Supplement S1. Fig. S1-S12, Tables S1-S3 Fig. S1. Map of ... - TSpace

Supporting information to the paper Hülsmann, L. et al. How to predict tree death from inventory data – lessons from a systematic assessment of European tree mortality models. Canadian Journal of Forest Research.

Supplement S1. Fig. S1-S12, Tables S1-S3 Fig. S1. Map of European tree mortality models and strict forest reserves in Germany and Switzerland. Fig. S2. Boxplot of pbias at the reserve level for each tree species. Fig. S3. Fixed effects of the influence of model and data characteristics on the square-root of |pbias|. Fig. S4. Multiple pairwise comparison of least-squares means and confidence intervals for different species from the linear mixed-effect model of the square-root of |pbias|. Fig. S5. Boxplot of |pbias| at the reserve level achieved by models with and without a covariate of growth for each tree species. Fig. S6. Boxplot of |pbias| at the reserve level achieved by models that were applied inside or outside the ecological zone in which the models were calibrated for each tree species. Fig. S7. |pbias| at the reserve level as a function of the census interval in the validation dataset. Fig. S8. Fixed effects of the influence of model and data characteristics on arcsine-transformed AUC. Fig. S9. Multiple pairwise comparison of least-squares means and confidence intervals for different species from the linear mixed-effect model of arcsine-transformed AUC. Fig. S10. Boxplot of AUC at the reserve level achieved by models with and without a covariate of growth for each tree species. Fig. S11. AUC at the reserve level as a function of the number of records in the validation dataset (Nval). Fig. S12a+b. Observed and predicted annual mortality rates as a function of DBH separately for each model-species combination. Table S1. Number of records per tree species and genus. Table S2. Minimum and maximum values of the tree, stand and site characteristics that were used as covariates in the mortality models. Table S3. Mortality models, related species and the model characteristics used to explain achieved model performance. 1

Supplement S2. Extended material and methods Tree characteristics Stand characteristics Site characteristics Model application Prediction bias Supplement S3. Table of coefficients for the validated mortality models (cf. additional .csv file). References

2

Supplement S1. Fig. S1-S6, Tables S1-S3

Fig. S1. Map of European tree mortality models and strict forest reserves in Germany and Switzerland. The location of the calibration dataset was estimated based on the information available from the publications. Number of reserves per respective validation dataset: Germany n = 22 and Switzerland n = 32.

3

Fig. S2. Boxplot of pbias at the reserve level for each tree species. Prevailing positive or negative pbias values indicate that for the respective species the models used for prediction tend to over- or underestimate tree mortality, respectively.

4

Fig. S3. Fixed effects of the influence of model and data characteristics on the square-root of |pbias|. Note that a ‘good’ model features low |pbias|. Positive and negative influences on performance are shown in blue and red, respectively. Note that the first level of all factors is the reference level, while the other levels are characterized by the shift relative to this reference level. The reference species is Abies.

5

Fig. S4. Multiple pairwise comparison of least-squares means and confidence intervals for different species from the linear mixed-effect model of the square-root of |pbias|. Different letters (a-e) indicate significant differences between species (p