The role of the residential urban forest in regulating throughfall: A case study in Raleigh, North Carolina, USA

Landscape and Urban Planning 119 (2013) 91–103

Contents lists available at ScienceDirect

Landscape and Urban Planning journal homepage: www.elsevier.com/locate/landurbplan

Research Paper

The role of the residential urban forest in regulating throughfall: A case study in Raleigh, North Carolina, USA Elina N.M. Inkiläinen a,∗ , Melissa R. McHale a , Gary B. Blank a , April L. James b , Eero Nikinmaa c a

North Carolina State University, North Carolina, USA Nipissing University, Ontario, Canada c University of Helsinki, Finland b

h i g h l i g h t s • • • • •

Urban residential forest reduced potential stormwater runoff by 9.1–21.4%. Throughfall amounts ranged greatly among yards with 41.5–88.3% canopy cover. Variability found in vegetation and throughfall between private and rented yards. Differences also found between front and backyards. Canopy cover (p < 0.0001) and coniferous trees (p = 0.017) influential variables.

a r t i c l e

i n f o

Article history: Received 6 April 2012 Received in revised form 2 July 2013 Accepted 6 July 2013 Keywords: Throughfall Urban residential forest Raleigh Ownership Stormwater regulation Landscape preferences

a b s t r a c t Overwhelming stormwater volumes, associated with deteriorating water quality and severe flooding in urbanizing cities, have become a great environmental and financial concern globally. Urban forests are capable of reducing the amount of stormwater runoff, in part, by regulating throughfall via canopy rainfall interception; however, the lack of stand-scale studies of urban throughfall hinders realistic estimates of the benefits of urban vegetation for stormwater regulation. Furthermore, urban forest characteristics that may be influencing rainfall interception are difficult to establish as these environments are extremely heterogeneous and managed, to a large extent, by private residents with varying landscape preferences. To quantify the amount of rainfall interception by vegetation in a residential urban forest we measured throughfall in Raleigh, NC, USA between July and November 2010. We analyzed 16 residential yards with varying vegetation structure to evaluate the relative importance of different descriptive measures of vegetation in influencing throughfall in an urban watershed. Throughfall comprised 78.1–88.9% of gross precipitation, indicating 9.1–21.4% rainfall interception. Canopy cover (p < 0.0001) and coniferous trees (p = 0.017) were the most influential vegetation variables explaining throughfall whereas variables such as leaf area index were not found significant in our models. Throughfall and vegetation characteristics varied significantly among yards (p < 0.0001), between front and back yards (p < 0.0001), and between rented and privately-owned yards (p = 0.001), suggesting a potentially significant role in stormwater regulation for urban residents. © 2013 Published by Elsevier B.V.

1. Introduction Stormwater runoff associated with an increased amount of impervious surfaces is the main cause of poor water quality, flooding, and deteriorating stream health in cities (Cappiella,

∗ Corresponding author. Tel.: +358503833747. E-mail addresses: elina.inkilainen@ecocity.fi (E.N.M. Inkiläinen), [email protected] (M.R. McHale), [email protected] (G.B. Blank), [email protected] (A.L. James), [email protected] (E. Nikinmaa). 0169-2046/$ – see front matter © 2013 Published by Elsevier B.V. http://dx.doi.org/10.1016/j.landurbplan.2013.07.002

Wright, & Schuler, 2005; Cunningham et al., 2009; Weijters et al., 2009). Urban forests, defined by Miller (1997) as all woody and associated vegetation in and around dense human settlements, have a great potential for reducing stormwater damage, by enhancing infiltration and evapotranspiration, as well as regulating the amount of throughfall reaching the ground via rainfall interception (Asadian & Weiler, 2009; Cappiella et al., 2005; McPherson, 1998; Xiao, McPherson, Simpson, & Ustin 1998; Xiao, McPherson, Ustin, Grismer, & Simpson, 2000b) (Table 1, Fig. 1). Rainfall interception is the proportion of rainfall that is intercepted by plant surfaces and evaporated directly back into the atmosphere (David, Valente,

92

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

Table 1 Definitions and acronyms. Acronym in text

Definition

P PNET TH TH, C

Gross precipitation Net precipitation = throughfall and stemflow Throughfall = the proportion of rainfall that penetrates or drips through a plant canopy (Hewlett, 1982) % Cumulative throughfall of gross precipitation (calculated by summing all throughfall values for the study period and dividing by cumulative gross precipitation) % Storm-based throughfall of gross precipitation (averaged from the means of each storm) Interception = the proportion of rainfall that is intercepted by plant surfaces and evaporated directly back into the atmosphere (David et al., 2005) Stemflow = the proportion of rainfall that flows down to the ground via stems Canopy cover Leaf area index Vertical structural complexity Coniferous tree cover Storm frequency index = variable indicating whether the storm was preceded by one or more rainless days Neighborhood

TH, S I ST CC LAI VSC CON F N

& Gash, 2005). Most studies have reported interception losses of 10 to 40% of gross precipitation, depending on meteorological factors and the type of vegetation (Crockford & Richardson, 1990; Llorens, Poch, Latron, & Gallart, 1997; Llorens & Domingo, 2007; Link, Unsworth, & Marks, 2004). Because of rainfall interception, throughfall is produced more gradually allowing more water to infiltrate the soil, reducing peaks in stormwater runoff (David et al., 2005; McPherson, 1998). Regardless of the great potential for reducing the adverse impacts of stormwater runoff where it is most needed, study of rainfall interception has been largely overlooked for urban areas. Urban forests differ in many ways from rural forests with regard to microclimate and tree architecture (Xiao et al., 1998, 2000b). In addition, the constant anthropogenic influence on urban forests presents higher probability of mechanical damage or stress caused by pollution, pests, and water availability (Asadian & Weiler, 2009; McPherson, 1998). Thus, a common perception in the field of environmental studies has been that urban forests may not fulfill the same functions as rural forests. More recently, however, it has been suggested that urban trees may in fact intercept higher amounts of rainfall compared to those in rural forests. Wider crowns and higher evaporation rates caused by wind and elevated temperatures have been found to produce lower throughfall magnitudes, producing

Fig. 1. Partitioning of gross precipitation into throughfall, stemflow, and rainfall interception (modified from Levia and Frost, 2006).

up to 60% less throughfall under individual tree crowns growing in urban areas (Asadian & Weiler, 2009; Xiao & McPherson, 2011; Xiao et al., 2000b). The amount of rainfall intercepted depends on the characteristics of both rainfall and vegetation in an area. For instance, the intensity and duration of rainfall and the frequency of storms have been highlighted as the main factors determining the efficiency of rainfall interception (David et al., 2005, 2006; Gash, 1979; Xiao, McPherson, Ustin, & Grismer, 2000a; Zeng, Shuttleworth, & Gash, 2000). Certain vegetative characteristics such as canopy storage capacity (i.e. the amount of water stored on foliage when the canopy is saturated) interact with rainfall patterns as well (Rutter, Kershaw, Robins, & Morton, 1971). Storms below canopy storage capacity produce less throughfall than storms exceeding canopy storage capacity. Less throughfall is produced when the canopy has sufficient time to dry in between storms (Gash, 1979). The functional type of vegetation greatly influences canopy storage capacity and the amount of rainfall intercepted. Conifers have been found to have higher leaf area index (LAI) than deciduous trees (Barbour, Burk, & Pitts, 1980). Interception losses of 20–40% have been reported in coniferous forests while 10–20% have been found in broadleaved forests (Crockford & Richardson, 1990; Link et al., 2004; Llorens et al., 1997; Llorens & Domingo, 2007). Seasonal changes in canopies also affect the amount of throughfall reduced via rainfall interception. Evergreen trees are capable of interception for a larger part of the year than deciduous trees and are thus especially important in regions with winter precipitation (Xiao et al., 1998). While the quantity of foliage is the most important indicator of canopy storage capacity, crown architecture has also been found to have an influence (Xiao et al., 2000b; Xiao & McPherson, 2011). Several models have been developed for estimating rainfall interception (Muzylo et al., 2009). Most models include one or more vegetation variables, with the most common metrics being canopy cover and LAI. To date, canopy cover has been more popular in predicting rainfall interception probably due to being relatively easy to measure (Bryant, Bhat, & Jacobs, 2005; Gash, 1979; Gash, Lloyd, &, Lachaud, 1995). Some researchers however stress the importance of LAI in estimating canopy storage capacity and the variable has been preferred in urban areas for modeling rainfall interception of individual tree crowns (van Dijk & Bruijnzeel, 2001; Xiao et al., 2000b). LAI may be a better indicator of rainfall interception at the crown-level but accurate measurement for heterogeneous canopies is challenging due to potential bias from direct sunlight penetrating the canopy, resulting in underestimations. This may be accentuated in characteristically heterogeneous urban forests.

