Seasonal contributions of vegetation types to suburban evapotranspiration



[1] Evapotranspiration is an important term of energy and water budgets in urban areas and is responsible for multiple ecosystem services provided by urban vegetation. The spatial heterogeneity of urban surface types with different seasonal water use patterns (e.g., trees and turfgrass lawns) complicates efforts to predict and manage urban evapotranspiration rates, necessitating a surface type, or component-based, approach. In a suburban neighborhood of Minneapolis–Saint Paul, Minnesota, United States, we simultaneously measured ecosystem evapotranspiration and its main component fluxes using eddy covariance and heat dissipation sap flux techniques to assess the relative contribution of plant functional types (evergreen needleleaf tree, deciduous broadleaf tree, cool season turfgrass) to seasonal and spatial variations in evapotranspiration. Component-based evapotranspiration estimates agreed well with measured water vapor fluxes, although the imbalance between methods varied seasonally from a 20% overestimate in spring to an 11% underestimate in summer. Turfgrasses represented the largest contribution to annual evapotranspiration in recreational and residential land use types (87% and 64%, respectively), followed by trees (10% and 31%, respectively), with the relative contribution of plant functional types dependent on their fractional cover and daily water use. Recreational areas had higher annual evapotranspiration than residential areas (467 versus 324 mm yr−1, respectively) and altered seasonal patterns of evapotranspiration due to greater turfgrass cover (74% versus 34%, respectively). Our results suggest that plant functional types capture much of the variability required to predict the seasonal patterns of evapotranspiration among cities, as well as differences in evapotranspiration that could result from changes in climate, land use, or vegetation composition.

1. Introduction

[2] Evapotranspiration is an important term of energy and water budgets in urban areas and it consequently plays a key role in management decisions related to conservation of water resources, mitigation of urban heat islands, and reduction of storm water runoff. Managing these ecosystem services requires accurate assessments of urban evapotranspiration rates, and indeed water vapor fluxes have been measured in cities representing different climates, geographical regions, and land use types [Balogun et al., 2009; Grimmond and Oke, 1999; Moriwaki and Kanda, 2004; Nemitz et al., 2002; Offerle et al., 2006; Spronken-Smith, 2002]. Collectively, these studies show that cities vary in magnitude and seasonality of evapotranspiration due to differences in climate, irrigation, and vegetation cover. For example, densely built-up city centers or downtown areas with high impervious surface cover tend to have lower evapotranspiration rates and smaller evaporative fractions [Grimmond et al., 2004; Nemitz et al., 2002] than residential areas with short buildings and high vegetation cover [Balogun et al., 2009; Grimmond et al., 1996, 2006; Moriwaki and Kanda, 2004].

[3] Recent studies suggest that plant growth form is another important factor controlling total evapotranspiration in urban areas, with turfgrass-dominated areas showing higher water fluxes than tree-dominated areas [Balogun et al., 2009; Kotani and Sugita, 2005; Offerle et al., 2006]. The spatial heterogeneity of urban landscapes, however, greatly complicates the extrapolation of these ecosystem-scale measurements to other urban and suburban areas, as vegetation composition varies widely within and between cities. The management of urban ecosystem services necessitates an approach that quantifies urban evapotranspiration rates according to the major land-surface types present, hereafter referred to as a component-based approach. Partitioning of ecosystem water fluxes among stand types or species has been previously investigated in natural ecosystems [Baldocchi et al., 2000; Oishi et al., 2008; Paco et al., 2009], yet this information is relatively lacking in urban ecosystems.

[4] Plant functional types provide a potentially useful approach for partitioning the seasonal patterns of urban evapotranspiration, as they represent major physiological and biophysical differences among plant species [Reich et al., 1997] and they also can be uniquely identified in urban landscapes using high-resolution satellite imagery [Tooke et al., 2009]. For example, evergreen needleleaf trees tend to have lower leaf-level transpiration rates than deciduous broadleaf trees [Givnish, 2002], while both tree types tend to have deeper roots and access to additional sources of water as compared to cool season grasses [Jackson et al., 1996; Ludwig et al., 2004]. Plant functional types also represent distinct seasonal patterns in physiological activity, or phenology, among plant species. For example, cool season turfgrasses, which are ubiquitous across temperate urban landscapes [Milesi et al., 2005], have higher physiological activity in the spring and fall [Zhang et al., 2007], while evergreen needleleaf and deciduous broadleaf trees show peak function in midsummer [Givnish, 2002]. Many studies have quantified evapotranspiration rates of turfgrasses [Feldhake et al., 1983, 1984; Zhang et al., 2007] and, to a lesser extent, of trees in urban areas [Bush et al., 2008; Whitlow and Bassuk, 1988; Whitlow et al., 1992], yet no work has examined the combined effects of these plant functional types on the seasonality of urban evapotranspiration.

[5] Here we report a 2 year study in which we made independent measurements of total ecosystem evapotranspiration and its main component fluxes to better understand how vegetation influences the spatial and seasonal variation in suburban evapotranspiration. We measured ecosystem evapotranspiration using an eddy covariance system mounted 40 m above ground, tree transpiration using heat dissipation sap flow sensors, and turfgrass evapotranspiration using a portable eddy covariance system mounted 1.35 m above ground. Our main objectives were to: (1) determine the magnitude and seasonal patterns of suburban evapotranspiration; (2) evaluate how well scaled component-based estimates match measured ecosystem evapotranspiration rates; (3) determine how the magnitude and seasonality of ecosystem evapotranspiration varies between residential and recreational land use types; and (4) examine the seasonal water use patterns of vegetation types and their influence on ecosystem evapotranspiration in recreational and residential land use types.

2. Methods

2.1. Site Information

[6] Our study was conducted in a first-ring suburban neighborhood in the Minneapolis–Saint Paul metropolitan area in east central Minnesota, United States (44°59′N, 93°11′W). Prior to rapid residential development in the 1950s, the prominent land use types in the area were farms and nurseries. The area has a cold temperate climate with mean annual temperature of 7.4°C and mean annual precipitation of 747 mm. Due to its location near the center of the metropolitan region, the study area was influenced by the urban heat island effect that has been documented for Minneapolis–Saint Paul [Sen Roy and Yuan, 2009; Todhunter, 1996; Winkler et al., 1981]. The relatively flat terrain made this landscape suitable for eddy covariance measurements from the KUOM broadcast tower located in the center of our approximately 7 ha study area (Figure 1). The area immediately adjacent to the tower consisted of residential land use (single-family, detached housing) to the northwest and northeast and recreational land use (golf course) to the southwest. Vegetated surfaces consisted primarily of forested patches, isolated trees, and open turfgrass lawns. Of the over 80 tree species identified in this area, the dominant canopy species were Acer negundo (boxelder), Acer saccharinum (silver maple), Fraxinus pennsylvanica (green ash), Gleditsia tricanthos (honey locust), Picea glauca (white spruce), Populus deltoides (cottonwood), Quercus alba (white oak), and Ulmus americana (American elm), with a mean tree height of 12 m.

Figure 1.

