Data Assimilation in Air Contaminant Dispersion Using a Particle Filter and Expectation-Maximization Algorithm

Rongxiao Wang,Xiaogang Qiu,Bin Chen,Sihang Qiu,Zhengqiu Zhu

doi:10.3390/atmos8090170

Abstract

The accurate prediction of air contaminant dispersion is essential to air quality monitoring and the emergency management of contaminant gas leakage incidents in chemical industry parks. Conventional atmospheric dispersion models can seldom give accurate predictions due to inaccurate input parameters. In order to improve the prediction accuracy of dispersion models, two data assimilation methods (i.e., the typical particle filter & the combination of a particle filter and expectation-maximization algorithm) are proposed to assimilate the virtual Unmanned Aerial Vehicle (UAV) observations with measurement error into the atmospheric dispersion model. Two emission cases with different dimensions of state parameters are considered. To test the performances of the proposed methods, two numerical experiments corresponding to the two emission cases are designed and implemented. The results show that the particle filter can effectively estimate the model parameters and improve the accuracy of model predictions when the dimension of state parameters is relatively low. In contrast, when the dimension of state parameters becomes higher, the method of particle filter combining the expectation-maximization algorithm performs better in terms of the parameter estimation accuracy. Therefore, the proposed data assimilation methods are able to effectively support air quality monitoring and emergency management in chemical industry parks.

Highlights

Air contaminant emissions and contaminant gas leakage incidents in chemical industry parks are significant events, which can pose a potential threat to public health
In Experiment 1, where the state parameters are only four dispersion coefficients, the data assimilation based on the particle filter is parameters are only four dispersion coefficients, the data assimilation based on the particle filter is effective for estimating the state parameters and improving the model prediction
When the dimension of state parameters becomes higher in Experiment 2, the estimation accuracy of the dimension of state parameters becomes higher in Experiment 2, the estimation accuracy of typical typical particle filter decreases because the particles with a high dimension are hard to converge to a particle filter decreases because the particles with a high dimension are hard to converge to a satisfactory satisfactory result

Summary

Introduction

Air contaminant emissions and contaminant gas leakage incidents in chemical industry parks are significant events, which can pose a potential threat to public health. Understanding the dispersion of air contaminants is essential to air quality monitoring and emergency responses to gas leakage incidents. Most conventional methods of predicting atmospheric dispersion rely on the atmospheric dispersion models (e.g., the Gaussian models and Lagrangian models) with some given input model parameters. Due to the dynamic and stochastic nature of atmospheric dispersion, it is impractical to measure these model parameters precisely, especially the meteorological data (e.g., the wind field). The source term (i.e., source location and release rate) is often unknown. These inaccurate or unknown parameters result in the inaccurate model prediction of air contaminant dispersion

Results

Discussion

Conclusion