A BAYESIAN NETWORK APPROACH FOR ECOGEOMORPHOLOGICAL MODELING FACING UNCERTAINTY: A CROSS-COMPARISON OF PRISTINE AND IMPACTED WETLANDS

Bayesian networks (BNs), also known as belief networks, belong to the broad family of probabilistic graphical models. A Bayesian network consists of a graphical structure and a probabilistic description of the relationships among the different variables of the analyzed system. The graphical structure explicitly represents cause and effect relationships, allowing a complex causal chain to be structured in a series of conditional relationships. Due to these characteristics, Bayesian networks are particularly useful for modeling complex environmental systems with multiple components related by different dependencies.

As a case study, we present a Bayesian network integrating the outputs of ecological and geomorphological models of a wetland ecosystem. Specifically we consider the Sian Ka'an wetland for comparative purposes with the Florida Everglades. In this case we consider all the variables averaged in the spatial domain. The variability of future climate, expressed by rainfall variability and different anthropic pressures, is described as uncertain nodes in the Bayesian network. All of the variables are characterized by multiple states, representing their future variability, in the form of probability distributions that are propagated to the model endpoint - the habitat value that expresses the overall quality of the wetland ecosystem.

The habitat value calculated under different scenarios, as a function of climate change and anthropic pressure, can reveal the linkages between these external stressors and ecogeomorphological dynamics. We plan also to compare the dynamics of the pristine Sian Ka'an wetland versus the highly impacted Florida Everglades under the same stressors. The Bayesian network can also inform public policies about the best use of wetland ecosystems in order to preserve ecosystem structure and function.