Fit nonstationary GEV with covariate (Stage 1)

NOAA sites get nonstationary GEV: mu(t) = mu0 + mu1 * (cov(t) - ref_value), sigma and xi stationary. All four parameters (mu0, mu1, log_sigma, xi) are returned for spatial modeling. ADCIRC sites get the usual stationary fit with three parameters (mu0, log_sigma, xi).

Usage

fit_gev_ns(dat, df, covariate, ref_value = NULL)

Arguments

dat

An evfuse_data object.

df

The raw data frame containing the covariate column.

covariate

String column name in df (e.g., "sst_regional_warm").

ref_value

Centering value for the covariate. Default: mean across all NOAA site-years. Centering improves numerical stability and makes mu0 interpretable as the location at the reference covariate value.

Value

An evfuse_stage1 object with 4-column theta_hat. NOAA rows contain (mu0, mu1, log_sigma, xi); ADCIRC rows contain (mu0, log_sigma, xi, NA). The vcov_list entries are 4x4 for NOAA and 3x3 for ADCIRC.

See also

fit_gev_detrended for the time-trend version that discards mu1 before spatial modeling.