NEWS

bdrc 2.0.1 (2026-02-17)

Fixed phantom observation variance bug in gplm0 and gplm: the phantom observation that centers the GP used zero variance, acting as an infinitely precise constraint that collapsed the posterior of b. Fixed by using the prior variance sig_b^2.
Fixed eta posterior scaling bug in plm and gplm: the standardized eta coefficients were not scaled by sigma_eta before applying the basis transformation P, resulting in incorrect B-spline coefficient posteriors.
Fixed sigma_beta back-transform bug in gplm0: stored values were sqrt(sigma_beta) instead of sigma_beta due to an incorrect sqrt(exp()) transform. This affected posterior summaries and plots for the sigma_beta parameter in the gplm0 model.
Updated reference in DESCRIPTION from arXiv preprint to the published Environmetrics paper (doi:10.1002/env.2711).

Integrated C++ via Rcpp and RcppArmadillo packages for significant performance improvements.
Multiple functions rewritten in C++ to speed up the MCMC sampling algorithm and various other tasks.
The "Deviance" posterior output has been transformed and renamed "Posterior log-likelihood". The Deviance was previously calculated as -2 times the Posterior log-likelihood.
The plot(tournament_obj, type = "Deviance") figure is now created by evaluating plot(tournament_obj, type = "boxplot").
The log-likelihood of the models is now computed with the log-transformed discharge observations (normally distributed), rather than with discharge on the real scale (log-normally distributed).
Pointwise WAIC values (WAIC_i) stored to the model objects.
Implemented standard error computations for WAIC and Delta_WAIC estimates.
Applied log-sum-exp trick in WAIC and Bayes factor calculations for numerical stability.
Added DOI links to references.
Revised summary() output for tournament objects.
Updated package vignettes to reflect recent changes and improvements.

The tournament function now supports information criteria (WAIC & DIC) as the model selection criteria, WAIC being the new default.
Six new real-world datasets from Iceland and Sweden containing paired observations of stage and discharge have be included in the package.
The package can now also be used in a user-friendly interactive rating curve builder Shiny application [https://bdrc.shinyapps.io/bdrc/].