Water Quality Risk Analysis Based On Bayerian Nteworks

For a long time, water environmental treatment is affected by uncertainty, which comes from the shortage of monitoring data and the limited knowledge of human. Uncertainty will always exist and can not be completely avoided. Then, the traditional deterministic methods show their unadaptability. Studying on the uncertainty methods to quantify the uncertainty of water quality prediction and make risk-based descisions has become a focus recently. Probability and statistics theory can be used to quantify the statistical uncertainty, and is the most important uncertainty analysis method. Uncertainty methods based on probability and statistics theory can be further divided into method based on mechanism model and method based on statistical model. The traditional Bayesian Network is data-driven statistical model, that is, the statistical relationships between viariables are estimated by data. In addtion, there is a special kind of Bayesian Network, which uses mechanism models to represent relationships between variabes. It is a couple of methods based on mechanism model and that based on statistical model.The subbasin of Qiantang River Basin—ongyang River Basin is taken as a study case, and the probability and statistics theory is used to quantify the risk of water quality exceeding standard values. Two Bayesian Networks—BN1and BN2are constructed. BN1is used to calculate the risks of water quality exceeding the standard values under different rainfall amount and temperature, and BN2is used to calculate the corresponding risks when considering the statistical distributions of rainfall and temperature. The main advantage of BN1is that it can be parameterized by little data, and that of BN2is that it can quickly calculate a great deal of risks. These two methods can be completment with each other’s advantages. In addition, for the period2020s, the possible impact of climate change on the risks of water quality exceeding the standard values in Dongyang River is further studied in purpose of helping decision-makers to assess future water quality and make preventive descisions. The main contents and research results are as follows:(1) A mechanism-model-based Bayesian Network—BN1, which connects control, intermediate and response viariables and uses mechanism modes (which are submodels) to stand for the relationships of viariables, is built. A new MCMC algorithm—RAM is applied on Bayesian parameter estimation of submodels(that is also parameter estimation of BN1). Then, MC simulations are carried out on BN1to calculate the risks of water quality exceeding standard values under different rainfall amount and temperature. DRAM can be used to make Bayesian estimation of point pollution dsicharge and water quality parameter at the same time. That can breakthrough the limitation of little data, and also quantify the uncertainty.(2) A data-based Bayesian Network—BN2is built. June and December are chosen as typical months, and the distributions of monthly average rainfall and temperature of June and Decmber are inputed to BN1to carry out MC simulations. A probability-based soft discretization approach is used to discretize the random results of BN1, then the discretized data is used to train BN2. At last, by the inference algorithm of BN2, the water quality risks when considering the statistical distributions of rainfall and temperature are calculated. Furthermore, the risks under more reservoir discharge are also calculated. The important advantage of BN2is its rapid reasoning algorithm, which can qulickly calculate all kinds of risks once parameter estimation is finished. Especially, when there are many times of MC simulation, or many risks need to be calculated, or complex mechanism models, BN2is significantly more efficient than BN1.(3) This paper applies the LARS-WG weather generator to simulate daily rainfall and temperature data of a single station under the A1B, A2and B1emission scenarios using the results of General Circulation Model HadCM3. The results show that the monthly average rainfall, monthly highest and lowest temperature will not change obviously, but the frequency of heavy rainfall and high temperature will possibly increase. Then, based on BN2, water quality risks when considering stochastic uncertainty of rainfall and temperature of June and December in2020s are calculated. Finally, a comparison of water quality risks between baseline period, A1B, A2and B1emission scenarios is made in order to provide a reference for decision-makers to prevent water quality from exceeding standard values.