Land cover classification derived from 2.4 m resolution QuickBird imagery for a 1 km radius surrounding the KUOM tall tower (indicated by the letter K at the origin) in a suburban neighborhood of Minneapolis–Saint Paul, Minnesota. The mobile tower in the turfgrass field is indicated by the letter T. The Lauderdale (L), Saint Paul (S), and Grove (G) sap flux sites are indicated by the letters corresponding to their names. The CTC sap flux site was located in a residential area 1.3 km south of the map edge and is not shown. The white lines represent isopleths of the average relative contributions of different land areas, based on the footprint model of Kljun et al. [2004], of the measured fluxes at the 40 m level of the tall tower for the years 2007 and 2008, following the approach presented by Rebmann et al. [2005]. Yellow lines delineate the residential and recreational land use areas that were used to bin the data for the analyses in Figure 8 and section 3.6. Stippled areas in the SE corner of the map indicate low herbaceous vegetation in experimental crop plots.

[7] We selected four sites with stands of trees suitable for sap flux measurements within our study area (Figure 1) [Peters et al., 2010]. Sites were selected to include representative tree species and sizes with the total number of sites limited by numerous logistical constraints (e.g., equipment, landowner permission, site suitability, signal cable lengths, and access to power). Trees at all sites were grown under park-like conditions with open canopies and a turfgrass ground cover. The lower branches of many trees were pruned, but canopies were mature and healthy with no obvious signs of disease or damage. Root systems were not likely restricted by impervious surfaces, as the nearest roads and sidewalks occurred beyond the drip line of all trees and often at distances > 10 m from each tree. Surface soils at all sites were moderately compacted with an average bulk density of 1.43 g cm−3, as is typical of urban soils [Craul, 1985]. Management regimes were considered low maintenance at all sites and included regular mowing, but had little or no irrigation and no fertilizer use based on personal communication with the landowners.

[8] We made eddy covariance measurements at a height of 1.35 m over a 1.5 ha turfgrass field that was located within the footprint of the KUOM tall tower (Figure 1). This site was representative of low-maintenance lawns in the area, as it was mowed to a height of 7 cm approximately once per week with clippings left to decompose on the surface and was not irrigated or fertilized. The C3 cool season turfgrass species, Poa pratensis (Kentucky bluegrass), Lolium perenne (perennial ryegrass) and Festuca arundinacea (tall fescue), dominated this site. Surface soils at this site were slightly compacted with an average bulk density of 1.22 g cm−3. While turfgrass management varied among lawns in our study area, numerous logistical constraints associated with making eddy covariance measurements in an urban area (e.g., landowner permission, security, adequate fetch) restricted our ability to simultaneously measure a highly managed lawn.

2.2. Meteorological Measurements

[9] To assess the environmental conditions in which our study occurred, we measured a suite of meteorological variables. At the turfgrass site, net radiation (RN) and incoming solar radiation (RS) were measured at 2 m above ground using a four-component net radiometer (model CRN1, Kipp and Zonen, Delft, Netherlands). Ground heat flux (G) was measured at 5 cm depth at two locations at the turfgrass site using heat flux plates (HFT-3.1, Radiation and Energy Balance Systems, REBS, Seattle, Washington, United States). An integrating platinum resistance temperature probe (STP-1, REBS, Seattle, Washington, United States) was used to measure storage of heat in the upper 5 cm of soil, which was added to the measurements recorded by the heat flux plates. Volumetric soil water content was measured at 10 cm depth (ECH2O, Decagon Devices, Inc., Pullman, Washington, United States) at the turfgrass site and calibrated relative to a set of gravimetric measurements made in the same soil type at the University of Minnesota climate station. These calibrations were conducted by oven-drying soil cores of a known volume that were collected throughout a 2 week dry-down period following a precipitation event. We measured RN (CNR1, Kipp and Zonen, Delft, Netherlands) from the KUOM broadcast tower at 150 m above ground. The 50% source area [Schmid et al., 1991] of our net radiation measurement was within 150 m of the tower (consisting of tree and grass patches on the golf course, three small buildings with driveways, a two-lane road intersection, and a small storm water pond), while the 90% source area was within 475 m of the tower (consisting of a mixture of golf course land use and several blocks of single-family detached housing). Precipitation data were obtained from the University of Minnesota climate station (<1 km from the KUOM tower site). Vapor pressure deficit (D) was calculated from the water vapor concentration and air temperature measurements at the KUOM tall tower and turfgrass sites.

2.3. Measuring Suburban Evapotranspiration and Its Component Fluxes

[10] Latent (LE) and sensible (H) heat fluxes over the suburban landscape were measured from the 40 m level on the KUOM broadcast tower using an eddy covariance system consisting of a 3-D sonic anemometer (CSAT3, Campbell Scientific, Inc., Logan, Utah, United States) and a closed-path infrared gas analyzer (LI-7000, LI-Cor, Lincoln, Nebraska, United States). The measurement height was > 2× the mean height of the trees, which overtopped the roofs of the houses. The sonic anemometer and the filtered air inlet for the gas analyzer were mounted on a 3 m boom that was oriented toward the northwest (315°). The gas analyzer and the data-acquisition system were housed in an insulated, thermostatically controlled, heated and ventilated rack-mount computer enclosure at the base of the tower. The water vapor channel of the gas analyzer was calibrated using a dew point generator (LI-640, LI-Cor, Lincoln, Nebraska, United States) in a nearby laboratory, with two different LI-7000 instruments cycled regularly between the lab and the tower. Air was sampled through a 9.5 mm I.D. pure FEP tubing at a rate of 23 SLPM and a bypass flow rate of 7.5 SLPM through 4 mm I.D. pure PTFE tubing and was delivered to the gas analyzer using a system of needle valves, mass-flow meters, and two rotary vane vacuum pumps (Gast models 1023 and 211 for the main and the bypass pumps, respectively, Benton Harbor, Michigan, United States). The digital data from the sonic anemometer was transmitted to the tower base using a pair of fiber optic media converters linked by multimode fiber optic cable to avoid RF interference from the broadcast tower. Data collected with the sonic anemometer and infrared gas analyzer were recorded at 10 Hz using in-house software running on a utility-grade UNIX server at the tower base (V120, Sun Microsystems, Santa Clara, California, United States).

[11] Thirty minute turbulent fluxes, including ecosystem evapotranspiration (Etotal), were computed using a slightly modified version of eth-flux [see Mauder et al., 2008]. Following Vickers and Mahrt [1997], a filter was implemented to remove spikes in the sonic data likely caused by data communication. A maximum of five values exceeding ± 6 standard deviations within a 300 s moving window were removed. This procedure was repeated three times for each 30 min block of data. The time lag due to the transport of air through the tubing was calculated using the peak of the cross-correlation function between the vertical wind velocity and the water vapor concentration. The mean lag for the water vapor signals was 11 s. All subsequent analyses were performed using the R (version 2.8) statistical language package [R Development Core Team, 2010]. The sonic temperature was corrected following Schotanus et al. [1983]. High-frequency losses in the fluxes were corrected following Eugster and Senn [1995] using a damping-loss constant of 0.28 s−1. Data were screened to remove periods when instruments were being serviced and wind directions were from the southeast (90 to 180°) to avoid interference from the tower structure. The land use associated with those wind directions was also atypical of the predominantly residential area because it included experimental crop plots cultivated by the University of Minnesota. Gaps in measurements from January 2007 to December 2008 represented 48% of the data, due to unfavorable wind directions from behind the tower (20%) and system maintenance or power outages (28%). We performed the minimum necessary level of filtering to maximize the amount of data available for comparisons of land use types. Due to the complexity of the suburban land surface and our focus on comparing scaled-up component fluxes to measured Etotal, we did not gap-fill the tall tower data for the analyses presented here.

