From explanatory to predictive network modeling of relationships among ecological indicators in the shallow temperate lagoon

Abstract

The aim of the study was to create a procedure and test it in a case study that would permit forming a predicted depiction of the cause-and-effect network of connections in the lagoon ecosystem. We propose the novelty procedure as a tool for predicting transformations in the structure of links within ecosystems and forecasting changes in the impact some ecological factors have on others. An attempt was made to combine the Structural Equation Modeling (SEM) and Artificial Neural Network (ANN) tools to obtain the most reliable model of the dependency structure among variables that would be based on prediction data from the initial structural model and processed with a backpropagation algorithm. The subject of the explanatory-predictive analysis in this work was the Vistula Lagoon (southern Baltic) ecosystem network: physical-chemical factors – zooplankton – phytoplankton – organic carbon forms. The SEM-ANN-SEMpredict modeling sequence produced satisfactory adjustment rates of the ANN models for training and validation trials for the majority of the organic parameters analyzed. Higher matching rates and a higher number of relationships were obtained with the SEMpredict model than with the SEM model based on observed data. The role of physical factors changed in the predictive model compared to the explanatory model. Wind speed changed the direction of the impact on the picophytoplankton from a negative to one that facilitated growth and influenced on the cyanobacteria biomass increase. The role of salinity in promoting the growth of phytoplankton groups in relation to the impact of basin depth was more prominent. The analysis of both the new predictive network of direct links and the total effect of these relationships permitted us to extend our knowledge about potential phenomena that can occur in the ecosystem studied. Innovative methodology combined SEM and ANN network models to analyze the structure and prediction of ecosystem network indicated important potential properties of ecological systems.