Bayesian models (baSAR) applied on luminescence data

This function allows the application of Bayesian models on luminescence data, measured with the single-aliquot regenerative-dose (SAR, Murray and Wintle, 2000) protocol. In particular, it follows the idea proposed by Combès et al., 2015 of using an hierarchical model for estimating a central equivalent dose from a set of luminescence measurements. This function is (I) the adoption of this approach for the R environment and (II) an extension and a technical refinement of the published code.

Usage

analyse_baSAR(
  object,
  XLS_file = NULL,
  aliquot_range = NULL,
  source_doserate = NULL,
  signal.integral,
  signal.integral.Tx = NULL,
  background.integral,
  background.integral.Tx = NULL,
  irradiation_times = NULL,
  sigmab = 0,
  sig0 = 0.025,
  distribution = "cauchy",
  baSAR_model = NULL,
  n.MCMC = 1e+05,
  fit.method = "EXP",
  fit.force_through_origin = TRUE,
  fit.includingRepeatedRegPoints = TRUE,
  method_control = list(),
  digits = 3L,
  distribution_plot = "kde",
  plot = TRUE,
  plot_reduced = TRUE,
  plot.single = FALSE,
  verbose = TRUE,
  ...
)

Arguments

object: Risoe.BINfileData, RLum.Results, list of RLum.Analysis, character or list (required): input object used for the Bayesian analysis. If a character is provided the function assumes a file connection and tries to import a BIN/BINX-file using the provided path. If a list is provided the list can only contain either Risoe.BINfileData objects or characters providing a file connection. Mixing of both types is not allowed. If an RLum.Results is provided the function directly starts with the Bayesian Analysis (see details)
XLS_file: character (optional): XLS_file with data for the analysis. This file must contain 3 columns: the name of the file, the disc position and the grain position (the last being 0 for multi-grain measurements).
Alternatively a data.frame of similar structure can be provided.
aliquot_range: numeric (optional): allows to limit the range of the aliquots used for the analysis. This argument has only an effect if the argument XLS_file is used or the input is the previous output (i.e. is RLum.Results). In this case the new selection will add the aliquots to the removed aliquots table.
source_doserate: numeric (required): source dose rate of beta-source used for the measurement and its uncertainty in Gy/s, e.g., source_doserate = c(0.12, 0.04). Parameter can be provided as list, for the case that more than one BIN-file is provided, e.g., source_doserate = list(c(0.04, 0.004), c(0.05, 0.004)).
signal.integral: vector (required): vector with the limits for the signal integral used for the calculation, e.g., signal.integral = c(1:5). Ignored if object is an RLum.Results object. The parameter can be provided as list, see source_doserate.
signal.integral.Tx: vector (optional): vector with the limits for the signal integral for the Tx curve. I f nothing is provided the value from signal.integral is used and it is ignored if object is an RLum.Results object. The parameter can be provided as list, see source_doserate.
background.integral: vector (required): vector with the bounds for the background integral. Ignored if object is an RLum.Results object. The parameter can be provided as list, see source_doserate.
background.integral.Tx: vector (optional): vector with the limits for the background integral for the Tx curve. If nothing is provided the value from background.integral is used. Ignored if object is an RLum.Results object. The parameter can be provided as list, see source_doserate.
irradiation_times: numeric (optional): if set this vector replaces all irradiation times for one aliquot and one cycle (Lx and Tx curves) and recycles it for all others cycles and aliquots. Please note that if this argument is used, for every(!) single curve in the dataset an irradiation time needs to be set.
sigmab: numeric (with default): option to set a manual value for the overdispersion (for LnTx and TnTx), used for the Lx/Tx error calculation. The value should be provided as absolute squared count values, cf. calc_OSLLxTxRatio. The parameter can be provided as list, see source_doserate.
sig0: numeric (with default): allow adding an extra component of error to the final Lx/Tx error value (e.g., instrumental error, see details is calc_OSLLxTxRatio). The parameter can be provided as list, see source_doserate.
distribution: character (with default): type of distribution that is used during Bayesian calculations for determining the Central dose and overdispersion values. Allowed inputs are "cauchy", "normal" and "log_normal".
baSAR_model: character (optional): option to provide an own modified or new model for the Bayesian calculation (see details). If an own model is provided the argument distribution is ignored and set to 'user_defined'
n.MCMC: integer (with default): number of iterations for the Markov chain Monte Carlo (MCMC) simulations
fit.method: character (with default): equation used for the fitting of the dose-response curve using the function plot_GrowthCurve and then for the Bayesian modelling. Here supported methods: EXP, EXP+LIN and LIN
fit.force_through_origin: logical (with default): force fitting through origin
fit.includingRepeatedRegPoints: logical (with default): includes the recycling point (assumed to be measured during the last cycle)
method_control: list (optional): named list of control parameters that can be directly passed to the Bayesian analysis, e.g., method_control = list(n.chains = 4). See details for further information
digits: integer (with default): round output to the number of given digits
distribution_plot: character (with default): sets the final distribution plot that shows equivalent doses obtained using the frequentist approach and sets in the central dose as comparison obtained using baSAR. Allowed input is 'abanico' or 'kde'. If set to NULL nothing is plotted.
plot: logical (with default): enables or disables plot output
plot_reduced: logical (with default): enables or disables the advanced plot output
plot.single: logical (with default): enables or disables single plots or plots arranged by analyse_baSAR
verbose: logical (with default): enables or disables verbose mode
...: parameters that can be passed to the function calc_OSLLxTxRatio (almost full support), readxl::read_excel (full support), read_BIN2R (n.records, position, duplicated.rm), see details.