[12] Measured Etotal should balance against the sum of its component fluxes such that:

equation image

where ET is evapotranspiration from trees, EG is evapotranspiration from turfgrass lawns, and EW is evaporation from open water (e.g., ponds and swimming pools). The focus of the KUOM flux study was to quantify fluxes from a residential landscape, which consists of trees and predominantly nonirrigated turfgrass lawns (Table 1). In carrying out the scaling exercise reported here, we wanted to understand how much of the imbalance between our scaled-up component fluxes and Etotal measured by the tall tower could be accounted for by processes and surface types we did not measure. Thus, we used modeled estimates of canopy interception, EG from irrigated lawns, and understory EG to provide a more robust assessment of our scaled-up measurements. Each of the surface type fluxes included in our scaled-up Etotal estimates was either independently measured or modeled as described below. Evaporation from bare soil and impervious surfaces was assumed to be negligible because bare soil represented a small fraction of the study area (∼7%, much of which was under tree canopies, based on FIA plot data) and impervious surfaces effectively prevent evaporation from soils below.

Table 1. Cover Types by Percentage of the Residential Land Use Area, Recreational Land Use Area, and Flux Footprint Climatology Areaa
Suburban LandscapeDeciduous Tree (%)Evergreen Tree (%)Nonirrigated Turfgrass (%)Irrigated Turfgrass (%)Open Water (%)Impervious Surfaces (%)
  • a

    Percent cover was determined from the land cover classification map shown in Figure 1 and does not include turfgrass under tree canopies. The footprint area was defined as the total area bounded by the 10% contour line (Figure 1).

Residential area376259222
Recreational area191116325
Footprint area2541835215

[13] ET is composed of tree transpiration (TT) and rainfall interception loss from tree canopies (IT) such that:

equation image

In our study area, Peters et al. [2010] found species' differences in TT per unit canopy area were largely explained by plant functional type differences in canopy structure and growing season length. Consequently, we used evergreen needleleaf and deciduous broadleaf plant functional types to estimate ET at the ecosystem scale, such that:

equation image

where ETeg is evapotranspiration from evergreen needleleaf trees and ETdec is evapotranspiration from deciduous broadleaf trees.

[14] TT was measured using Granier-type heat dissipation sap flow [Granier, 1987; Lu et al., 2004] on a total of 37 trees (17 trees in 2007 and 20 different trees in 2008) across four sites. Trees represented seven different genera (Fraxinus, Juglans, Picea, Pinus, Quercus, Tilia, and Ulmus) and two plant functional types (evergreen needleleaf and deciduous broadleaf) [Peters et al., 2010]. Measurement periods ran from May to November in 2007 and April to November in 2008. The software package BaseLiner (version 2.4.1, Hydro-Ecology Group, Duke University) was used to calculate sap flux density per conducting sapwood area. TT per unit canopy area was calculated as the product of each tree's sap flux density and the ratio of cross-sectional sapwood area to projected canopy area. Cross-sectional sapwood area was determined from multiple cores per tree using visual estimates apparent immediately after removal from the trunk [Lu et al., 2004]. Adjustments were made for radial variation in sap flow according to Pataki et al. [2010].

[15] IT was estimated using a tree-based adaptation of the Rutter canopy interception model [Rutter et al., 1975; Valente et al., 1997]. The crown of each tree was considered a closed canopy and interception loss calculated on a canopy-area basis. Following Wang et al. [2008], we modeled crown storage capacity of both evergreen needleleaf and deciduous broadleaf trees as a function of leaf area index (LAI) with a specific leaf storage of 0.2 mm. The seasonal pattern of LAI was modeled using piecewise logistic equations fit to stand-level LAI measurements (LAI-2000, LI-Cor, Lincoln, Nebraska, United States) that were collected biweekly at sap flux sites from prior to leaf-out to after senescence in 2007 and 2008 [Peters and McFadden, 2010]. Maximum LAI values for evergreen needleleaf and deciduous broadleaf trees were determined from midsummer LAI measurements on each study tree in 2008 [Peters et al., 2010] and LAI ranges were determined from seasonal LAI patterns observed in homogeneous stands of evergreen needleleaf and deciduous broadleaf trees in 2006 [Peters and McFadden, 2010]. Modeled LAI of evergreen needleleaf trees ranged from a minimum of 7.8 m2 m−2 in winter to a maximum of 8.8 m2 m−2 in summer, while modeled LAI of deciduous broadleaf trees ranged from a minimum of 0 m2 m−2 in winter to a maximum of 5.5 m2 m−2 in summer. IT was modeled every half-hour by calculating potential evapotranspiration (EP) in kg H2O m−2 s−1 using the Priestley-Taylor equation [Priestley and Taylor, 1972]:

equation image

where α is a constant of 1.26, RN is net radiation in MJ m−2 s−1, G is ground heat flux in MJ m−2 s−1, λ is latent heat of vaporization in MJ kg−1, Δ is the slope of the saturation vapor pressure temperature curve in kPa °C−1, and γ is the psychrometric constant in kPa °C−1.

[16] Through our subsequent analyses comparing measured and component-based estimates of Etotal, we determined it was important to include irrigated and nonirrigated turfgrass vegetation types when scaling up EG to the suburban ecosystem. Because most tree canopies in our study area had a turfgrass understory, we also included two understory turfgrass vegetation types in our estimate of EG, such that:

equation image

where EGnirr is evapotranspiration from nonirrigated turfgrass lawns, EGirr is evapotranspiration from irrigated turfgrass lawns, EGunder_eg is evapotranspiration from turfgrass growing below evergreen needleleaf tree canopies, and EGunder_dec is evapotranspiration from turfgrass growing below deciduous broadleaf tree canopies.

[17] EGnirr was measured using the eddy covariance method over a 1.5 ha turfgrass field located in the footprint of the KUOM tall tower [Hiller et al., 2010]. Water vapor and sensible heat fluxes were measured using an eddy covariance system consisting of a 3-D sonic anemometer (CSAT3, Campbell Scientific, Logan, Utah, Unites States) and an open-path infrared gas analyzer (LI-7500, LI-Cor, Lincoln, Nebraska, United States) inclined at an angle of 30° toward the west (270°), both mounted at a height of 1.35 m above the ground on a portable meteorological tripod (905, Met One Instruments, Grants Pass, Oregon, United States). Wind velocity, air temperature, and scalar concentrations of water vapor were recorded at 20 Hz using a data logger (CR5000, Campbell Scientific, Logan, Utah, United States) and fluxes were computed over 30 min periods with the eth-flux software by applying a time lag, if needed, and calculating the covariance between the vertical wind speed and the water vapor concentration. No correction for self-heating of the LI-7500 [Burba et al., 2008] was applied in the water vapor flux calculations because the effect was estimated to be < 1% during cold winter periods and negligible during the growing season period. The turbulence data processing followed the same procedures as for the tall tower except that the open-path gas analyzer data were corrected for changes in air density and sensible heat following Webb et al. [1980] and high-frequency losses in the fluxes were corrected using Moore's [1986] transfer functions for line averaging and sensor separation. Gaps in measurements from January 2007 to December 2008 represented 42% of the data, due to unfavorable wind directions from behind the tower (0–180°), power outages, and unsatisfied quality criteria [Foken and Wichura, 1996]. Gaps in EGnirr fluxes were filled using multiple linear regression models of RN and air temperature with 2 week windows and separate daytime and nighttime fitted coefficients.

