton biomass are shown in Fig 6. (B) Strength of the trophic cascade at a given temperature was estimated by taking the log ratio of algal biomass (estimated as chlorophyll a concentration) in the presence of predators and grazers (AGP) versus the algal abundance in the presence of grazers only (AG) (Eq 4, Table 2). Lines represent fixed effects of temperature from the full model (Table 2), centered on the grand mean of all recorded ecosystem temperatures (Eq 13). Gray shading and symbols indicate the week, from week 2 (July 10) to week 9 (August 28), 2012. Data for these figures may be found at https://doi.org/10.5281/zenodo.2652579 in GarzkeAllwks.csv. A, algae only; AG, algae and grazers; AGP, algae, grazers, and predators; Chla, chlorophyll a; LMM, linear mixed effects model.
Fig 3.
Comparison of estimated temperature dependences of phytoplankton biomass (MB), NEP, and ER for communities with algae only (A), algae and grazers (AG), and algae, grazers, and predators (AGP).
Composite estimates of temperature dependences are as shown in Figs 2A and 5 (following Methods, Eq 14). No temperature dependence is indicated by the dashed line, and the vertical gray dotted lines indicate 0.65 and 0.32 eV, expected temperature dependences of algal photosynthesis and respiration, and −0.65 and −0.32 as expectations for the temperature dependence of phytoplankton total biomass. Data for these figures may be found at https://doi.org/10.5281/zenodo.2652579 in GarzkeAllwks.csv. A, algae only; AG, algae and grazers; AGP, algae, grazers, and predators; ER, ecosystem respiration; NEP, net ecosystem production.
Table 1.
Model selection results for LMMs of phytoplankton biomass.
The full statistical model (Methods, Eq 12) related ln(chlorophyll a) to ecosystem trophic structure (Zj) and average ecosystem temperature over the entire experimental period (TM), while accounting for effects of temperature variation over time (weekly average temperature [Twj]) and with ecosystem identity as a random effect. We compared models using likelihood ratios (LogLik), AICC, Akaike weights (w), and δAICC weights. The model was fit to 240 observations in 30 groups. The full model (PBF) includes all terms, and models representing alternate hypotheses excluded terms indicated by “NA.” Values indicate model-estimated coefficients. Coefficients were pooled (Methods: Statistical analysis) to estimate slopes and intercepts for Figs 2 and 3.
Table 2.
Model selection results for trophic cascade analysis.
We used LMMs with terms for average temperature for ecosystem j in week w (Twj), weeks 2–9 (Wk), and their interaction. We treated the power level (e.g., 100 W, 200 W, etc.), our temperature treatment, as a random effect to account for repeated measures on ecosystems over time. We compared models using likelihood ratios (LogLik), AICC, Akaike weights (w), and δAICC weights. The model was fit to 79 observations in 10 groups. The full model (TCFull) includes all terms, and models representing alternate hypotheses excluded terms indicated by “NA.” Coefficients were pooled (Methods: Statistical analysis) to estimate slopes and intercepts for Fig 2.
Fig 4.
(A) Total zooplankton density (ln(ind)/10 L), comprising Daphnia and copepod taxa, declined with increasing temperature but not with predator presence. (B) Daphnia density (ind/L) declined with predators (gray dashed line, versus black line indicating trend with no predators) (Table 4), and (C) copepod spp. density (ln(ind)/10 L) declined with temperature but not predators (Table 5). Lines are regressions, with ecosystem as a random effect for ecosystems with predators (gray lines) and without predators (black solid line). Each data point is an observed total zooplankton density for crustacean taxa (Daphnia and copepods) in each ecosystem on a sampling date. Data for these figures may be found at https://doi.org/10.5281/zenodo.2652579 in GarzkeAllwks.csv. ind, individuals.
Table 3.
Results of model selection for zooplankton abundance in ecosystems with grazers (AG) and with grazers and predators (AGP). We used linear regressions (Methods: Statistical analysis). Models included terms for weekly average temperature (Twj), ecosystem trophic treatment (Zj) and their interaction, and a random effect for ecosystem identity. We modeled 120 observations in 20 groups (ecosystems). We compared models using likelihood ratios (LogLik), AICC, Akaike weights (w), and δAICC weights. NA indicates that the term was not included in the model.
Table 4.
Daphnia density: Results of model selection for Daphnia abundance in ecosystems with grazers and with grazers and predators.
