The categories are a guide, not an absolute classification. The lower category plot is an average of all the raw spectra that went into that category. Any particular spectrum may be missing some of the labeled peaks and have other peaks not labeled in the category plot. It is even easy to find examples of spectra where the category name is a poor description of that particular mass spectrum. The categorization process worked best on spectra with more than a few peaks.
The categories are in order of frequency (category 1 is the most common). The frequency is partly a function of where the aircraft was flying: during the WB-57F Aerosol Mission (WAM), the stratospheric categories are frequent just because the aircraft spent most of its time in the stratosphere.
Both the mass spectra and categories are displayed here at a resolution of 0.25 mass units. The original mass spectra were recorded at a much higher resolution and extended to higher masses.
Each spectrum was automatically assigned to a category based on regression tree analysis. In this method, each spectrum is initially assigned to a different category. All categories are correlated with all other categories. The best correlated categories are averaged and new correlations are computed. The process of combining the best correlated categories is continued until there are a reasonable number of categories left. For computational reasons, blocks of 2500 to 4100 spectra were partially categorized and then the process was continued with the entire data set. The point at which to stop combining categories was done on a sliding scale such that the more common categories had to be more highly correlated with each other in order to be combined.
Positive and negative spectra were categorized separately and each aircraft mission or ground field deployment was categorized separately. Some of the final set of categories (about 70) were manually combined when physically reasonable. This was particularly important for mineral categories where the details of the mass spectra were highly variable but some of the salient peaks (SiO2-, ...) were consistent.
The correlation coefficients for the categories were calculated using a combination of a list of linearly scaled peak areas and logarithmically scaled raw data. Only the average of the log scaled raw data are displayed in the category graphs on this web site.
If there is a rare category that looks like noise, that is probably exactly what it is. A few spectra had baseline shifts that put noise above the threshold used for categorization. Such noise does not correlate well with anything else so these spectra tended to end up in their own categories.
There are, however, a few gems of rare but very interesting categories.
The relative humidities listed next to each spectrum are calculated with respect to liquid water, not ice. (The liquid relative humidity is given because it is relevant for aerosol hygroscopic growth.) At temperatures near 200 K, ice is stable at liquid water relative humidities greater than about 50%. Relative humidities recorded in either ice or water clouds are overestimated because the the water measurements included condensed water.