Training BRIN models

With this module you can fit BRIN models on temperatures data and registered phenological dates as training set. For example below we train a BRIN model on Montpellier temperatures and bud burst dates of Chasselas variety registered at the Domaine de Vassal (43.327676, 3.565170). The file Data_BB.xlsx is available in the github repository.

using Phenology

#TN (daily minimal) and TX (daily maximal) temperatures in Montpellier, from 1946 to 2022.
date_vec, x = Common_indexes(joinpath(stations_folder, "TN_Montpellier.txt"), joinpath(stations_folder, "TX_Montpellier.txt"))

#Importing bud burst data.
df_vassal_chasselas = @chain begin
    DataFrame(XLSX.readtable(joinpath(@__DIR__, "Data_BB.xlsx"), "data_Tempo"))
    @subset(:variete_cultivar_ou_provenance .== "Chasselas")
    @subset(:nom_du_site .== "INRA Domaine de Vassal")
end
date_vecBB = Date.(df_vassal_chasselas.date)

#Fitting model
model = BRIN_Model(date_vec, x, date_vecBB)

You can also put the years and the days of the year doy in the example as an integer (the number of days between the 1ˢᵗ of January and the date) of the bud burst in two separate vectors :

years = df_vassal_chasselas.annee
doy = df_vassal_chasselas.jour_de_l_annee
model = BRIN_Model(date_vec, x, years, doy)

The method BRIN_Model(x_vec::AbstractVector, n_train::AbstractVector) takes as arguments two vectors of same length, for which each element is associated to a year :

• x_vec is a vector of series of yearly temperatures between the 1ˢᵗ of August of the precedent year and 30ᵗʰ of July of the current year.
• n_train is a vector of numbers of days between the 1ˢᵗ of August of the precedent year associated and the bud burst date.

Example :

n_train = [doy[i] + length(Date(year - 1, 8, 1):Date(year - 1, 12, 31)) for (i,year) in enumerate(years)] #doy + nb of days between 1ˢᵗ of August and the last day of the year
x_vec = [x[Date(year - 1, 8, 1) .<= date_vec .< Date(year, 8, 1), :] for year in years]
model = BRIN_Model(x_vec, n_train)