run_ensemble_bayes: Run an ensemble of offline land models and complete Bayesian...
In JGCRI/gcamland: GCAM Land Allocation Module

Description Usage Arguments Details Value Output Author(s)

Parameter combinations are selected by generating a quasi-random sequence and mapping it to a specified range for each parameter. Then, each parameter set is run through the offline land model in each of the Perfect, Adaptive, HybridLinearAdaptive, HybridPerfectAdaptive, and Linear variants. (I.e., if N parameter sets are selected, then 5N scenarios are run.)

run_ensemble_bayes(
  N = 500,
  aOutputDir = "./outputs",
  skip = 0,
  lpdf = get_lpdf(1),
  lprior = uniform_prior,
  aType = "Hindcast",
  aIncludeSubsidies = FALSE,
  aDifferentiateParamByCrop = FALSE,
  aSampleType = "LatinHyperCube",
  aTotalSamplesPlanned = 500,
  logparallel = NULL
)

`N`	Number of parameter sets to select
`aOutputDir`	Output directory
`skip`	Number of iterations to skip (i.e., if building on another run.)
`lpdf`	Log-likelihood function. Used only if Bayesian posteriors are being run.
`lprior`	Log-prior probability density function. Used only if Bayesian posteriors are being run.
`aType`	Scenario type: either "Reference" or "Hindcast"
`aIncludeSubsidies`	Boolean indicating subsidies should be added to profit
`aDifferentiateParamByCrop`	Boolean indicating whether all crops should use the same expectation parameters
`aSampleType`	String indicating what type of sampling, currently only "LatinHyperCube" and "Sobol" are supported
`aTotalSamplesPlanned`	Number of samples planned. For Latin Hypercube, we need to know the total before we start.
`logparallel`	Name of directory to use for parallel workers' log files. If `NULL`, then don't write log files.

If the scenario type is "Hindcast", then after each model has been run, the Bayesian analysis will be run so that its results can be stored with the rest of the ScenarioInfo structure.

List of ScenarioInfo objects for the ensemble members

The model results are written to a series of files in the specified output directory. The list of ScenarioInfo objects is written to a file called bayes-scenario-info.rds in the output directory. This file can be loaded with a command such as scenaro_list <- readRDS('output/bayes-scenario-info.rds'). These objects contain links to the model output files, as well as the posterior probability density tables, if the Bayesian analysis was run.

KVC November 2017

JGCRI/gcamland documentation built on Oct. 6, 2020, 5:30 p.m.