---
title: "Prediction support, surface uncertainty, and contour uncertainty"
description: "Distinguish geometric support, model-conditional uncertainty, method spread, and pointwise contour bands."
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Prediction support, surface uncertainty, and contour uncertainty}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup}
library(potentiomap)
data("synthetic_wells")
p <- ps_make_points(synthetic_wells[1:16, ], "x", "y", "gw_elevation",
                    "well_id", "EPSG:26916")
fit <- suppressWarnings(ps_interpolate(p, methods = "OK", grid_res = 350,
                                      support = TRUE, return = "result"))
```

```{r uncertainty}
fit$support$summary
u <- ps_surface_uncertainty(fit, approach = "kriging_variance")
u$method_manifest
band <- ps_contour_uncertainty(u, 168, method = "gaussian_pointwise",
                               accept_gaussian = TRUE)
band$level_manifest
```

Support categories are not confidence classes. Kriging variance is conditional
on the fitted covariance model, and the contour product is pointwise rather
than a simultaneous confidence region.