Disruption of ecological networks in lakes by climate change and nutrient fluctuations

The study investigates how climate warming and fluctuations in nutrient supply (phosphorus) reorganize ecological interaction networks in lakes. Human impacts, including climate change and pollution, alter the nature and strength of species interactions, which are a fundamental yet often overlooked component of biodiversity change. Temperate lakes in Switzerland have experienced managed re-oligotrophication (reduction of phosphorus inputs) since the 1970s alongside long-term warming. The central research questions are: how do network properties such as connectance and interaction strength vary over time and across environmental gradients of temperature and phosphorus; how do these drivers jointly affect the direction and strength of trophic control (top-down vs bottom-up); and which interaction types and guilds are most sensitive to these changes. Understanding these dynamics is important for predicting ecosystem stability, biodiversity maintenance, and for guiding lake management under global change.

Prior work has suggested that ecological networks vary in space and time in both the number and strength of interactions and that these properties can signal large-scale ecosystem changes with implications for stability. However, empirical knowledge of whole-network reorganization under multiple stressors is limited due to scarce long-term datasets and the prevalence of methods and theories that assume fixed interactions or focus on a single interaction type. Theoretical and experimental studies indicate that warming can reduce trophic transfer efficiency and alter food-web structure, and that nutrient levels modulate trophic controls. Nonetheless, the non-additive, state-dependent effects of warming and nutrient dynamics on entire community networks in natural systems remain poorly quantified.

The authors analyzed 20–43 years (1977–2020) of monthly plankton abundance data from ten peri-alpine Swiss lakes, alongside monthly measurements of water-column temperature and dissolved inorganic phosphorus (phosphate). Plankton taxa (phyto- and zooplankton) were aggregated into a conceptual network of up to 15 trophic guilds (nodes) based on taxonomy, body size, and feeding behavior, spanning three trophic levels (invertebrate predators, omnivores, large and small herbivores, mixotrophs, and primary producers; with phytoplankton guilds split by size/coloniality when possible). Some lakes lacked small grazer data (rotifers, ciliates), and certain taxa (nauplii; small single-cell cyanobacteria) were excluded due to methodological constraints. Abundance data were standardized (units harmonized), winsorized at the 99% quantile to reduce outlier influence, and environmental time series were interpolated for occasional missing values. To reconstruct time-varying causal networks and quantify interaction strengths, the study used empirical dynamic modeling (EDM). Convergent cross-mapping (CCM) was applied in a 60-month moving window to infer causal associations between guild time series after correcting for seasonality using seasonal surrogate null models. Significant links were those where local cross-map skill (rho) exceeded 95% of surrogates; negative rhos were set to zero. Network connectance was computed as the percentage of realized causal links relative to all possible links among nodes, and mean interaction strength was the average rho among significant links. The embedding dimension (E) was selected via simplex projection (E=2–15), and CCM convergence was evaluated using subsets (20% vs 50% of data). The same EDM framework tested feedback between temperature and phosphate over full time series. To model how environmental drivers shape network properties and trophic controls, the authors used multivariate S-maps (locally weighted linear models within EDM) to predict network connectance, mean interaction strength, and the balance of top-down vs bottom-up links as functions of water temperature, phosphate, and lake morphometrics (depth at sampling site and total volume). This enabled exploration of nonlinear, state-dependent effects of temperature-phosphate combinations across lakes. Trophic control metrics were computed as differences between top-down and bottom-up connectance and strengths (positive indicates top-down control). Interaction types (trophic, non-trophic, hybrid) were quantified for prevalence and strength, with Wilcoxon tests used for pairwise comparisons across lakes.