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

Another variable that may be influential in modeling throughfall in urban systems is the Shannon diversity index (Shannon, 1948; ref: McElhinny, Gibbons, Brack, & Bauhus, 2005). Although the most popular use of this index has been to estimate species diversity in forests, it is similarly useful in quantifying the vertical structural complexity of vegetation. In fact, Calder (1996) found that more canopy layers allow more interception through the gradual wetting of the canopy. Landscape heterogeneity, characteristic of urban systems, is largely a sum of the way people manage their individual, privatelyowned plots of land. Residents manage their yards according to landscape preferences that are reflected in vegetation structure. These preferences have been suggested to depend on cultural norms (Stamps, 1999; Yu, 1994) and the type of neighborhood people live in (Nassauer Wang, & Dayrell, 2009). The perceived pressure from neighbors to conform to the style of the neighborhood may result in differences in vegetation structure between front yards and more concealed backyards (Larsen & Harlan, 2006; Nassauer et al., 2009). These societal influences need to be considered in studies of urban rainfall interception and throughfall in order to truly understand the factors affecting stormwater benefits provided by urban trees. In this study, we aimed to quantify the amount of throughfall and interception occurring in an urban residential forest managed by multiple residents. Secondly, we set out to evaluate a set of vegetation variables in order to determine the best predictors of throughfall and interception. Finally, we tested whether individual resident’s landscape designs or the ownership structure of yards had an impact on the generation of throughfall. Specifically, we expected to see high variability in vegetation characteristics and throughfall between privately-owned and rented yards and between front and back yards. These results will help us understand the value of urban residential forests for throughfall regulation and the ways in which residents’ landscape choices affect the amount of throughfall produced. Our findings will inform urban land-use planning and future research on measuring and modeling rainfall interception and throughfall in urban conditions.

2. Materials and methods 2.1. Study area The City of Raleigh, North Carolina, is one of the fastest growing regions in the USA; the population has grown by 46.3% from 2000 to 403,892 in 2010 (United States Census Bureau, 2011). The city has maintained a high tree cover of 55% (Unpublished data, 2013) owing to the gradual development from agricultural areas to lowintensity residential areas that allow the existence of forest-like vegetation in urban areas. According to Raleigh Department of City Planning (http://www.raleighnc.gov/cp), single-family residential areas accounted for 34.1% of land surface in 2007, making it the single largest land use type - and a significant holder of the valuable urban forest resource in Raleigh. Development is also spreading rapidly in the region. As other sprawling cities, Raleigh will most likely experience a considerable growth in residential land use in the future. Thus, the relative importance of residents’ landscape choices on stormwater regulation may increase in the future in Raleigh and other similar regions. The experiment took place in the Beaverdam Creek residential watershed, located in Raleigh (latitude: N35◦ 48 22 ; longitude W78◦ 40 17 , Fig. 2). The watershed of approximately 8 km2 is characterized by relatively high canopy coverage, dispersed building pattern, and large lot size. The study area was chosen to represent a typical low-intensity residential area in the region. Forests

93

in the residential watershed comprise a mix of deciduous oaks and pines with a characteristically dense understory of shrubs and vines. Exotic ornamental species are also common. The average elevation in the study area is below 80 m above sea level (USGS, 2009). The region is influenced by humid subtropical climate, characterized by moderate spring and fall temperatures, warm to hot summers, and mild winters. Average summertime temperatures range between 19 and 34 ◦ C with highs around 38 ◦ C (NOAA, 2011). Winter averages fall typically within 8–12 ◦ C with lows commonly above or slightly below the freezing point. Precipitation occurs throughout the year but large, high intensity storms are especially common during the summer (NOAA, 2011). Between 1981 and 2010, average annual precipitation was 1096 mm and monthly averages ranged between 74 mm in April and 120 mm in July. The relative humidity is especially high between late summer and fall (Southeast Regional Climate Center, 2007). The average wintertime snowfall in Raleigh is 152 mm (NOAA, 2011). 2.2. Experimental design We chose neighborhoods in close proximity to the western branches of Beaverdam Creek (Figs. 2 and 3) and randomly selected potential properties with varying canopy cover and vegetation structure within these neighborhoods. Residents living in these neighborhoods were approached and 16 yards were identified for measuring gross precipitation and throughfall. Nine of these yards were owned by private residents and the remaining seven were rental properties, collaboratively managed by the landlord and tenants. The average size of yard lots (excluding house) was 1342 ± 1026 m2 . The 16 yards were divided into three neighborhoods based on location, average size of lot, and ownership. Neighborhoods one (N1, yards 1–4) and two (N2, yards 10–12) consisted of relatively large, privately owned properties with generally high canopy cover. Neighborhood three (N3, yards 5–9; 13–16) included two privately owned and seven rental properties with lower canopy cover and smaller lot size compared to the other two neighborhoods. The average sizes of lots in N1, N2, and N3 were 2777 ± 958 m2 , 1481 ± 129 m2 , and 659 ± 248 m2 , respectively. We laid a 10 m × 10 m grid over each yard to ensure a sampling density of one measuring point per 100 m2 . Each yard had 4 to 35 measuring points, depending on its size, resulting in a total sample size of 206 points. The measuring point in each 100 m2 grid cell was selected by blindly throwing an object and marking the random spot where the object landed. Front yards, with an average size of 497 ± 255 m2 , were generally smaller than backyards with an average size of 854 ± 810 m2 . In setting up the research, we respected homeowners’ wishes by excluding from the effective study area driveways and other frequently used areas close to buildings. These factors resulted in fewer measuring points in front yards (N = 78), where most driveways were located, compared to backyards (N = 128). Buckets (26 l in volume with a diameter of 0.30 m) were used to measure throughfall and gross precipitation for the 206 measuring points. We raised the opening of the bucket 0.36 m above the ground to minimize splashing of water into and out of the bucket. Buckets were stabilized with metal stakes to decrease the probability of getting knocked over; however, occasionally some buckets were knocked over by wind, animals, people, and the flooding of a stream. Due to the gradual addition of yards to the study area between July 28 and September 12, the number of observations differs from storm to storm. Miscommunication between tenants and the landlord led to the removal of 28 measuring points located in six yards

94

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

Fig. 2. Beaverdam Creek watershed in Raleigh, North Carolina, USA. Inset includes the outline of the State of North Carolina (top right), the urban boundaries of the city of Raleigh, and topographically defined Beaver Creek watershed. A dashed white ellipse around the western branches of Beaver Creek indicates the study area.

on October 22, 2010. These yards (5, 7, 8, 13, 14 and 15; Fig. 3) were located in neighborhood three. 2.3. Measurements and calculations 2.3.1. Components of rainfall Throughfall and gross precipitation were measured between July 28 and November 17, 2010. Six of the twenty recorded storms were considered unrepresentative of the entire study area due to containing data from less than 60% of 206 measuring points or

having a magnitude less than 0.4 mm and thus being prone to large measurement errors. Excluding unrepresentative or biased data to assure the reliability of the remaining dataset is common practice in rainfall interception research (e.g. Bryant et al., 2005; Pypker, Unsworth, & Bond, 2006; Shinohara et al., 2010; Xiao & McPherson, 2011). Thus, 14 storms were selected for further analyses (Table 2). We conducted the measurements on the day following each storm. Thus, the date of the storm refers to the date when data were collected, regardless of when the storm started.

Fig. 3. Study area with 16 residential yards at a close proximity to Beaverdam Creek (left); Close-up of yards 1–4 with measuring points (middle); Measuring point under a Dogwood (right). GIS data sources: Wake County Government, NC.

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

95

Table 2 Average gross precipitation (P), throughfall (TH), cumulative TH, % storm-based TH), % interception (I), assuming 0.5% and 2% stemflow (ST), and storm frequency index (F, 1 = 0 days from previous storm; 2 = one or more days from previous storm) across storms. Storm

Na

date 12-Aug-10 19-Aug-10 20-Aug-10 24-Aug-10 12-Sep-10 20-Sep-10 30-Sep-10 15-Oct-10 21-Oct-10 27-Oct-10 28-Oct-10 05-Nov-10 07-Nov-10 17-Nov-10

a b

P (mm) Mean

184 138 125 195 198 205 206 200 201 170 175 172 176 173

TH (mm) Std dev

TH, C (%)

TH, S (%)

I (%)b

F

Mean

Std dev

Mean

Mean

Std dev

(ST = 0.5)

(ST = 2)

2.6 22.2 14.8 54.3 2.3 82.0 98.3 11.1 1.8 16.7 6.5 14.4 2.8 10.5

0.8 1.6 2.3 4.3 0.7 0.8 1.7 0.3 0.2 1.1 0.7 0.3 0.3 0.5

1.8 18.7 14.5 50.0 1.2 74.4 90.1 10.0 0.9 14.0 5.1 11.8 1.6 8.1

1.3 4.0 3.5 10.2 0.9 14.0 18.7 2.3 0.6 3.9 1.9 2.8 0.9 1.9

70.8 84.4 98.1 92.1 54.5 90.7 91.6 90.0 50.0 83.9 78.2 82.4 58.4 77.2

69.6 84.7 98.9 92.3 52.7 90.7 91.6 90.0 51.1 83.6 77.7 82.3 58.9 77.4

44.0 18.0 23.4 18.1 35.0 17.1 18.9 20.3 35.3 21.0 25.6 19.1 31.2 18.8

28.7 15.1 1.4 7.4 45.0 8.8 7.9 9.5 49.5 15.6 21.3 17.1 41.1 22.3

27.2 13.6 −0.1 5.9 43.5 7.3 6.4 8.0 48.0 14.1 19.8 15.6 39.6 20.8

25.7

31.6

22.9

30.1

88.9

78.1

30.1

21.4

19.9

2 2 1 1 2 2 2 2 2 2 1 2 2 2

Number of observations. Based on storm-based TH (TH, S).

