Changes between Version 2 and Version 3 of EwEugHintsForFittingModelsToTimeSeriesReferenceData


Ignore:
Timestamp:
2010-01-30 16:47:11 (9 years ago)
Author:
varunr
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • EwEugHintsForFittingModelsToTimeSeriesReferenceData

    v2 v3  
    11== 3.8 Hints for fitting models to time series reference data == 
    2 A critical step in development of credible models for policy analysis is to show that they can at least reproduce observed historical responses to disturbances such as fishing. See also [[Time series fitting in Ecosim.htm|Time series fitting in Ecosim]]. See [[Time series.htm|Time series]] for instructions for setting up time series data into your model. 
     2A critical step in development of credible models for policy analysis is to show that they can at least reproduce observed historical responses to disturbances such as fishing. See also [wiki:EwEugTimeSeriesFittingInEcosim Time series fitting in Ecosim]. See [wiki:EwEugTimeSeries Time Series] for instructions for setting up time series data into your model. 
    33 
    44This demonstration necessarily involves an iterative exercise in ‘fitting’ the model to data, by correcting parameter estimates and time series forcing information so as to show what values (or ranges of values, or alternative hypotheses about key processes) could explain the observed historical patterns.  For any such fitting exercise, it is critical to have as long a reference period, with as many different disturbance patterns, as is possible to assemble. Note though, that only where a time series is used to ‘drive’ the model, (i.e. fishing mortalities and effort series) is it necessary to have information for all years in the time series. Estimates of relative abundance, catches, etc. are not required for all years. Short reference data series carry little information about responses to some disturbances, and hence ability of a model to fit such short series is no test at all of its ability to make useful predictions about disturbances not represented in the reference data.  In more vivid terms, many model errors (structure and parameter values) will only reveal themselves (make themselves evident through strong departures of predicted from observed patterns) when the model is challenged to reproduce very long time series of responses.