mwtraf {paltran}R Documentation

Dynamic adjustment of training sets - moving windows reconstruction

Description

The moving window method identifies the best number of nearest neighbours (window size) for each fossil diatom assemblage and the best transfer function based on the error statistic for each sample of the test set (e.g. sediment core)

Usage

mwtraf(..., method = "wapls", rplot = "TRUE", mwsize = c(40, 60, 80, 100, 120, 140, 160), mw_type = "dca", ncomp = 4, rmsep.incl = TRUE, env.trans = FALSE, spec.trans = FALSE)

Arguments

... x,y,z: required: species training set (x) as matrix and related environmental parameter (y). optional: test set(z) - species data from a sediment core. For all data: row- and columnnames are required! Otherwize the sample specific training set and the related environmenal parameter can not be found!
method currently only wapls is running: choosing ncomp=1 a simple wa is done, for each sample specific training set.
rplot TRUE/FALSE should the results plotted?
mwsize vector of the number of samples within each window: default runs only with large data sets, e.g. combined TP data set from the EDDI data base
mw_type How should the nearest neighborous be found? curently only a dca is running, chord distance and NMDS is in preperation.
ncomp number of components that should bee extract using wa-pls
rmsep.incl should the RMSEP used for modelselection?
env.trans should the environmental parameter bee transformed? "sqrt" for square root and "log10" for logarithm to the basis 10 are possible choices, default is FALSE.
spec.trans should the species data bee transformed? "sqrt" for square root and "log10" for logarithm to the basis 10 are possible choices, default is FALSE.

Details

WARNINGS: One run of a WA-PLS with a normal sample size takes about 20 seconds. Using mwtraf, for each sample 7 WA-PLS runs (default) are calculated, that takes 140 seconds. The reconstruction for a whole sedimet (80-100 samples) core can take 5 h and more! Please try first wih a smal test set, before running the whole reconstruction! Data must be organised in the same way as running da cca or dca in package vegan! (see examples) This approach needs large training sets, like the combinded TP data set from EDDI.

Value

sample.performance perforamnce of the sample specific wa-pls model
reconstruction reconstructed environmental parameter

Note

choosing mwsize=30 and ncomp =1 this approach equals the locally weighted WA regression (Battarby et al. 2006).

Author(s)

Sven Adler, sven.adler2@uni-rostock.de, University Rostock, Institute for Biosciences, General and Systematic Botany, Germany

References

Huebener, T., Dressler,M., Schwarz,A.,Langner, K., Adler,S. 2007. Dynamic adjustment of training sets (`moving windows` reconstruction) by using transfer functions in paleolimnology -a new approache, J. o. Paleolimnology. DOI 10.2007/s10933-007-9145-7

See Also

wa, wapls, wa package analogue

Examples


data(dud.df)
data(train_set.MV)
data(train_env.MV)

fit1<-mwtraf(train_set.MV,train_env.MV,dud.df[1:3,],mwsize = c(40, 60, 80))
fit1

[Package paltran version 1.0-0 Index]