We measured the volume of throughfall using a 2-l container allowing for measurement accuracy of 25 ml, that is ±0.4 mm. The actual measurement accuracy of field measurements depends on multiple factors, including potential bias caused by high winds, animals occasionally knocking over buckets, splash, and evaporation of water from the buckets. Because the climate in Raleigh is characterized by high humidity, especially during the night time, it is unlikely that significant evaporation losses occurred before we took our measurements in the morning. Six control measuring points under open sky were identified for collecting gross precipitation in yards 2, 3, 6, 9, 12, and 16. The strong correlation (R = 0.997) among measurements from control points indicated little bias caused by adjacent trees and buildings in the yards or the ∼750 m distance between the northern and southern groupings of yards (Fig. 3). Thus, data collected in these yards were applicable to forested yards, lacking open areas, in the same neighborhood. We assumed stemflow of 0.5 to 2% of gross precipitation based on the work by Bryant et al. (2005) in rural forests of similar plant communities, located in the southeastern US. Guevara-Escobar, González-Sosa, Véliz-Chávez, Ventura-Ramos, & Ramos-Salinas, 2007 found 2% stemflow under an isolated urban Ficus tree and Xiao et al. (1998) simulated stemflow of 0.6% for deciduous urban forests in California, agreeing well with our estimates. Rainfall interception was calculated from a simple mass balance equation often used to describe rainfall partitioning (Crockford & Richardson, 2000; Horton, 1919; Xiao et al., 2000b): I = P − TH − ST

(1)

where I is rainfall interception, P, TH, and ST are gross precipitation, throughfall, and stemflow, respectively (Table 1). To allow better comparison with other studies we will report the % throughfall of gross precipitation calculated from cumulative data and storm averages. The “cumulative” % throughfall (TH, C) is calculated by summing all throughfall values for the study period and dividing by cumulative gross precipitation. The “storm-based” % throughfall (TH, S) is averaged from the means of each storm. 2.3.2. Canopy cover For each yard, we measured canopy cover on two occasions during the study period. Measurements were taken between October 1 and October 4, 2010 and again between December 2 and December 17, 2010. We used a spherical densiometer (Lemmon, 1956) for

measuring canopy cover. Four measurements were taken from different directions above the opening of each bucket. 2.3.3. Leaf area index For each yard, we measured photosynthetically active radiation (PAR) for calculating LAI by using a Sunfleck PAR Ceptometer (Decagon Pullman, WA) on three occasions throughout the study period: between August 27 and September 1, 2010; between October 18 and October 29, 2010; and on November 28, 2010. PAR was measured below (Qi ) and above (Qo ) the canopy under clear skies. Measurements were carried out between 11:00 and 15:00 to avoid errors caused by tall trees blocking the radiation which may happen when the zenith angle of the sun departs greatly from 0◦ (directly overhead). We took the above canopy readings in a sports field within a distance of 700 m from each measuring point and repeated them at least twice in a period of 3 h. We obtained the actual LAI values from the following equation: LAI = − ln(Qi /Qo )k−1

(2)

The light-extinction coefficient (k) can be calculated from the following equation when leaf angle distribution is assumed to be spherical: k = 1/(2 cos )

(3)

The zenith angle of the sun  is calculated as follows:  = arccos(sin L sin D + cos L cos D cos 0.2618(t − to ))

(4)

where L is latitude, D is the solar declination, t is the time (hours), and to is the time of solar noon. The constant 0.2618 converts hours to radians. The Beer–Lambert law assumes that foliage is randomly distributed in space and that there is a spherical distribution among leaf inclination angles (Jarvis & Leverenz, 1983). Violations of this assumption may yield underestimations of LAI caused by bias from direct sunlight penetrating the canopy. Large gaps in canopy cover, characteristic of urban forests, thus complicate the measurement of LAI at the level of a measuring point. We took several measurements above the opening of each bucket and each yard included at least four measuring points. Thus, the averaged values within and across yards should provide reliable estimates of LAI. The temporal resolution of the measured PAR data was further increased with data from Lake Wheeler Road weather station (latitude: 35.72816; longitude: −78.67981, 10,000 m south from the northernmost measuring point) after assuring the comparability of

96

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

the two datasets via regression analyses. We created a regression plot of the weather station data and applied it to our own above canopy measurements from the same time interval to interpolate Qo for each measurement of Qi . 2.3.4. Vertical structural complexity We measured vertical structural complexity (VSC) twice during the study to evaluate the vertical arrangement and layering of shrubs and trees above each of our measuring points. First measurements were taken between September 24 and October 7 and second measurements between December 9 and December 19. We visualized a rectangle of 27 m3 (3×3×3 m) centered on each bucket and estimated the fraction of vegetation (C) within this ‘box’. The assessment was repeated vertically throughout the canopy. The measurements were repeated towards the end of the study period to account for seasonal shifts in leaf area or changes induced by natural thinning of plants, storm damage, mechanical damage, or pruning. Minor changes in vegetation structure may not be detectable via visual estimation which becomes more challenging in dense forests consisting of multiple canopy layers. To allow the comparison among measurements taken from different measuring points at different times it is essential that one person conducts the measurements each time despite the possibility of systematic bias. The vertical structural complexity index (VSC) was calculated using the Shannon–Weiner equation (Shannon, 1948;ref: McElhinny et al., 2005): VSC =

n   1 i=1

Ci

/ln(



1 ) Ci

(5)

where Ci is the fraction of vegetation for each vertical layer i. 2.3.5. Other vegetation parameters We also measured the quantity, distance and direction of trees, shrub masses, and nearest buildings from each measuring point. These variables were measured within a plot of six meters in radius (∼113 m2 ) from the measuring point. When no trees or shrubs were present in the plot, we measured the nearest tree or shrub outside the plot. To evaluate the importance of each functional plant group in rainfall partitioning, we estimated the percentage cover of coniferous, broadleaved evergreen, and broadleaved deciduous tree and shrub cover within the plot around each measuring point, totaling six vegetation groups. Deciduous plants shed leaves at wintertime; evergreen plants maintain at least some leaf cover throughout the year. Thus, the group “evergreen broadleaved trees” includes trees such as the southern live oak (Quercus virginiana) and evergreen Magnolia species whereas the group “deciduous broadleaved trees” would include the white oak (Q. alba). Following Nowak et al. (2008, p. 348), we considered trees with a shrubby appearance as shrubs when they had a diameter at breast height (DBH) lower than 2.5 cm or had more than six stems. 2.3.6. Statistical analyses Storms were classified into those with 24 or more rainless hours preceding the time of measurement (dry canopy) and those with less than 24 h (wet canopy). We created a dummy variable, the storm frequency index (F), to test for the importance of the initial dryness of the canopy on throughfall using ANOVA. To assess the effects of ownership on throughfall and vegetation structure, we performed ANOVA comparing privately-owned and rented yards. This research is a late growing season study, focusing on the time when trees had leaf cover. To assess the seasonal changes in vegetation and throughfall, we divided the data into summer storms and fall storms. Following previous studies such as Wang, Endreny, & Nowak, 2008 we set the threshold for fall to November 7

