In weecology/portalcasting: Model and Forecast Portal Rodent Dynamics

  knitr::opts_chunk$set(
    collapse = TRUE,
    comment = "#>"
)

  library(portalcasting)
  vers <- packageVersion("portalcasting")
  today <- Sys.Date()

  models_controls <- prefab_models_controls( )
  models_names    <- unlist(mapply(getElement, prefab_models_controls(), "metadata")["print_name", ])

  nmodels        <- length(models_names)
  nmodels_text   <- as.character(english::english(nmodels))

This vignette shows how to set up and run your own Portal Predictions directory.

Installation

First things first, make sure you have the current version of portalcasting installed from GitHub:

install.packages("remotes")
remotes::install_github("weecology/portalcasting")
library(portalcasting)

Create a Portal Predictions directory

The setup_dir() function creates and populates a standard Portal Predictions directory that includes resources, data, models, fits, and forecasts subdirectories. By default, setup_dir() downloads the most up-to-date version of the Portal Data archive from GitHub and the NMME downscaled weather forecasts into the resources subdirectory; unpacks relevant components; prepares the newmoons, rodents, covariates, covariate forecast, and metadata data files (and places them in the data subdirectory); and populates the models subdirectory with scripts for the r nmodels_text existing models (r models_names).

Specialized versions of setup_dir are tailored for local model exploration, development, and testing ("setup_sandbox") and for use in the real-deal pipeline ("setup_production"). The settings are fairly similar, although setup_sandbox has extra rigid argument checking turned off and extra verbose messaging turned on. In addition, it does not download the repository archive resource (only the Portal Data and NMME resources). setup_production provides a robust starting point for a user interested in seeing the range of what the package can do, with specific error checking and verbose messaging. setup_production does download the entire Portal Predictions archive from GitHub. Note that downloading the full directory does take a few minutes.

There are many arguments available for the user to tailor the setup of the directory, which can be found on the help pages (?setup_dir and ?directory_settings). Perhaps the most important argument is main which allows the user to point the directory to any particular location. The default is main = ".", which is basically "the present directory". A common implementation would be to create a directory in the home directory of a computer (indicated by "~") and within a named folder, say "portalcast_directory", which would be done with main = "~/portalcast_directory", or by setting main <- "~/portalcast_directory" and then using main = main throughout the code:

main <- "~/portalcast_directory"
setup_dir(main = main)

The models and datasets arguments to setup_<> functions allow for control over which models and rodent data sets are included in the directory, and new models and datasets can be added as long as they are given controls in the new_dataset_controls or new_model_controls arguments. Many additional arguments for altering directory set up are available indirectly via the settings argument and wrapped into the directory_settings functions.

Run a forecast

The portalcast() function controls the running of potentially multiple models across various species in different data sets for a given time setup (defined in directory_settings and time_settings specifically). This can involve what we have classically called "forecasts" (from the present time step) and "hindcasts" (from a time step before the current one), although the data must be set accordingly (using time_settings) for each time point prior to invoking portalcast().

Presently, the preloaded model set includes r nmodels_text models: r models_names. The jags and GARCH models all take a bit longer to run, so for the purposes of illustration, we will only run ESSS, AutoArima, and NaiveArima models, which we indicate through input to the models argument:

portalcast(main = main, models = c("ESSS", "AutoArima", "NaiveArima"))

If the user does not specify the models, all of the prefab models are run. Note that we need to point portalcast to the directory of interest via the main argument. This allows us to go between different directories from the same R session with relative ease, but it does mean that main is a key argument in nearly all functions. Indeed, main is the only argument for many portalcasting functions.

Plot the results

Presently two plotting types are available for visualizing the data and model results: time series and point-in-time plots for forecasts. The functions for both of these figure types point directly to the forecast metadata file that allows for flexible selection of which specific model, data set, and end moon (forecast origin) to use, as well as selection via specific identifiers (when multiple versions of a model are run).

Time series plots are constructed using plot_forecast_ts:

plot_forecast_ts(main = main, dataset = "controls")

Point-in-time prediction plots are constructed using plot_forecast_point, and default to the next step ahead in time:

plot_forecast_point(main = main, dataset = "controls")

Reading in data to the R console from the directory

A series of read_<name> functions are available for simple loading of the data sets into R from the directory. A generalized read_data function includes an argument for which data set to load ("rodents" [and then which specific data set], "covariates", "climate_forecasts", "newmoons", or "metadata"), and each of those data sets also has a specific function, such as read_moons. read_forecasts_metadata has a function itself, but is not called via read_data currently.

read_data(main = main, data_name = "rodents")
read_data(main = main, data_name = "rodents_dataset", dataset = "all")
read_data(main = main, data_name = "rodents_dataset", dataset = "controls")
read_data(main = main, data_name = "covariates")
read_data(main = main, data_name = "climate_forecasts")
read_data(main = main, data_name = "newmoons")
read_data(main = main, data_name = "metadata")

read_rodents(main = main)
read_rodents_dataset(main = main)
read_covariates(main = main)
read_climate_forecasts(main = main)
read_newmoons(main = main)
read_metadata(main = main)
read_forecasts_metadata(main = main)

Reading in forecast output to the R console from the directory

Presently, four functions are available for interfacing with saved forecast output.

select_forecasts provides a simple interface to the forecast metadata file with quick filtering:

select_forecasts(main = main, models = "AutoArima")

And read_forecast_tab, read_model_forecast, and read_forecasts_metadata each read in the associated output from a given forecast, as indicated by its forecast_id, which is displayed in the output from select_forecasts:

read_forecast_table(main = main)
read_model_forecast(main = main)
read_forecast_metadata(main = main)

weecology/portalcasting documentation built on Jan. 31, 2024, noon