extRemes (version 1.65)

dclust: Decluster Data by Runs Declustering

Description

Decluster data by assuming that exceedances belong to the same cluster if they are separated by fewer than 'r' (run length) values below a given threshold.

Usage

dclust(xdat, u, r, cluster.by = NULL, verbose=getOption("verbose"))

Arguments

xdat

a single numeric vector of data to be declustered.

u

single number or vector of thresholds.

r

run length

cluster.by

If there are blocks implying a natural clustering that is to be preserved (e.g., if data cover several years, but only for a single season), this is a vector defining the blocks to ensure that clusters do not cross over from one block to another.

verbose

logical whether to field progress information to screen or not.

Value

• A list with components:
• xdat.dc
  Maximums from each cluster with additional filler values below the given threshold u in order to maintain the same length as the original data vector xdat. This is for compatability with extRemes GUI data object of class ev.data.
• ncluster
  The number of clusters found by runs declustering.
• clust
  numeric vector giving the clusters.

Details

This function applies runs declustering to automatically decluster a dataset. To ensure that clusters do not cross natural or decided boundaries, use the cluster.by option. That is, suppose data are measured only in the summer, say from June 1 through August 1. In such a case, it is perhaps not desired to have a value from August 1, 2003 and June 1, 2004 in the same cluster. To account for this, create a cluster.by vector defining years in order to keep clusters within years. For the example of data from June 1 to August 1 (62 days), a vector like c(rep(1, 62), rep(2, 62), ..., rep(n, 62)) should be used for the cluster.by argument.

This function will return a vector of the same length as the original data vector, but with maximums from each cluster followed by filler numbers that are below the given threshold, u.

Missing values are not handled. The function will still run, but the results will be questionable.

References

Coles, S. (2001) An Introduction to Statistical Modeling of Extreme Values. London: Springer-Verlag, 208pp.

Examples

# Load a dataset.
data(Tphap)

plot( Tphap[,"MaxT"])
abline( h=115)

# Decluster using a threshold of 115 degrees and a run length of 'r=1'.
temp <- dclust(xdat=Tphap[,"MaxT"], u=115, r=1, cluster.by = Tphap[,"Year"])
temp[["ncluster"]] # See how many clusters were found.

# Now do the same as above, but with a run length of 3 for comparison.
# Note: 'r=2' gives same clusters as 'r=1' for these data.
temp2 <- dclust(xdat=Tphap[,"MaxT"], u=115, r=3, cluster.by = Tphap[,"Year"])
temp2[["ncluster"]]