| Abstrakt: |
Urban forest inventories are conducted to obtain important information regarding trees and the ecosystem services they provide in the urban environment. This information is derived from a sample of ground plots within which all trees are measured. It is common that some of the selected sample plots are unable to be measured, primarily due to lack of permission to access privately owned lands. At a minimum, nonresponse results in decreased sample sizes and eroded confidence in information derived from the inventory. Further, concern is raised regarding sample bias and its associated effects on analytical outcomes when the nonresponse cannot be considered random with respect to population characteristics. In this study, data from urban forest inventories conducted in 33 cities across the U.S. were used to assess amounts of nonresponse present. Total amounts of nonresponse ranged from 1.7% in Madison, WI to 36.8% in Bridgeport, CT. Examinations of potential nonrandom occurrence of nonresponse were conducted using plot location information in conjunction with various digital map layers and notable trends were found in relation to median income, median age, percent canopy cover, percent residential ownership, and percent impervious surface. In contrast, there appeared to be little correlation between nonresponse and percent of English-only speaking households. As a caveat to overly broad interpretation, in Houston, TX it was shown there was no correlation between nonresponse and median income and the relationship with percent canopy cover was opposing that of the general trend. The magnitude of potential nonresponse bias in estimates of tree biomass and tree frequency was investigated by substituting predicted values for nonresponse plots, with underestimation of both attributes ranging from approximately 1% to 10%. Thus, urban forest inventory practitioners should evaluate city-specific circumstances to effectively mitigate potential bias resulting from nonresponse in the sample. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink