Fig. 6. The structure of the CART model for the Blackthroated Green Warbler. The model recursively partitions the observations into groups (or “terminal nodes”) such that the proportion of total deviance attributable to betweengroup differences is maximized, subject to parsimony constraints. The model covariates are defined in the text. The “main split” is the test “September NDVI < 0.680” which partitions the data into a left subtree where the condition holds, and a right subtree where it does not. The left subtree is partitioned further into three terminal nodes, each with low mean abundance (Mean) and high prediction reliability (Reliability, explained in the text). Within the right subtree, terminal node 7 has the highest mean abundance at 0.938 individuals per standardized pointcount. It is defined as those locations with September NDVI ≥ 0.68, Seasonality < 4.284, and standard deviation of mean total July precipitation < 39.423. The numerical values of the decision criteria, group means, and reliability indices are bootstrap estimates. The covariates are defined in the text.