[18] EGirr was calculated using the Priestly Taylor equation (equation (4)) with α equal to 0.87. Also known as the crop coefficient, α ranges from 0.6 to 1.14 for nonstressed, cool season grasses [Brown et al., 2001; Carrow, 1995; Ervin and Koski, 1998; Zhang et al., 2007]. While α is known to vary seasonally, the exact patterns depend on species, soil type, and environmental conditions. In an effort to model water loss from a typical irrigated lawn in our study area, we chose a median α value among those reported in the literature.

[19] EGunder_eg and EGunder_dec were modeled as linear functions of the solar radiation passing through each tree canopy [Eastham and Rose, 1988; Feldhake et al., 1985]. Radiation passing through the canopy (RV) in W m−2 was calculated using the Beer-Lambert law [Campbell and Norman, 1998] such that:

equation image

where RS is incident solar radiation at the top of the canopy in W m−2 and k is an attenuation coefficient. LAI was modeled separately for evergreen needleleaf and deciduous broadleaf trees as described above. Following another urban study [Wang et al., 2008], we set k equal to 0.7 for both tree types. Calculated RV was used to predict EGunder_eg and EGunder_dec from monthly linear regression models of measured EGnirr and RS (R2 ranged from 0.76 to 0.94 across the 2007 and 2008 growing seasons). EW, which included a few small ponds, water traps on the golf course, and residential swimming pools, was calculated using the Priestly Taylor equation (equation (4)) with α equal to 1.26.

2.4. Scaling Up Component Fluxes

[20] We assessed the land cover in the study area using satellite imagery, aerial photography, a Geographic Information System (GIS) land use database (ArcMap, version 9.3, ESRI, Redlands, California, United States), and an urban forest inventory. A land cover classification was produced using QuickBird (2.4 m resolution) multispectral imagery acquired on 26 July 2006 and leaf-off, true color aerial imagery (0.15 m resolution) acquired by Ramsey County on 8 April 2003. Preparation of the QuickBird imagery included orthorectification using a 10 m digital elevation model, georectification using the Ramsey County survey control network, masking of water bodies using a GIS data layer, and normalized difference vegetation index (NDVI) thresholding to isolate the vegetated fraction of the image. The vegetation classes were then extracted using the following steps. First, an unsupervised isodata classification was used to separate areas covered by trees, turfgrass, and shadow. Second, the shadowed area was masked and reclassified using a supervised classifier. Third, the tree class was separated into deciduous broadleaf and evergreen needleleaf tree classes using a green and red normalized band ratio from the leaf-off color aerial imagery. Fourth, irrigated and nonirrigated turfgrass areas were separated using a minimum-distance supervised classification. Fifth, swimming pools were classified using a simple ratio of the blue to the near infrared bands of the QuickBird imagery, within the residential land use area. We assessed the accuracy of the land cover classification using the subplot centers of 150 randomly selected field plots that we surveyed in 2005 and 2006 using the U.S. Forest Service Forest Inventory and Analysis (FIA) urban forest protocol [U.S. Department of Agriculture, 2005]. The overall per-pixel accuracy of the classification was 82%, with similar accuracies of ∼80% for the deciduous tree, evergreen tree, and turfgrass classes, and 93% for the impervious surface class.

[21] Footprint models have been widely used to estimate the source area of eddy covariance measurements of Etotal [Kljun et al., 2004; Schmid, 1994]; however, these models assume that measurements are made over extended homogeneous surfaces and they have not been rigorously validated in spatially complex urban landscapes [Vesala et al., 2008a]. Although our study area contains considerable spatial heterogeneity in surface types, it does not violate these model assumptions as much as a densely built-up city center or downtown area because it has a relatively flat topography, a dense tree canopy that overtops most of the buildings, and lacks deep street canyons caused by tall buildings. Therefore, to allow for comparisons between measured Etotal and scaled component fluxes, we used a parameterized version of a Lagrangian stochastic footprint model [Kljun et al., 2004] to estimate the crosswind integrated footprint of each half-hourly measurement of Etotal. We ran the footprint model using a fixed boundary layer depth of 1000 m and a roughness length calculated from the eddy covariance measurements. As the Kljun et al. [2004] parameterization does not predict two-dimensional flux source areas, we used the following approach based on Barcza et al. [2009] to obtain a spatial sample of the fractional cover of land surface types within the source area. First, we used the footprint model to calculate distance from the tall tower to the maximal contribution point as well as the distances corresponding to every tenth percentile of the footprint for each half-hourly measurement of Etotal. Second, we plotted the 10 points of the flux footprint along the mean wind direction on the land cover map in a GIS and calculated the fractional cover of each component class within a 1 ha circular area (radius = 56 m) around each point. Third, we calculated the average of the 10 sampled areas, each weighted by the crosswind-integrated footprint corresponding to that point, to obtain the final estimate of the fractional land cover that contributed to each half-hourly flux measurement. To estimate the fractional cover of turfgrass underneath the tree canopy, we calculated the difference between the percent turfgrass cover from field measurements on FIA plots (which included understory vegetation) and the percent turfgrass cover derived from the satellite data (which could not detect the understory). On average, 50% of the tree-covered area had a turfgrass understory and this value was used to represent the fractional cover of understory turfgrass for the entire footprint area.

[22] For comparison with measured Etotal, component fluxes were converted from a per cover area basis to a per ground area basis by multiplying ETeg, ETdec, EGnirr, EGirr, EGunder_eg, EGunder_dec, and EW by the fractional cover estimates and summing the terms. Comparisons between measured and component-based Etotal methods were restricted to half-hour periods in 2007 and 2008 when friction velocity u* > 0.2 m s−1 and scaled Monin-Obukhov length zm/L ≥ −200 and ≤1. These direct comparisons between the summed component fluxes and measured Etotal from the tall tower were possible only on a half-hourly basis because gaps in the tall tower time series (meaning we lacked both the measured Etotal fluxes and the footprint information needed to scale up the component-based Etotal) inhibited our ability to make comparisons over longer integrations such as monthly or annual sums. We conducted an error analysis of the component-based estimates of Etotal by propagating errors associated with the land cover classification, the measured and modeled component fluxes [Taylor, 1982]. EGnirr was assumed to have a relative error of 20%, which is typical of eddy covariance measurements in general [Baldocchi et al., 1988]. EGirr was assumed to have a relative error of 16%, based on the published range of α reported for cool season turfgrass species [Brown et al., 2001; Carrow, 1995; Ervin and Koski, 1998; Zhang et al., 2007]. For both evergreen and deciduous trees, TT was assumed to have a relative error of 40%, based on the measured variation in annual sums among individual trees. IT was assumed to have a relative error of 40% and 23% for deciduous and evergreen trees, respectively, based on measured variation in LAI. Because EW, EGunder_dec, and EGunder_eg were minor components of Etotal and we did not have a strong quantitative basis for estimating their respective errors, they were excluded from this error analysis.

