Effective management of river ecosystems requires knowledge of the interrelationships between biological, chemical and geomorphological processes and patterns. This is a complex challenge, and there are significant gaps in our understanding of these interrelationships. For example, the response of biological communities to geomorphological changes in rivers is particularly poorly understood. These knowledge gaps are compounded by the lack of coherent biological, chemical and geomorphological datasets for many rivers, limiting the extent to which traditional data analysis and modelling techniques can be applied.
Here we describe the application of a new technique, Structural Equation Modelling (SEM), to the investigation of biological, chemical and geomorphological data collected from rivers across England and Wales. The SEM approach is a multivariate statistical technique enabling simultaneous examination of direct and indirect relationships across a network of variables. Further, SEM allows a-priori conceptual or theoretical models to be tested against available data. For example, a- priori models can be developed in collaboration with river managers and then evaluated using SEM as part of participatory modelling projects. This is a significant departure from the solely exploratory analyses which characterise other multivariate techniques. Bayesian approaches can also be applied within the SEM framework, offering the opportunity to address challenges such as incomplete datasets and non-normal data distributions. Such challenges are common in the analysis of spatial patterns associated with riverine ecosystems.
We took biological, chemical and geomorphological data collected by the Environment Agency for 700 sites in rivers across England and Wales, and created a single, coherent dataset suitable for SEM analyses. Biological data cover benthic macroinvertebrates, chemical data relate to a range of standard parameters ( e. g. BOD, dissolved oxygen and phosphate concentration), and geomorphological data cover factors such as river typology, substrate material and degree of physical modification. We developed a number of a-priori theoretical models based on existing understanding of river ecosystems. These models were able to explain correctly the variance and covariance shown by the datasets, proving to be a relevant representation of the processes involved. The models explained around 80% of the variance in indices describing benthic macroinvertebrate communities. Dissolved oxygen was of primary importance, but geomorphological factors, including river habitat type and degree of habitat degradation, also had significant explanatory power. The model produced new insights into the relative importance of chemical and geomorphological factors for river macroinvertebrate communities. The SEM technique proved powerful, for example able to deal with the co-correlations that are common in rivers due to multiple feedback mechanisms.
In this paper we also examine how SEM could be used to guide data collection and support decision-making (DM) for river ecosystems. We highlight the benefits of a Bayesian approach to solving SEM, especially in the context of supporting DM. We demonstrate how both simple and more complex a-priori conceptual models can be used in SEM. We explore whether greater complexity, which may add credibility to a model, increases explanatory power compared to relatively simple models. We examine how subjective judgements that are inherent to the development of a-priori models, for example relating to the separation between individual habitat types, influence the outcomes and interpretations of SEM analyses. Our experience highlights the importance of close collaboration with potential users throughout each step of the SEM framework, and we examine how this collaboration might be put into practice.