Changes between Version 3 and Version 4 of EwEugHintsForFittingModelsToTimeSeriesReferenceData


Ignore:
Timestamp:
2010-01-30 17:00:58 (14 years ago)
Author:
varunr
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • EwEugHintsForFittingModelsToTimeSeriesReferenceData

    v3 v4  
    88The basic idea in this procedure is as follows.  Set up an Ecosim model and reference time series (of forcing inputs like fishing rates, and indices of temporal system response like relative biomasses and estimated total mortality rates). Examine the simulated and observed time patterns of response indices, look for groups that show large discrepancies in time pattern (trend), with particular emphasis on groups that have high biomass and are important prey or predator for other groups.  As an example, sardines and anchovy in a Benguela model (Shannon et al., 2004) showed upward trend in data but not in initial simulation results.  Focus in turn on each such group, and examine alternative hypotheses for the discrepancy (by varying appropriate parameters to see if the model fit is improved).  The following are common hypotheses that should be examined in roughly the order listed: 
    99 
    10   Bad trend data — it is possible that the model predictions are sound, but that the trend data are misleading for some reason, (e.g., increasing catchability in CPUE indices).Incomplete or incorrect forcing data, especially for fishing mortality rates—Ecosim-simulated patterns for exploited species will obviously not track observed patterns if those patterns have been caused by fishing, but no good time pattern of fishing mortalities (or at least fishing efforts) have been provided.Inappropriate vulnerability parameters for the group’s prey — low [[Vulnerabilities flow control.htm|vulnerability settings]] (e.g., the 2.0 default) for a group’s column in the vulnerability matrix (of its prey to it) can cause two errors: (i) failure of the group to increase following reductions in mortality (due to preventing the group from increasing its total food intake); (ii) and/or failure of the group to decrease following increases in mortality, due to overestimates of compensatory Q/B responses to decreased intraspecific competition.  Check this by clicking the ‘sensitivity of SS to vulnerabilities’ button in the [[Fit to time series.htm|Fit to time series]] form to determine whether vulnerability changes would cause changes in goodness of fit, and consider using the fitting interface to search for improved vulnerability estimates. See [[Effect of P B Z and vulnerability.htm|Effect of P/B (Z) and vulnerability for time series fitting for more]] information.Incorrect P/B (Z) setting in Ecopath for the group—it is common to see P/B, i.e. Z values set far too large in the Ecopath inputs, resulting in low EE and hence low sensitivity of a group to changes in mortality agents.  Check the simulated time plot of total, fishing, and predation mortality rate components on the Ecosim [[Run ecosim.htm|Run Ecosim]] form to see if the total mortality rate and its partitioning among factors are reasonable. See [[Effect of P B Z and vulnerability.htm|Effect of P/B (Z) and vulnerability for time series fitting for more]] information.Changes in system productivity—in some systems we have seen correlated declines or increases across a variety of species, despite differences among species in harvesting impacts, which might be explained by changes in basic productivity due to factors like upwelling.  The Ecosim Fit to time series form can be used to ‘reconstruct’ an apparent temporal pattern in primary productivity, by fitting the model to time series for all groups while varying a time series of productivity ‘anomalies’.Trophic mediation effects—evaluate the possibility that changes in consumption and mortality have been caused by ‘indirect’ or ‘mediation’ effects, such as groups providing hiding places for other groups or driving behaviour of groups so as to make those groups more or less vulnerable to other predators.  In systems that have benthic and pelagic primary producers, note that shading effects by phytoplankton on benthic plants are not represented explicitly in Ecosim, and must be modelled as mediation effects (by [[Mediation.htm|setting up a mediation function]] that causes negative effects on benthic plant production as phytoplankton biomass increases).  This is also the case with turbidity and decreased foraging efficiency of visual predators that can be caused by phytoplankton. 
     10  Bad trend data — it is possible that the model predictions are sound, but that the trend data are misleading for some reason, (e.g., increasing catchability in CPUE indices).Incomplete or incorrect forcing data, especially for fishing mortality rates—Ecosim-simulated patterns for exploited species will obviously not track observed patterns if those patterns have been caused by fishing, but no good time pattern of fishing mortalities (or at least fishing efforts) have been provided.Inappropriate vulnerability parameters for the group’s prey — low [wiki:EwEugVulnerabilitiesFlowControl vulnerability settings] (e.g., the 2.0 default) for a group’s column in the vulnerability matrix (of its prey to it) can cause two errors: (i) failure of the group to increase following reductions in mortality (due to preventing the group from increasing its total food intake); (ii) and/or failure of the group to decrease following increases in mortality, due to overestimates of compensatory Q/B responses to decreased intraspecific competition.  Check this by clicking the ‘sensitivity of SS to vulnerabilities’ button in the [[Fit to time series.htm|Fit to time series]] form to determine whether vulnerability changes would cause changes in goodness of fit, and consider using the fitting interface to search for improved vulnerability estimates. See [[Effect of P B Z and vulnerability.htm|Effect of P/B (Z) and vulnerability for time series fitting for more]] information.Incorrect P/B (Z) setting in Ecopath for the group—it is common to see P/B, i.e. Z values set far too large in the Ecopath inputs, resulting in low EE and hence low sensitivity of a group to changes in mortality agents.  Check the simulated time plot of total, fishing, and predation mortality rate components on the Ecosim [[Run ecosim.htm|Run Ecosim]] form to see if the total mortality rate and its partitioning among factors are reasonable. See [[Effect of P B Z and vulnerability.htm|Effect of P/B (Z) and vulnerability for time series fitting for more]] information.Changes in system productivity—in some systems we have seen correlated declines or increases across a variety of species, despite differences among species in harvesting impacts, which might be explained by changes in basic productivity due to factors like upwelling.  The Ecosim Fit to time series form can be used to ‘reconstruct’ an apparent temporal pattern in primary productivity, by fitting the model to time series for all groups while varying a time series of productivity ‘anomalies’.Trophic mediation effects—evaluate the possibility that changes in consumption and mortality have been caused by ‘indirect’ or ‘mediation’ effects, such as groups providing hiding places for other groups or driving behaviour of groups so as to make those groups more or less vulnerable to other predators.  In systems that have benthic and pelagic primary producers, note that shading effects by phytoplankton on benthic plants are not represented explicitly in Ecosim, and must be modelled as mediation effects (by [[Mediation.htm|setting up a mediation function]] that causes negative effects on benthic plant production as phytoplankton biomass increases).  This is also the case with turbidity and decreased foraging efficiency of visual predators that can be caused by phytoplankton. 
    1111 
    1212If none of these hypotheses produces predicted patterns similar to the data, look closely at the Ecosim predicted patterns of change in consumption, growth, and mortality rates, and try to evaluate how these rates would have to change in order to produce observed trend patterns.  Examine the observed time series for other groups, particularly prey and predators of the group under study, to see if those time series suggest changes in trophic conditions (growth, mortality) that have not yet been captured by the model due to inappropriate parameter settings for the other groups.