larscaspersen/phenoflex_toyproblem

An example of the current fitting routine using PhenoFlex

This is an example with flowering data of apple cultivar “Blanquina” for the period 1987 - 1997 and 2004 - 2020 in Asturias, Spain. Hourly weather data were generated from daily temperature observations using the chillR function stack_hourly_temps(). Phenology and temperature data can be found in the “data” folder.

In the “code”, folder you can find a script called “example_fit.R”. It contains the usual routine we use to prepare the data and then carry out the first (out of usually more than 10) round of fitting using the function phenologyFitter() from the chillR package. The function requires several arguments:

• par.guess contains the initial guess for the parameters. The order should be (yc, zc, s1, Tu, E0, E1, A0, A1, Tf, Tc, Tb, slope).
• modelfn is the wrapper function for the phenoflex model. It can also be customised, so that some parameters are for example kept constant.
• bloomJDays is a vector of the Julian days of the flowering.
• SeasonList is a list with the hourly temperature data organized in the same order as the provided phenology data.
• lower contains the lower bounds of the model parameters.
• upper contains the upper bounds of the model parameters.
• seed to make results reproducible.
• control is a list of further variables directly passed to the simulation annealing function currently used in phenoflex (GenSA::GenSA()). Most interesting is “maxit” which defines the maximum number of iterations. “max.time” could be also of interest, because it specifies the time the optimizer should run in seconds, which can be used instead of “max.it”

The fitting result is checked on RMSE. The temperature response curve resulting from the new set of parameters is inspected using some helper functions, which you can find in the file “helper_functions.R”.