evfit: Fit an extreme value distribution to observations

Description

Fits an extreme value distribution using L-moments to the values provided. In the presence of zero flow observations a mixed distribution is fitted.

Usage

evfit(x, distribution, zeta = NULL, check = TRUE,
      extreme = c("minimum", "maximum"))

Value

An object of class 'evfit' containing the L-moments and the estimated parameters is returned. Objects of class 'evfit' are basically a list with the following elements:

values: the values x used for fitting.
freq.zeros: a character vector of length one. Frequency of zero flow observations.
is.censored: logical, if the censored time was used for fitting.
parameters: a list as long as distribution containing the estimated parameters for each distribution.
lmom: sample L-moments of the censored series (only containing non-zero values).
extreme: character vector of length one, indicating what kind of extreme value was fitted.
T_Years_Event: optional. If quantiles have been computed they a stored in a matrix with return periods in rows and distributions in columns.

Arguments

x: numeric vector. Data which is an extreme value distribution is fitted to.
distribution: A character vector of distributions to fit. Basically all distributions provided by Hosking's lmom-package and their reversed counterparts can be chosen. See reversing.
zeta: numeric vector of length one for manually setting a lower bound. Only a few distributions allow for a lower bound, namely 'gpa', 'ln3', 'wak' and 'wei'. The default value of NULL results in not bounding the distribution, therefore the parameter zeta is estimated.
check: logical, should check_distribution get called?
extreme: character vector of length one. Can be either 'minimum' or 'maximum'. Helps to choose a correct distribution.

Details

This function is vectorized over distribution.

According to paragraph 7.4.2 of the WNO manual, special care has to be taken in the presence of zero flow observations. A cdf called G(x) is fitted to the non-zero values of the original time series.

If a distribution is fitted which allows for finite lower bound (zeta), and zeta is estimated being negative, estimation is repeated constraining zeta = 0. If this behavior is not desired, the parameter zeta has to be set explicitly.

Examples

Run this code

data("ngaruroro")
ng <- as.xts(ngaruroro)
minima <- as.vector(apply.yearly(ng$discharge, min, na.rm = TRUE))
evfit(x = minima, distribution = c("wei", "gevR"),
      extreme = "minimum")