Woody plant mortality algorithms: description, problems and progress

Abstract

Mortality as a basic process should be an integral part of any model of woody plant dynamics. But progress in developing mortality algorithms is slow. The aim of this paper is to provide a state-of-the-art description of mortality algorithms in published plant growth models, their structure, and the problems they face. The subsequent objective is to assess how models attempt to solve the problems, and provide some insight on how modelling mortality can be improved. Sixty-one models, concerned with forest yield and dieback and forest/shrubland dynamics, were reviewed, the greatest proportion being the gap type. The algorithms could be broadly divided into stochastic and deterministic. Despite considerable variability, the basic premise of every algorithm was the same: beyond a specified threshold a plant either dies (deterministic), or has a greater probability of dying (stochastic). Factors used to predict mortality include size, age, competition, carbon balance (growth) and abiotic influences. Some algorithms were random, and one-third used more than one predictor. Although the classification is far from clear-cut, mortality algorithms can be placed in one or more of the following classes: carbon-based, abiotic/age-based, competitive, gap-type, statistically fitted and progressive-stress type. It is clear both that mortality is difficult to model, and that there is no best way to model it for all applications. The complexity of environmental stresses and lack of process information of woody plant mortality has however led to the wide use of empirical algorithms. These have many problems, including data intensity and spatial and temporal specificity. Such problems render mortality simulation prone to error and weak in response to environmental change. Test data and procedures also are lacking. In the literature suggested improvements to the problem include the use of mechanistic algorithms. But this approach also has associated problems. So here it is suggested here that the use of a biologically reasonable predictor in a simple stochastic algorithm, used within a mechanistic model of plant growth, is a preferable improvement. The approach provides a mechanistic-empirical consensus. A shift in emphasis towards modelling for exploration and explanation, rather than to predict with high levels of accuracy is also suggested.