Value

Function returns results numerically and graphically:

———————————–
[ NUMERICAL OUTPUT ]
———————————–

RLum.Results-object

slot: @data

Element	Type	Description
`$summary`	`data.frame`	statistical summary, including the central dose
`$mcmc`	`mcmc`	coda::mcmc.list object including raw output
`$models`	`character`	implemented models used in the baSAR-model core
`$input_object`	`data.frame`	summarising table (same format as the XLS-file) including, e.g., Lx/Tx values
`$removed_aliquots`	`data.frame`	table with removed aliquots (e.g., `NaN`, or `Inf` `Lx`/`Tx` values). If nothing was removed `NULL` is returned

slot: @info

The original function call

————————
[ PLOT OUTPUT ]
————————

(A) Ln/Tn curves with set integration limits,
(B) trace plots are returned by the baSAR-model, showing the convergence of the parameters (trace) and the resulting kernel density plots. If plot_reduced = FALSE for every(!) dose a trace and a density plot is returned (this may take a long time),
(C) dose plots showing the dose for every aliquot as boxplots and the marked HPD in within. If boxes are coloured 'orange' or 'red' the aliquot itself should be checked,
(D) the dose response curve resulting from the monitoring of the Bayesian modelling are provided along with the Lx/Tx values and the HPD. Note: The amount for curves displayed is limited to 1000 (random choice) for performance reasons,
(E) the final plot is the De distribution as calculated using the conventional (frequentist) approach and the central dose with the HPDs marked within. This figure is only provided for a comparison, no further statistical conclusion should be drawn from it.

Please note: If distribution was set to log_normal the central dose is given as geometric mean!

Details

Internally the function consists of two parts: (I) The Bayesian core for the Bayesian calculations and applying the hierarchical model and (II) a data pre-processing part. The Bayesian core can be run independently, if the input data are sufficient (see below). The data pre-processing part was implemented to simplify the analysis for the user as all needed data pre-processing is done by the function, i.e. in theory it is enough to provide a BIN/BINX-file with the SAR measurement data. For the Bayesian analysis for each aliquot the following information are needed from the SAR analysis. LxTx, the LxTx error and the dose values for all regeneration points.

How is the systematic error contribution calculated?

Standard errors (so far) provided with the source dose rate are considered as systematic uncertainties and added to final central dose by:

$$systematic.error = 1/n \sum SE(source.doserate)$$

$$SE(central.dose.final) = \sqrt{SE(central.dose)^2 + systematic.error^2}$$

Please note that this approach is rather rough and can only be valid if the source dose rate errors, in case different readers had been used, are similar. In cases where more than one source dose rate is provided a warning is given.

Input / output scenarios

Various inputs are allowed for this function. Unfortunately this makes the function handling rather complex, but at the same time very powerful. Available scenarios:

(1) - object is BIN-file or link to a BIN-file

Finally it does not matter how the information of the BIN/BINX file are provided. The function supports (a) either a path to a file or directory or a list of file names or paths or (b) a Risoe.BINfileData object or a list of these objects. The latter one can be produced by using the function read_BIN2R, but this function is called automatically if only a file name and/or a path is provided. In both cases it will become the data that can be used for the analysis.

