Compare fitted variogram models

Fits each requested gstat variogram candidate independently and preserves initial values, fitted parameters, singular status, weighted fitting error, warnings, and errors. Smallest variogram SSE is not proof of best predictive performance.

Usage

ps_variogram_compare(
  variogram,
  models = c("Sph", "Exp", "Gau", "Mat"),
  initial = NULL,
  fit_method = 7,
  anisotropy = NULL,
  validation = NULL,
  select = FALSE,
  selection_metric = c("validation_rmse", "weighted_sse")
)

Arguments

variogram

A potentiomap_variogram or gstat empirical variogram.

models

Candidate model abbreviations.

initial

Optional model or named model-specific initial values.

fit_method

gstat fit method.

anisotropy

Optional explicit anisotropy result or c(angle, ratio).

validation

Optional compatible validation result/table.

select

Select a model using an explicitly supplied criterion.

selection_metric

Validation RMSE or weighted variogram SSE.

Value

A potentiomap_variogram_comparison with every fit and ranking.

Examples

data("synthetic_wells")
pts <- ps_make_points(synthetic_wells, "x", "y", "gw_elevation",
                      "well_id", "EPSG:26916")
cmp <- ps_variogram_compare(ps_variogram(pts), c("Sph", "Exp"))
#> Warning: 2 lag-direction bin(s) contain fewer than five pairs.
cmp$ranking
#>     candidate_id model status singular weighted_sse nugget partial_sill range
#> 1 variogram_0001   Sph failed       NA           NA     NA           NA    NA
#> 2 variogram_0002   Exp failed       NA           NA     NA           NA    NA
#>   kappa warning_count error_count validation_rmse rank
#> 1    NA             0           1              NA   NA
#> 2    NA             0           1              NA   NA
# Weighted SSE alone is not a predictive-performance guarantee.