A few samples of the ceiling height climatology at SFO. The threshold used to identify a low ceiling (green) and a "burnoff" (blue) are also shown. The twenty days from mid-July obviously have a different climatology than do the twenty days from mid-January. The julian day is in GMT, so 4 AM PST is at the midpoint of the tic marks.
As seen above, the summer data often shows the onset of low ceiling at night, and a burnoff the following morning. Here burnoff is defined as the ceiling rising to, or above, 4000 feet, which in most cases is the same time as rising to infinity. The onset of low ceiling is defined as a ceiling at or below 2000 feet.
Days were found which showed a low ceiling at 4 AM. The time of onset of low ceiling (with low ceiling unbroken to 4 AM), and the time of subsequent burnoff were calculated for these days. Time of onset before midnite was recorded as -1. Time of burnoff both at or after 1900 PST was recorded as 19. Records for which there was missing data at 4 AM, or for which missing hourly data prevented the calculation of an onset or burnoff time, were thrown out. The years 1969-1973 and 1975-1981 had to be thrown out entirely because the data was spaced every three hours. Reanalysis data was available only from 1953. We thus produced a data set of 917 events which showed a low ceiling at 4 AM during July and August from 1948-1998, for which the burnoff time could be found in continuous hourly data, and for which reanalysis data was available.
This entire project (almost) can be reproduced by executing the dothis script within sfoburnoff.tar.gz (11 Mb).
A similar project that predicts the probability of being burned-off at 10 AM, rather than predicting the hour of burnoff, is bundled in sfoprob.tar.gz (6 Mb).