[XLS_file = NULL]

If no XLS file (or data frame with the same format) is provided the functions runs an automatic process that consists of the following steps:

Select all valid aliquots using the function verify_SingleGrainData
Calculate Lx/Tx values using the function calc_OSLLxTxRatio
Calculate De values using the function plot_GrowthCurve

These proceeded data are subsequently used in for the Bayesian analysis

[XLS_file != NULL]

If an XLS-file is provided or a data.frame providing similar information the pre-processing steps consists of the following steps:

Calculate Lx/Tx values using the function calc_OSLLxTxRatio
Calculate De values using the function plot_GrowthCurve

Means, the XLS file should contain a selection of the BIN-file names and the aliquots selected for the further analysis. This allows a manual selection of input data, as the automatic selection by verify_SingleGrainData might be not totally sufficient.

(2) - object RLum.Results object

If an RLum.Results object is provided as input and(!) this object was previously created by the function analyse_baSAR() itself, the pre-processing part is skipped and the function starts directly with the Bayesian analysis. This option is very powerful as it allows to change parameters for the Bayesian analysis without the need to repeat the data pre-processing. If furthermore the argument aliquot_range is set, aliquots can be manually excluded based on previous runs.

method_control

These are arguments that can be passed directly to the Bayesian calculation core, supported arguments are:

Parameter	Type	Description
`lower_centralD`	numeric	sets the lower bound for the expected De range. Change it only if you know what you are doing!
`upper_centralD`	numeric	sets the upper bound for the expected De range. Change it only if you know what you are doing!
`n.chains`	integer	sets number of parallel chains for the model (default = 3) (cf. rjags::jags.model)
`inits`	list	option to set initialisation values (cf. rjags::jags.model)
`thin`	numeric	thinning interval for monitoring the Bayesian process (cf. rjags::jags.model)
`variable.names`	character	set the variables to be monitored during the MCMC run, default: `'central_D'`, `'sigma_D'`, `'D'`, `'Q'`, `'a'`, `'b'`, `'c'`, `'g'`. Note: only variables present in the model can be monitored.

User defined models

The function provides the option to modify and to define own models that can be used for the Bayesian calculation. In the case the user wants to modify a model, a new model can be piped into the function via the argument baSAR_model as character. The model has to be provided in the JAGS dialect of the BUGS language (cf. rjags::jags.model) and parameter names given with the pre-defined names have to be respected, otherwise the function will break.

FAQ

Q: How can I set the seed for the random number generator (RNG)?

A: Use the argument method_control, e.g., for three MCMC chains (as it is the default):

method_control = list(
inits = list(
 list(.RNG.name = "base::Wichmann-Hill", .RNG.seed = 1),
 list(.RNG.name = "base::Wichmann-Hill", .RNG.seed = 2),
 list(.RNG.name = "base::Wichmann-Hill", .RNG.seed = 3)
))

This sets a reproducible set for every chain separately.

Q: How can I modify the output plots?

A: You can't, but you can use the function output to create own, modified plots.

Q: Can I change the boundaries for the central_D?

A: Yes, we made it possible, but we DO NOT recommend it, except you know what you are doing!
Example: method_control = list(lower_centralD = 10))

Q: The lines in the baSAR-model appear to be in a wrong logical order?

A: This is correct and allowed (cf. JAGS manual)

Additional arguments support via the ... argument

This list summarizes the additional arguments that can be passed to the internally used functions.

Supported argument	Corresponding function	Default	Short description
`threshold`	verify_SingleGrainData	`30`	change rejection threshold for curve selection
`sheet`	readxl::read_excel	`1`	select XLS-sheet for import
`col_names`	readxl::read_excel	`TRUE`	first row in XLS-file is header
`col_types`	readxl::read_excel	`NULL`	limit import to specific columns
`skip`	readxl::read_excel	`0`	number of rows to be skipped during import
`n.records`	read_BIN2R	`NULL`	limit records during BIN-file import
`duplicated.rm`	read_BIN2R	`TRUE`	remove duplicated records in the BIN-file
`pattern`	read_BIN2R	`TRUE`	select BIN-file by name pattern
`position`	read_BIN2R	`NULL`	limit import to a specific position
`background.count.distribution`	calc_OSLLxTxRatio	`"non-poisson"`	set assumed count distribution
`fit.weights`	plot_GrowthCurve	`TRUE`	enables / disables fit weights
`fit.bounds`	plot_GrowthCurve	`TRUE`	enables / disables fit bounds
`NumberIterations.MC`	plot_GrowthCurve	`100`	number of MC runs for error calculation
`output.plot`	plot_GrowthCurve	`TRUE`	enables / disables dose response curve plot
`output.plotExtended`	plot_GrowthCurve	`TRUE`	enables / disables extended dose response curve plot

