Key components in the assessment of the consequences of intentional or accidental release of CBRN materials are the atmospheric conditions responsible for the transport and dispersion. Traditionally the atmospheric inputs are obtained from downscaled climatological averages, which at best do not fully represent the range of possible conditions and at worst may actually be misleading with respect to actual atmospheric conditions. To improve the range of atmospheric conditions used in consequence assessments, we use an unsupervised machine learning technique based upon Kohonen neural networks (Self-Organizing Maps) to identify the atmospheric patterns in the historical record. The historical record is generated with a high-resolution limited area atmospheric model that re-analyzes the past 30 years based upon global data and local observations. The Self-Organizing Map is trained using daily patterns of multivariate output, such as low-level winds and stability, from the atmospheric model re-analysis.

The number of patterns that are identified by the Self-Organizing Map must be specified a priori, therefore the analysis typically overestimates the total number of patterns resulting in many patterns that bear close similarity to each other. In order to overcome this limitation we apply an optimization stage based upon hierarchical trees to identify the minimum number of unique atmospheric patterns. We can then identify a reasonable number of cases that represent the full range of possible atmospheric conditions. We also compare our method to similar Growing Hierarchical Self-Organizing Map and Neural Gas techniques.