We used linear regressions (Methods: Statistical analysis). Models included terms for weekly average temperature (Twj), ecosystem trophic treatment (Zj) and their interaction, and a random effect for ecosystem identity. We compared models using likelihood ratios (LogLik), AICC, Akaike weights (w), and δAICC weights. We modeled 120 observations in 20 groups (10 AGP ecosystems with predators, and 10 AG ecosystems without predators). NA indicates that the term was not included in the model.
Table 5.
Copepod density: Results of model selection for copepod spp. abundance in ecosystems with grazers and with grazers and predators.
We used linear regressions (Methods: Statistical analysis). Models included terms for weekly average temperature (Twj), ecosystem trophic treatment (Zj) and their interaction, and a random effect for ecosystem identity. We compared models using likelihood ratios (LogLik), AICC, Akaike weights (w), and δAICC weights. We modeled 120 observations in 20 groups (10 AGP ecosystems with predators, and 10 AG ecosystems without predators). NA indicates that the term was not included in the model.
Fig 5.
The effect of mean ecosystem temperature on (A) NEP and (B) net ER for three community types that varied in their trophic interactions: (i) algae-only (A), (ii) algae + grazers (AG), and (iii) algae + grazers + notonectid predators (AGP).
Black lines indicate the among-ecosystem effects of temperature, modeled by Eq 5 using hierarchical regressions fit to among-ecosystem variation in temperature, after taking into account within-group variation temperature effects (light lines) (Tables 1, 6 and 7). Temperature dependences within and among tanks were estimated by best model or best model set (Tables 1, 6 and 7, Methods: Statistical analyses). Temperature in Celsius is shown for comparison only; models were fit to inverse temperature. All measured data points to which models were fitted are shown in Fig 6. Temperatures within tanks declined over time (S2 Fig). Data for these figures may be found at https://doi.org/10.5281/zenodo.2652579 in GarzkeAllwks.csv. A, algae only; AG, algae and grazers; AGP, algae, grazers, and predators; ER, ecosystem respiration; NEP, net ecosystem production.
Fig 6.
The effect of ecosystem temperature (Twj) on (A) phytoplankton biomass, (B) NEP, and (C) net ER for three community types that varied in their trophic interactions: (i) algae-only (A), (ii) algae + grazers (AG), and (iii) algae + grazers + notonectid predators (AGP).
There were 10 ecosystems (j) in each trophic treatment, and each ecosystem was sampled 8 times (once per week from weeks 2 to 9). Each week is indicated by a symbol shape, and ecosystem identities within weeks are distinguished by shades of gray. In a single model (Eq 13), we considered effects of temperature within ecosystems over time, as well as among-ecosystem variation in mean temperature (Figs 2 and 5). Blue lines are fit to the 8 observations (points) from each ecosystem (one from each week), and their slope indicates within-ecosystem temperature effects estimated from best models in Tables 2, 6 and 7. Black lines indicate the modeled among-ecosystem effects of temperature (Tables 1, 6 and 7; Figs 2A and 5). Temperature in Celsius is shown for comparison only; models were fit to inverse temperature. Temperatures within tanks declined over time (S2 Fig). Data for these figures may be found at https://doi.org/10.5281/zenodo.2652579 in GarzkeAllwks.csv. A, algae only; AG, algae and grazers; AGP, algae, grazers, and predators; ER, ecosystem respiration; NEP, net ecosystem production.
Table 6.
Results of model comparisons for effects of temperature and time on NEP based on AIC weight (w) and δAICC values.
Nested versions of the full model (Methods, Eq 12). Response variables are modeled as functions of temperature Twj for each tank j on week w relative to the mean temperature for tank j over all weeks (T in Kelvin), and trophic structure (Zj). Models included a random effect for the experimental unit—tanks with and without predators received the same power inputs. See Methods for additional details on modeling. The model was fit to 219 observations in 30 groups. The full model (NEPF) includes all terms, and models representing alternate hypotheses excluded terms indicated by “NA.” Values indicate model-estimated coefficients. Coefficients were pooled (Methods: Statistical analysis) to estimate slopes and intercepts for Figs 3 and 5.
Table 7.
Results of model comparison for effects of temperature and time on ER based on AIC weight (w) and δAICC values.
Nested versions of the full model (Methods, Eq 12). Response variables are modeled as functions of temperature Twj for each tank j on week w relative to the mean temperature TM for tank j over all weeks (T in Kelvin), and trophic structure (Zj). Models included a random effect for the experimental unit—tanks with and without predators received the same power inputs. The model was fit to 240 observations in 30 groups. See “Methods: Statistical analyses” for additional details on modeling.