when the first temperatures below freezing-point occurred (NOAA, 2011), and deciduous trees began rapidly losing leaves. Thus, we recorded 12 summer storms (Aug 12–Nov 5) and two fall storms (Nov 7, Nov 17). Variables influencing throughfall were selected for a linear model by using the forward stepwise selection method (˛ = 0.05). We adopted Mallow’s Cp, adjusted R2 , and root mean square error (RMSE) as the main selection criteria for selecting influential variables. Due to the uneven variability in residuals among different storm magnitudes in the data, i.e. heteroscedasticity, we used the Kruskal–Wallis test, a nonparametric alternative to ANOVA, to compare throughfall (mm) among yards, neighborhoods, and between front and backyards. For the % storm-based TH, calculated from throughfall (mm) and gross precipitation (mm), we were able to use regular ANOVA tests. SAS® 9.1 Software (SAS, 2011) and JMP® 8 Software (JMP, 2011) were used for calculating descriptive statistics and restricted maximum likelihood (REML) correlations, for performing ANOVA tests, and for selecting a linear regression model. 3. Results 3.1. Throughfall regulation The 20 measured storms occurred between July 28 and November 17, 2010 and delivered 466.9 mm of gross precipitation to the study area. We excluded six storms containing less than 60% of measuring points and used the 14 remaining storms with cumulative gross precipitation of 340.2 mm for further analyses (August 12–November 17; Table 2). Storm magnitude ranged from 1.8 ± 0.2 mm on October 21 to 98.3 ± 1.7 mm on September 30. Summer and fall 2010 were generally consistent with long-time observations of higher precipitation averages in summer-early fall compared to late fall-early winter in Raleigh. September 2010, however, was wetter (182.6 mm) than the 30-year monthly averages recorded in Raleigh (112.0 mm) and October (36.1 mm) was dryer than long-time averages (90.2 mm) (Raleigh State Weather Station: latitude: 35.79444, longitude: −78.69889). No snow was recorded during the study period. Cumulative throughfall from all storms accounted for 418.9 mm or 89.1% of gross precipitation (TH, C) or 79.1 ± 30.6% of stormbased TH (TH, S). Of this, the 14 storms selected for further analyses produced 302.3 mm of throughfall, that is, 88.9% of cumulative TH or 78.1 ± 30.1% of storm-based TH. Especially considering the large standard deviation related to these numbers (e.g. see Fig. 4A) we consider the 14 storms representative of all 20 storms measured. TH (C) and TH (S) in N1, N2, and N3 were 88.0% and 76.2 ± 27.3% (N1), 87.3 and 74.2 ± 32.2% (N2), and 93.0 and 86.9 ± 32.7% (N3), respectively. Assuming stemflow from 0.5 to 2%, estimated rainfall interception was 10.5–9.0% (TH, C) or 21.4–19.9% (TH, S; Table 2). Cumulative TH (TH, C) decreased from an average of 89.6 during summer to 73.1% during fall. Storm-based TH (TH, S) decreased from 80.4 ± 24.7 during summer to 68.1 ± 25.0 during fall. 3.2. Vegetation characteristics Table 3 summarizes average vegetation characteristics at the study site. Average canopy cover in the study site yards was 67.0 ± 29.1% which can be considered very high for urban areas. However, as buildings and driveways were excluded from the effective study area, the resulting canopy coverage should be considered as an estimate of the canopy-level coverage of a low-intensity residential area. The average canopy cover in N1, N2, and N3 was 71.1 ± 61.4%, 71.9 ± 27.3%, and 51.8 ± 31.9%, respectively.

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

97

Fig. 4. ANOVA: The percentage of storm-based throughfall (TH, S) (A) among yards (F(15) = 14.3, p<0.0001); (B) between privately-owned (1) and rented (2) yards (F(1) = 37.3, p<0.0001); (C) between wet (24>) and dry (24≤) canopies (F(1) = 79.9, p<0.0001), and (D) between front and backyards (F(15) = 10.9, p = 0.001). Horizontal grey lines indicate 95% confidence intervals of the mean.

Average LAI during the study period was 1.9 ± 1.3 and was 2.2 ± 1.3 in N1, 1.9 ± 1.1 in N2, and 1.4 ± 1.3 in N3, withinneighborhood variability in LAI being especially high in N3. Vertical structural complexity index in the whole study area was 35.3 ± 24.6 and averaged at 38.8 ± 25.7 in N1, 38.4 ± 22.7 in N2, and 23.5 ± 19.8 in N3. An average of 524 ± 481 trees were found per ha with highest numbers of trees found in N1 (640 ± 514) and N2 (539 ± 465), and lowest in N3 (222 ± 206).

The mean% coniferous tree cover was 22.5 ± 33.0%, neighborhood averages being 24.9 ± 33.2% (N1), 23.3 ± 31.3% (N2), and 15.7 ± 33.4% (N3). Broadleaved evergreen tree cover was 5.5 ± 16.7%, averaging at 2.5 ± 7.2% in N1, 8.5 ± 18.9% in N2, and 9.9 ± 26.5% in N3, showing high within-neighborhood variability. Broadleaved deciduous trees averaged 72.0 ± 36.5% in the study area, 72.6 ± 35.3% in N1, 68.2 ± 36.4% in N2, and 74.4 ± 39.3% in N3.

Table 3 Average LAI, VSC, trees per ha, % canopy cover (CC), % coniferous (CON), % broadleaved evergreen (BE), and % broadleaved deciduous (BC) trees and shrubs, and % TH across study period by yard.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 a b

Na

Nb

LAI

VSC

22 30 19 35 5 7 5 4 6 15 15 16 8 4 4 11

301 407 265 405 45 93 44 35 81 194 179 208 72 44 36 109

3.2 1.6 1.8 2.2 0.6 1.1 1.6 2.7 1.1 2.2 1.4 2.0 2.2 2.1 1.8 1.1 1.9

44.5 25.4 32.3 52.2 13.3 29.1 38.0 32.6 25.6 41.8 41.2 32.9 21.0 31.6 19.4 12.5 35.3

Number of measuring points per yard. Number of observations.

Trees/ha

1032 197 699 755 71 322 285 326 219 734 224 628 155 137 287 201 524

CC (%)

88.3 54.8 63.0 80.2 41.5 53.1 63.5 68.1 43.3 85.5 60.3 69.2 69.9 67.8 58.4 30.9 67.0

Trees (%)

Shrubs (%)

CON

BE

BC

CON

BE

BC

41.0 9.8 0.3 44.2 0.0 34.5 0.0 6.9 14.8 50.5 0.0 17.9 18.1 0.0 22.5 18.3 22.5

2.1 0.0 0.5 6.6 20.0 0.5 15.2 4.7 12.1 9.5 2.3 12.8 0.0 0.0 1.8 25.0 5.5

56.9 90.2 99.2 49.2 80.0 65.0 84.8 88.4 73.1 39.9 97.7 69.2 81.9 100.0 75.8 56.7 72.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.8 2.0 0.0 0.0 0.0 0.0 5.6 0.7

69.4 46.5 43.9 48.0 86.4 46.1 55.0 59.3 45.1 56.0 15.1 70.1 30.0 0.0 33.0 50.9 49.4

30.6 53.5 56.1 52.0 13.6 53.9 45.0 40.7 54.9 40.2 82.9 29.9 70.0 100.0 67.0 43.5 50.0

TH, C (%)

TH, S (%)

88.3 91.5 88.5 84.0 94.8 94.5 98.3 98.2 91.9 83.2 92.4 86.8 88.4 89.3 97.2 90.6 88.9

70.5 84.4 83.2 67.7 87.0 82.3 90.9 85.7 82.1 63.3 81.2 78.4 87.9 86.6 94.1 90.4 78.1

98

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

Small amounts of coniferous shrub cover occurred in N2 (1.9 ± 9.4%) and N3 (1.1 ± 4.2%), no cover in N1, and an average of 0.7 ± 5.0% in the whole study area, again showing very high variability. Evergreen shrub cover averaged at 49.4 ± 37.3% with highest amounts found in N1 at 51.5 ± 35.2%, followed by N2 (48.4 ± 36.3%), and N3 (45.1 ± 42.5%). Broadleaved deciduous shrub cover had an average of 50.0 ± 37.1% in the study area, with the following neighborhood averages: 48.5 ± 35.2% (N1), 49.7 ± 36.2% (N2), and 53.8 ± 42.1% (N3). We observed some seasonal variability in vegetation between summer and fall periods although the differences remained relatively small. Mean canopy coverage in the study site during summer and fall periods was 69.0 ± 29.1% and 54.9 ± 26.1%, respectively. Mean canopy cover for the entire study period was 67.0%, reflecting the domination of the summer period when leaf area remained high. LAI experienced a larger shift from 2.1 ± 1.3 during summer to 1.1 ± 0.6 during fall, averaging at 1.9 for the entire study period. The vertical structural complexity of vegetation increased towards the end of the study period, from 34.5 ± 23.9 to 40.4 ± 27.9 with an average of 35.3. 3.3. Variables explaining throughfall Canopy cover had the strongest negative correlation with the storm-based TH (R = −0.33), followed by the number of trees per ha (R = −26), VSC index (R = −0.25), % coniferous (R = −0.22) trees, and LAI (R = −0.15). The remaining variables with a negative correlation with throughfall, i.e. % broadleaved evergreen trees, % coniferous and broadleaved evergreen shrubs, and the distance to trees and buildings had a lower correlation with the storm-based TH. As could be expected, % broadleaved deciduous trees (R = 0.20) and shrubs (R = 0.10) had a positive correlation with storm-based TH. We generated a linear empirical model including the most influential parameters for estimating throughfall (TH): TH = 3.548 + 0.929P − 0.051CC − 0.002((P − 25.744) (CC − 67.028)) − 0.011CON − 0.998F

