Deriving a Bayesian Network to Assess the Retention Efficacy of Riparian Buffer Zones

摘要: Bayesian networks (BN) have increasingly been applied in water management but not to estimate the efficacy of riparian buffer zones (RBZ). Our methodical study aims at evaluating first BN predict RBZ retain sediment and nutrients (dissolved, total, particulate nitrogen phosphorus) from widely available variables (width, vegetation, slope, soil texture, flow pathway, nutrient form). To evaluate influence parent nodes how number states affects prediction errors, we used a predefined general structure, collected 580 published datasets North America Europe, performed classification tree analyses multiple 10-fold cross-validations different BNs. These errors ranged 0.31 (two output states) 0.66 (five states). The outcome remained unchanged without least influential (flow vegetation). Lower were achieved when had more than two states. influenced most strongly error as its lowest highest better predicted intermediate While derived BNs could support or replace simple design guidelines, they are limited for detailed predictions. More representative data on vegetation additional like preferential will probably improve predictive power.