The main file is md.codas (produced by the perl script md_stats.pl). This file
is large (940,868 events), so it has been compressed with gzip. The columns
are: event ID; origin date and time; Md magnitude; Md subsource; Md NSTA (from
column 101 of hypoinverse summary line); number of weighted codas in DB
(NCDA);  number of weighted codas with nsamples > 0 (NNZ) number of weighted
codas with nsample == 0 (NWZ); number of codas apparently missing from DB
(NSTA - NCDA = Nmiss); number of waveform segments associated with event
(NWF).

Then I used awk to break out several groups of events from this master list.

no_codas: Mds with no weighted codas in DB, 148 events. For events with
waveforms, these codas could be recovered by scanning waveforms again. For the
others, if the codas can be found in old hypoinverse archive files, they
should be added to the DB.

zero_samples: Mds with all codas having nsamples = 0, 97,433 events. For
events with waveforms (167), these codas could be recovered by scanning
waveforms again. For the others, changing the coda "algorithm" to "HAND" may
be necessary.

md.codas_missCodas: Mds where some codas apparently are missing from DB
(Nmiss > 0). Or else we don't understand what column 101 of hypoinverse
summary line really means.

md.codas_noWF: Md events with no associated waveforms, 194,632 events. No
specific action for these; I've already tried collecting event waveforms from
the NCEDC continuous archive. But note that these events extended all the way
up to 2001.

md3_fewcodas: events with Md >= 3 and less than 3 codas, 1039 events. The
cutoffs I used are arbitrary, but I wanted to find large events that seemed to
have anomalously few codas. Some of these are aftershocks of large events, so
probably no more codas are possible. But other events indicate some kind of
problem that might warrant a review. Automatic scanning of waveforms might fix
some of these problems, but could make a mess of others.


Pete Lombard
9 October 2019