Propagate parameter uncertainty of TPC fits using bootstrap with residual resampling

Usage

predict_curves(
  temp = NULL,
  dev_rate = NULL,
  fitted_parameters = NULL,
  model_name_2boot = NULL,
  propagate_uncertainty = TRUE,
  n_boots_samples = 100
)

Source

The dataset used in the example was originally published in Satar & Yokomi (2022) under the CC-BY-NC license

Arguments

temp: a vector of temperatures used in the experiment. It should have at least four different temperatures and must contain only numbers without any missing values.
dev_rate: a vector of estimated development rates corresponding to each temperature. These rates are calculated as the inverse of the number of days to complete the transition from the beginning of a certain life stage to the beginning of the following at each temperature. It must be numeric and of the same length as temp.
fitted_parameters: a tibble obtained with fit_devmodels() function, including parameter names, estimates, standard errors, AICs, and objects (fitted_models) using the nls.multstart::nls_multstart() approach.
model_name_2boot: A vector of strings including one or several TPC models fitted by fit_devmodels(). Contrarily to that function, model_name_2boot = "all" is not allowed in this function due to the slow bootstrapping procedure. We recommend applying this function only to a small pre-selection of models (e.g., one to four) based on statistical and ecological criteria with the help of plot_devmodels() function.
propagate_uncertainty: A logical argument that specifies whether to propagate parameter uncertainty by bootstrap with residual resampling. If FALSE, the function returns predictions from the fitted TPCs for the selected model(s). If TRUE, bootstrap is applied using residual resampling to obtain multiple TPCs as detailed in vignettes of the rTPC package. Defaults to TRUE.
n_boots_samples: Number of bootstrap resampling iterations (default is 100). A larger number of iterations makes the resampling procedure more robust, but typically 100 is sufficient for propagating parameter uncertainty, as increasing n_boots_samples will increase computation time for predicting resampled TPCs.

Value

A tibble object with as many curves (TPCs) as the number of iterations provided in the n_boots_samples argument if propagate_uncertainty = TRUE minus the bootstrap samples that could not be fitted (i.e., new nonlinear regression models did not converge for them). Otherwise, it returns just one prediction TPC from model fit estimates. Each resampled TPC consists of a collection of predictions for a set of temperatures from temp - 20 to temp + 15 with a resolution of 0.1°C and a unique identifier called boots_iter. In addition to the uncertainty TPCs, the estimated TPC is also explicit in the output tibble.

References

Angilletta, M.J., (2006). Estimating and comparing thermal performance curves. J. Therm. Biol. 31: 541-545. (for model selection in TPC framework)

Padfield, D., O'Sullivan, H. and Pawar, S. (2021). rTPC and nls.multstart: A new pipeline to fit thermal performance curves in R. Methods Ecol Evol. 12: 1138-1143.

Rebaudo, F., Struelens, Q. and Dangles, O. (2018). Modelling temperature-dependent development rate and phenology in arthropods: The devRate package for R. Methods Ecol Evol. 9: 1144-1150.

Satar, S. and Yokomi, R. (2002). Effect of temperature and host on development of Brachycaudus schwartzi (Homoptera: Aphididae). Ann. Entomol. Soc. Am. 95: 597-602.

Examples

if (FALSE) { # interactive()
data("aphid")

fitted_tpcs <- fit_devmodels(temp = aphid$temperature,
                             dev_rate = aphid$rate_value,
                             model_name = "all")

plot_devmodels(temp = aphid$temperature,
               dev_rate = aphid$rate_value,
               fitted_parameters = fitted_tpcs,
               species = "Brachycaudus schwartzi",
               life_stage = "Nymphs")

# Obtain prediction TPCs with bootstraps for propagating uncertainty:
boot_tpcs <- predict_curves(temp = aphid$temperature,
                            dev_rate = aphid$rate_value,
                            fitted_parameters = fitted_tpcs,
                            model_name_2boot = c("lactin2", "briere2", "beta"),
                            propagate_uncertainty = TRUE,
                            n_boots_samples = 10)

boot_tpcs
}