[23] To compare measured Etotal from a residential neighborhood and a recreational land use area in a golf course, we delineated two regions of interest (Figure 1) based on the Ramsey County GIS database. We restricted land use comparisons to periods when the 10% flux contribution fell exclusively within one of the two land use types during daytime conditions (RS > 10 W m−2). Annual Etotal from recreational and residential land use areas was determined by scaling component fluxes according to their respective land cover fractions shown in Table 1.

3. Results

3.1. Land Cover

[24] Open turfgrass lawns were the dominant land cover type in our suburban study area, representing 53% of the tall tower footprint climatology, followed by 29% tree cover, and 15% impervious surface cover (Table 1). Of the tree-covered area, 86% was composed of deciduous broadleaf species and 14% of evergreen needleleaf species. 45% of trees had a DBH < 20 cm and 74% of trees had a projected canopy area < 40 m2 (Figure 2). A turfgrass understory was present in 52% of tree-covered areas. Vegetation cover varied considerably between the two dominant land use types, such that turfgrass cover dominated the recreational area and tree cover dominated the residential area. The recreational area had 40% higher turfgrass cover, 54% more irrigated turfgrass cover, 23% less tree cover, and 13% less impervious surface cover than the residential area. The difference in percent cover of irrigated turfgrass between residential and recreational areas was confirmed with personal observations during the FIA field inventory that most homes did not have automatic irrigation systems.

Figure 2.

Probability density functions of tree (a) diameter at 1.4 m height (DBH) and (b) projected canopy area in a suburban area of Minneapolis–Saint Paul, Minnesota. Trees were sampled (n = 164) using FIA urban forest protocols and were located within the tall tower footprint area, defined as the total area bounded by the 10% contour line (Figure 1).

3.2. Environmental Conditions

[25] The seasonal patterns of RN, air temperature, and D were relatively similar between the 2 years of study, but there were several periods with higher D in the summer of 2007 compared to 2008. The seasonal patterns of precipitation and soil moisture varied considerably between 2007 and 2008 (Figure 3). Not only was total annual rainfall higher in 2007 (749 mm) and more similar to the 30 year (1971–2000) average (747 mm) than 2008 (550 mm), the seasonal distribution of rainfall varied as well [National Climatic Data Center (NCDC), 2004]. In 2007, the greatest amount of precipitation occurred in September (170 mm), whereas in 2008 it occurred in April (109 mm). August was the wettest month (103 mm) over the 30 year average [NCDC, 2004]. Consequently, rainfall patterns in 2007 were relatively more similar to the long-term average than in 2008. Air temperature and D were also higher in late spring and early summer in 2007 compared to 2008. Soil moisture reached a growing season (April–November) low of 17% on average in May of 2007 compared to a low of 11% on average in August of 2008.

Figure 3.

Environmental conditions in a suburban area of Minneapolis–Saint Paul, Minnesota from 1 January 2007 to 31 December 2008. (a) Daily totals of net radiation (RN) were measured at the 150 m level of the KUOM tower. (b) Mean daily air temperature (Tair) and mean daytime vapor pressure deficit (D) were measured at the 40 m level of the KUOM tower. (c) Soil water content (SWC) was measured at 10 cm depth at the turfgrass site. Precipitation (bars) was measured at a nearby (<1 km) climate station. Gaps in soil water content data indicate time periods when the instrumentation failed.

3.3. Ecosystem Evapotranspiration

[26] Etotal from the suburban landscape, measured from the 40 m level on the KUOM tower, varied seasonally across the 2 years of study (Figure 4). On average, daytime (RS > 10 W m−2) sums of Etotal were near zero in winter in both 2007 and 2008 and showed increased rates from April to October. In April and May, we observed higher daytime sums of Etotal on average in 2007, but in July and August we observed higher daytime Etotal on average in 2008. Daily Etotal peaked in June of 2007 at an average rate of 2.7 mm d−1 and in July of 2008 at an average rate of 3.1 mm d−1. The prevailing northwest wind direction led Etotal to be relatively more representative of residential than recreational land use areas (Figure 1). We consider this systematic bias when interpreting these results.

Figure 4.

Daytime (RS > 10 W m−2) sums of ecosystem evapotranspiration (Etotal) per month measured at the 40 m level of the KUOM tower over a suburban area of Minneapolis–Saint Paul, Minnesota. Dark lines represent median daytime sums of Etotal, the box represents the 75th and 25th percentiles, and the dotted lines represent values within a 1.5 interquartile range. n varies from 4 to 20 days per month. The gap in the fall of 2008 indicates a period when the instrumentation was out of service following a lightening strike.

[27] In midsummer (June–August), the eddy covariance system at the 40 m level of the tall tower had an energy budget closure of 83%. This imbalance did not vary with wind direction or between residential and recreational land use types. To estimate the energy budget closure at the tall tower we used G measured at the turfgrass site, which did not contain any impervious surfaces. Therefore, it is likely that G was an underestimate for the suburban landscape and this would have contributed to the energy imbalance. The average midday (1100–1500 h) evaporative fraction (LE/RN) and Bowen ratio (H/LE) were 0.40 and 0.67, respectively, in midsummer. We observed a higher midsummer energy budget closure of 94% with the 1.35 m portable eddy covariance system at the turfgrass site, and with a higher midday evaporative fraction (0.45) and lower Bowen ratio (0.58).

3.4. Component Evapotranspiration Fluxes

[28] The magnitude and seasonal patterns of evapotranspiration varied considerably among the plant functional types represented in this study. Across the 2008 growing season (April–November), average daily EGnirr was higher than either ETeg or ETdec on a per cover area basis (Figure 5a). EGnirr and ETeg were similar at the beginning and end of the growing season and both remained > 1 mm d−1 on average for at least 7 months in 2008, compared to only 4 months for ETdec. All three plant functional types showed maximum daily evapotranspiration rates in June, when day length was at its maximum. Midsummer daily evapotranspiration in 2008 also varied among the four turfgrass components represented in this study (Figure 5b) with average daily EGirr (3.10 mm d−1) higher than EGnirr (2.99 mm d−1), EGunder_dec (0.07 mm d−1), and EGunder_eg (0.01 mm d−1). The differences in evapotranspiration rates among plant functional types measured and modeled in 2007 were consistent with these results (data not shown).

Figure 5.

(a) Average daily sums of measured evapotranspiration per cover area for nonirrigated turfgrass (squares, EGnirr), evergreen needleleaf trees (triangles, ETeg), and deciduous broadleaf trees (circles, ETdec) per month across the 2008 growing season in a suburban area of Minneapolis–Saint Paul, Minnesota. EGnirr was measured using a portable eddy covariance tower, and ETeg and ETdec were calculated from heat dissipation sap flow measurements and a canopy interception model. (b) Average summertime (June–August) daily sums of evapotranspiration per cover area in 2008 for measured and modeled vegetation components. Evapotranspiration from irrigated turfgrass (EGirr) was modeled using the Priestley-Taylor equation, and evapotranspiration from understory turfgrass (EGunder_eg and EGunder_dec) was modeled as a linear function of the solar radiation passing through the tree canopy, as estimated by the Beer-Lambert law. Error bars are ± 1 standard error, and n varies from 8 to 31 days per month.

