Measuring urban tree loss dynamics across residential landscapes

Abstract

The spatial arrangement of urban vegetation depends on urban morphology and socio-economic settings. Urban vegetation changes over time because of human management. Urban trees are removed due to hazard prevention or aesthetic preferences. Previous research attributed tree loss to decreases in canopy cover. However, this provides little information about location and structural characteristics of trees lost, as well as environmental and social factors affecting tree loss dynamics. This is particularly relevant in residential landscapes where access to residential parcels for field surveys is limited. We tested whether multi-temporal airborne LiDAR and multi-spectral imagery collected at a 5-year interval can be used to investigate urban tree loss dynamics across residential landscapes in Denver, CO and Milwaukee, WI, covering 400,705 residential parcels in 444 census tracts. Position and stem height of trees lost were extracted from canopy height models calculated as the difference between final (year 5) and initial (year 0) vegetation height derived from LiDAR. Multivariate regression models were used to predict number and height of tree stems lost in residential parcels in each census tract based on urban morphological and socio-economic variables. A total of 28,427 stems were lost from residential parcels in Denver and Milwaukee over 5years. Overall, 7% of residential parcels lost one stem, averaging 90.87 stems per km². Average stem height was 10.16m, though trees lost in Denver were taller compared to Milwaukee. The number of stems lost was higher in neighborhoods with higher canopy cover and developed before the 1970s. However, socio-economic characteristics had little effect on tree loss dynamics. The study provides a simple method for measuring urban tree loss dynamics within and across entire cities, and represents a further step toward high resolution assessments of the three-dimensional change of urban vegetation at large spatial scales.

Keywords: Multi-temporal LiDAR; Remote sensing; Socio-ecological systems; Urban ecology; Urban forestry; Vegetation dynamics.

Figure 1:

Study areas in Denver, CO (A) and Milwaukee, WI (B). Light gray polygons represent the census tracts considered in the study, bold lines are the counties’ perimeters.

Figure 2:

Example of tree stems lost within residential parcels in Denver, CO detected from the difference between the canopy height models (-ΔCHMs) calculated for the years 2008 and 2013 (A). Satellite images are from Google Earth™ and taken in July 2007 (B) and June 2014 (C), before and after the 5-year reference periods. Each dot represents the position of a tree stem lost, whereas the its radius is proportional to stem height. The complete tree datasets for Denver and Milwaukee are available as supplementary material.

Figure 3:

Stem height frequency distribution for all the trees (height >5 m) lost within residential parcels in Denver, CO and Milwaukee, WI in a 5-year period.

Figure 4:

Bivariate relationship between the total number of tree stems lost per unit of residential area and A) the average woody vegetation volume, and B) woody vegetation cover within residential parcels in the initial year (i.e. 2008 and 2010 for Denver and Milwaukee, respectively). Dots represent parcel-based values averaged at census tract level.

Figure 5:

Average number of tree stems lost (A) within a 5-year period per unit of residential area and their average height (B) in relation to the decade of maximum housing development of neighborhood (i.e. census tract) in Denver, CO and Milwaukee, WI. Numbers above bars represent the number of census tracts analyzed, errors bars represent standard errors of the mean.