Summarise an AEME ensemble across members

run_aeme_ensemble() stores every member's full model output on the aeme object. Turning that into a plot or a skill score means, each time, looping over the members and pulling the variable out with get_var() - the expensive step. ensemble_summary() does that extraction once and returns a tidy object holding

stats - one row per date (and depth, and model, and variable) with the ensemble mean, sd, n and the quantiles named in probs;
members - the extracted long frame the statistics were built from (Date, depth, value, Model, var_sim, ens), kept unless keep_members = FALSE;
obs - the matching lake observations, depth-aligned to depths.

plot_ensemble() accepts the result directly, so a summary built once can be replotted for several variables, depths and intervals for free.

as.data.frame() returns the stats table - the long per-date summary, ready to write out or join to something else.

Usage

ensemble_summary(
  aeme,
  model,
  vars_sim = "HYD_temp",
  depths = NULL,
  probs = c(0.025, 0.25, 0.5, 0.75, 0.975),
  remove_spin_up = TRUE,
  add_obs = TRUE,
  depth_ref = c("surface", "bottom"),
  obs_tol = 0.5,
  keep_members = TRUE
)

# S3 method for class 'aeme_ensemble_summary'
as.data.frame(x, row.names = NULL, optional = FALSE, ...)

Arguments

aeme: an aeme object with ensemble output, or an ensemble_summary().
model: character; model(s) to plot. Optional when aeme is an ensemble_summary().
vars_sim: character; one or more AEME variables to summarise.
depths: numeric; depths (m, referenced to depth_ref) at which to summarise depth-resolved variables. NULL (default) pools all layers for a depth-resolved variable - matching the historical plot_ensemble() behaviour - and is the only sensible value for a 1-D variable such as LKE_lvlwtr or HYD_thmcln.
probs: numeric; quantile probabilities for the summary bands. 0.5 is always added so a median is available.
remove_spin_up: logical; remove spin-up period from plot. Default is TRUE.
add_obs: logical; add observations to plot
depth_ref: character; "surface" (default) or "bottom", passed to get_var(): whether depths are metres below the surface or metres above the lake bed (see get_var() for how the bottom follows the water level). Observations are placed in the same reference before they are matched to depths. Observations recorded below the surface are the usual case; those recorded as a height above the bed (see add_obs()) are used as they are for "bottom". An observation in the other reference is converted with the height of the water surface above the bed on its date: from the lake level observations, interpolated in time, or where there are none the ensemble median of the modelled level.
obs_tol: numeric; half-width (m) of the window used to attach observations to each requested depth. Default 0.5.
keep_members: logical; keep the extracted per-member frame on the result. Default TRUE. FALSE drops it - smaller object, but plot_ensemble(type = "line") and future member-level scores then can't be computed from it.
x: an aeme_ensemble_summary.
row.names, optional: passed to the data.frame method; ignored.
...: unused.

Value

An aeme_ensemble_summary object: a list of stats, members (or NULL) and obs (or NULL), with probs, depths, depth_ref, vars_sim, vars_2d, n_members and period attributes.

Usage

Arguments

Value

See also