Note

If you provide more than one BIN-file, it is strongly recommended to provide a list with the same number of elements for the following parameters:

source_doserate, signal.integral, signal.integral.Tx, background.integral, background.integral.Tx, sigmab, sig0.

Example for two BIN-files: source_doserate = list(c(0.04, 0.006), c(0.05, 0.006))

The function is currently limited to work with standard Risoe BIN-files only!

Function version

0.1.33

How to cite

Mercier, N., Kreutzer, S., 2024. analyse_baSAR(): Bayesian models (baSAR) applied on luminescence data. Function version 0.1.33. In: Kreutzer, S., Burow, C., Dietze, M., Fuchs, M.C., Schmidt, C., Fischer, M., Friedrich, J., Mercier, N., Philippe, A., Riedesel, S., Autzen, M., Mittelstrass, D., Gray, H.J., Galharret, J., Colombo, M., 2024. Luminescence: Comprehensive Luminescence Dating Data Analysis. R package version 0.9.26. https://r-lum.github.io/Luminescence/

References

Combès, B., Philippe, A., Lanos, P., Mercier, N., Tribolo, C., Guerin, G., Guibert, P., Lahaye, C., 2015. A Bayesian central equivalent dose model for optically stimulated luminescence dating. Quaternary Geochronology 28, 62-70. doi:10.1016/j.quageo.2015.04.001

Mercier, N., Kreutzer, S., Christophe, C., Guerin, G., Guibert, P., Lahaye, C., Lanos, P., Philippe, A., Tribolo, C., 2016. Bayesian statistics in luminescence dating: The 'baSAR'-model and its implementation in the R package 'Luminescence'. Ancient TL 34, 14-21.

Further reading

Gelman, A., Carlin, J.B., Stern, H.S., Dunson, D.B., Vehtari, A., Rubin, D.B., 2013. Bayesian Data Analysis, Third Edition. CRC Press.

Murray, A.S., Wintle, A.G., 2000. Luminescence dating of quartz using an improved single-aliquot regenerative-dose protocol. Radiation Measurements 32, 57-73. doi:10.1016/S1350-4487(99)00253-X

Plummer, M., 2017. JAGS Version 4.3.0 user manual. https://sourceforge.net/projects/mcmc-jags/files/Manuals/4.x/jags_user_manual.pdf/download

Author

Norbert Mercier, IRAMAT-CRP2A, Université Bordeaux Montaigne (France)
Sebastian Kreutzer, Institute of Geography, Heidelberg University (Germany)
The underlying Bayesian model based on a contribution by Combès et al., 2015. , RLum Developer Team

Examples


##(1) load package test data set
data(ExampleData.BINfileData, envir = environment())

##(2) selecting relevant curves, and limit dataset
CWOSL.SAR.Data <- subset(
  CWOSL.SAR.Data,
  subset = POSITION%in%c(1:3) & LTYPE == "OSL")

if (FALSE) { # \dontrun{
##(3) run analysis
##please not that the here selected parameters are
##choosen for performance, not for reliability
results <- analyse_baSAR(
  object = CWOSL.SAR.Data,
  source_doserate = c(0.04, 0.001),
  signal.integral = c(1:2),
  background.integral = c(80:100),
  fit.method = "LIN",
  plot = FALSE,
  n.MCMC = 200

)

print(results)


##XLS_file template
##copy and paste this the code below in the terminal
##you can further use the function write.csv() to export the example

XLS_file <-
structure(
list(
 BIN_FILE = NA_character_,
 DISC = NA_real_,
 GRAIN = NA_real_),
   .Names = c("BIN_FILE", "DISC", "GRAIN"),
   class = "data.frame",
   row.names = 1L
)

} # }