Development of a mechanistic model (ERIMO-I) for analyzing the temporal dynamics of the benthic community of an intermittent Mediterranean stream

To evaluate the role of the benthic community within headwater stream ecosystems, it is crucial to understand the mechanisms of the processes dominating their turnover rates and temporal dynamics. To analyze the benthic community dynamics of an intermittent Mediterranean stream (Fuirosos, Spain), we developed a mechanistic model that describes the most important ecosystem components (five functional feeding groups of invertebrates, periphyton, and benthic organic matter) under consideration of important transformation processes. The model is an extension of that developed for application to the prealpine River Sihl (Switzerland). A combination of prior knowledge from the literature and information from field observations within a Bayesian framework is used to constrain plausible ranges of the estimated model parameters and to evaluate the uncertainty of the model outcome. The Bayesian inference resulted in a realistic description of most of the main features of the system, offered insights into the degree of information the data contains about model parameters, and also made it possible to quantify the dependence structure of the parameter estimates. Local and global sensitivity analyses revealed that rate parameters for growth and death of invertebrate functional feeding groups are most influential on the model results and additional information on these parameters would be most helpful to reduce output uncertainty. The study clearly points out the importance of organic matter dynamics, as allochthonous organic matter is the main energy source in this Mediterranean headwater stream. Hydrological disturbances of the system such as droughts and floods lead to complex colonization patterns which would require additional data to support a more detailed description by an extended model. Furthermore, the study shows that it is possible to deal with a complex ecological model with eight state variables and more than 60 parameters in a Bayesian framework.