Ensembles are sets of time series data, where each member of the data set represents a different “possibility” for a given location and parameter. Rather than creating a large number of forecast runs to represent each separate possibility in an ensemble, a new forecast run type was added to collect the separate runs into a group with a single identity. An ensemble forecast run represents a set of discrete runs, one for each “possibility” in the ensemble. Since most of the models in the CAVI modeling suite do not fully support an ensemble alternative or run, the processing of the ensemble into separate runs is handled by the CAVI rather than by the user or the models. Therefore, the CAVI manages the ensemble runs and implement efficiency strategies of its own rather than rely on the efficiencies that may or may not be in the individual models.

HEC-DSS has a data grouping concept called collections which is used for storing and retrieving sets of data for a given location and parameter. A collection is a group or set of HEC-DSS time-series datasets that share a common location (A & B Parts), parameter (C Part), and time-step (E Part). Any set of HEC-DSS time-series records for a given location, parameter, and time-step can be put into a collection, however, when applied to an ensemble, the records grouped into a collection must span a common time window.

As such, ensembles are stored to HEC-DSS as collections. The collection concept has only been implemented for regular or irregular interval, point-based, time-series data. It has not been implemented for gridded data. Currently ensemble time series can only be read as the point-based precipitation time series for HEC-HMS and streamflow time series for HEC-ResSim. The following sections provide an example of an ensemble compute in CAVI using streamflow time series as input to HEC-ResSim.

The collection concept is implemented through an F Part naming convention. The F Part of each member of a collection has a collection ID string prepended to a common version label (F Part). The collection ID string consists of “C:” followed by a 6-digit collection member number (ID) and ending with a pipe symbol “|”. The “|” is used as a delimiter, it separates the collection ID string from the standard F Part string that specifies the “version” (uniqueness) of the data. The ID number must be 6 characters long and may be simple integers or alpha-numeric strings, but must be unique within the collection. Like the other pathname parts, the “version” string following the | will be the same for all members in the collection. 


Introduced in CWMS 3.4, the Ensemble Forecast Processor (EFP) allows the user to process various metrics from ensemble flows at specified locations in the watershed, and then use those metrics to influence other model alternatives in the forecast. The EFP plugin is flexible and can be placed throughout the program order of the CAVI forecast to allow for a variety of uses. 


Also first introduced in CWMS 3.4, the Ensemble Viewer allows for the user to view ensemble time series. The viewer also has the ability to view ensemble metrics that summarize the ensembles as a single summary time series, single value for each ensemble, or as a single summary value.