Plot comprehensive model trajectories for one location

Renders the multi-panel "Model trajectories" figure for a single location from a persisted mosaic_trajectories artifact (written by run_MOSAIC() when control$predictions$capture_trajectories = TRUE). Each panel shows the comprehensive internal-state channels over time — surveillance fit (reported cases/deaths), burden, incidence drivers, force of infection, the SVEIR + W compartments, and sanity checks (mass balance, population, epidemic fraction) — with a uniform-thinned set of actual posterior member trajectories (the spaghetti, conveying spread) and a bold central line (see Central line). Observed surveillance points are overlaid only on the reported_cases/reported_deaths panels.

Usage

plot_model_trajectories(
  trajectories,
  location,
  output_dir,
  prefix = "trajectories",
  verbose = TRUE
)

Arguments

trajectories

A mosaic_trajectories object (from readRDS("2_calibration/trajectories_ensemble.rds")).

location

Character. ISO/location code to plot; must appear in trajectories$location_names.

output_dir

Directory to write the figures into (created if needed).

prefix

File prefix. Default "trajectories" — outputs <prefix>_<location>.pdf and <prefix>_<location>_p1.png.

verbose

Logical. Print progress. Default TRUE.

Value

Invisibly, a named list of the file paths written.

Details

Pure read-render: this function never runs the engine, never re-simulates, and never re-weights — it consumes only the compact artifact (per-channel central series + thinned actual lines, both already reduced over the best subset). It is the trajectory analogue of plot_model_ensemble.

Central line: for reported_cases / reported_deaths the bold line is the ensemble central series following control$predictions$central_method (by default the weighted median for cases and the weighted mean for deaths) and is bit-identical to the cases/deaths prediction plots' central line – it is reduced from the same engine-level member trajectories over the same final displayed member set and weights (no re-simulation). The thinned lines are engine-level member trajectories, not observation-level draws. True deaths (disease_deaths) follow the deaths channel's method so the two deaths panels are comparable. All other channels (compartments, FOI, incidence, derived) have no prediction-plot counterpart and use the weighted median.

Weighting note: the central series and lines are reduced over the final displayed member set and weights: the candidate best subset (is_best_subset / weight_best) when subset optimization is off, and the optimized subset (is_best_subset_opt / weight_best_opt) when control$predictions$optimize_subset = TRUE. Channels are captured at sim time (stream-to-disk) and reduced over the optimized members with NO re-simulation. The CFR(t) panel carries a dashed step line at the posterior median reported CFR for each year when trajectories$cfr_refs is present.

See also

render_MOSAIC_figures, plot_model_ensemble.