Spatially implemented Bayesian network model to assess environmental impacts of water management

Abstract

AbstractBayesian networks (BNs) have become a popular method of assessing environmental impacts of water management. However, spatial attributes that influence ecological processes are rarely included in BN models. We demonstrate the benefits of combining two‐dimensional hydrodynamic and BN modeling frameworks to explicitly incorporate the spatial variability within a system. The impacts of two diversion scenarios on riparian vegetation recruitment at the Gila River, New Mexico, USA, were evaluated using a coupled modeling framework. We focused on five individual sites in the Upper Gila basin. Our BN model incorporated key ecological drivers based on the “recruitment box” conceptual model, including the timing of seed availability, floodplain inundation, river recession rate, and groundwater depths. Results indicated that recruitment potential decreased by >20% at some locations within each study site, relative to existing conditions. The largest impacts occurring along dynamic fluvial landforms, such as side channels and sand bars. Reductions in recruitment potential varied depending on the diversion scenario. Our unique approach allowed us to evaluate recruitment consequences of water management scenarios at a fine spatial scale, which not only helped differentiate impacts at distinct channel locations but also was useful for informing stakeholders of possible ecological impacts. Our findings also demonstrate that minor changes to river flow may have large ecological implications.