**7. Conclusions**

The problem of analyzing multivariate spatiotemporal laws of air quality data is challenging due to the innate data complexity and latent associations. In this paper, we have presented our design of an innovative visual analysis system, AirInsight, to address this problem. Our system supports multivariate spatiotemporal pattern exploration and abnormal case analysis. A visual analysis framework and a spatiotemporal anomaly detection strategy are designed. In the analysis process, we also propose an interpretable dimensionality reduction algorithm CLSP and a clustering algorithm NHC that can diagnose noises. Several coordinated views and novel intuitive glyph designs are included in this system to provide rich contextual information. We also described three case studies and user evaluations to demonstrate that our work enables the user to explore multivariate patterns, trace time-varying processes, compare different cities, and find abnormal timestamps and cities.

As a wider variety of big data are collected, artificial intelligence provides an effective way to handle interdisciplinary issues. Automated algorithms can give users answers to complex questions. However, finding out what causes such results is not an easy task, which often requires integrating contextual information, triggering a wide use of visualization. Based on this, we propose the visual analysis system that combines both the automation algorithms and the interactive visual representations to mine and interpret the potential features in big data. In the future, we will further explore the effective combination of artificial intelligence and visual analysis, such as developing more interpretable automation algorithms, assisting users in adjusting model parameters through visualization, and so on.

**Author Contributions:** conceptualization, H.Z.; methodology, K.R.; software, Y.L. and Z.L.; writing–original draft preparation, D.Q.

**Funding:** This research was funded by National Natural Science Foundation of China under Grant grant number 41671379.

**Acknowledgments:** Thanks to the experts who provided requirements and user feedback for our work, as well as the participants who actively participated in the evaluation of the system.

**Conflicts of Interest:** The authors declare no conflict of interest.
