ATMDATA: Automatic aTMospheric Data Analysis Tools based on AI technologies

We developed feature-engineering-based machine learning classifiers to effectively differentiate between new-particle formation (NPF) event and non-event days at the SMEAR II station in Hyytiälä, Finland. These classifiers utilize a range of time series features derived from multi-modal log-normal distributions. The machine learning models demonstrated over 90% accuracy in identifying NPF event and non-event days, and approximately 80% accuracy in further classifying the days into specific subcategories, such as class Ia, class Ib, class II, and non-event.

One of the key findings of the study is the classifiers’ exceptional ability to consistently identify all class Ia event days, where particle growth and formation rates are measured with high confidence. The study also conducted a comparative analysis between feature-engineering-based machine learning methods and image-based deep learning techniques. The results indicate that, with well-crafted feature engineering, machine learning not only matches but can also surpass deep learning approaches, particularly in situations where time efficiency is paramount.

The findings of this study set the stage for advancing the use of AI tools in automating atmospheric data analysis. This progress is crucial for deepening our understanding of atmospheric aerosols and their effects on human health, air quality, and climate.

Contact Person:
Martha Arbayani Zaidan, SMIEEE, Ph.D., Docent
Academy Research Fellow | Data Scientist | Head of IRON group
Department of Computer Science (CS) and Institute for Atmospheric & Earth System Research (INAR)/Physics, University of Helsinki, Finland
www.marthazaidan.com
www.helsinki.fi/iron

References:

Zaidan, M.A., Haapasilta, V., Relan, R., Junninen, H., Aalto, P.P., Kulmala, M., Laurson, L. and Foster, A.S., 2018. Predicting atmospheric particle formation days by Bayesian classification of the time series features. Tellus B: Chemical and Physical Meteorology, 70(1), pp.1-10.

Zaidan, M.A., Haapasilta, V., Relan, R., Paasonen, P., Kerminen, V.M., Junninen, H., Kulmala, M. and Foster, A.S., 2018. Exploring non-linear associations between atmospheric new-particle formation and ambient variables: a mutual information approach. Atmospheric Chemistry and Physics, 18(17), pp.12699-12714.

Laarne, P., Amnell, E., Zaidan, M.A., Mikkonen, S. and Nieminen, T., 2022. Exploring non-linear dependencies in atmospheric data with mutual information. Atmosphere, 13(7), p.1046.