Fig 1.
Conceptual overview of hypothesized bottom-up causal relationships within our study system.
We tested the following a priori hypotheses: seabirds increase nutrients of coastal leaves (H1) and algal turfs (H2). Higher turf nutrients lead to faster turf growth (H3) and higher turf cover (H4). Faster turf growth leads to higher herbivore productivity (H5) and biomass (H6). Higher turf cover leads to higher herbivore productivity (H7) and biomass (H8). Higher turf nutrients lead to higher herbivore productivity (H9) and biomass (H10). For hypotheses H9–H10, turf nutrients could affect herbivores either directly (due to increased food quality, lines labeled H9, H10) and/or indirectly via their effects on turf growth (i.e., food replenishment, lines labeled H3–H5, H3–H6) or turf cover (i.e., food quantity, lines labeled H4–H7, H4–H8). Therefore, we compared the direct effect (food quality only) to the total effect (food quality + food quantity + food replenishment) of turf nutrients on herbivore biomass and productivity to elucidate the pathway(s) by which seabird nutrients in algal turfs influence consumers. These causal pathways were tested within broader directed acyclic graphs (DAGs), which included potential confounding variables (S1A Fig). Based on our initial findings, we built an alternative DAG in which H7 and H8 are reversed to test the hypothesis that herbivore productivity and biomass have negative effects on turf cover via top-down controls (S1B Fig).
Fig 2.
Effect of seabird biomass on nutrient uptake by terrestrial plants and marine turf algae.
(A) Seabird biomass (kg/ha) across the five study sites as a function of rat invasion status. (B, C) Effect of seabird biomass on δ15N (a proxy for seabird-provided nutrients) in coastal plants (B) and marine turf algae (C). Main plots—points represent partial residuals from Bayesian models, lines represent estimated conditional effects, and shading represents 75% Bayesian credible intervals. Insets—posterior predictive distributions for effect of seabird biomass on response (both log-transformed). Point represents median estimate and lines represent 95% and 75% highest posterior density intervals (HPDIs). The data underlying this figure can be found in https://doi.org/10.5281/zenodo.15485420.
Fig 3.
Effect of seabird nutrients in turf algae on (A) turf algal productivity and (B) turf algal cover.
Main plots—Points represent partial residuals from Bayesian models, lines represent estimated conditional effects, and shading represents 75% Bayesian credible intervals. Insets—Posterior predictive distributions for effect of seabird nutrients in turf on log-transformed turf growth (A) or turf cover (B). Point represents median estimate and lines represent 95% and 75% highest posterior density intervals (HPDIs). The data underlying this figure can be found in https://doi.org/10.5281/zenodo.15485420.
Fig 4.
Effect of turf productivity on (A) herbivore productivity and (B) herbivore biomass, and effect of turf cover on (C) herbivore productivity and (D) herbivore biomass.
Main plots—Points represent partial residuals from Bayesian models, lines represent estimated conditional effects, and shading represents 75% Bayesian credible intervals. Insets—Posterior predictive distributions for effect on log-transformed response. Point represents median estimate, and lines represent 95% and 75% highest posterior density intervals (HPDIs). The data underlying this figure can be found in https://doi.org/10.5281/zenodo.15485420.
Fig 5.
Direct and total effect of turf nutrients on herbivore growth and herbivore biomass.
Direct effects are driven by food quality (turf nutrients) alone, while total effects also include indirect pathways via food replenishment (turf productivity) and food availability (turf cover). Main plots—Points represent partial residuals, lines represent estimated conditional effects, and shading represents 75% Bayesian credible intervals. Insets—Posterior predictive distributions for effect on log-transformed response. Point represents median estimate and lines represent 95% and 75% highest posterior density intervals (HPDIs). The data underlying this figure can be found in https://doi.org/10.5281/zenodo.15485420.
Fig 6.
Causal pathways of seabird nutrient inputs on turf algae and herbivorous fishes.
Results are from Bayesian models testing for causal pathways based on an alternative DAG assuming both bottom-up and top-down controls in this system (S1B Fig). (A–F) Posterior predictive distributions for each causal pathway, point represents median estimate and lines represent 95% and 75% HPDIs. Distributions are colored to show positive effects in green and negative effects in red. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.15485420.