Overcoming data sparseness and parametric constraints in modeling of tree mortality: a new nonparametric Bayesian model

Accurately describing patterns of tree mortality is central to understanding forest dynamics and is important for both management and ecological inference. However, for many tree species, annual survival of most individuals is high, so that mortality is rare and, therefore, difficult to estimate. Furthermore, tree mortality models have potentially complex suites of covariates. Here, we extend traditional and recent approaches to modeling tree mortality and propose a new nonparametric Bayesian method. Our model is constrained to both reflect and distinguish known relationships between mortality and its two key covariates, diameter and diameter increment growth, but it remains sufficiently flexible to capture a wide variety of patterns of mortality across these covariates. Our model also allows incorporation of outside information in the form of priors, so that increased mortality of large trees can always be formally modeled even when data are sparse. We present results for our nonparametric Bayesian mortality model for maple (Acer spp.), holly (Ilex spp.), sweet gum (Liquidambar styraciflua L.), and tulip-poplar (Liriodendron tulipifera L.) populations from North Carolina, USA.