An aerosol classification scheme for global simulations using the K-means machine learning method

Abstract

Abstract. The K-means machine learning algorithm is applied to climatological data of seven aerosol properties from a global aerosol simulation using EMAC-MADE3. The aim is to partition the aerosol properties across the global atmosphere in specific aerosol regimes; this is done mainly for evaluation purposes. K-means is an unsupervised machine learning method with the advantage that an a priori definition of the aerosol classes is not required. Using K-means, we are able to quantitatively define global aerosol regimes, so-called aerosol clusters, and explain their internal properties and their location and extension. This analysis shows that aerosol regimes in the lower troposphere are strongly influenced by emissions. Key drivers of the clusters' internal properties and spatial distribution are, for instance, pollutants from biomass burning and biogenic sources, mineral dust, anthropogenic pollution, and corresponding mixtures. Several continental clusters propagate into oceanic regions as a result of long-range transport of air masses. The identified oceanic regimes show a higher degree of pollution in the Northern Hemisphere than over the southern oceans. With increasing altitude, the aerosol regimes propagate from emission-induced clusters in the lower troposphere to roughly zonally distributed regimes in the middle troposphere and in the tropopause region. Notably, three polluted clusters identified over Africa, India, and eastern China cover the whole atmospheric column from the lower troposphere to the tropopause region. The results of this analysis need to be interpreted taking the limitations and strengths of global aerosol models into consideration. On the one hand, global aerosol simulations cannot estimate small-scale and localized processes due to the coarse resolution. On the other hand, they capture the spatial pattern of aerosol properties on the global scale, implying that the clustering results could provide useful insights for aerosol research. To estimate the uncertainties inherent in the applied clustering method, two sensitivity tests have been conducted (i) to investigate how various data scaling procedures could affect the K-means classification and (ii) to compare K-means with another unsupervised classification algorithm (HAC, i.e. hierarchical agglomerative clustering). The results show that the standardization based on sample mean and standard deviation is the most appropriate standardization method for this study, as it keeps the underlying distribution of the raw data set and retains the information of outliers. The two clustering algorithms provide similar classification results, supporting the robustness of our conclusions. The classification procedures presented in this study have a markedly wide application potential for future model-based aerosol studies.

A global climatology of ice-nucleating particles under cirrus conditions derived from model simulations with MADE3 in EMAC

“…k-means clustering; Hartigan and Wong, 1979) to identify regions dominated by specific INP types or by possible competition mechanisms between different INPs. This technique has recently been demonstrated for the analysis of global aerosol simulations by Li et al (2022). Following the results presented in this study, cirrus cloud and the resulting climate modifications induced by the ice-nucleating particles described here could be analysed in order to provide further insight into these INP-induced climate effects.…”

Transport patterns of global aviation NOx and their short-term O3 radiative forcing – a machine learning approach

“…Kassomenos et al (2010), in their calculation of the optimal cluster count for the K-means method, stress the volatility of this value based on one's choice of evaluation metric as well as on the method, like trial and error with manual verification. The optimal value found in this study, which is calculated based on the L method, is likely to vary if another method is employed, like the combined use of the sum of squared errors (SSE) with the silhouette coefficients, as was done by Li et al (2022). Another important note on the robustness of the clustered results pertains to the strong dependence of the horizontal transport distributions on the atmospheric circulation, i.e., on the wind field and more specifically on the dominant westerlies in both the Northern and Southern Hemisphere.…”

“…One of the challenges of applying clustering techniques is determining the optimal number of clusters, which, in the case of QuickBundles, is determined via the user-defined threshold θ . It is typical to combine the evolution of a sum of squared errors (SSE) function with silhouette coefficients to arrive at the most efficient number of clusters, as has been done in a recent aerosol clustering study with a K-means method (Li et al, 2022). This is a grouping technique that seeks to distribute data points across an optimally chosen number of clusters based on the minimization of a withincluster sum of squares (Hartigan and Wong, 1979).…”

“…Within aviation, aircraft routes have been grouped according to methods like the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) or the Hierarchical DBSCAN (HDBSCAN) as applied by Corrado et al (2020), Basora et al(2017), Olive and Basora (2019), and Olive and Morio (2019). Within atmospheric sciences, other techniques like K-means clustering have been used to systematically categorize different types of aerosol regimes at distinct altitudes and regions (Li et al, 2022). Another example is Kassomenos et al (2010), who applied several clustering approaches like K-means and self-organizing maps (SOMs) to isolate the pathways with highest PM 10 concentrations.…”

A global climatology of ice nucleating particles at cirrus conditions derived from model simulations with EMAC-MADE3

“…The exact code version used to produce the result of this paper is archived at the German Climate Computing Center (DKRZ) and can be made available to members of the MESSy community upon request. The model setup and the simulation data analysed in this work are available at http://doi.org/10.5281/zenodo.6834299(Beer, 2022).https://doi.org/10.5194/acp-2022-529 Preprint. Discussion started: 4 August 2022 c Author(s) 2022.…”