(6)

where P is gross precipitation, CC is % canopy cover, CON is % coniferous trees, and F is storm frequency index, explaining the initial wetness of the canopy. Other variables, such as VSC index (p = 0.1361), LAI (p = 0.8101), the number of trees per ha (p = 0.8884) were not influential at the 0.05 significance level. The linear model confirmed that P (p < 0.0001) accounted for most variability in TH. In addition, the estimate of the coefficient of 0.93 for P indicates that a significant proportion of rainfall was reduced by vegetation. All vegetation variables that were included in the model had a negative relationship with TH. Every additional percentage unit of CC (p < 0.0001) decreased TH by 0.05 mm and every unit of CON (p = 0.017) decreased TH a further 0.01 mm. The shared influence of CC and P (p < 0.0001) was also found significant with an estimate of −0.002. The storm frequency index F (p = 0.0054) affected TH as 1.00 mm less throughfall was produced when the storm was preceded by one or more rainless days. In other words, a wet canopy produced a millimeter more throughfall than a dry one. A sensitivity analysis of the model revealed that at small storm magnitudes the model was especially sensitive to changes in the initial dryness of the canopy. We also conducted an ANOVA to compare differences between dry and wet canopies and found significant differences (F(1) = 79.9, p < 0.0001; Fig. 4C). Fig. 5 shows the actual vs. predicted throughfall using the described model. The empirical model explained 94% of the variability in throughfall (Adj. r2 = 0.9440, p < 0.0001, RMSE = 7.1281). The addition of vegetation variables CC and CON increased the explained variability and decreased the RMSE, albeit the

Fig. 5. Actual vs. Predicted throughfall (Adj. r2 = 0.9440, p<0.0001, RMSE = 7.1281) for a model including vegetation parameters. Dashed line indicates 1:1. Horizontal dashed line shows the mean of response (22.9 mm).

improvement was relatively low compared to storm magnitude alone (Adj. r2 = 0.9383, p < 0.0001, RMSE = 7.4823). As seen in Fig. 5, the variability in throughfall increases as storm magnitude increases, indicating that the residuals show an increasing trend towards larger values, otherwise known as heteroscedasticity. 3.4. Throughfall regulation at yard-level The 16 yards presented significant variability in vegetation characteristics. Average canopy cover at yard-level had a very high variability from 30.9 ± 31.6% (yard 16) to 88.3 ± 10.4% (yard 1) (Table 3). LAI varied from 0.6 ± 1.1 (yard 5) to 3.2 ± 1.3 (yard 1). The highest vertical structural complexity was found in yard 4 with the score 52.2 ± 24.5 and lowest complexity occurred once again in yard 16, with the score 12.5 ± 8.1. The largest amount of coniferous tree cover was found in yards 1, 4, and 10 with 41.0 ± 29.4%, 44.2 ± 32.1%, and 50.5 ± 27.2% cover, respectively. Yards 2, 3, 11, and 14 had more than 90% broadleaved deciduous tree cover and all except for yards 5 and 16 had less than 20% broadleaved evergreen tree cover. Significant differences were observed in the amount of throughfall (mm) generated in yards, verified by the Kruskal–Wallis test (ChiSq(15) = 31.1, p = 0.0086). We used ANOVA for detecting differences in the average % storm-based TH and detected significant variability among yards (Fig. 4A; F(15) = 14.3, p < 0.0001). The range in storm-based TH among yards was 63.3 ± 33.2% to 94.1 ± 45.9% with lowest and highest values found in yards 10 and 15, with 85.5 ± 9.6% and 58.4 ± 39.0% canopy cover, respectively (Table 3). We expected vegetation characteristics and throughfall amounts to vary with ownership structure. Indeed, ANOVAs revealed significant differences among rented and privatelyowned yards with regard to all vegetation variables, with the largest deviations mentioned below. Lower % canopy cover was found in rented versus privately-owned yards, reported as the difference between the two means (−10.6%; F(1) = 41.6, p < 0.0001). A significantly lower % coniferous tree cover was found in rented yards (−14.6%; F(1) = 61.3, p < 0.0001) coupled with significantly higher % broadleaved deciduous tree cover (12.0%; F(1) = 33.4, p < 0.0001), compared to privately-owned yards. Less broadleaved evergreen shrub cover was found in rented yards

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

(−7.3%; F(1) = 11.8, p = 0.0006) and higher broadleaved deciduous shrub cover (8.1%; F(1) = 14.7, p = 0.0001) compared to privatelyowned yards. Also, large differences were observed in the number of trees per ha between rented and privately-owned properties (374 trees/ha; F(1) = 200.0, p < 0.0001). Higher amounts of stormbased TH were observed in rented (87.1 ± 35.3%) compared to privately-owned properties (76.7 ± 28.9%; F(1) = 37.3, p < 0.0001; Fig. 4B). As hypothesized, significant differences in throughfall and vegetation characteristics were also found between front and backyards. Storm-based TH was found to be 4.1% larger in front yards (80.7 ± 31.6%) compared to backyards (76.6 ± 29.1%) as seen in Fig. 4D (F(1) = 10.9, p = 0.001) and cumulative TH was 3.5% higher in front yards compared to backyards (87.8%). We found significant differences between front and backyards in the variables % canopy cover (58.5 ± 29.3%; 72.0 ± 27.8%; F(1) = 131.9, p < 0.0001), LAI (1.5 ± 1.2; 2. 2 ± 1.3; F(1) = 185.1, p < 0.0001), VSC (30.4 ± 22.1; 38.2 ± 25.5; F(1) = 61.0, p < 0.0001), % coniferous trees (15.3 ± 31.4; 26.7 ± 33.2%; F(1) = 71.9, p < 0.0001), % broadleaved deciduous trees (79.1 ± 67.8%; 67.8 ± 36.1%; F(1) = 56.9, p < 0.0001), % coniferous shrubs (1.8 ± 8.1%; 0.0 ± 0.4%; F(1) = 73.6, p < 0.0001), % broadleaved evergreen shrubs (45.2 ± 39.3%; 51.8 ± 35.9%; F(1) = 18.5, p < 0.0001), % broadleaved deciduous shrubs (53.0 ± 39.0%; 48.2 ± 35.9%; F(1) = 10.1, p < 0.0015), and number of trees per ha (301 ± 331; 654 ± 506; F(1) = 361.6, p < 0.0001). The % broadleaved evergreen trees was not found significantly different between front (5.6 ± 19.1%) and backyards (5.8 ± 15.0%). 4. Discussions 4.1. Throughfall regulation by the urban residential forest 4.1.1. Rainfall partitioning Our first goal was to find out by how much throughfall was reduced by the urban forest located in a residential area in Raleigh between late summer and fall. Cumulative TH was 88.9% or 78.1% when assessed through storm averages. In other words, the primarily deciduous broadleaved urban residential forest with an average canopy cover of 67.0% reduced TH via I by 9.1–10.6% (cumulative) or 19.9–21.4% (storm-based), assuming 2–0.5% ST. 4.1.2. Comparison to other urban forests Previous urban studies have reported TH percentages ranging between 38.1 and 93.4% of P (Table 4). The wide range of throughfall values is mostly explained by the different scales of these studies as some researchers have focused on individual trees (crown-level) whereas others have considered the entire canopy (stand-level). The direct comparison between results obtained at crown- and stand-level would be faulty as the proportion of direct TH is naturally much lower when limiting the area of interest under the drip-line of a single tree crown as opposed to the level of the entire canopy. Thus, we will compare our findings with other stand-scale studies conducted in urban areas. In the Mediterranean climate of Sacramento, CA, Xiao et al. (1998) found TH percentages between 86.4 and 93.4%, comparing well with our findings of 88.9% of cumulative TH. Xiao et al. (1998) noted that less TH (86.4%) was generated in the ‘suburban sector’ dominated by broadleaved evergreen trees compared to the ‘city sector’ dominated by broadleaved deciduous trees (93.3%). Apart from maintaining foliage throughout the rainy winter seasons, evergreen trees also tend to have higher LAI than do deciduous trees (Xiao et al., 1998). Thus, the differences between the two land use sectors were related to dominant vegetation types and the seasonal rainfall patterns of the region. In our study, the role of broadleaved evergreen trees in the amount of rainfall

