They use whatever data they have available while maintaining the data density as feasible. It's not radiosonde launches that kill it, it's the satellite obs -- thousands per model run, largely over the poles, that they can't use. Ten radiosonde obs don't make a large difference; much of the other data are used in normal times as well. This comes from discussions with people at NCEP.

The other suggests that the airplane dropsondes were not input into the model, perhaps because there weren't any around/just prior to 12z or perhaps because of a data ingest issue. Not sure on that one, though.