Predicting tree mortality from growth data: how virtual ecologists can help real ecologists

Authors


*Correspondence author. E-mail: bjoern.reineking@env.ethz.ch.

Summary

  • 1Tree growth and mortality are key elements of forest dynamics, and thus are of great concern for forest managers. It is widely accepted that tree mortality can be predicted using tree growth data. Several approaches have been proposed for modelling the growth-mortality relationship, differing in terms of data sources and model flexibility. However, little is known about their ability to reliably reconstruct the shape of the real growth-mortality relationship due to a lack of long-term data.
  • 2We adopted a ‘virtual ecology’ approach to this problem, simulating forests with either of two a priori specified growth-mortality relationships. Different sampling regimes in these virtual forests resulted in virtual tree-ring data, forest inventory data, or a combination of both. We used eight existing or newly developed models of different structural flexibility to analyse the growth-mortality relationship. The accuracy of the different model outputs, i.e. the deviation from the a priori specified growth-mortality relationships, was quantified with the Kullback-Leibler distance.
  • 3For all data sources, reliable growth-mortality models could be identified. The highest accuracies were found for tree-ring based models, which require only a small sample size (60 dead trees). High model accuracies were also found for forest inventory based models, starting at sample sizes of 500 trees.
  • 4Flexible statistical approaches turned out to be superior to less flexible models only for large sample sizes (totally 2000 trees). The additional use of Bayesian statistics, specifically designed for small sample sizes, led to high model accuracies only when model flexibility was constrained.
  • 5Synthesis. Our study shows that simulated experiments are a powerful tool for selecting reliable approaches to analyse ecological processes such as tree mortality. Reliable models are fundamental for gaining novel ecological insights into the growth-mortality relationship of tree species. The use of more accurate growth-mortality relationships in forest succession models would allow for strongly improved projections of past and future forest dynamics. Our study provides the theoretical basis for a sound estimation of such growth-mortality models, and it also provides guidelines for efficient sampling schemes in real forests.

Ancillary