99

intercepted was not significant, possibly due to the short duration of the study period from late summer to fall. We did, however, find that % broadleaved deciduous trees was positively correlated with % storm-based TH, given that this influence was not found significant in the regression model. Wang et al. (2008, Table 4) found that rainfall interception in Maryland, USA, accounted for 18.4% of P, resulting in net precipitation (PNET) , i.e. TH and ST, of 81.6%. These results were simulated using the model UFORE-Hydro (currently iTree-Hydro) in an urban watershed dominated by deciduous vegetation (22% of watershed), located in the humid subtropical climate. Our storm-based results from similar climate are comparable with theirs as we found PNET of 78.6–79.1% (assuming ST of 0.5–2%) whereas our cumulative TH would yield PNET as high as 89.4–90.9%. The considerably higher CC of 67.0% in our study site should result in much lower PNET , compared to the 22% coverage by Wang et al. (2008). It should be noted that the model used land cover data to estimate canopy coverage, which may result in an underestimation of actual coverage, especially when assessing deciduous forests during leaf-off. 4.1.3. Comparison to rural forests Comparing our findings with rural forests within the same humid subtropical climate zone helps us understand the relative importance of urban forests in reducing throughfall. Similar vegetation structure assessed through canopy cover, functional groups, and LAI among other variables should yield similar throughfall percentages. Regardless of having higher canopy cover of 67.0%, the urban residential forest in this study produced only 4% less stormbased throughfall and 6.9% more cumulative throughfall of gross precipitation than a broadleaved deciduous rural forest (TH = 82.0%) in the same climate zone with lower canopy cover of 52% (Bryant et al., 2005, Table 4). Bryant et al. (2005) recorded 724.8 mm of precipitation at their hardwood study site near Columbia, southwest Georgia, between April 2001 and June 2002. The authors report average annual precipitation of 830 mm which is considerably less than the typical averages for Colombia, GA (1200 mm; http://average-rainfall-cities.findthedata.org/l/76/Columbus-Ga) and Raleigh, NC (1100 mm; http://www.climate-zone.com/climate/ united-states/north-carolina/raleigh/). These unusually low precipitation averages measured by Bryant et al. (2005), resulting in underestimated throughfall percentages, are concluded to be unrepresentative of the climate zone and thus not directly comparable to this study. For more detailed comparisons with rural forests with regard to vegetation structure, we will focus on subpopulations, i.e. neighborhoods with similar canopy cover and functional vegetation groups. In a study of a rural deciduous broadleaved forest in Japan, Deguchi, Hattori, & Park, 2006 measured 77.1% of TH (Table 4). LAI was reported to be 3.1 during the study period, thus higher than the average of 1.9 found at our urban study site. Neighborhood one with 2.2 LAI and 72.6% deciduous broadleaved tree cover produced 88.0% of cumulative TH. Storm-based TH was 76.2%, very similar to 77.1% by Deguchi et al. (2006). Cumulative TH was 10.9% higher than that measured by Deguchi et al. (2006). However, considering the lower LAI in our urban forest stand, it is difficult to determine whether the difference is real or caused by differences in vegetation structure. 4.2. Factors controlling the performance of vegetation We found some seasonal changes in vegetation structure but the changes were relatively small due to the restricted study period from August to mid-November, 2010. The average % TH (C, S) decreased moderately towards the end of the study. We believe this effect to be caused by seasonal rainfall patterns and meteorological conditions that may obscure changes caused by the decreasing

100

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

Table 4 Relevant urban and rural studies. Land use (scale)

Climate (species/functional type)

Canopy cover (%)

LAI

P (mm)

TH (%)

I (%)

Author

Urban (stand) Urban (stand) Urban (stand) Urban (stand) Urban (stand) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Urban (crown) Rural (stand) Rural (stand) Rural (stand) Rural (stand)

Humid subtropical (broad-leaved deciduous) Humid subtropical (broad-leaved deciduous) Humid subtropical (deciduous) Mediterranean (broad-leaved deciduous d ) Mediterranean (broad-leaved evergreend ) Mild oceanic (Pseudotsuga menziesii) Mild oceanic (Thuja plicata) Semiarid (Ficus benjamina) Mediterranean (Pyrus calleryana) Mediterranean (Quercus suber) Mediterranean (Jacaranda mimosifolia) Mediterranean (Tristania conferta) Mediterranean (Ginkgo biloba) Mediterranean (Liquidambar styraciflua) Mediterranean (Citrus limon) Humid subtropical (mixed hardwood-conifer) Humid subtropical (deciduous broadleaved) Humid subtropical (deciduous broad-leaved) Humid subtropical

67.0 67.0 22.0c na na na na na na na na na na na na 74.0 52.0 na na

1.9 1.9 4.3 na na na na na 7.0 3.4 na na 5.2 4.7 3.0 na na 3.1 na

340.2 340.2 1029 393.2 433.2 377.0 377.0 152.0 441.0 700.0 570.0 570.0 728.2 728.2 728.2 684.9 724.8 3857.2 4934.8

88.9a 78.1b na 93.4 86.4 50.1 46.2 38.1 77.0 58.0 na na 73.8 81.6 70.9 80.9a 82.0 77.1 90.0

10.6–9.1a 21.4–19.9b 18.4 6.0 13.0 49.1 60.9 59.5 15.0 27.0 15.3 66.5 25.2 14.3 27.0 18.6a 17.4 16.8 10.0

This study This study Wang et al. (2008) Xiao et al. (1998) Xiao et al. (1998) Asadian and Weiler (2009) Asadian and Weiler (2009) Guevara-Escobar et al. (2007) Xiao et al. (2000b) Xiao et al. (2000b) Xiao and McPherson (2002) Xiao and McPherson (2002) Xiao and McPherson (2011) Xiao and McPherson (2011) Xiao and McPherson (2011) Bryant et al. (2005) Bryant et al. (2005) Deguchi et al. (2006) Lin et al. (2000)

a b c d

% Cumulative TH and I. % Storm-based TH and I. Based on the 5% of tree cover on impervious and 17% or tree cover on pervious land cover reported by Wang et al. (2008). Dominance by leaf surface area.

leaf area. Fall and winter in the study area are characterized by smaller storms that occur generally at longer intervals compared to larger high intensity convective storms coupled with higher relative humidity during summers (Table 2). These seasonal changes can greatly affect the initial wetness of the canopy and the rate of evaporation from the wet canopy (Gash, 1979; Rutter et al., 1971), and consequently, total percentage of TH. The strong linear relationship found between P and TH shows that the amount of TH produced depends profoundly on storm magnitude. The shared effect of P and CC was also found influential in our model (p < 0.0001), demonstrating how the performance of a canopy in reducing TH depends on both the incident precipitation and % CC. The frequency of re-wetting cycles has been highlighted as the most important factor determining the amount of rainfall intercepted (David et al., 2006; Zeng et al., 2000) due to its effect on the initial wetness of the canopy. Indeed, we also found that the initial wetness of the canopy, represented by the index F of time passed since the previous storm, was one of the most influential predictors of TH generation (p = 0.0054). Specifically, a dry canopy in the beginning of a storm produced on average one mm less TH compared to a wet canopy. We measured CC twice during the study period: in early October and December. The first storm data analyzed is from August 12. In order to account for all seasonal changes in vegetation cover between August and December it would have been ideal to take the first measurements in July but this was not possible for us in the beginning of the study. This did not seem to affect our model though as CC performed well in predicting the generation of TH. Vegetation cover in the study area remained relatively high until late fall in 2010 which allowed for the application of vegetation data measured in early October to TH data measured in mid-August. However, it is likely that having more data on seasonal changes in % CC would have further increased the performance of the model in estimating TH and is recommended for future studies. 4.3. Influential vegetation variables We aimed to determine the most influential vegetation variables affecting the generation of throughfall in the urban residential yards of various levels of canopy cover and different vegetation

types. Our results indicate that canopy cover is the single most important vegetation variable in explaining % TH and was therefore included in the linear model estimating throughfall (p < 0.0001). The actual influence in throughfall was moderate, as every additional unit of canopy cover (%) decreased throughfall by 0.05 mm. The % CON was selected for the linear model (p = 0.017) whereas other functional plant groups were not found influential in predicting TH in the urban forest. Considering the timing of this study late in the growing season, the influence of functional groups may have been skewed in favor of conifers. However, such an effect would also have favored evergreen broadleaves. Also, our results agree well with findings of other researchers, as coniferous trees are known to provide higher interception rates than broadleaves (Crockford & Richardson, 1990; Link et al., 2004; Llorens et al., 1997; Llorens & Domingo, 2007). VSC index had the second highest correlation with % TH but our regression analyses found neither VSC index (p = 0.1361) nor LAI (p = 0.8101) influential. The remaining variables (i.e. the number of trees per ha and the distance to trees and buildings in relation to measuring point) had either a negligible or no influence on the amount of TH generated. Our results indicated that CC is a more influential variable in predicting throughfall than LAI. CC may be a more robust predictor compared to LAI whose accurate measurement is especially challenging in urban forests with high within-stand heterogeneity. The successful use of LAI in studies of isolated urban trees (Xiao & McPherson, 2011; Xiao et al., 2000b) suggests that LAI may provide a better estimate of TH production at the crown-level compared to stand-level studies. Some previous studies in rural forests have also found that significant changes in LAI did not result in expected changes in throughfall proportion (Bellot, 1998; Deguchi et al., 2006), agreeing with our results. We tested performance of the VSC index for estimating TH in this study. Even though VSC index was not selected for the model, we found that the variable had a stronger correlation with % TH than LAI, a previously established variable for predicting TH. More research is needed to determine whether VSC may in some cases be a better indicator of TH generation than LAI. According to our experience, the visual estimation of VSC became more challenging in dense forests where it is difficult to obtain a full view of vertical

