A data mining approach to predictive vegetation mapping using probabilistic graphical models

摘要: Abstract This paper develops a novel method to model and predict the spatial distribution of vegetation types in Swaziland using physiographic bioclimatic variables. The uses data mining approach implemented within probabilistic graphical match two observed hierarchical levels vegetation. classification Bayesian networks (BN) parameterization is based on expectation-maximization (EM) algorithm. tested random sample mapped allowed for identification key environmental variables that are most important capturing distribution. We show while elevation geology explaining patterns both models, influence climatic other at differ. overall predicted classes was very similar their map. Overall error rate found be 9.35% 16 4.9% one with 5 classes, indicating excellent accuracy despite complex landscape study area. Possible sources some limitations discussed conclusions drawn including suggestions further investigation.