Visual Analytics Of Spatiotemporal Multivariate Air Pollution Data

In recent years,air pollution has become a major environmental problem that endangers human health.One of the keys to improving air quality lies in the coordinated regional governance and the coordinated control of pollutants.Whether studying simulation models from the perspective of atmospheric science or mining environmental big data features from the perspective of data science,it is difficult to provide a comprehensive solution by relying solely on the knowledge of a single discipline.Moreover,inevitable uncertainty in the model,and the large scale and high complexity of pollution data sets,bring challenges to the collaborative analysis of air pollution.Therefore,based on interdisciplinary science,we combine the traditional models with scientific nature and artificial intelligence methods with characterization capabilities,to explore new analyzing solutions for air pollution distribution patterns,transport uncertainty,and the evolution of potential pollution sources.In this paper,we analyze air pollution data and the back trajectory model based on visualization and data mining methods.We first extract pollution patterns and anomalous events from the historical data,and then quantify the uncertainty of the back trajectory model.Based on both,we propose a potential pollution source identification method that integrates data features and models effectively.From three perspectives of time,space,and multivariate,we analyze the evolution of air pollution and pollution sources.We design three multi-view collaborative visual analytic systems,which enable exploring and mining the latent information interactively and visually.The contributions of this paper are as follows.(1)We propose a visual analytics system for air pollution to extract potential patterns and detect anomalies.Air pollution monitoring data is a typical spatio-temporal multivariate complex data and it is difficult to achieve progressive mining of the data using only independent quantitative metrics or algorithms.Therefore,we propose a visual analysis system to explore patterns and anomalies in the data.In order to create a visual space where multivariate information can be identified,a composite least square projection algorithm is proposed to effectively reduce information loss during the process and achieve the perception of multivariate distributions.Next,we propose a multivariate pattern extraction algorithm and a joint spatio-temporal anomaly detection method,which can identify regular patterns and reveal hidden multiple anomalous events.Based on this,glyphs for both regular patterns and anomalies are designed to summary the rich contextual information and track different timevarying states.We develop application case studies of air pollution monitoring data to demonstrate that the methods and system can guide users to interactively explore association patterns of pollutants,quickly identify periods or areas of abnormal pollution change,analyze periodic time-varying trends in pollution,and synergistic change patterns of urban agglomerations.(2)We propose a visual analysis method and system for trajectory trends and uncertainty of the backward trajectory model.The backward trajectory model is simple and widely used to identify air pollution sources,but the simulated trajectories are in three-dimensional geographic space,and the movement trends are difficult to observe.Besides,the uncertainty of the model is often ignored in the analysis process,so we propose a visual analysis system to address these issues.Based on neural network models,we propose a deep representation method of threedimensional back trajectories to measure the shape similarity of trajectories and find movement trends with spatiotemporal independence.Further,we quantify the uncertainty of the ensemble simulation of the back trajectory model and analyze the potential correlation between uncertainty and geographical locations.A glyph is designed to assist users in clearly exploring the movement trends and comparing uncertainty distributions within geographical neighborhoods.In this paper,the system is applied to the backward trajectory model of air pollution,which can show the simulated trajectory movement trends and quantify the uncertainty of the model,analyze the potential correlation between uncertainty and external factors such as geographic neighborhood and weather systems,and explore the causes of uncertainty in conjunction with interactive views.(3)We propose a visual analysis method and system for the identification and evolution of potential pollution sources.The current pollution tracing methods based on the backward trajectory model have low accuracy,and the differences in the distribution of pollution sources among different pollutants and different time periods cannot be ignored,so a visual analysis method integrating data features is proposed to explore potential air pollution sources.Based on the back trajectory model and air pollution data,we identify potential pollution sources with pollution transmission and co-occurrence relationships.After that,we construct a dynamic multivariate geographic network of air pollution impacts.Focusing on the network,we propose a diversity measures method for geographical distribution features.Based on it,the synergistic distribution patterns of pollution sources are extracted and the importance is evaluated.Finally,we design a multi-level glyph and a multi-view collaborative analysis framework to realize the cross-validation of the backward trajectory model and data features in the identification of pollutant sources,and finally explore the patterns of pollution source distribution and multivariate time-varying evolution patterns.In this paper,we conduct comparative experiments on real and synthetic datasets to verify the accuracy of methods in terms of data projection,pattern extraction,and feature learning.The usability and effectiveness of visual analytics systems are demonstrated through several application cases in real air pollution.The visual analytics of multivariate spatial and temporal air pollution data,the study of pollution distribution features,transport uncertainty and pollution source evolution patterns,can provide directional support and a scientific basis for air pollution targeting treatment and reduction measures establishment.