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

arrangement of branches. Thus, we recommend this method for sparse forests such as managed urban parks with good visibility. 4.4. Landscape preferences and throughfall Our final goal was to evaluate the significance of people’s landscape choices in affecting the amount of TH, i.e. potential stormwater runoff, produced in the study area. The 16 yards representing low-intensity residential areas in Raleigh differed significantly with regard to CC, LAI, the quantity of trees, and dominating functional plant groups. Differences were also found among neighborhoods and between front and backyards. Backyards are less visible to the street, which may partly explain why they tended to have higher CC, LAI, and more trees per ha. The differences in vegetation structure between front and backyards resulted in moderate reductions (4.1%) in TH in backyards. N1 and N2 consisted of privately-owned, relatively large properties with considerably higher CC, LAI, VSC, share of coniferous tree cover, and number of trees per ha compared to N3, consisting of small and mostly rented properties. These factors are likely explanations for the similar % storm-based TH produced in N1 (76.2%) and N2 (74.2%) and considerably higher amounts produced in N3 (86.9%). Our regression and correlation analyses showed that the combination of the above-mentioned characteristics resulted in lower TH for the larger, privately owned properties. Larsen & Harlan (2006) suggested landscape preferences as the main driver of vegetation structure in residential areas. Specifically, these preferences have been found to guide the management choices between front and backyards. In this study, front yards facing the street were typically kept relatively open with few, often planted, trees and shrubs, and managed lawn. The preferred design for front yards may reflect the type of neighborhood in question. Front yards in a neighborhood tend to resemble one other, so departing from common standards may be perceived negatively by neighbors (Nassauer et al., 2009; Zmyslony & Gagnon, 2000). The more remote backyards often consisted of more forest-like plant assemblages. These “natural-like” forest patches were sometimes mixed with planted exotic species, reflecting the variety in vegetation structure, characteristic of urban residential forests. Management of backyards has been found to reflect people’s personal preferences and choices to a greater extent than front yards do (Larsen & Harlan, 2006). In the study area, this was realized as more variable functions and activities for backyards compared to front yards. Three of 16 backyards had extensive vegetable gardens that require irrigation and fertilization. Some backyards had fruit trees that often need regular pruning. Four home-owners had pets that were always kept in backyards. One home-owner mentioned removing unwanted shrub cover by the adjacent, sloped stream bank because the vegetation was considered to look unattractive and to complicate walking across the site. These factors may have considerable local effects on the potential of vegetation in reducing throughfall and stormwater runoff and should be further investigated. Although we did not collect qualitative data on landscape preferences and their effects on the management of vegetation, we were able to show that ownership structure is a major driver of vegetation characteristics in neighborhoods. We also showed that there were significant differences between front and back yards in landscaping patterns, with some statistically significant influence on the quantity of throughfall. 4.5. Management strategies for regulating throughfall An actual example of successful vegetation structure for regulating throughfall in the study area is provided by yards 4 and 10. The specific vegetation characteristics in these yards include CC of 80.2 ± 16.9% and 85.5 ± 9.6%, respectively. Of this, 44.2 ± 32.1%,

101

and 50.5 ± 27.2% was formed by coniferous canopy, respectively. Relatively high LAI of 2.2 ± 1.1 and 2.2 ± 0.7 was measured in these yards. High VSC was found in the canopy, described by the indices of 52.2 ± 24.5 and 41.8 ± 21.2 in yards 4 and 10. In addition, 755 ± 399 and 734 ± 433 trees were recorded per ha in these yards, more than ten times the quantity found in yard 5 with 71 ± 36 trees per hectare. Yards 4 and 10 produced 84.0 and 83.2% of cumulative TH, compared to the average of 88.9%, and 67.7 ± 28.5 and 63.3 ± 33.2% of storm-based TH compared to 78.1% for the whole study area. It is important to clarify here that stormwater damages created by frequent, large summer storms, characteristic of the climate zone, cannot be sufficiently mitigated by increasing forest cover. Based on our linear regression model, a storm of 55 mm falling on canopy coverage of 67.0% with 22.5% coniferous tree cover (study area average), would produce 89.1–91.0% TH falling on dry or wet canopy, respectively. Increasing forest cover to 85% (+18%) and the share of coniferous trees to 50% (+27.5%) would produce 85.2–87.1% TH on dry or wet canopy, respectively, that is, 3.7–1.8% less TH than the urban forest currently in the study area. Thus, at the landscape-level, measures to regulate throughfall by increasing forest cover may offer relatively small additional reductions in throughfall and potential stormwater runoff. Additionally, stormwater damages are directly linked to impervious areas which tend to be less extensive in low-intensity residential areas with relatively high canopy cover, such as Beaverdam Creek Watershed. Throughfall regulation by residential forests may nevertheless offer local solutions for urban stormwater mitigation. Residents may contribute e.g. by enhancing beneficial forest characteristics for throughfall regulation above impervious surfaces, such as pavements and rooftops, and by preserving vegetation cover especially close to stream banks. The reductions can be further enhanced by promoting multiple canopy layers of preferably evergreen shrub cover beneath the tree canopy. Residents who prefer open front yards may preserve forest-like vegetation in more remote sections of backyards. Our results from a low-intensity residential area in Raleigh suggest that neighborhoods consisting of rental properties tend to have lower canopy cover consisting of less beneficial vegetation for throughfall regulation compared to privately-owned yards. Further studies should be directed at examining the underlying societal factors driving residents’ landscape designs and local throughfall generation. These factors influence urban hydrology and need to be considered in revising cities’ stormwater strategies and legislations.

5. Conclusions Our results from a low-intensity residential area in the humid subtropical climate indicate that vegetation has a significant influence on the regulation of throughfall and potential stormwater runoff. Results also show that residents can considerably affect the process. The performance of vegetation in reducing throughfall depends profoundly on the magnitude of the incident storm and the frequency of storms. Canopy cover was found to be more influential in predicting throughfall than LAI, whose significant change from summer to fall period failed to considerably influence throughfall in our models. Our results compare well with most urban stand-scale studies that have been based on actual observations. Comparisons with rural studies in the same climate zone were harder to interpret due to differences in precipitation patterns and vegetation structure. The small number of comparable studies in both urban and rural settings hindered the thorough evaluation of our results against

102

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103

other studies. More urban stand-scale studies in different climate zones are needed to understand fully the potential role of urban forests in stormwater regulation. Significant variability was found among yards and between front and backyards, indicating that residents’ landscape preferences in fact influence the amount of throughfall generated. Ownership structure also had a significant effect on vegetation structure and throughfall generation at the neighborhood scale. According to our calculations, increasing forest cover and the amount of coniferous trees would provide a moderate reduction in throughfall and potential stormwater runoff. Acknowledgements This study was conducted as master’s thesis research at the North Carolina State University and the University of Helsinki. We want to thank all colleagues who contributed to this work, especially the Urban Ecology Lab Group and Drs. John Frampton, Fikret Isik, George Hess, and Tim Albaugh. We owe our gratitude to the generous residents of Beaverdam Creek Watershed and the Atlantis Program for making this research physically and financially possible. Finally, we thank the National Science Foundation for their support of this research through the Triangle, NC, Urban Long Term Research Area—exploratory award (BCS-0948229). References Asadian, Y., & Weiler, M. (2009). A new approach in measuring rainfall interception by urban trees in coastal British Columbia. Water Quality Research Journal of Canada, 44(1) Barbour, M. G., Burk, J. H., & Pitts, W. D. (1980). Terrestrial plant ecology. Menlo Park, CA: Benjamin/Cummings. Bellot, J. (1998). Stemflow and throughfall determination in a resprouted Mediterranean holm-oak forest. Annales des Sciences Forestières (0003-4312), 55(7), 847. Bryant, M. L., Bhat, S., & Jacobs, J. M. (2005). Measurements and modeling of throughfall variability for five forest communities in the southeastern US. Journal of Hydrology, 312(2005), 95–108. Calder, I. (1996). Dependence of rainfall interception on drop size: 1. Development of the two-layer stochastic model. Journal of Hydrology, 185, 363–378. Cappiella, K., Wright, T., Schuler, T., 2005. Urban watershed forestry manual. Part 1: Methods for increasing forest cover in a watershed. Center for Watershed Protection. Crockford, R. H., & Richardson, D. P. (1990). Partitioning of rainfall in a eucalypt forest and pine plantation in southeastern Australia: II stemflow and factors affecting stemflow in a dry sclerophyll eucalypt forest and a Pinus radiata plantation. Hydrological Processes, 4, 145–155. Crockford, R. H., & Richardson, D. P. (2000). Partitioning of rainfall into throughfall, stemflow and interception: effect of forest type, ground cover and climate. Hydrological Processes, 14(16–17), 2903–2920. Cunningham, M. A., O’Reilly, C. M., Menking, K. M., Gillikin, D. P., Smith, K. C., Foley, C. M., Belli, S. L., Pregnall, A. M., Schlessman, M. A., & Batur, P. (2009). The suburban stream syndrome: evaluating land use and stream impairments in the Suburbs. Physical Geography, 30(3), 269–284. David, J. S., Valente, F., & Gash, J. H. C. (2005). Evaporation of intercepted rainfall. In M. G. Anderson (Ed.), Encyclopedia of hydrological sciences (pp. 627–634). Chichester: John Wiley. David, T. S., Gash, J. H. C., Valente, F., Pereira, J. S., Ferreira, M. I., & David, J. S. (2006). Rainfall interception by an isolated evergreen oak tree in a Mediterranean savannah. Hydrological Processes, 20, 2713–2726. Deguchi, A., Hattori, S., & Park, H. T. (2006). The influence of seasonal changes in canopy structure on interception loss: application of the revised Gash model. Journal of Hydrology, 318, 80–102. Gash, J. H. C. (1979). An analytical model of rainfall interception by forests. Quarterly Journal of the Royal Meteorological Society, 105, 43–55. Gash, J. H. C., Lloyd, C. R., & Lachaud, G. (1995). Estimating sparse forest rainfall interception with an analytical model. Journal of Hydrology, 170, 79–86. Guevara-Escobar, A., González-Sosa, E., Véliz-Chávez, C., Ventura-Ramos, E., & Ramos-Salinas, M. (2007). Rainfall interception and distribution patterns of gross precipitation around an isolated Ficus benjamina tree in an urban area. Journal of Hydrology, 333, 532–541. Hewlett, J. D. (1982). Principles of forest hydrology. Athens, GA: University of Georgia Press. Horton, R. E. (1919). Rainfall interception. Monthly Weather Review, 47, 603–623.

