Effect of Plot and Sample Size on Timing and Precision of Urban Forest Assessments David J. Nowak, Jeffrey T. Walton, Jack C. Stevens, Daniel E. Crane, and Robert E. Hoehn Abstract. Accurate field data can be used to assess ecosystem services from trees and to improve urban forest management, yet little is known about the optimization of field data collection in the urban environment. Various field and Geographic Information System (GIS) tests were performed to help understand how time costs and precision of tree population estimates change with varying plot and sample sizes in urban areas using random sampling approaches. Using one-tenth acre (0.04 ha) plots, it is estimated that, on average, approximately three plots per day can be measured with plot data collected on several variables for all trees greater than 1 in (2.54 cm) in diameter along with general plot, ground cover, and shrub data. A field crew of two people can gather approximately 200 one-tenth acre (0.04 ha) plots during a 14 week summer field season depending on city traffic, city area, and tree cover conditions. These 200 plots typically yield approximately a 12% relative standard error on the total number of trees. Key Words. Tree measurement urban forest monitoring urban forest sampling. Measuring the urban forest structure (i.e., species composi- tion, number of trees, tree sizes and locations, tree health) can give managers and planners a basis with which to develop and evaluate programs for managing urban trees and forests throughout a city. In addition, long-term monitoring of urban forest structure can provide essential data related to rates and factors of change affecting population totals, tree mortality, tree planting and natural regeneration, tree health, and species changes. An accurate quantification of urban forest structure is also needed to assess the various ecosystem services and values pro- vided by the urban forest. Urban vegetation, particularly trees, provides numerous benefits that can improve environmental quality and human health in and around urban areas. These ben- efits include improvements in air and water quality, building energy conservation, cooler air temperatures, reductions in ul- traviolet radiation, and many other environmental and social benefits (Nowak and Dwyer 2007). By having accurate infor- mation on urban forest structure, managers can understand what the current urban forest provides in terms of various en- vironmental benefits and also alter the structure of the urban forest (e.g., tree plantings, species and site selections, and tree maintenance and removals) to enhance these benefits in the fu- ture. One of the best ways to assess the entire urban forest is through sampling procedures. However, varying sample and plot sizes affect total cost (time) of data collection and the precision of the urban forest estimate. The purpose of this ar- ticle is to illustrate, based on field data collection tests, how plot and sample size of randomly located circular plots in urban areas can affect data collection time, number of permissions needed to access plots, and precision of tree cover and total tree population estimates. These types of data have been lack- ing related to urban forest sampling and can be useful in devel- oping sampling schemes to help provide desired precision of estimates and understand the costs associated with obtaining that precision. METHODS Effect of Plot Size on Data Collection Time and Total Population Estimate Precision To estimate the effect of plot size on time needed to collect field data and on total population estimates, a random sample of 26 residential plots (from a total of 100 residential plots that were measured and analyzed using the Urban Forest Effects [UFORE] model in Syracuse, NY, U.S. [Nowak and Crane 2000 Nowak and O���Connor 2001]) were measured and timed using a field crew of two people. Crews were trained before field data col- lection and were experienced in urban forest field data collec- tion. For each plot, permission was obtained from the lot owner (where the plot center was located) by knocking on the front door of the lot residence. If the plot encompassed more than one lot, additional lot owners were contacted for permission if trees in those additional lots were located within the plot boundary. On each plot, all UFORE variables (i-Tree 2007) were col- lected on concentric one-twenty-fourth acre (24 ft radius circle), one-tenth (37.2 ft radius), and one-sixth acre plots (48.1 ft ra- dius) (0.0168 ha [7.3 m radius], 0.04 ha [11.3 m radius], and 0.067 ha [14.7 m radius] plots, respectively). These variables include several tree variables (e.g., species, diameter at breast height, crown, and health parameters) on all trees greater than 1 in (2.54 cm) in diameter at breast height (4.5 ft [1.37 m]) and general plot information (e.g., location, plot center, tree and shrub cover), ground cover types, and general shrub types and dimensions. Electronic distance measuring devices were used to record trees distances from plot center and tree heights. Data collection also included measures of general plot slope and as- pect. Data collection was cumulatively timed moving from the smallest to largest plot and number of access permissions needed was recorded. Average measurement time, number of lots ac- cessed, and number of trees along with associated standard errors were assessed for each plot design. In addition, an estimated total number of trees in the residential area was calculated and com- pared with an estimate using 100 one-tenth acre (0.04 ha) plots 386 Nowak et al.: Effect of Plot and Sample Size on Timing and Precision Arboriculture & Urban Forestry 2008. 34(6):386���390. ��2008 International Society of Arboriculture

