Multi-stemming and size enhance survival of dominant tree species in a frequently typhoon-disturbed forest

Questions Quantifying tree species persistence through recurrent disturbances is of crucial importance for understanding forest dynamics in typhoon-prone regions. We ask the following: (a) What are the major determinants of dominant tree survival in frequently typhoon-disturbed forests? (b) Are survival determinants different between small and large trees? Location A subtropical old-growth forest located in Fushan, Taiwan (24 degrees 45 ' 34 '' N, 121 degrees 33 ' 58 '' E), with frequent typhoon disturbances. Methods Data were from three consecutive censuses of a 25-ha permanent forest plot that censused trees >= 1 cm in diameter every five years. The survival of three dominant tree species was modeled using generalized additive model and boosted trees with abiotic and biotic predictors. We evaluated model performance using validation data obtained from the two available census intervals. Results Model validations showed that multi-stemming and tree size enhanced the survival of large and small trees, respectively. For the most dominant species, multi-stemming had a consistently positive effect on survival irrespective of diameter classes. Abiotic factors and conspecific density had little effect on tree survival. Furthermore, evaluating model performance based on the data used in the model construction (i.e., training data) overestimated the predictive ability of survival models. Conclusions We showed that the survival determinants for the three most dominant species at Fushan changed from tree size for small trees to multi-stemming for large trees. The results suggest that the dominant species in this frequently typhoon-disturbed forest have the stature and architectural traits to persist, and thereby maintain their dominance and shape the forest physiognomy. Our approach illustrates how datasets from different census periods can be used in model validation to better assess model performance.