Prediction of Return Levels for a NSGEV object

Source: R/predictNSGEV.R

Prediction of Return Levels for a NSGEV object

# S3 method for NSGEV
predict(
  object,
  period = NULL,
  newdata = NULL,
  resample = FALSE,
  RLType = c("exceed", "average"),
  confInt = c("delta", "none", "proflik"),
  confLevel = 0.95,
  trace = 0,
  ...
)

Arguments

object

A NSGEV model.

period

A numeric vector giving the periods at which the Return Levels will be computed.

newdata

A data frame containing the covariates needed. Each row represents a block with unit duration.

resample

Not used yet. See the RL function. For now, the RL period that are greater than nrow(newdata) are discarded in the computation.

RLType

The type of Return Level as in RL.

confInt

The type of Confidence Interval (CI) to compute. The value "delta" leads to using the delta method, and "proflik" leads to using the profile-likelihood method.

confLevel

The Confidence Levels at which the CI will be computed. For "proflik" CI, only one level can be given for now.

trace

Level of verbosity.

...

Not used yet.

Value

A list with several elements.

RL

A data frame with the predicted RLs and the related CIs, with one row by block.

PsiStar

When confInt is equal to "proflik", this element is a matrix with its row \(i\) giving the value \(\boldsymbol{psi}^\star\) of the vector of parameters that maximizes the Return Level \(\rho(T)\) with period \(T = T_i\) under the constraint on the log-likelihood. Since most often the rows are close enough, a significant reduction of the computing time could be achieved in the near future by using the same value of \(\boldsymbol{\psi}\) for all the Return Periods.

Note

For the profile-likelihood method, the determination of the confidence intervals is quite slow because a constrained optimization problem is solved for each period.

Future versions might allow different durations across blocks by using dedicated arguments. For now it must be kept in mind that the periods are understood as multiple of a constant block duration. So if newdata has 100 rows the maximal Return Period that can be used without resampling is 100.

Author

Yves Deville

Examples

example(as.NSGEV.fevd)
#> 
#> a.NSGE> require(extRemes)
#> 
#> a.NSGE> ## see the examples for extRemes::fevd.
#> a.NSGE> data(PORTw)
#> 
#> a.NSGE> ## fit a GEV model
#> a.NSGE> fit <- fevd(x = TMX1, data = PORTw,
#> a.NSGE+             location.fun = ~AOindex, scale.fun = ~AOindex,
#> a.NSGE+             units = "deg C")
#> 
#> a.NSGE> ## coerce
#> a.NSGE> ns <- as.NSGEV(fit)
L.delta <- predict(ns, period = seq(from = 10,  to = 68, by = 10),
             RLType = "exceed")
#> Warning: Recycling array of length 1 in vector-array arithmetic is deprecated.
#>   Use c() or as.vector() instead.
#> Warning: Recycling array of length 1 in vector-array arithmetic is deprecated.
#>   Use c() or as.vector() instead.
#> Warning: Recycling array of length 1 in vector-array arithmetic is deprecated.
#>   Use c() or as.vector() instead.
#> Warning: Recycling array of length 1 in vector-array arithmetic is deprecated.
#>   Use c() or as.vector() instead.
#> Warning: Recycling array of length 1 in vector-array arithmetic is deprecated.
#>   Use c() or as.vector() instead.
#> Warning: Recycling array of length 1 in vector-array arithmetic is deprecated.
#>   Use c() or as.vector() instead.
plot(U_95 ~ period, data = L.delta$RL, ylim = c(20, 27), type = "l")
lines(quant ~ period, data = L.delta$RL)


## much slower, but much trustier!
if (FALSE) {
L.proflik <- predict(ns, period = seq(from = 10,  to = 68, by = 10),
             RLType = "exceed", confInt = "proflik") 
plot(U_95 ~ period, data = L.proflik$RL, ylim = c(20, 30), type = "l")
lines(quant ~ period, data = L.proflik$RL)
}