climate_histories: Create Bclim climate_histories

Description

Runs a number of algorithms to create climate histories for a given set of layer clouds (from layer_clouds and a set of chronologies

Usage

climate_histories(layer_clouds,
chronology,
time_grid,
n_mix=10,
mix_warnings=FALSE,
n_chron=2000,
keep_parameters=TRUE,
control_mcmc=list(iterations=100000, burnin=20000, thinby=40, report=100),
control_chains=list(v_mh_sd=2, phi1_mh_sd=1, phi2_mh_sd=10, v_start=statmod::rinvgauss(layer_clouds$n_layers-1,2,1), Z_start=sample(1:n_mix, layer_clouds$n_layers, replace=TRUE), phi1_start=rep(3,layer_clouds$n_dimensions), phi2_start=rep(20,layer_clouds$n_dimensions)),
control_priors=list(phi1_dl_mean=rep(1.275,layer_clouds$n_dimensions), phi1_dl_sd=rep(0.076,layer_clouds$n_dimensions), phi2_dl_mean=rep(4.231,layer_clouds$n_dimensions), phi2dl_sd=rep(0.271,layer_clouds$n_dimensions)))

Arguments

layer_clouds

An object of class layer_clouds obtained from layer_clouds

chronology

A set of chronologies given as a matrix. These should be provided in thousands of years before present. See details below

time_grid

The time grid on which to create the climate histories

n_mix

The number of mixture components for the Mclust mixture algorithm.

mix_warnings

Whether to display warnings related to the mixture algorithm

n_chron

The number of chronologies to use

keep_parameters

Whether to keep latent parameters or not. Useful for convergence checking so default is TRUE

control_mcmc

A list containing elements that control the MCMC, including the number of iterations, the size of the burn-in period, the amount to thinby, and how often for the algorithm to report its progress.

control_chains

A list containing elements that control the starting values of the parameters (v_start, Z_start, phi1_start and phi2_start) and the Metropolis-Hastings proposal standard deviation for v, phi1 and phi2.

control_priors

A list containing the prior parameters for the volatilities, given by phi1 and phi2, both of which should be the log-mean and log-sd of the log-normal distribution. The values provided here are for the GISP2 ice core for the period 0 to 10k years BP.

Value

v.store: Samples of the posterior estimated volatilities
chron.store: Samples of the used chronologies
c.store: Samples of the posterior estimated climates
z.store: Samples of the posterior mixture indices
phi1: Values used for the IG prior on v for each climate dimension
phi2: Values used for the IG prior on v for each climate dimension
chron.loc: A character string giving the location of the chronology file
nchron: The number of chronologies in the chronology file
parameters: A list containing further latent parameter values for convergence checking (only if keep_parameters is TRUE)

Details

This function takes the layer_clouds produced by layer_clouds uses a set of algorithms to produce cliamte histories on the provided time grid. The full details are in the paper referenced below. The options listed above allow quite a detailed level of control over the behaviour of the algorithm, and convergence should be checked using suitable means (see e.g. the R package boa or coda).

One of the key inputs to this function is a chronology. This should be a matrix of n_chron by n_layers containing sample chronologies as produced by, e.g. the R package Bchron. These are used by the climate_histories function to take account of chronological uncertainty. In the (unlikely) event that there is no chronological uncertainty, the rows of the chronologies can be identical.

References

Parnell, A. C., et al. (2015), Bayesian inference for palaeoclimate with time uncertainty and stochastic volatility. Journal of the Royal Statistical Society: Series C (Applied Statistics), 64: 115–138.