Factor analysis of long-term NO3- chemical ionization mass spectrometer (CIMS) dataset from Tvärminne coastal station

Atmospheric trace gases is a very wide term, which involves tens to hundreds of thousands of distinct chemical compounds. Not only does this pose significant challenges when attempting to measure these compounds but inferring the interconnections between individual species from such data is a monumental task. This is especially true for analyzing ambient data from a chemical ionization mass spectrometer from an entire year. Often some factorization method is applied to reduce the dimensions that must be considered when analyzing either gas or particle phase data and especially PMF has become a very popular tool for source apportionment of atmospheric mass spectral datasets. However, the computations required for PMF take a significant amount of time, and running the factorization for a full year of data would require a lot of time and resources. Therefore, this work focuses on using the faster Non-Negative Matrix Factorization (NNMF) algorithm to accomplish what PMF does, but in a fraction of the time. Specifically, bin-NNMF, a method analogous to the one described by Zhang et al. (2019), was used in this work. The key distinction between PMF and NNMF is that NNMF does not accept an error matrix, denoting the uncertainty of each separate data point in the input matrix. To still account for uncertainties, the rows and columns of the input matrix were instead weighted. Using this approach, which allows for faster experimenting with factorization outcomes and can handle the whole dataset without issue, the NO3-CIMS data for the entire year of 2024 was analyzed.

While work is still ongoing to further investigate the dataset, the analysis so far shows that the instrument has operated stably during the studied time-period. Several sets of NNMF runs with different weighting schemes, and between one and twelve output factors have been conducted. This would be extremely time consuming, or even impossible, using the PMF algorithm. The output factors all seem useful for further interpretation of the data, with slight variations based on the chosen weighting scheme. In general, the factors present distinct temporal patterns, and the spectral chemistry seems to make sense. Looking at the factors in connection to wind direction, many factors also exhibit clear directionality, as one might expect from a successful factor analysis. Especially for a coastal site the directional separation may be crucial for further data interpretation. For example, factors corresponding to organics from land or sea respectively were identified along with factors for sulfuric acid, iodic acid, and methanesulfonic acid, mostly originating from the sea. Therefore, NNMF seems to offer a viable alternative to the commonly used PMF analysis and provides a powerful tool for understanding long term mass spectral data.