Effects on the accuracy of orchid SDMs
The AUC values under all modeling strategies exceeded 0.8, with both
Kappa and TSS values exceeding 0.4, indicating that our constructed
models outperformed the random model and performed well in the
prediction orchids habitat (see Appendix S2.1 for details of the model
accuracy results).
We compared the accuracy of the orchid lifeforms classification models
constructed based on physiological traits with the models generated by
all-data, obtaining 27 groups of comparative data under different
approaches for modeling and verifying, of which 15 groups (55.6%)
showed that the accuracy of the classification modeling was higher than
that of the all-data models. However, we also observed that in the RF
models, only the R-t model had higher Kappa values than the all-data
model, while in the other comparison groups, the accuracy showed equal
or slightly lower values (ranging from 0.001-0.09). When we temporarily
discounted the kind of models, the proportional improvement of model
accuracy by orchids lifeforms classification was able to reach more than
70%. We performed a T-test on these differences and found general
significant differences between classification modeling and all-data
modeling (see Fig. 1). The situation was quite distinct when we examined
the effect of human influences on the accuracy of the models. Based on
the constructed models, we created 32 groups of comparative data, 23 of
which showed (71.9%) that the accuracy of the orchid model with the
inclusion of the HI factor was higher than that of the model without
this variable, and in those contrasts groups, the maximum difference did
not exceed 0.04. However, we performed T-test in these comparisons, and
significance was present in only two groups (M- t and G-all), which did
not appear to be a generalization seemingly (see Fig. 2).