gumbelEstThreshold: Find the Optimal Threshold for the Gumbel Model

In usnistgov/potMax: Empirical Calculation of the Distribution of the Maximum from a Peaks Over Threshold Process

Description

Estimate the threshold to use for the 2D extremal Poisson process when the tail length parameter for that model is exactly zero.

Usage

gumbelEstThreshold(x, lt, n_min, n_max, progress_tf = TRUE)

## S3 method for class 'declustered_series'
gumbelEstThreshold(x, lt, n_min, n_max,
  progress_tf = TRUE)

## Default S3 method:
gumbelEstThreshold(x, lt, n_min, n_max,
  progress_tf = TRUE)

Arguments

x	An S3 object of class declustered_series or a numeric vector.
lt	(numeric scalar) The length of the time series in units of time (seconds, minutes, hours, etc.).
n_min	The minimum number of thresholded observations to include
n_max	The maximum number of thresholded observations to include
progress_tf	Display a progress bar if TRUE, else not.

Details

A sequence of candidate thresholds is generated, and the threshold that minimizes the maximum vertical distance of the points of the W plot (described in gumbelWplot) to the 45^\circ line is selected. See the vingette for more details.

Value

An S3 object of class thresholded_series with elements $seleted_threshold, $lt, $y, $checked_thresholds, and $w_stats. The element $y is a numeric vector containing the actual observations (NOT differences from the threshold) that exceed the selected threshold. The element $w_stats contains the maximum vertical distance from points to the 45^\circ line of the W plot for the corresponding $checked_threshold.

Methods (by class)

• declustered_series:

• default:

Examples

## Not run: 
complete_series <- -jp1tap1715wind270$value

declustered_obs <- decluster(complete_series)

thresholded_obs <- gumbelEstThreshold(x = declustered_obs,
                                     lt = 100,
                                     n_min = 10,
                                     n_max = 100)

## End(Not run)

usnistgov/potMax documentation built on May 3, 2019, 2:38 p.m.