|
388 Nowak et al.: Effect of Plot and Sample Size on Timing and Precision
<br />lation estimate (Table 2). Average time per plot increased from
<br />approximately 62 min (SE = 7.4) for a one -twenty-fourth acre
<br />plot (0.017 ha) to 106 min (SE = 14.0) for a one -sixth acre
<br />(0.067 ha) plot. Number of permissions (lots) also increased from
<br />an average 1.9 (SE = 0.1) to 3.1 (SE = 0.2), and number of trees
<br />measured per plot increased from 2.6 (SE = 1.1) to 6.5 (SE =
<br />1.5). All three plot sizes produced total population estimates with
<br />a sampling error within 1 SE of the estimated population total of
<br />251,000 trees, but as plot size increased, the total estimate moved
<br />closer to the 251,000 estimate and SE decreased (Table 2). The
<br />trend of the overall estimate decreasing with plot size (Table 2)
<br />suggests that the sample size was not large enough for the two
<br />smaller plot sizes. The effect of increasing the number of plots
<br />for the smaller plots sizes such that the total sample area remains
<br />the same among all plots sizes remains to be investigated.
<br />A similar pattern occurred when accessing tree cover from
<br />digital maps using plot sizes that ranged from one -twenty-fourth
<br />acre (0.017 ha) to one-fourth acre (0.1 ha), including an FIA
<br />cluster plot. Number of permissions increased and percent of plot
<br />in parcel with plot center decreased as plot size increased (Table
<br />3). The one-fourth acre (0.1 ha) plot produced the closest esti-
<br />mate of actual tree cover value and had the lowest SE and rela-
<br />tive SE. The one -twenty-fourth acre (0.017 ha) plot produced the
<br />estimate farthest from the actual tree cover value, although it was
<br />still within 1 SE from the true mean and had the highest SE and
<br />relative SE. The FIA cluster design, which is being used in the
<br />urban forest health monitoring program (Cumming et al. 2008),
<br />produced estimates of tree cover with a slightly higher SE
<br />and relative SE than a one -sixth acre (0.067 ha) single plot design.
<br />The FIA plot design also required nearly double the permissions
<br />of the one -sixth acre (0.067 ha) plot design (Table 3).
<br />Average time to set up a residential plot (e.g., gain permission
<br />and establish plot center) was 15.6 min (SE = 1.9); average time
<br />needed to estimate cover types was 12.8 min per plot (SE =
<br />1.1). Thus, the average fixed time per plot was approximately 30
<br />min. The average time to record all measurements on one tree
<br />was 12.2 min (SE = 0.9).
<br />Effect of Sample Size on Total Population
<br />Estimate Precision
<br />The relative standard error (RSE) of total number of trees drops
<br />significantly with the first 50 to 100 one -tenth acre (0.04 ha)
<br />plots established, from 54.1% RSE with 10 plots to 17.1% RSE
<br />at 100 plots. After approximately 100 plots, the RSE continues to
<br />drop, but a reduced rate per additional plot (Figure 1). The av-
<br />erage RSE for 200 plots is 12.1% (Table 1).
<br />DISCUSSION
<br />The key to assessing urban forests is to determine the optimal
<br />number of plots and plot size needed to gain the desired preci-
<br />sion of an estimate at minimal cost. Unfortunately, there is not
<br />much information in the literature on costs of urban field plots
<br />and structural variability across the urban forest. A general rule
<br />of sampling is increasing the plot size and number of plots tends
<br />to increase precision, but at increased cost. Data presented in this
<br />article begin to reveal the increases in precision and time costs
<br />associated with different sample designs for sampling trees in
<br />urban areas.
<br />Assuming an average tree density of 204 trees per acre of
<br />urban tree cover (504 trees/ha cover) (Dwyer et al. 2000) and a
<br />national average tree cover of 27.1% (Nowak et al. 2001), the
<br />average time to set up and measure a one -tenth acre (0.04 ha)
<br />urban plot in the United States would be approximately 95 min
<br />(five plots per 8 hr day). However, this estimate does not include
<br />travel time. The longer the distance between plots and the slower
<br />the traffic, the fewer the number of plots that can be measured
<br />per day. This estimate also includes plot permissions; however,
<br />plots on several land uses often do not require permission and
<br />access setup time could be reduced. Also, the fewer the trees per
<br />plot or fewer variables measured, the more plots can be mea-
<br />sured per day. A reasonable estimate of average number of one -
<br />tenth acre (0.04 ha) plots per day for a field crew of two people
<br />would be approximately three plots per day for a full suite of tree
<br />and plot measurements in a midsized city.
<br />Number of plots per day will vary by the amount of tree cover
<br />in a region because when tree cover increases, the amount of
<br />time measuring trees increases. In desert regions, urban tree
<br />cover averages 9.3% (Nowak et al. 2001) and average plot setup
<br />and measurement time would be approximately 51 min. In grass-
<br />lands (urban tree cover averages 17.8%), average plot time
<br />would be approximately 72 min. In forested areas (urban tree
<br />cover averages 34.4%), average plot time would be approxi-
<br />mately 113 min. Again, these estimates do not include travel
<br />time or office time needed to establish plot locations and maps.
<br />The standard UFORE model sampling approach establishes
<br />approximately 200 one -tenth acre (0.04 ha) circular plots in ran-
<br />domized grid or stratified random sample. The selection of 200
<br />plots was based on an estimated amount of plots that could be
<br />surveyed by field crew of two people during a summer season
<br />(14 weeks), given an average data collection rate of three plots
<br />per day. In some cities with high tree cover and/or traffic vol-
<br />umes, data collection will take longer than 14 weeks. In addition
<br />to data collection time, there are also costs associated with es-
<br />tablishing the locations of the plots, transportation, equipment,
<br />data entry or data transfer, and data analysis and reporting costs.
<br />The use of 200 one -tenth acre (0.04 ha) plots produces a
<br />reasonable population estimate if a 12% RSE is acceptable to the
<br />user. Depending on the desired precision, a smaller sample size
<br />may provide adequate estimates of the urban forest population.
<br />Table 2. Average time, number of lots accessed, trees per plot, and total population estimate from 26 residential plots
<br />measured in Syracuse, New York, U.S. using different plot sizes.
<br />Time (min) No. of lots
<br />Plot size (ac) Mean SE' Mean SE' Mean
<br />1/24 (0.017 ha) 61.8 7.4 1.9 0.1 2.6 1.1
<br />1/10 (0.04 ha) 84.1 9.9 2.8 0.2 4.6 1.3
<br />1/6 (0.067 ha) 106.1 14.0 3.1 0.2 6.5 1.5
<br />Standard error.
<br />''Actual estimated number based on 100 one -tenth acre (0.04 ha) plots is 251,000 trees (SE = 35,000).
<br />'Relative standard error.
<br />No. trees per plot
<br />SE' Range
<br />0-27
<br />0-33
<br />0-34
<br />No. of residential trees
<br />Estimate"
<br />429,998
<br />316,968
<br />267,922
<br />SE' RSE"
<br />178,366 41.5
<br />90,708 28.6
<br />61,220 22.8
<br />©2008 International Society of Arboriculture
<br />
|