Smooth a ground-plot cloud for display

Fits a thin-plate spline through the observations and returns it on a dense age grid with a pointwise confidence band. This is a VISUAL AID ONLY: nothing in the scoring path consumes it, and growth_reference_curves() continues to build the scored ground-plot reference by binning. Keeping the two separate is deliberate – swapping the scored reference changes every biomass_max_est that rests on plots, which is a calibration decision rather than a plotting one.

Usage

growth_smooth_observations(
  obs,
  bin = 20L,
  site = NULL,
  k = NULL,
  n_grid = 200L,
  level = 0.95
)

Arguments

obs

A tibble with age and aboveground_c_mg_ha.

bin, site

As in growth_bin_observations(); used only to collapse repeated visits, not to summarize.

k

Integer. Spline basis dimension. NULL derives one from the number of occupied bins, capped at 5, which keeps the fit from chasing individual plots.

n_grid

Integer. Number of ages at which to evaluate the fit.

level

Numeric. Confidence level for the band.

Value

A tibble with age, value, lo, hi, and a k attribute; zero rows when there are too few distinct observations to fit.

Details

What it is for is judging the binned series. A bin holding one plot is drawn at the same visual weight as a bin holding thirty, and the straight lines between bins imply a trajectory the plots may not support; a fit over the whole cloud shows how much of that movement is real. Where the band is wide, the binned points nearby are not evidence of anything.

Observations are collapsed by location and bin first, exactly as growth_bin_observations() does, so the fit and the binned series rest on the same evidence and any difference between them is the summarizing method rather than the sample.

The fit is on the IDENTITY scale. A log link is the obvious response to right-skewed biomass, but with a handful of plots at the old end it extrapolates violently – in the network this was built against it lifted one species' curve to 347 Mg C/ha against a binned maximum of 238 – so the skew is left to the confidence band to express. The band is clamped at zero, since negative aboveground carbon is not a state a stand can be in.

No prediction is returned outside the observed age range: a spline given no data has nothing to say, and a curve drawn past the last plot invites the reader to believe otherwise.

See also

Other growth calibration helpers: extract_landscape_cohort_structures(), growth_add_objective(), growth_auto_window(), growth_best_candidates(), growth_bin_observations(), growth_calibration_design(), growth_calibration_partition(), growth_calibration_work_root(), growth_climatic_distance(), growth_climatic_weight(), growth_expand_over_pseudo_species(), growth_factorial_ratio_grid(), growth_fitting_windows(), growth_identifiability(), growth_inflation_factor(), growth_mortality_onset_frac(), growth_plot_palette(), growth_pseudo_species_name(), growth_reference_curves(), growth_reference_inflection(), growth_score_fit(), growth_scoring_for(), growth_structure_cell_curves(), growth_structure_cohort_table(), growth_structure_design(), growth_structure_summary(), growth_window_for(), plot_growth_calibration(), plot_growth_candidate(), plot_growth_factorial_sensitivity(), plot_growth_structures(), read_growth_scoring(), read_landscape_cohort_structures(), scale_linetype_growth_reference(), write_growth_review_bundle()