StrathE2E2 (version 3.3.0)

e2e_plot_sens_mc: Plot diagnostic data from sensitivity and Monte Carlo analyses.

Description

The function generates diagnostic plots from either a parameter sensitivity analysis or a Monte Carlo simulation of credible intervals of model outputs.

Usage

e2e_plot_sens_mc(model, selection = "SENS", use.example = FALSE)

Arguments

model

R-list object defining the model configuration, compiled by the e2e_read() function

selection

Text string from a list identifying source of data to bed merged. Select from: "SENS", "MC", referring to sensitivity analysis or Monte Carlo analysis. Remember to include the phrase within "" quotes.

use.example

Logical. If TRUE use pre-computed example data from the internal North Sea model rather than user-generated data (default=FALSE).

Value

Graphical display in a new graphics window.

Details

In both cases, the inputs required are a model list object created by the e2e_read() function defining the model configuration and relevant model.ident argument to point to the required files, and a logical argument to determne the origin of data files (either saved as csv files in the user workspace, or example data for the North Sea model provided with the package.

Details relating to sensitivity analysis plot:

The function reads processed results generated by the function e2e_run_sens(), or the function e2e_process_sens_mc(...,selection="SENS"), and plots the Elementary Effect mean (x-axis; magnitude of sensitivity) against Elementary Effect standard deviation (y-axis; strength of interactions between parameters).

Processed results from the function e2e_run_sens(), or the function e2e_process_sens_mc(...,selection="SENS"), are stored in the csv file ../Modelname/Variantname/sorted_parameter_elementary_effects-*.csv where * represents the model run identifier (model.ident) text embedded in the R-list object created by the e2e_read() function. The path to this file is relative to the results folder specified in teh e2e_read() function call.

Each symbol in the plot represents a single parameter in the model. The parameters are colour-coded to indicate 6 different types - fitted and fixed parameters of the ecology model, fishing fleet model parameters, fishery harvest ratios, environmental drivers, and physical configuration parameters.

The plot also shows a wedge formed by the two dashed lines. These correspond to +/-2 standard errors of the mean, so for points falling outside of the wedge there is a significant expectation that the distribution of elementary effects is non-zero. For points falling within the wedge the distribution of elementary effects is not significantly different from zero.

For details of how the Elementary Effect values are derived for each parameter see help(e2e_run_sens)

Details relating to Monte Carlo analysis plot:

The function creates a plot showing the credible distributions of ecology model parameters based on the results from the e2e_run_mc() function. These distributions are formed from the input distributions to the Monte Carlo process, weighted by the likelihood of observed target data on the state of the ecosystem given each combination of parameter values.

Post-processed data from the e2e_run_mc() function are stored in the file ../Modelname/Variantname/CredInt/CredInt_processed_PARAMETERresults_data-*.csv, where * represents the model run identifier (model.ident) text embedded in the R-list object created by the e2e_read() function. The path to this file is relative to the results folder specified in teh e2e_read() function call.

Each parameter in the plot is scaled to its baseline value as (value - baseline)/baseline. Ideally, the baseline is the maximum likelihood model developed by application of the optimization functions available in the package (e.g. e2e_optimize_eco()). Each parameter is then represented by a box and whisker plot which shows the distribution of credible parameter values around zero, i.e. around the baseline. The median of the credible values distribution for each parameter is shown my a black tick-mark. The box spans the 50 of simulated values). Whisker lines span the 99

The individual parameters are identified by numbers (rather than text names). These numbers correspond to the column numnbers in the file ../Modelname/Variantname/CredInt/CredInt_processed_parameters-*.csv . Details of the parameters associated with each identification number are available as a dataframe by using the function e2e_get_parmdoc().

The input distribution of parameter values to the Monte Carlo process is drawn from a random uniform distribution with limits specified in the monte_carlo control file for the model setup (located in a sub-folder of the /Param/control folder). This distribution is shown by a red box and whisker at the bottom of the plot. Given the random uniform input we expect the quartiles (limits of the box) to be symetrical and located mid-way between zero and the upper and lower extremes. Vertical red lines show the expected limits of the quartiles boxes if model results were completely insensitive to individual parameter values.

The extent to which individual parameter distributions deviate from the random uniform input is an indication of their sensitivity in the model. Parameters whose distributions are tightly confined around zero (the baseline value) are highly sensitive.

For some parameters, in particular the preference parameters, their credible distributions may span a wider range than the inputs. This may seem unexpected, but arises because within each feeding guild the preference parameters are not independent of each other. The preferences within each guild must sum to 1. Hence, during the Monte Carlo process, after drawing new values of the preference values they are all rescaled to sum to 1, which may mean that some of them will have been varied by more than the original coefficient of variation of the input random uniform.

For details of how the distribution of credible output values from StrathE2E are calculated see the help information for the e2e_run_mc() function.

References

For details on the sensitivity analysis method see: Morris, M.D. (1991). Factorial sampling plans for preliminary computational experiments. Technometrics, 33, 161-174.

For a review of sensitivity analysis methods including the Morris Method see: Wu, J. et al. (2013). Sensitivity analysis of infectious disease models: methods, advances and their application. J R Soc Interface 10: 20121018, 14pp.

See Also

e2e_read, e2e_run_mc, e2e_run_sens, e2e_get_parmdoc

Examples

# NOT RUN {
# These examples requires the Strathe2E2examples supplementary data package.

# Load the 1970-1999 version of the North Sea model supplied with the package:
if(require(StrathE2E2examples)){
    model <- e2e_read("North_Sea", "1970-1999")
# Plotting illustrated here using the example data sets, but you could run the 
# functions e2e_run_sens() or e2e_run_mc() to generate some results instead.

# Plot the results of a parameter sensitivity analysis. Each point represents one parameter.
# The x-axis represents the sensitivity of the overall likelihood of the observed data to 
# The y-axis represents the extent to which each parameter interacts with others.
    e2e_plot_sens_mc(model, selection="SENS", use.example=TRUE)  
}

#---------------------------------------------------------------------------

# Plot the credible distributions of ecology model parameters based on the results of a
# Monte Carlo analysis. These distributions are formed from the input distributions to 
# the Monte Carlo process, weighted by the likelihood of observed target data
# on the state of the ecosystem given each combination of parameter values.
if(require(StrathE2E2examples)){
    e2e_plot_sens_mc(model, selection="MC", use.example=TRUE)    # for Monte Carlo results

# To direct the graph output to a file rather than the screen, wrap the
# plot_Monte_Carlo_parameter_distributions() function call in a graphical device call:
# or jpeg("plot.jpg"), png("plot.png")
    pdf(file.path(tempdir(), "plot.pdf"),width=9,height=4)
    e2e_plot_sens_mc(model, selection="SENS", use.example=TRUE)
    dev.off()
}

#---------------------------------------------------------------------------

# For a comprehensive workflow through the generation of sensitivity and
# Monte Carlo analysis data see:
    help(e2e_run_sens)
    help(e2e_run_mc)

# }