Fig. 7

a The Mean Squared Error loss observed during Neural Network cross-validation, for both the train and test set at each epoch, where maximum epoch size is 1500 and b the Neural Network architecture of the best-fit model between the environmental variables (Relative Humidity-RH, Temperature-T, Vapour Pressure Deficit –VPD, Solar Radiation-RAD and Wind Speed-WS) and the average TR time series of cluster C1 (Y1), C2 (Y2) and C3 (Y3). The thicker lines imply variables with greater impact on the TR values than the thinner ones, and black lines indicate positive influence while grey imply negative effect. The neurons in the input, hidden and output layers are annotated with I, H and O, respectively. The bias neurons are B1 and B2, and can be considered similar to intercepts in linear regression models