Development of a new methodology to predict cascading stressors through aquatic ecosystems: Posterior Predictive Meta-analytic Network

• Applicant: Dr. Willem Kaijser
• Subject Area: Hydrogeology, Hydrology, Limnology, Urban Water Management, Water Chemistry, Integrated Water Resources Management ; Ecology and Biodiversity of Plants and Ecosystems
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 554796071

Multiple stressors cascade through ecosystems in complex ways. For instance, nutrient input in lakes increases chlorophyll-a concentrations, initializing (i) turbidity causing the decline in small macrophytes by light restrictions and (ii) the decrease in carbon dioxide due to photosynthesis, which limits macrophytes that can only use carbon dioxide. While multiple studies address different chain parts, it is not yet possible to meta-analytically combine individual linear, curve-, and non-linear functions into a single causal network. This project aims to generate and apply a novel method that integrates the results of multiple studies addressing different stressor-response relationships to account for the cascading of stressors through ecosystems. While different model types exist to address stressor pathways through an ecosystem, they all have shortcomings to address this specific question. Mechanistic models address theoretical concepts and make predictions. However, they lack a meta-analytic and flexible approach. Bayesian Structural Equation Models incorporate different pathways, but have no direct meta-analytic component. Bayesian Belief Networks (BBN) are flexible and can incorporate new relations, but lack the ability to include meta-analytic information and curve-, and non-linear functions. Bayesian Meta-Analytic Structural Equation Models (BMASEM) rely on standardized effect-sizes, but cannot predict given a stressor gradient. I will develop a new method, the posterior predictive meta-analytic network (PPMN), that combines the advantages of the previously described approaches, while circumventing some of their shortcomings. This new method is flexible and can formalize multiple types of information with the goal of using it. This model relies on absolute effect sizes (parameters) extracted from literature that are meta-analytically combined. In principle, it is an extension of a BBN and BMASEM, using linear, curve-, and non-linear functions. Synthesizing these meta-analytic functions into a network generalizes yet unexplored direct and indirect causal relations between the response and stressors, and can be used to predict probable responses. Using this basic structure, networks can be built to model how stressors cascade through an ecosystem. I will develop a formal and working syntax of the PPMN model in R, which will be available on GitHub and eventually CRAN. In parallel, I will gather information to create a full network of eutrophication in shallow lakes, from the input of nutrients to the shift in macrophyte growth forms, and to invertebrate and fish species. Finally, I will use the model structure and collected data to model from top-down (predicting biota and biochemical changes from nutrient input) and from bottom-up (predicting nutrient input from observed biota). Once established, the model structure will be applicable to many other questions regarding the cascading of stressors through ecosystems.

• DFG Programme: Research Grants
• International Connection: Finland, Netherlands
• Cooperation Partners: Dr. Janne Alahuhta; Professor Dr. Eric-Jan Wagenmakers
• Co-Investigator: Professor Dr. Florian Hartig