Plot predicted concentration — plot_concentration • EpiSewer

Plots measured and/or predicted concentrations over time. The predicted concentrations are taken from a fitted EpiSewer model.

Usage

plot_concentration(
  results = NULL,
  measurements = NULL,
  flows = NULL,
  include_noise = TRUE,
  normalized = FALSE,
  median = FALSE,
  mark_outliers = FALSE,
  forecast = TRUE,
  forecast_horizon = NULL,
  date_margin_left = 0,
  date_margin_right = 1,
  facet_models = FALSE,
  facet_direction = "rows",
  base_model = "",
  model_levels = NULL,
  concentration_col = "concentration",
  flow_col = "flow",
  date_col = "date",
  outlier_col = "is_outlier",
  type = "time",
  intervals = c(0.5, 0.95)
)

Arguments

results: Results object returned by EpiSewer() after model fitting. Can also be a named list with results from different model runs, in which case all results are plotted together and distinguished by colors.
measurements: A data.frame with observed measurements, which will be plotted alongside the predicted values (useful to assess model fit). Can also be used with results = NULL, in which case only the observed measurements are plotted.
flows: A data.frame with observed flows. If normalized = TRUE, this will be used to normalize the estimated and observed concentrations to the median flow.
include_noise: If TRUE (default), concentrations including measurement noise are shown. If FALSE, only the expected concentrations are shown.
normalized: If TRUE, the estimated and observed concentrations are normalized to the median flow. Thereby, the noise in concentrations that is due to variation in flow is essentially removed. This is especially useful for assessing forecasting performance when the future flow values were not known during model fitting.
median: Should the estimated median be shown, or only the credible intervals? Default is FALSE to avoid over-interpretation of the median.
mark_outliers: If TRUE, outliers in the measurements are highlighted in red. See also argument outlier_col below.
forecast: Should forecasted concentrations be shown? Default is true. This requires that the model was fitted with a forecast horizon, see model_forecast(). Because concentrations depend on the flow volume, accurate concentration forecasts are only possible if flow values beyond the estimation date are provided. This is typically only possible when estimating retrospectively. If flow values beyond the estimation date are not available, the forecasted concentrations will be based on the median flow volume.
forecast_horizon: How many days into the future should forecasts be plotted? Note that this is restricted by the forecast horizon specified during model fitting, see horizon_assume().
date_margin_left: By how many days into the past should the plot be expanded? Can also be negative to cut off some of the earliest dates.
date_margin_right: By how many days into the future should the plot be expanded? Can also be negative to cut off some of the latest dates. By default, this is 1 to improve the visibility of the latest date.
facet_models: Should the plot be faceted by model? Default is FALSE.
facet_direction: How should the facetting be done? Either in different "rows" (default) or in different "columns".
base_model: Name of the base model (in the named list provided to results) which should be compared to the other models. This model will be plotted in black and will not be part of the legend.
model_levels: A character vector with the names of the models to be included. The colors and legend will be ordered according to the order in model_levels.
concentration_col: Name of the column in the measurements data.frame which contains the measured concentration that should be plotted.
date_col: Name of the date column in the measurements data.frame.
outlier_col: Name of a logical column in the measurements data.frame which identifies outlier measurements (for example added by mark_outlier_spikes_median()).
type: If "time" (default), the concentration time series is plotted. If "pp_check", the observations are ordered by concentration and plotted against the predicted concentration (useful for posterior predictive checks).

Value

A ggplot object showing predicted and observed concentrations over time. Can be further manipulated using ggplot2 functions to adjust themes and scales, and to add further geoms.

Details

When plotting a posterior predictive check (type="pp_check"), each observed concentration is shown together with the 95% Credible Interval of the posterior predictive distribution for that observation. The observations are ordered by concentration, which helps to visualize bias and variance of the predictions as a function of the concentration.