Jarvis, P. G., & Leverenz, J. W. (1983). Productivity of temperate, deciduous and evergreen forests. In O. L. Lange, P. S. Nobel, C. B. Osmond, & H. Ziegler (Eds.), Physiological plant ecology IV Encyclopedia in plant physiology, NS (12D) (pp. 233–280). Berlin: Springer. JMP® 8 Software | JMP 01.12.2011 http://www.jmp.com/. Larsen, L., & Harlan, S. L. (2006). Desert dreamscapes: residential landscape preference and behavior. Landscape and Urban Planning, 78, 85–100. Lemmon, P. E. (1956). A spherical densiometer for estimating forest overstory density. Forest Science, 2, 314–320. Levia, D. F., Jr., & Frost, E. E. (2006). Variability of throughfall volume and solute inputs in wooded ecosystems. Progress in Physical Geography, 30(5), 605–632. Lin, T.-C., Hamburg, S. P., King, H.-B., & Hsia, Y.-J. (2000). Throughfall patterns in a subtropical rain forest of Northeastern Taiwan. Journal of Environmental Quality, 29, 1186–1193. Link, T. E., Unsworth, M., & Marks, D. (2004). The dynamics of rainfall interception by a seasonal temperate rainforest. Agricultural and Forest Meteorology, 124, 171–191. Llorens, P., Poch, R., Latron, J., & Gallart, F. (1997). Rainfall interception by a Pinus sylvestris forest patch overgrown in a—Mediterranean mountainous abandoned area. 1. Monitoring design and results down to the event scale. Journal of Hydrology, 199, 331–345. Llorens, P., & Domingo, F. (2007). Rainfall partitioning by vegetation under Mediterranean conditions: a review of studies in Europe. Journal of Hydrology, 335, 37–54. McElhinny, C., Gibbons, P., Brack, C., & Bauhus, J. (2005). Review: forest and woodland stand structural complexity: its definition and measurement. Forest Ecology and Management, 218, 1–24. McPherson, E. G. (1998). Development, structure and sustainability of Sacramento’s urban forest. Journal of Arboriculture, 24(4), 174–190. Miller, R. W. (1997). Urban forestry: planning and managing urban greenspaces (2nd ed., pp. p27). New Jersey: Prentice Hall. Muzylo, A., Llorens, P., Valente, F., Keizer, J. J., Domingo, F., & Gash, J. H. C. (2009). A review of rainfall interception modeling. Journal of Hydrology, 370(1–4), 191–206. Nassauer, J. I., Wang, Z., & Dayrell, E. (2009). What will the neighbors think? Cultural norms and ecological design. Landscape and Urban Planning, 92(3–4), 282–292. http://dx.doi.org/10.1016/j.landurbplan.2009.05.010 NOAA (2011). NOWData—NOAA Online Weather Data.http://www.nws.noaa.gov/ climate/xmacis.php?wfo=rah (accessed on 10.10.2011). Nowak, D. J., Crane, D. E., Stevens, J. C., Hoehn, R. E., Walton, J. T., & Bond, J. (2008). A ground-based method of assessing urban forest structure and ecosystem services. Arboriculture and Urban Forestry, 34(6), 347–358. Pypker, T. G., Unsworth, M. H., & Bond, B. J. (2006). The role of epiphytes in rainfall interception by forests in the Pacific Northwest. II. Field measurements at the branch and canopy scale. Canadian Journal of Forest Research, 36, p825. Rutter, A. J., Kershaw, K. A., Robins, P. C., & Morton, A. J. (1971). A predictive model of rainfall interception in forests. 1. Derivation of the model from observations in a plantation of Corsican Pine. Agricultural Meteorology, 9, 367–384. SAS® 9.1 Software | SAS 12/01/2011 http://www.sas.com/software/sas9/. Shannon, C. E. (1948). A mathematical theory of communication. The Bell System Technical Journal, 27, 379–423. Shinohara, Y., Onozawa, Y., Chiwa, M., Kume, T., Komatsu, H., & Otsuki, K. (2010). Spatial variations in throughfall in a Moso bamboo forest: sampling design for the estimates of stand-scale throughfall. Hydrological Processes, 24, p255. Southeast Regional Climate Center, (2007). Relative Humidity (%) for Selected Cities in the Southeast. http://www.sercc.com/climateinfo/historical/avgrh.html (accessed on 5.11.2011). Stamps, A. E. (1999). Demographic effects in environmental aesthetics: a metaanalysis. Journal of Planning Literature, 14(2), 155–175. Unpublished data. (2013). Urban Morphology drives the homogenization of urban tree cover in Baltimore, MD and Raleigh, NC. Manuscript submitted for publication. USGS. (2009). 1-Arc Second National Elevation Dataset. U.S. Geological Survey USGS, Sioux Falls, SD. http://seamless.usgs.gov (accessed 10.10.2010). United States Census Bureau. (2011). http://censusviewer.com/city/NC/Raleigh van Dijk, A., & Bruijnzeel, L. (2001). Modelling rainfall interception by vegetation of variable density using an adapted analytical model. Part 2. Model validation for a tropical upland mixed cropping system. Journal of Hydrology, 247, 239–262. Wang, J., Endreny, T. A., & Nowak, D. J. (2008). Mechanistic simulation of tree effects in an urban water balance model. Journal of The American Water Resources Association, 44(1), 75–85. Weijters, M. J., Janse, J. H., Alkemade, R., & Verhoeven, J. T. A. (2009). Quantifying the effect of catchment land use and water nutrient concentrations on freshwater river and stream biodiversity. Aquatic Conservation (1052–7613), 19(1), p104. Xiao, Q. F., & McPherson, E. G. (2002). Rainfall interception by Santa Monica’s municipal urban forest. Urban Ecosystems, 6, 291–302. Xiao, Q., & McPherson, E. G. (2011). Rainfall interception of three trees in Oakland, California. Urban Ecosystems, 14(4), 755–769. http://dx.doi.org/ 10.1007/s11252-011-0192-5

E.N.M. Inkiläinen et al. / Landscape and Urban Planning 119 (2013) 91–103 Xiao, Q. F., McPherson, E. G., Simpson, J. R., & Ustin, S. L. (1998). Rainfall interception by Sacramento’s urban forest. Journal of Arboriculture, 24, 235–244. Xiao, Q. F., McPherson, E. G., Ustin, S. L., & Grismer, M. E. (2000). A new approach to modeling tree rainfall interception. Journal Geophysical Research, 105, 29173–29188. Xiao, Q. F., McPherson, E. G., Ustin, S. L., Grismer, M. E., & Simpson, J. R. (2000). Winter rainfall interception by two mature open-grown trees in Davis, California. Hydrological Processes, 14, 763–784.

103

Yu, K. (1994). Cultural variations in landscape preference: comparisons among Chinese sub-groups and Western design experts. Landscape and Urban Planning, 32, 107–126. Zeng, N., Shuttleworth, J., & Gash, J. H. C. (2000). Influence of temporal variability of rainfall on interception loss. Part 1. Point analysis. Journal of Hydrology, 228, 228–241. Zmyslony, J., & Gagnon, D. (2000). Path analysis of spatial predictors of frontyard landscape in an anthropogenic environment. Landscape Ecology, 15(4), 357–437.