[29] IT represented 27% and 32% of annual ETdec and ETeg, respectively. On summer days with rain events, daily ETdec and ETeg were 44% and 22% higher, respectively, than on days without rain events. Although precipitation events were originally filtered out of the EGnirr measurements, gap-filled EGnirr showed a 22% decline on summer days with rain events. We were unable to separate evaporation and transpiration components of turfgrass because the eddy covariance measurements used at the turfgrass site represent only total evapotranspiration.

3.5. Comparing Component-Based and Eddy Covariance Measured Etotal

[30] Compared to component-based estimates of Etotal in which all open-canopy turfgrass was assumed to be nonirrigated (EGnirr), component-based estimates that included a modeled irrigated turfgrass (EGirr) vegetation type showed better agreement with measured Etotal, particularly in summer when winds were from the southwest (R2 = 0.66 and 0.73, respectively, Figure 6). Including EGirr reduced the summer imbalance from an underestimate of 7% to 3% in the southwest quadrant, but had little effect on the imbalance in spring and fall or when winds were from the northwest or northeast. For the remainder of our analyses, all component-based estimates of Etotal consequently included modeled EGirr.

Figure 6.

Comparison of half-hourly measured and scaled component sums of ecosystem evapotranspiration (Etotal) in a suburban area of Minneapolis–Saint Paul, Minnesota. Component sums were constructed either (a) without a separate irrigated turfgrass component (EGirr) or (b) with EGirr. Etotal was measured at the 40 m level of the KUOM tower. Data shown are from summer (June–August) in 2007 and 2008 when winds were from the southwest. Lines represent the 1:1 line.

[31] Overall, component-based estimates of Etotal underestimated measured Etotal by an average of 3%, with the component-based approach explaining 83% of the variation in measured fluxes when a linear regression model was fit through the origin (y = 1.03x, P < 0.001). Thirty minute estimates of component-based Etotal had a mean relative error of 21%. The imbalance between the two methods, however, varied seasonally with the component-based approach resulting in a 20% overestimate of measured fluxes in spring, an 11% underestimate in summer, and a 9% overestimate in fall (Figure 7). The largest overestimates of measured Etotal tended to occur when winds were from the southwest, while underestimates tended to occur when winds were from the northwest or northeast and during periods of high RN (>800 W m−2), high H fluxes (>175 W m−2), and high D (>2 kPa). Scaling up TT differences according to xylem anatomy types (conifer, ring-porous, and diffuse-porous) instead of plant functional types reduced the overall imbalance such that measured Etotal was overestimated by 1% and varied seasonally from a 25% overestimate in spring to a 6% underestimate in summer. The imbalance was also relatively unaffected by the method used to sample the land cover fractions associated with each half-hourly flux. Using only the maximum contribution point or the nonweighted mean of the 10 footprint contribution points changed the overall imbalance by −0.007% and +0.003%, respectively.

Figure 7.

Comparison of half-hourly measured and scaled component sums of ecosystem evapotranspiration (Etotal) in a suburban area of Minneapolis–Saint Paul, Minnesota during (a) spring (April and May), (b) summer (June–August), and (c) fall (September–November). Etotal was measured at the 40 m level of the KUOM tower. Component sums were constructed as ET + EG + EW. Data shown are from 2007 and 2008. Lines represent the 1:1 line.

3.6. Comparing Etotal From Two Suburban Land Use Types

[32] Measured Etotal varied in magnitude and seasonality between the two suburban land use types represented in our study (Figure 8). Recreational areas had higher average daytime Etotal than residential areas in the spring and fall, as well as during the dry period in June 2007. In addition, recreational areas had a lower average midday Bowen ratio in midsummer than residential areas (0.47 and 0.88, respectively). Daytime Etotal was relatively similar between the two land use types in late summer (August and September) and in winter.

Figure 8.

Average daytime ecosystem evapotranspiration (Etotal) per month measured over two land use types, recreational (open circles) and residential (solid circles), in a suburban area of Minneapolis–Saint Paul, Minnesota. Etotal was measured at the 40 m level of the KUOM tower. Data shown were restricted to daytime (RS > 10 W m−2) periods without precipitation. Error bars are ± 1 standard error, and n varies from 17 to 270 half-hourly data points per month.

[33] Annual sums of Etotal, which were based on the sum of scaled component fluxes, were higher on average from the recreational area (467 mm yr−1) compared to the residential area (324 mm yr−1) (Figure 9a). Despite large differences in total annual rainfall between 2007 and 2008, the interannual variation in annual Etotal was relatively small for both residential and recreational land uses. In 2007 and 2008, annual Etotal from recreational areas represented 62% and 85% of annual precipitation, respectively, while annual Etotal from residential areas represented 42% and 61% of annual precipitation, respectively. EG was the largest component of annual Etotal in both recreational and residential areas (87% and 64%, respectively), followed by ET (10% and 31%, respectively), and EW (3% and 5%, respectively). EGirr was a much larger component of annual EG in recreational than residential areas (83% versus 23%, respectively), while ETdec was the dominant component of annual ET in both land use types (95% and 83%, respectively) because deciduous trees were more abundant than evergreen trees in the suburban area we studied.

Figure 9.

(a) Annual precipitation and annual evapotranspiration of scaled component fluxes, trees (ET), turfgrass (EG), and open water (EW), from the residential (res.) and recreational (rec.) land use areas (Figure 1). ET includes evapotranspiration from evergreen needleleaf (ETeg) and deciduous broadleaf trees (ETdec). EG includes evapotranspiration from irrigated (EGirr), nonirrigated (EGnirr), and understory turfgrass (EGunder_eg and EGunder_dec). The average proportional contribution of vegetation components, ETdec (circles), ETeg (triangles), EGirr (open squares), and EGnirr (solid squares), to ecosystem evapotranspiration (Etotal) from (b) residential and (c) recreational land use areas per month. Data shown include only daytime periods (RS > 10 W m−2) in 2008, and n varies from 254 to 892 half-hourly data points per month.

[34] The contribution of different vegetation types to Etotal in 2008, as determined from the sum of scaled component fluxes, varied seasonally in both land use types (Figures 9b and 9c). EGirr had the highest proportional contribution to Etotal in July, while the maximum contribution from EGnirr occurred in spring (April and May) and fall (October and November). The largest contribution from ETdec occurred in August and September, while the highest contribution from ETeg, albeit small, was in April and November, particularly in residential areas. Data for vegetation contributions in 2007 were consistent with these results (data not shown).

