The classification of clouds now in use was proposed in 1802 by an amateur, adopted within a decade, and has survived two centuries of instrumentation that might have been expected to supersede it. The persistence is worth examining, because the scheme has no theoretical foundation whatever. It classifies by appearance -- heaped, layered, fibrous, and the combinations of these -- and it was devised before anything was understood about how clouds form.
A classification without a theory is ordinarily a liability. The categories of such a scheme divide a continuum at places chosen for convenience, and as understanding accumulates the divisions are found to cut across the processes that matter, and the scheme is then abandoned in favour of one that does not. That is the history of most descriptive taxonomies proposed in the same period.
The cloud scheme escaped because its categories turned out to correspond to something. A heaped cloud indicates convection: air rising in a column, cooling, and condensing at the height where it becomes saturated. A layered cloud indicates the opposite condition, a stable atmosphere with a whole stratum lifted or cooled uniformly. A fibrous cloud indicates ice, and therefore altitude. The visual distinctions the classifier drew are visible precisely because the underlying processes differ, and an observer sorting by appearance was sorting by mechanism without knowing it. Anyone can do the sorting, and that is part of the point. Stand in a field on a warm afternoon and watch the white heaps build over the same patch of ground, flat at the base and piled above. The flat base is where the rising air first cools enough for its water to show, and the piling is the air still going up. Come back on a grey still morning and the sky is one sheet from edge to edge, with no heaping at all. Look up on a cold clear day and the high white streaks are drawn out and feathered. Three looks, three mechanisms, and no instrument of any kind.
The correspondence is not perfect and the exceptions are instructive rather than decisive. Certain formations present the appearance of one category and the mechanism of another, and these are the cases where observers disagree, exactly what one would expect if the appearance is a proxy. Nobody proposes to replace the scheme on that account, because a proxy that is right in the great majority of cases and identifiable from the ground remains more useful than a mechanistic classification requiring measurements the observer cannot make.
The general lesson is that a descriptive category is not automatically superficial. Where the visible feature is produced by the process one wants to identify, description and explanation coincide, and a scheme built by looking can outlast the theory that would have justified it.
