Predict a `pgpTList` Object.

Usage

# S3 method for pgpTList
predict(
  object,
  newdata = NULL,
  lastFullYear = FALSE,
  trace = 0,
  which = c("param", "max"),
  probExcM = c(0.05, 0.01, 0.005, 0.001, 5e-04, 1e-04),
  ...
)

Arguments

object: A pgpTList object representing a list of Poisson-GP fitted models
newdata: An optional object giving the "new" dates for which the simulation will be done. It can simply be an object with class "Date" or an object with class "dailyMet". Depending on the class of newdata, the variables required for the prediction (such as sine waves) will be recomputed or not. When newdata is not given or is NULL, object$data is used.
lastFullYear: Logical, used only when newdata is not provided or is NULL. When TRUE, only the last full year in newdata will be used.
trace: Integer level of verbosity.
which: The kind of result wanted.
probExcM: A vector of (small) probabilities of exceedance for the maximum $M$ of the marks on the prediction period. For each probability $p$ , the value of the quantile $m$ of $M$ with probability $1 - p$ will be computed. Note that the quantile may be NA if the value $F_M(m)$ of distribution function of $M$ is lower than $1 -p$ when $m := \textrm{max}_t\{u_t\}$ because the POT model can only provide the tail distribution of $M$ .
...: Not used yet.

Value

An object with class "predict.pgpTList" A data frame in long format. Among the columns we find Date,

tau and u and the NHPP parameters muStar,

sigmaStar and xiStar. The column sigma

contains the GP scale parameter.

Note

Remind that the NHPP parameters do not depend on the threshold, although their estimates obviously do. They can be used to assess the sensitivity too the threshold choice.

Caution

This method still may change.