4. Discussion

4.1. Suburban Evapotranspiration Rates

[35] Ecosystem evapotranspiration (Etotal) observed in this study fell within the range of values represented for other suburban residential areas around the world [Balogun et al., 2009; Grimmond and Oke, 1999, 2002; Moriwaki and Kanda, 2004; Offerle et al., 2006; Spronken-Smith, 2002; Vesala et al., 2008b]. The evaporative fraction (0.40) we observed was most similar to that measured in older suburban areas in North America, such as Chicago [Grimmond et al., 1994]. Among suburban areas previously studied in North America, the fraction of available energy used for evaporation ranges from 0.22 to 0.46, with the highest values associated with recently developed exurban areas dominated by open turfgrass lawns and trees with small canopies and the lowest values associated with older suburbs dominated by mature tree canopies [Balogun et al., 2009]. With 29% tree cover and 53% turfgrass cover, the footprint climatology area of our study had a higher vegetation cover than is typical of North American cities [Grimmond and Oke, 2002; Nowak et al., 1996]. Although it has been suggested that advected energy from impervious surfaces can partially compensate for less vegetation cover overall in urban compared to rural ecosystems [Oke, 1979], the Etotal measured in this study remained less than values measured in nonurban ecosystems of North America. For example, growing season Etotal rates > 3 mm d−1 are commonly observed at hardwood forests in northern Wisconsin [Cook et al., 2004], Indiana [Schmid et al., 2000], Michigan [Schmid et al., 2003], North Carolina [Oishi et al., 2008], and Tennessee [Wilson et al., 2001]. Additionally, the annual Etotal of 467 and 324 mm yr−1 found at our suburban site was lower than values reported for rainfed maize and soybean crops in Illinois (547 and 660 mm yr−1, respectively) [Falge et al., 2001], tallgrass prairie in Oklahoma (637 mm yr−1) [Burba and Verma, 2005], mixed deciduous forest in Tennessee (571 mm yr−1) [Wilson et al., 2001], and mixed hardwood forest in North Carolina (604 mm yr−1) [Oishi et al., 2008].

[36] The seasonality of suburban Etotal observed in this study was more similar to patterns found in temperate hardwood forests from southern latitudes than northern latitudes of North America. For example, in a mixed deciduous forest in Oak Ridge, Tennessee, Wilson et al. [2001] observed Etotal rates that were significantly above zero from April to October, similar to patterns found in our study. Additionally, in a deciduous maple-beech to oak-hickory transition forest in south central Indiana, Schmid et al. [2000] found increased Etotal from May to October. In contrast, elevated Etotal occurred only from May to September in an upland hardwood forest in northern Wisconsin [Cook et al., 2004], and from late May to September in a hardwood-boreal transition forest in the lower peninsula of Michigan [Schmid et al., 2003]. These comparisons suggest that the urban heat island effect, in addition to the high percent cover of turfgrasses that are active in the spring and fall, may have played an important role extending the growing season at our suburban study site, relative to surrounding rural areas. This is consistent with analyses that find increases in annual net primary productivity in urban areas, particularly in cold regions, which have been attributed to extended growing seasons caused by the urban heat island effect [Imhoff et al., 2004].

4.2. Scaling Component-Based Fluxes

[37] Our comparison of measured and component-based estimates of Etotal showed that component-based approaches can capture much of the seasonal and spatial variability in suburban Etotal with a similar accuracy (79%) to that of eddy covariance measurements (∼80%) [Baldocchi et al., 1988]. Similar to studies in forest ecosystems, we found that component-based estimates of Etotal slightly underestimated measured values, particularly at high flux rates in summer [Bovard et al., 2005; Hogg et al., 1997; Oishi et al., 2008; Wilson et al., 2001]. Unlike these forest studies, however, we found that component-based estimates tended to overestimate measured fluxes in spring and fall. Several potential sources of error may explain these discrepancies between methods, including (1) errors associated with the measurements or models used to estimate component fluxes, (2) a mismatch in the spatial footprints associated with the measured and scaled fluxes, (3) undersampling species' differences in water use, and (4) the spatial heterogeneity of urban microclimates.

[38] The measurements and models used to estimate components of Etotal in this study all contribute some degree of uncertainty when scaling up Etotal. Previous studies from forest ecosystems have found that heat dissipation sap flux techniques may underestimate transpiration, particularly during periods of high radiation [Bovard et al., 2005; Hogg et al., 1997; Oishi et al., 2008; Wilson et al., 2001], and could explain the component-based underestimates we observed during summer. The lack of energy balance closure observed at the tall tower and turfgrass sites, although typical of eddy covariance systems in general [Wilson et al., 2002], suggests a systematic underestimate of both Etotal and EG. While an underestimate of EG may help explain the summer imbalances, it is contrary to explaining the overestimates observed in spring and fall. Depending on the seasonal variation in LAI among trees in our study area, the modeled seasonal LAI patterns also represent a source of error in our estimates of IT, EGunder_eg, and EGunder_dec. As comparisons between component-based and measured estimates of Etotal were restricted to periods with high quality data and exclude most rainfall events, it is unlikely that the IT component significantly contributed to the observed imbalances. The seasonal pattern of understory turfgrass evapotranspiration, however, is consistent with the observed imbalances. EGunder_eg and EGunder_dec were highest in spring and fall when tree LAI was low, and EGunder_eg and EGunder_dec were lowest in midsummer when tree LAI was high (data not shown). It is also likely that using a constant coefficient equal to 0.87 in the Priestley-Taylor equation to model EGirr led to an overestimate of Etotal in spring and fall and an underestimate in summer. Studies show that α varies seasonally in nonstressed, cool season grasses, with lowest values in spring and fall and highest values in summer [Brown et al., 2001; Ervin and Koski, 1998; Zhang et al., 2007].

[39] Even in urban areas with relatively flat topography, flux footprints are particularly difficult to assess due to the spatial heterogeneity of surface types and complex transport of scalars, and can lead to the mischaracterization of footprint areas when scaling up component fluxes [Vesala et al., 2008a, 2008b]. While we did not attempt to replicate the 2-D source area associated with each measured half-hourly value of Etotal, we did find that component-based estimates of Etotal were not greatly affected by different methods of sampling the land cover classification map. Consequently, we believe the misrepresentation of the footprint area to be a relatively minor source of error in our component-based estimates of Etotal.

[40] Species-rich ecosystems, such as ours, additionally complicate efforts to scale up measurements made on only a few individuals, species, or sites, as water use can be highly variable among tree [Kumagai et al., 2005; Oren et al., 1998; Pataki and Oren, 2003] and turfgrass species and among management regimes [Zhang et al., 2007]. Although the effect was relatively small, including TT differences among xylem anatomy types resulted in a 4% reduction in the overall imbalance compared to using the coarser plant functional type categories. Evapotranspiration from unmeasured understory plants has additionally been suggested to contribute to observed imbalances in forest ecosystems [Bovard et al., 2005; Oishi et al., 2008]. It is unlikely that evapotranspiration from common suburban understory vegetation types, such as woody shrubs, vegetable or flower gardens, contributed significantly to the Etotal in our study area because their cover represented less than 2% of the landscape. Although the seven tree genera and one turfgrass lawn we studied provides a limited sampling of species' differences in water use, the relatively good match that we found between component-based and measured Etotal suggests that this group of plant functional types captured much of the species' variation in water use within the suburban landscape we studied.

[41] Management practices, including irrigation, fertilization, mowing height, and canopy shading can also significantly impact rates of water loss from turfgrass [Feldhake et al., 1983; Zhang et al., 2007]. The large differences in daily evapotranspiration between our measured nonirrigated turfgrass and our modeled values of irrigated and understory turfgrass suggest that irrigation management effects were captured in seasonality, but fertilizer and mowing practices were not accounted for. Occasional irrigation by homeowners, particularly during periods of low soil moisture in midsummer, was also not accounted for and may contribute to the observed underestimate of measured fluxes in summer.