to illustrate how plot size affects the total tree and standard error estimate. Average plot time for field plot setup, cover estimates, and measurements per tree were used to estimate how average field measurement time would likely vary as tree cover changes. In a separate analysis, an additional test of plot size and plot design was conducted using GIS tree cover, land use, and parcel data for the city of Syracuse. Five hundred points were randomly distributed throughout the city. At each point, the following seven different plot sizes or designs were constructed around the point using GIS: 1) one-twenty-fourth acre (0.017 ha) circular plot 2) one-twelfth acre (0.034 ha) circular plot 3) one-tenth (0.04 ha) circular plot 4) one-eight acre (0.05 ha) circular plot 5) one-sixth acre (0.067 ha) circular plot 6) one-fourth acre (0.1 ha) circular plot and 7) four one-twenty-fourth acre (0.017 ha) circular plots (cluster plot) using the USDA Forest Service For- est Inventory and Analysis (FIA) plot design (USDA Forest Service 2000). With this cluster plot design, three subplots were established 120 ft (36.6 m) from the center subplot at 120��, 240��, and 360�� azimuths. For each of the plot sizes and designs, total amount of tree cover within the plot was assessed using a 2 ft (0.61 m) resolu- tion tree cover map (Myeong et al. 2003), and the number of parcels and associated number and area of land uses in each parcel within the plot design was recorded using a digital land use parcel map. The average amount of permissions required for each plot design was categorized among three classes: 1) per- mission required (residential land use parcels) 2) permission questionable���uncertain if crew would need to obtain permis- sion (commercial/industrial, institutional, utility/transportation parcels) and 3) no permission needed (greenspace, street right- of-ways, and vacant parcels) to assess how permissions would vary based on plot size and design. The average percent of plot area within the parcel that contained the plot center was also calculated. This calculation was done to help determine how much of the plot area would require the crew to move to an additional parcel and how much of that extra plot space would require additional permissions. Mean tree cover and standard error for each plot design were calculated and compared with the actual tree cover as classified by the tree cover map. Effect of Sample Size on Total Population Estimate Precision To determine the effect of sample size on the standard error estimate for the total tree population, sample data from 14 cities were analyzed using the UFORE model (Nowak and Crane 2000 Nowak et al. 2002) (Table 1). For each city, population total, standard error (SE), and relative SE were calculated. The relative SE is a measure of estimated reliability and is the ratio of SE to the estimate, in this case, population total (SE/total �� 100) (US Department of Health and Human Services, Centers for Disease Control and Prevention 2007). Eleven of the cities were sampled using a stratified random sampling approach, and three using a randomized grid approach, which was used to facilitate long-term monitoring of urban forest change. Standard error for each city was standardized to a population size of 200 plots using the formula: SE standard deviation/���n. The average SE using 200 plots was calculated for the 14 cities and used to illustrate how SE of the total tree population estimate will vary as sample size varies between 10 and 500 plots. RESULTS Effect of Plot Size on Data Collection Time and Total Population Estimate Precision Increasing plot size from a one-twenty-fourth acre (0.017 ha) plot to a one-sixth acre (0.067 ha) plot nearly doubled the amount of time needed to measure the plot variables, but also nearly cut in half the relative standard error for the total popu- Table 1. Estimates of total number of trees and standard errors from 14 cities analyzed using the UFORE model.z City Number of trees Year No. plots 200 ploty Samplex Total SE SE RSE Atlanta, GAw 9,415,000 749,000 1997 205 758,000 8.1 Str. random Baltimore, MDv 2,571,000 494,000 2004 200 494,000 19.2 Str. random Boston, MAw 1,183,000 109,000 1996 217 114,000 9.6 Str. random Freehold, NJu 48,000 6,000 1998 144 5,000 10.1 Str. random Jersey City, NJu 136,000 22,000 1998 220 23,000 16.7 Str. random Minneapolis, MNt 979,000 165,000 2004 110 122,000 12.5 Random grid Moorestown, NJu 583,000 53,000 2000 206 54,000 9.3 Str. random Morgantown, WVs 658,000 79,000 2004 136 65,000 9.9 Str. random New York, NYw 5,212,000 719,000 1996 206 729,000 14.0 Str. random Philadelphia, PAw 2,113,000 211,000 1996 210 216,000 10.2 Str. random San Francisco, CAr 668,000 98,000 2004 194 97,000 14.5 Random grid Syracuse, NYv 876,000 119,000 2001 197 119,000 13.5 Str. random Washington DCq 1,928,000 224,000 2004 201 224,000 11.6 Random grid Woodbridge, NJu 986,000 97,000 2000 215 100,000 10.2 Str. random zAverage relative standard error 12.1%. yEstimated standard error (SE) and relative standard error (SE/total �� 100 RSE) using a sample of 200 one-tenth acre (0.04 ha) plots. xStr. random stratified random sample random grid randomized grid sample. wData collection by ACRT, Inc. vData collection by U.S. Forest Service. uData collection by New Jersey Department of Environmental Protection. tData collection by Davey Resource Group. sData collection by West Virginia University. rData collection by city personnel. qData collection by Casey Trees Endowment Fund. Arboriculture & Urban Forestry 34(6): November 2008 387 ��2008 International Society of Arboriculture