Estimate W via nonparametric block bootstrap (stationary)

Implements the bootstrap procedure from Section 2.4 of Russell et al. (2020). Resamples years (preserving spatial dependence) and re-fits stationary GEV at each site, then estimates the covariance of the stage-1 estimators. For a nonstationary alternative that accounts for NOAA sea level trends, see bootstrap_W_detrended.

Usage

bootstrap_W(dat, B = 500, log_scale = TRUE, seed = NULL)

Arguments

dat

An evfuse_data object.

B

Number of bootstrap replications (default 500).

log_scale

Logical. If TRUE, work with log(sigma) (default TRUE).

seed

Random seed for reproducibility.

Value

A list with components:

W_bs

The raw bootstrap covariance matrix (Lp x Lp).

Gamma

The bootstrap matrix of MLEs (B x Lp).

n_failures

Number of (site, bootstrap) combinations that failed.

Note

A small fraction of bootstrap resamples may produce degenerate data that causes warnings or convergence failures in extRemes::fevd. Failed fits are recorded in n_failures and excluded via pairwise-complete covariance estimation (failure rates are typically well below 1 percent).

See also

bootstrap_W_detrended for the nonstationary version.