A conceptual model for site-level ecology of the giant gartersnake ( Thamnophis gigas ) in the Sacramento Valley, California
Giant gartersnakes (Thamnophis gigas) comprise a species of semi-aquatic snakes precinctive to marshes in the Central Valley of California (Hansen and Brode, 1980; Rossman and others, 1996). Because more than 90 percent of their historical wetland habitat has been converted to other uses (Frayer and others, 1989; Garone, 2007), giant gartersnakes have been listed as threatened by the State of California (California Department of Fish and Game Commission , 1971) and the United States (U.S. Fish and Wildlife Service, 1993). Giant gartersnakes currently occur in a highly modified landscape, with most extant populations occurring in the rice - growing regions of the Sacramento Valley, especially near areas that historically were tule marsh habitat (Halstead and others, 2010, 2014). In ricelands and managed marshes, many operational decisions likely affect the health and viability of giant gartersnake populations. Land-use decisions, including the management of water, aquatic vegetation, terrestrial vegetation, and co-occurring species, have the potential to affect giant gartersnakes. Little is known, however, about the effects of these types of decisions on the viability of giant gartersnake populations. Bayesian network models are a useful tool to help guide decisions with uncertain outcomes. These models require the articulation of what experts think they know about a system, and facilitate learning about the hypothesized relations (Marcot and others, 2001; Uusitalo , 2007). Bayesian networks further provide a clear visual display of the model that facilitates understanding among various stakeholders (Marcot and others, 2001; Uusitalo , 2007). Empirical data and expert judgment can be combined, as continuous or categorical variables, to update knowledge about the system (Marcot and others, 2001; Uusitalo , 2007). Importantly, Bayesian network models allow inference from causes to consequences, but also from consequences to causes, so that data can inform the states of nodes (values of different random variables) in either direction (Marcot and others, 2001; Uusitalo , 2007). Because they can incorporate both decision nodes that represent management actions and utility nodes that quantify the costs and benefits of outcomes, Bayesian networks are ideally suited to risk analysis and adaptive management (Nyberg and others, 2006; Howes and others, 2010). Thus, Bayesian network models are useful in situations where empirical data are not available, such as questions concerning the responses of giant gartersnakes to management.