I.4.2. ->ENSB - Ensemble processing
 
                                                    
 Primary ICPs:   [ENSMBLc] [TAPESc] [DAYSc]
 
 Secondary ICPs: <TIMAVGc>
 
   In some experiments, the Model is run a number of times over the same time
period - with different initial conditions, for example.  It may be desirable
to process such a set of runs as a group.  This is referred to as "ensemble
processing", and each Model run is referred to as a "member" of the ensemble.
 
   When ensemble processing is requested [ENSMBLc], the relationship between
the requested days list [DAYSc] and the requested tapes list [TAPESc] is
modified.  Instead of treating the list of tapes as a progression in time
(where each day is assumed unique), each tape is assumed to belong to a
different member of the ensemble, and each requested day is processed from each
requested tape.  Note that the use of a simple, one-dimensional list of input
tapes restricts ensemble processing to one input tape per member.
 
   When time averaging is also requested (*TAVG) <TIMAVGc>, the "time average"
statistics computed are actually ensemble statistics, where each requested day
is grouped (averaged) with the same day from each of the specified input tapes.
 
   It is possible to define one ensemble average as comparison Case A <TAPESA>
<DAYSA>, another ensemble average as Case B <TAPESB> <DAYSB>, and compute the
difference between the two ensemble averages.  Ensemble processing is imple-
mented in a way that minimizes the amount of memory needed, but this results in
inefficient dataset positioning.  See (*MSFH) for a discussion.
 
   The capabilities provided by this option are somewhat limited, and the
implementation is not particularly efficient for the kind of complex processing
often required for forecast studies.  See (*LSDP) for an alternative which may
be more appropriate.