The application of Bayesian networks to evaluate risks from multiple stressors to water quality of freshwater ecosystems

It is difficult to predict and manage the ecological consequences of multiple water quality stressors on our freshwater systems. This is due to the dynamism of the source-stressor-response relationships and multiple factors including lack of data, complex impact pathways and risks, and uncertainties that are difficult to parameterise. We present a risk-based probability modelling approach using a Bayesian network (BN), to manage multiple water quality stressors at multiple spatial scales. We illustrate the use of this approach, by evaluating the probable ecological effects of altered water quality associated with multiple sources in three case study rivers in South Africa. Water quality and land use activity were used to describe conceptual risk pathways, parameterise the BNs and model the probable consequences of multiple water quality stressors. The BN model demonstrated that the endpoints that were selected for the study reflected the risks associated with the levels of the input water quality variables. The model further demonstrated that the electrical conductivity BN was just as sensitive as the more complex salt model. The BN model was further able to accurately represent risks to all systems irrespective the water quality data base size. This approach can contribute towards more sustainable water resource management.

多重水质胁迫因子对淡水生态系统造成的生态影响难以预测与管控。究其原因，在于源-胁迫因子-响应关系的动态特性，以及数据匮乏、影响路径与风险复杂、难以参数化的不确定性等多重因素的共同作用。本研究提出一种基于风险的概率建模方法，借助贝叶斯网络（Bayesian Network, BN），在多空间尺度下管控多重水质胁迫因子。我们以南非三条案例研究河流为对象，评估多来源引发的水质变化所可能产生的生态效应，以此说明该方法的应用方式。研究采用水质与土地利用活动刻画概念性风险路径，完成贝叶斯网络的参数化，并对多重水质胁迫因子的潜在影响进行建模。该贝叶斯网络模型结果显示，本研究选取的生态终点能够反映输入水质变量水平对应的风险等级。模型同时表明，电导率子贝叶斯网络的灵敏度可与更为复杂的盐类模型相媲美。无论水质数据库规模大小，该贝叶斯网络模型均可精准表征所有相关系统面临的风险。该方法可为可持续水资源管理工作提供有效支撑。