Figure 6
Influence of remaining variables on likelihood of macrophyte presence. Likelihood values were calculated as predictions in which one variable’s value gradually increased, while the remaining variables’ values were fixed at their median value. Random forests were trained with data of which 18% was missing, without outlier or correlated variable removal and followed by median imputation, while applying 5-fold cross-validation. In total, 10 repetitions were performed and likelihood values were averaged (black lines), with grey zones representing the standard deviation over these 10 repetitions. The number of individual trees was equal to 100 for each macrophyte.