- Water temperature causally influences phosphate levels (negative effect expected), but phosphate does not causally influence temperature, indicating unidirectional control of nutrient availability by warming across lakes. - Network properties are dynamic: connectance and interaction strengths vary over time and across lakes. During re-oligotrophication, connectance increased significantly in 2/5 lakes (e.g., Lake Zurich +4.2%, Spearman R=0.35, P<0.001). During recent accelerated warming, connectance decreased significantly in 6/8 lakes (e.g., Lake Zurich −14.8%, R=−0.78, P<0.001). Mean interaction strength showed less variability and lake-specific trends. - S-map models reveal nonlinear, interactive effects of temperature and phosphate on network connectance and link strength. Warming generally reduces connectance and interaction strength, with reductions particularly pronounced under high phosphate conditions. Model uncertainty increases where the temperature–phosphate parameter space is less populated (e.g., at simultaneously high values). - Trophic controls: Top-down links are more frequent than bottom-up in 9/10 lakes, but in about half the lakes bottom-up links are stronger on average. Re-oligotrophication slightly increased top-down prevalence relative to bottom-up in 3 lakes, whereas warming generally decreased top-down relative to bottom-up links, indicating a shift toward bottom-up control as warming progresses. - S-map predictions indicate that under high temperatures and high phosphate, plankton networks tend to be bottom-up controlled; under low to intermediate temperatures, top-down control is more prevalent and stronger in most lakes. - Interaction types: Hybrid links (which can be trophic or non-trophic depending on conditions) are the most prevalent across lakes, whereas non-trophic links are strongest on average; hybrid links tend to be weaker. Pairwise Wilcoxon tests show significant differences among interaction-type prevalence and strengths. - Key guilds: Small grazers (rotifers, ciliates, mixotrophic flagellates) and colonial cyanobacteria are highly connected and influential in trophic controls. Ciliates and cyanobacteria show strong sensitivity to long-term temperature changes. Cyanobacteria are strongly driven by phosphate and temperature and broadly connected, implying pervasive network effects. - Environmental context: From 2010 to 2020, average water-column temperatures rose by 0.4–1.7 °C across study lakes, comparable to increases over the previous six decades. Warming reduces vertical mixing and phosphorus resuspension, further constraining nutrient supply and reinforcing warming’s indirect influence on networks.

The findings demonstrate that climate warming, through both direct metabolic effects and indirect regulation of phosphorus availability, reorganizes plankton interaction networks in temperate lakes. Warming generally diminishes connectance and interaction strength, potentially weakening stabilizing feedbacks and altering energy flow pathways. The observed and modeled shifts from top-down to bottom-up control under warmer and nutrient-rich conditions indicate that consumer dynamics become increasingly governed by resource availability, increasing sensitivity to nutrient inputs. This has implications for eutrophication risks, bloom dynamics, and management strategies, where nutrient controls become even more critical under warming. The prominence of hybrid interactions and the central roles of small grazers and cyanobacteria suggest that intermediate consumers and generalists are key indicators of structural changes affecting ecosystem stability and predictability. Nonlinear, state-dependent relationships captured by EDM highlight that ecosystem responses to multiple stressors are non-additive and context-dependent, underscoring the need for dynamic, mechanistic inference over static correlations.

This study provides empirical, network-level evidence that climate warming and nutrient fluctuations jointly and nonlinearly disrupt ecological interactions in lakes. Using long-term data and equation-free modeling (EDM with CCM and S-maps), the authors show that warming typically reduces network connectance and shifts trophic control toward bottom-up regulation, especially at high phosphate levels. Hybrid interactions are widespread, non-trophic links are strongest, and small grazers and colonial cyanobacteria emerge as sensitive, broadly connected indicators of network reorganization. These insights offer tools for diagnosing and anticipating climate impacts on community dynamics and can inform lake management by emphasizing nutrient reduction under warming scenarios. Future research should expand long-term, high-frequency monitoring (including small grazers), extend analyses to other regions and ecosystems, integrate additional stressors (e.g., pollutants, hydrological changes), refine network reconstructions at finer taxonomic resolution, and explore forecast horizons and early-warning indicators derived from dynamic network properties.

- Data gaps and methodological changes: Some lakes lacked small grazer data; sampling protocols and depths changed over time; post-oligotrophication nutrient sampling became less frequent in several lakes, requiring linear interpolation of many values; and abundance data were winsorized to mitigate outliers. - Conceptual aggregation: Taxa were aggregated into guilds to standardize across lakes and reduce classification errors, potentially obscuring species-specific interactions; certain groups (nauplii, small single-cell cyanobacteria) were excluded. - Modeling choices: A 60-month moving window was used (though robustness to window size was tested); CCM relies on convergence and seasonal surrogate corrections; negative rho values were set to zero; and S-map predictions are less certain outside observed temperature–phosphate ranges, particularly at high values. - Generalizability: Lakes are from a single geographic/climatic region with shared histories, which may limit extrapolation to other systems. Correlative trends alone cannot disentangle interacting drivers, necessitating the EDM framework but still subject to data coverage constraints.