corr.rwl.seg {dplR}

Compute Correlations between Series

Description

Computes the correlation between each tree-ring series in a rwl object.

Usage

corr.rwl.seg(rwl, seg.length = 50, bin.floor = 100, n = NULL,
             prewhiten = TRUE, pcrit = 0.05, biweight = TRUE,
             method = c("spearman", "pearson","kendall"),
             make.plot = TRUE, label.cex = 1, floor.plus1 = FALSE,
             master = NULL,
             master.yrs = as.numeric(if (is.null(dim(master))) {
                              names(master)
                          } else {
                              rownames(master)
                          }),
             ...)

Arguments

rwl

a data.frame with series as columns and years as rows such as that produced by read.rwl.

seg.length

an even integral value giving length of segments in years (e.g., 20, 50, 100 years).

bin.floor

a non-negative integral value giving the base for locating the first segment (e.g., 1600, 1700, 1800 AD). Typically 0, 10, 50, 100, etc.

n

NULL or an integral value giving the filter length for the hanning filter used for removal of low frequency variation.

prewhiten

logical flag. If TRUE each series is whitened using ar.

pcrit

a number between 0 and 1 giving the critical value for the correlation test.

biweight

logical flag. If TRUE then a robust mean is calculated using tbrm.

method

Can be either "pearson", "kendall", or "spearman" which indicates the correlation coefficient to be used. Defaults to "spearman". See cor.test.

make.plot

logical flag indicating whether to make a plot.

label.cex

numeric scalar for the series labels on the plot. Passed to axis.cex in axis.

floor.plus1

logical flag. If TRUE, one year is added to the base location of the first segment (e.g., 1601, 1701, 1801 AD).

master

NULL, a numeric vector or a matrix-like object of numeric values, including a data.frame. If NULL, a number of master chronologies, one for each series in rwl, is built from rwl using the leave-one-out principle. If a vector, the function uses this as the master chronology. If a matrix or data.frame, this object is used for building the master chronology (no leave-one-out).

master.yrs

a numeric vector giving the years of series. Defaults to names or rownames of master coerced to numeric type.

...

other arguments passed to plot.

Details

This function calculates correlation serially between each tree-ring series and a master chronology built from all the other series in the rwl object (leave-one-out principle). Optionally, the user may give a master chronology (a vector) as an argument. In the latter case, the same master chronology is used for all the series in the rwl object. The user can also choose to give a master data.frame (series as columns, years as rows), from which a single master chronology is built.

Correlations are done for each segment of the series where segments are lagged by half the segment length (e.g., 100-year segments would be overlapped by 50-years). The first segment is placed according to bin.floor. The minimum bin year is calculated as ceiling(min.yr/bin.floor)*bin.floor where min.yr is the first year in either the rwl object or the user-specified master chronology, whichever is smaller. For example if the first year is 626 and bin.floor is 100 then the first bin would start in 700. If bin.floor is 10 then the first bin would start in 630.

Correlations are calculated for the first segment, then the second segment and so on. Correlations are only calculated for segments with complete overlap with the master chronology. For now, correlations are Spearman’s rho calculated via cor.test using method = "spearman".

Each series (including those in the rwl object) is optionally detrended as the residuals from a hanning filter with weight n. The filter is not applied if n is NULL. Detrending can also be done via prewhitening where the residuals of an ar model are added to each series mean. This is the default. The master chronology is computed as the mean of the rwl object using tbrm if biweight is TRUE and rowMeans if not. Note that detrending can change the length of the series. E.g., a hanning filter will shorten the series on either end by floor(n/2). The prewhitening default will change the series length based on the ar model fit. The effects of detrending can be seen with series.rwl.plot.

The function is typically invoked to produce a plot where each segment for each series is colored by its correlation to the master chronology. Green segments are those that do not overlap completely with the width of the bin. Blue segments are those that correlate above the user-specified critical value. Red segments are those that correlate below the user-specified critical value and might indicate a dating problem.

Value

A list containing matrices spearman.rho, p.val, overall, bins, rwi, vector avg.seg.rho, numeric seg.lag, seg.length, pcrit, label.cex. An additional character flags is also returned if any segments fall below the critical value. Matrix spearman.rho contains the correlations for each series by bin. Matrix p.val contains the p-values on the correlation for each series by bin. Matrix overall contains the average correlation and p-value for each series. Matrix bins contains the years encapsulated by each bin. The vector avg.seg.rho contains the average correlation for each bin. Matrix rwi contains the detrended rwl data, the numerics seg.lag, seg.length, pcrit, label.cex are from the oroginal call and used to pass into plot.crs.

Author(s)

Andy Bunn. Patched and improved by Mikko Korpela.

See Also

corr.series.seg, skel.plot, series.rwl.plot, ccf.series.rwl, plot.crs

Examples

library(utils)
data(co021)
crs <- corr.rwl.seg(co021, seg.length = 100, label.cex = 1.25)
names(crs)
## Average correlation and p-value for the first few series
head(crs$overall)
## Average correlation for each bin
crs$avg.seg.rho