Improvement Of Aerosol Numerical Modeling By Using Sensitivity Ananlysis And Uncertainty Anlaysis

Due to highly varied uncertainties in input data（e.g.emissions,boundary conditions,and meteorological data）and simplified treatments of atmospheric chemistry and physics,complex air quality models（AQMs）,such as CMAQ and CAMx,contain large uncertainties in applications.Uncertainty analysis is a widely used method for model improvement since it is capable of identifying the key uncertainty sources among numerous model inputs and provide an effective pathway to improve model simulations.However,the current methodology still has limitations in accurately and efficiently conducting uncertainty analysis for complex AQMs due to the numerous model inputs,the large computational requirement of the traditional Monte Carlo Method（MCM）and the high uncertainty brought by the Reduce-Form Method（RFM）.To address these limitations,this thesis devised two refined uncertainty propagation methods and a new framework for efficient uncertainty analysis for complex AQMs.The framework was applied to uncertainty analysis of PM_2.5 simulations in the Pearl River Delta（PRD）region as a case study.Based on the identification of the key uncertainty sources,a new data fusion method was used to enhance the boundary conditions of in-flow domains to improving PM_2.5 simulations in the PRD region.Below are the key conclusion drawn in the thesis.（1）This thesis established a quantitative uncertainty analysis framework,which consists of sensitivity analysis,inputs uncertainty quantification,uncertainty propagation,Bayesian Monte Carlo uncertainty adjustment,uncertainty assessment and uncertainty attribution,for complex AQMs.This thesis proved to be feasible in finding the key uncertainty sources of complex AQMs and thus direct an effective pathway for model improvements.（2）By combining several sets of local sensitive coefficients at different emission conditions,this study developed a new uncertainty propagation method,the stepwise-based HDDM-RFM,to overcomes the limitation of the traditional HDDM-RFM in predicting nonlinear responses to large perturbations of inputs（>50%）.Evaluations reveal that the new method obviously improves the accuracy of uncertainty propagations in the case with large uncertainty inputs involved.（3）Incorporating the HDDM with SRSM framework（HDDM-SRSM）can dramatically reduce the model runs required in SRSM models formulation and thus increase the efficiency of uncertainty propagation.Compared with the traditional SRSM method,the HDDM-SRSM boosts the efficiency of uncertainty propagation by approximate 60% while maintaining high accuracy.（4）On the basis of previous data fusion methods and bias kriging adjustment,this study devised a new hybrid data fusion method.The new hybrid data fusion method is a combination of bias kriging adjustment,which could overcome the limitation of the previous data fusion method in spatial bias adjustment,and the previous data fusion method.Cross-validation illustrates that the new hybrid method dramatically promoted the spatial accuracy of fused fields.（5）Applying the framework of uncertainty analysis to PM_2.5 simulations in the PRD region,this study finds that the average relative uncertainty in PM_2.5 simulations associated with meteorology,emission rates and boundary conditions is 24%²8%,which can explain much of the differences between observed PM_2.5 and simulated PM_2.5 concentration.Among the parametric uncertainty inputs considered in this study,wind speed,PM_2.5 boundary conditions,and PM_2.5 emissions are identified as the major sources of parametric uncertainty for PM_2.5 simulations in the PRD region.Data fusion method can effectively enhance boundary conditions and consequently improve regional modeling.Evaluation metrics of a case study in PRD region show that the NME declines by 6%-10% and R increase by 0.13-0.18 by enhancing the boundary conditions in the third modeling domain.Taking regional PM_2.5 simulation as an example,the thesis conducted a systematic study on uncertainty analysis and improvement of AQMs.The framework that established in this thesis not only could provide a methodology reference for studies on uncertainty quantification in numerical models,AQMs improvement,probabilistic air quality forecasting,probabilistic risk assessment and quick response models but also is helpful in advancing model applications from deterministic simulations to probabilistic simulations.