Predictors of mortality for juvenile trees in a residential urban-to-rural cohort in Worcester, MA
This paper explores predictors of juvenile tree mortality in a newly planted cohort in Worcester, MA, following an episode of large-scale tree removal necessitated by an Asian Longhorned Beetle (Anoplophora glabripennis, ALB) eradication program. Trees are increasingly seen as important providers of ecosystem services for urban areas, including: climate moderation and thus reduction in heating/cooling costs; air and water filtration; carbon uptake and storage; storm water runoff control; and cultural and aesthetic values. Many cities have initiated tree planting programs to receive these benefits, typically seeking to complement existing urban forest. Conversely, Worcester's reforestation program was necessary to offset the loss of approximately 30,000 trees removed to eradicate the invasive pest ALB. Since then, more than 30,000 juvenile trees have been planted to offset the loss, creating the opportunity to study a highly dynamic urban forest. Tree planting effectiveness is contingent on high survivorship rates, particularly during the establishment phase during the first five years after planting. Using a large data set including biophysical and sociodemographic variables, this research uses Conditional Inference Trees (CIT), a machine learning technique, to explore predictors of mortality. The most important variables as determined by CIT were used to create a logistic regression to predict mortality. This analysis was run for all trees, and for several subsets of the sample based on tree type and season and year of planting, yielding twenty individual models. Results indicated that the following variables are important predictors of mortality during establishment, in descending order: adjacent home/building age, proportion renter occupancy, days since tree planted, tax parcel size, number of trees planted on property, and tax parcel value. Of these variables, proportion renter occupancy and days since tree planted were most frequently found to be significant in the logistic regression modeling.