Rating curve fits

Every rc_*() constructor returns a list with class c("rc_<method>", "rating_curve"). These elements are part of the interface:

Details

gaugings

A tibble of the gaugings used, with columns discharge and stage, after dropping any with a missing value.

curve_parameters

The estimated parameters of the curve, as a named list with one element per parameter type. For a multi-segment curve an element holds one value per segment (for k, per breakpoint); elements may differ in length where a parameter is fixed by how the segments join. Empty for rc_loess(), which has no parameters. coef() returns the same estimates as a flat named vector.

settings

The arguments the fit was made with, including any user-supplied variances, aligned with gaugings. Enough to refit.

variance

For constructors with a variance argument: the variance scheme, with any estimated parameter filled in, such as the power estimated under var_power() (fit$variance$exponent).

Use fitted(), residuals(), coef() and predict() rather than reaching into the fit for those quantities.

Fits also carry elements that depend on how they are computed, and that may change as the methods for modelling the scatter develop: model, the underlying model object, such as from stats::nls(); rse, the residual standard error (on the log scale for the log-scale fits); and, for constructors with a variance argument, weights_used, the weights the final model was fitted with, and irls (for iteratively reweighted least squares): NULL, or under var_prop(), which refits in rounds, a list giving the number of rounds (iterations) and whether they converged. Some constructors carry further elements, documented on their own help pages.