Skip to contents

Predict both spatial intercept and spatial slope at new locations

Source: R/spatial_interpolation.R
predict_spatial_dual_gp.Rd

The spatial calibration carries two independent GPs. Both share knot coordinates and a single length scale parameter (ls_intercept_km == ls_slope_km in posterior (two names for the same draw), but have distinct sigma_intercept_spatial and sigma_slope_spatial, and distinct z_intercept_spatial[*] and z_slope_spatial[*] knot effects.

Usage

predict_spatial_dual_gp(coords_new, knot_coords, draws, scaling, metric)

Arguments

coords_new

matrix(n_obs, 2) of (lon, lat) in DEGREES.

knot_coords

matrix(n_knots, 2) of (lon, lat) in DEGREES.

draws

data.frame of posterior draws (subset of leafwax_posterior$draws). Must contain columns z_intercept_spatial[1..n_knots], z_slope_spatial[1..n_knots], sigma_intercept_spatial, sigma_slope_spatial, and one of ls_intercept_km / ls_slope_km.

scaling

list with lon_mean, lon_sd, lat_mean, lat_sd, d2H_sd.

metric

character; the metric the posterior was FITTED under, one of "chordal" or "standardized" (typically model$metadata$spatial_metric). Required — a posterior must be predicted with the metric it was fitted under. "standardized" additionally needs scaling's lon/lat mean & sd.

Value

list with two matrices, each n_draws x n_obs: intercept (additive contribution to beta_0 in standardized d2H_wax space) and slope (additive contribution to the local \(\beta_{\delta^2 H_p}\) slope).