Classification trees: An alternative non‐parametric approach for predicting species distributions

Abstract

Abstract. The use of Generalized Linear Models (GLM) in vegetation analysis has been advocated to accommodate complex species response curves. This paper investigates the potential advantages of using classification and regression trees (CART), a recursive partitioning method that is free of distributional assumptions. We used multiple logistic regression (a form of GLM) and CART to predict the distribution of three major oak species in California. We compared two types of model: polynomial logistic regression models optimized to account for non‐linearity and factor interactions, and simple CART‐models. Each type of model was developed using learning data sets of 2085 and 410 sample cases, and assessed on test sets containing 2016 and 3691 cases respectively. The responses of the three species to environmental gradients were varied and often non‐homogeneous or context dependent. We tested the methods for predictive accuracy: CART‐models performed significantly better than our polynomial logistic regression models in four of the six cases considered, and as well in the two remaining cases. CART also showed a superior ability to detect factor interactions. Insight gained from CART‐models then helped develop improved parametric models. Although the probabilistic form of logistic regression results is more adapted to test theories about species responses to environmental gradients, we found that CART‐models are intuitive, easy to develop and interpret, and constitute a valuable tool for modeling species distributions.

Spatial prediction of species distribution: an interface between ecological theory and statistical modelling

“…Ejrnaes (2000) suggests that this may be a more important effect than deviations from the symmetric unimodal response (Austin and Gaywood, 1994). Vayssieres et al (2000) also examined the shape of species responses during a comparison of GLM and CART methods using three Oak species. Two of the three species were fitted with models including cubic polynomials and interaction terms.…”

The Relative Importance of Biotic and Abiotic Controls on Young Conifer Growth After Fire in the Klamath-Siskiyou Region

“…Austin et al 1994, Franklin 1998Vayssières et al 2000). In this context, classification recursively partitions an environmental space so as to maximize the homogeneity of the response variable within subspaces.…”

Predicción de la potencialidad de los bosques esclerófilos españoles mediante redes neuronales artificiales

“…Algunos ejemplos se observan en las nuevas tendencias centroeuropeas y norteamericanas en la predicción de hábitat con la asunción previa de cambio climático (Huntley et al, 1995;Sykes et al, 1996;Iverson et al, 1998;Iverson et al, 1999;Parmesan et al, 1999;Pearson et al 2 000;Berry et al, 2001). Los modelos realizados en los últimos años intentan probar, a su vez, nuevas metodologías, por ejemplo, un análisis de componentes principales usado para la predicción de hábitat (Robertson et al, 2001), introduciendo algunas variantes en el ENFA ("Ecological-Niche Factor Analysis") (Hirzel et al, 2001), donde sólo se usan las presencias para predecir las distribuciones potenciales (Hirzel et al, 2002); uso de los datos fitosociológicos bibliográficos en la predicción de la vegetación bajo los efectos del cambio climático (Duckworth et al, 2000); árboles de clasificación para predecir las distribuciones de las especies (Vayssières et al, 2000), y otros que insertan modificaciones en métodos predictivos clási-cos, como el realizado mediante regresiones logís-ticas por Augustin et al, (2001), entre otros. De forma paralela aparecen trabajos cuyo objetivo principal es la comparación entre diferentes técni-cas predictivas (Guisan et al, 1999;Paruelo & Tomasel, 1997;Hilbert & Ostendorf, 2001).…”