The Wasp Water Quality Model Based On Uncertainty Analysis Research

Water environment as an open giant system, is full of uncertainty factors from all aspects, including not only its internal physics, chemistry and biology factors like diffusion of pollutants in the water, migration, precipitation, biological decomposition and their comprehensive effects, but also the external effects coming from nature and human activities which brought great difficulties for learning the rule of water quality change.In order to improve the understanding of the water environment and make more sensible decisions, it is needed to use water quality model to quantify the mechanical process. After nearly a century of efforts, the current water quality model has been widely used in water quality management to reflect the water quality change rule. But because of the complexity of water environment, water quality model still exist great uncertainties, on one hand, the effect of uncertainty affect the model’s simulation, on the other hand also restricts the application of model in practice. So how to reduce the uncertainty of simulation, has become an important direction of current model research work.Currently, the aim of parameter calibration is used to describe the characteristic of the rule in different water bodies. But in a mechanical model, many coupled parameters which have complex influence, often lead to a phenomena named "equifinality ", which means the parameter combination at the same time were able to "satisfy" the precision of the model. However each parameter set of the water quality model corresponds to one or another process, so by the general likelihood function is hard to judge the real process and is difficult to measure the contribution of each sub-process. Meanwhile, mechanical model would lose the ability to simulate transformation process which will enlarge the risk of prediction in water quality management.In order to reduce the uncertainty of model simulation, based on the Simulink environment, the WASP model is decomposed as its framework, so that can monitor each module in the simulation process; In terms of improvement of model structure, the segment of Yunlin village to Yangwa of North canal is taken as an example, using the measured data to fit the change of the channel parameters by regression analysis to reduce the uncertainty coming from the model structure; Moreover the COD module was established based on artificial neural network, to add function to WASP model; Using the Monte Carlo method, the4indictors:DO, CBOD, ammonia nitrogen and nitrate nitrogen were simulated, and uncertainty generated by each module are analyzed; Through global sensitivity analysis by Sobol and EFAST method, which are commonly used to determine the sensitive parameters of the model, the1st order sensitivity and the total sensitivity of each parameters were evaluated, and the sensitivity of the parameters under the conditions of different temporal and spatial changes was also discussed; By in-situ monitoring and laboratory simulation results, a priori knowledge base is established to constrain the parameter distribution according to the constraint conditions. By the method of multi-objective GLUE method, the model parameters uncertainty can be reduced; Finally, with Monte Carlo simulation, according to the structure of the WASP, the characteristic of the water quality changes is discussed.The main conclusions can be summarized as:(1) The module inside the WASP model can be modified with the data-driven models to fit the local characteristics, and also can supplement the WASP model simulation function;(2) the model’s overall output uncertainty is different over time, and each sub-process in the different time also presents different uncertainties. WASP has an obvious phenomena of "equifinality", only by the traditional method for global optimization rate is not enough to reflect the process of mutual transformation between each constituent in water body, should be re-verify by experimental method;(4) through the global sensitivity analysis, SOD, E12, E1R, fD5, fon and GP1, PNH3, KNIT, anc, K12were found as sensitive parameters; Compared the different between total sensitivity and the first-order sensitivity index, it is found that sensitive parameters in WASP model is coupling strongly, which means that one factor at a time method is error-prone when applied in WASP’s SA; WASP parameter sensitivity will be affected obviously by temporal (seasonal) change, but will not by spatial change;(5) through the modification of the model, parameter selection under the constraint of priori knowledge base and multi-objective calibration, the uncertainty of the model is significantly reduced and can better reflect the water quality transformation of the north canal; Based on WASP framework, the characteristic of the water quality transformation can be explained by Monte Carlo simulation on each module.