Figure 1.
Effect of records’ availability and spatial distribution on model fit.
Effect of records availability and spatial distribution on model fit based on the AUC evaluation of the different algorithms. For the AUC evaluation, we present the back-transformed mean values estimated using Linear Mixed Effect models for each algorithm. The first column presents the results with relation to the number of records and the second with relation to the records distribution.
Table 1.
Results of the Linear Mixed Effect models for the AUC, Kappa, IFK, FGM and DFAC (deviance from average variable contribution).
Figure 2.
Effect of records’ availability and spatial distribution on geographical consistency.
Effect of records availability and spatial distribution on geographical consistency of the different algorithms. For each spatial scale (small scale –Kappa; medium scale – IFK; and large scale - FGM), we present the back-transformed mean values estimated using Linear Mixed Effect models for each algorithm. The first column presents the results with relation with the number of records and the second with relation with the records distribution. For clarity of comparisons, ANN results are presented separately whenever its values were much lower than those obtained for other algorithms. See Tables 1 and S5 for further statistical information.
Figure 3.
Consistency of the variables’ contribution to the model.
Variability of the contribution of each environmental variable (i.e. deviance from the average variable contribution to the model) for each algorithm. In the Y axis higher deviance represents a lower consistency in the contribution values given by the algorithm to the different variables across runs. The values for variable “B04” in the ANN algorithm go to 80% and other variables present outliers going beyond the 40%, however, for plotting convenience we show only the deviance up to the 40%. See Table 1 and Tables S8 and S9 for further statistical information.
Table 2.
Summary of the algorithms’ performance across analyses and the different aims for which they attain better results (for more details see Figs. 1, 2, 3).
Figure 4.
Framework for analysing the algorithms adequacy for modelling our species distribution by means of model fit, binary predictions similarity and selection of variables importance. These results are analysed across algorithms by means of Linear Mixed Effects models (LME), which will aid in the selection of the most suitable algorithm for modelling our species distributions.