Stochastic models for conifer tree crown profiles

We perform a feasibility study of using stochastic models to describe the profile of tree crowns and to capture the stochastic nature of the tree crown form for five conifer species of the Sierra Nevada. In 70% of the cases investigated we found that we could model tree crown profiles as a quadratic or cubic trend in conjunction with a simple autoregressive moving average model (ARMA). In the remaining cases we used a quadratic or cubic trend in conjunction with white noise. These stochastic ARMA models are visually and statistically an improvement over using Euclidean geometric crown profile models. Competing models were judged by using Akaike's Information Criterion (AIC) to achieve a parsimonious model. It was found that first-order moving average MA(1) models or first-order autoregressive AR(1) models were adequate for modeling the majority of the cases studied and that these models were qualitatively similar. MA(1) models were preferred over the AR(1) models because less information is required to simulate them.