[42] Finally, microclimate variability in urban areas [Bonan, 2000; Byrne et al., 2008; Peters and McFadden, 2010] may represent a relatively more important source of error in urban component-based estimates of Etotal than in natural ecosystems. The spatial heterogeneity of urban surface types with different energy balances and heat storage capacities results in the advection, or lateral transport, of heat from paved surfaces with high sensible heat fluxes to vegetated surfaces with high latent heat fluxes [Oke, 1979; Spronken-Smith et al., 2000]. Advection can be an important energy source to irrigated urban parks and lawns, causing elevation of EG rates by up to 35% [Feldhake et al., 1983; Oke, 1979; Spronken-Smith et al., 2000]. Trees grown over asphalt can transpire 30% more water than trees grown over turfgrass [Kjelgren and Montague, 1998], while some tree species initiate stomatal closure and have reduced rates of water loss [Kjelgren and Montague, 1998; Montague and Kjelgren, 2004]. In addition, sparsely planted trees can have TT rates two to three times higher than those of densely planted trees [Hagishima et al., 2007] and windbreaks can reduce EG rates by 25% [Danielson and Feldhake, 1981].

[43] Our study trees had relatively open growth forms compared to forest-grown trees, yet they were all grown under more park-like conditions as compared to trees in more densely built-up urban areas. Consequently, our TT measurements likely underestimate ET rates of trees growing in complete isolation or near paved surfaces. This idea was further supported by the large underestimates of Etotal we observed during periods of high net radiation, high sensible heat fluxes, and high vapor pressure deficit from northwest and northeast wind directions. In contrast, the turfgrass site was a relatively large lawn that was less likely to have been exposed to advected sources of energy compared to smaller residential lawns surrounded by sidewalks, driveways, and streets. It is difficult to determine whether our EG measurements provide a systematic overestimate or underestimate of water loss from residential yards. As microclimate effects are an important source of uncertainty in predicting component-based Etotal in urban and suburban ecosystems, future studies should focus on how best to account for this variability when scaling.

4.3. Contributions of Trees and Turfgrasses to Suburban Etotal

[44] The relative contributions of trees and turfgrass to annual Etotal was driven largely by fractional cover and plant functional type differences in daily water use. Studies from forest ecosystems similarly show that the relative contribution of different tree species to Etotal is driven both by species' abundances on the landscape, often measured as contributions to total basal area or leaf area, and by species-specific differences in water use [Tang et al., 2006; Wullschleger et al., 2001]. Despite their higher rates of water use than deciduous broadleaf trees, evergreen needleleaf trees had a nearly negligible contribution to Etotal in our study area due their relatively small fractional cover of the landscape. Turfgrass, however, due its high cover and high daily EG rates across the growing season, represented the largest proportional contribution (87% and 64% in recreational and residential areas, respectively) to annual Etotal in our study area. Pasture grasses in natural savannah ecosystems, by contrast, represent a much smaller component (20–44%) of Etotal than trees [Baldocchi and Xu, 2007; Paco et al., 2009]. While these studies are from savanna ecosystems with similar percent tree cover to our study area, it is important to note they are from areas with a Mediterranean climate and very different species compositions.

[45] The proportional contribution of different vegetation types to suburban Etotal varied seasonally due to different patterns in physiological activity among plant functional types. While evergreen needleleaf and deciduous broadleaf trees both have maximum physiological function in the middle of summer in temperate ecosystems, the evergreen leaf habit and high cold tolerance allows evergreen needleleaf trees to remain physiologically active over a longer growing season than either deciduous broadleaf trees or cool season turfgrasses [Catovsky et al., 2002; Givnish, 2002; Havranek and Tranquillini, 1995]. Correspondingly, we observed the highest proportional contribution to Etotal from deciduous broadleaf trees in late summer and, albeit small, the highest contributions from evergreen needleleaf trees in April and November when deciduous trees were leafless. Our observation of a midsummer low in the contribution to Etotal from nonirrigated turfgrass is consistent with the midsummer dormancy of cool season turfgrasses [Feldhake et al., 1984; Fry and Huang, 2004; Zhang et al., 2007].

4.4. Land Use Comparisons

[46] Similar to other urban studies, we found that the magnitude of Etotal varied between land use types according to differences in the total cover and composition of vegetation [Frey et al., 2010; Grimmond and Oke, 1999; Offerle et al., 2006; Spronken-Smith, 2002; Vesala et al., 2008b]. Seasonal patterns of Etotal were also highly influenced by the phenology of the dominant vegetation type. For example, the turfgrass-dominated recreational area not only showed higher annual Etotal than the deciduous tree-dominated residential area, but it specifically showed higher rates of Etotal in spring and fall when cool season turfgrasses are most active. In addition, our study extends these previous findings to show that turfgrass management, particularly through irrigation inputs, is another important factor influencing the spatial and seasonal variability of suburban Etotal. The relatively high evaporative fluxes we measured from the recreational area during the dry conditions in June of 2007 suggest that increased irrigation during periods of low soil moisture can significantly alter Etotal from urban areas. While homeowners also maintain irrigated turfgrass lawns, management intensity can vary widely among individual landowners [Larson et al., 2009]. To predict suburban Etotal, these results suggest it will be important to know not just the total vegetation cover in urban landscapes, but also the fractional cover of different vegetation types and turfgrass management practices. In addition, it is likely that changes in land use, vegetation composition, or turfgrass management practices, particularly related to irrigation use, will lead to changes in total water fluxes from urban and suburban ecosystems.

[47] Given the growing interest among cities in using “green infrastructure” to manage problems such as storm water runoff and energy demand, it is important for urban planners and designers to consider the trade-offs among different potential landscape configurations [Mitchell et al., 2008]. For example, in arid regions the water conservation benefits of low water use species must be weighed against the cooling benefits of high water use species [Larson et al., 2009; Shashua-Bar et al., 2009]. In contrast, cities in mesic regions with frequent rainfall events may value the storm water interception benefits of vegetation with large canopies relatively more than other ecosystem services related to Etotal [Wang et al., 2008]. Although we found that cool season turfgrasses had higher rates of water loss than evergreen needleleaf and deciduous broadleaf trees per unit cover area, the overall cooling effect of turfgrass was not necessarily higher than that of trees. This is because trees also provide cooling by intercepting solar radiation and shading the ground beneath canopies [Peters and McFadden, 2010]. In the arid communal settlement of Midreshet Ben Gurion, Israel, for example, Shashua-Bar et al. [2009] found that turfgrass lawns shaded by tree canopies resulted in both a greater reduction in air temperature and a 50% reduction in water use compared to unshaded turfgrass lawns.

[48] By approximating runoff as the difference between measured precipitation and annual component-based Etotal, we estimate runoff to represent 26% and 48% of annual precipitation on average from the recreational and residential areas in our study, respectively. While these estimates neglect irrigation inputs, changes in storage, and groundwater recharge and, thus, should be used with caution, they fall within the range of runoff fractions (2–59%) previously reported for watersheds in the Minneapolis–Saint Paul metropolitan area [Brezonik and Stadelmann, 2002]. These results are in agreement with studies showing that increased pervious surface area in urban regions reduces direct runoff [Brezonik and Stadelmann, 2002; Haase, 2009; Mitchell et al., 2008; Wang et al., 2008]. Although it is beyond the scope of our study to offer specific planting recommendations for use in green infrastructure, this study advances our ability to quantify the relative contributions of different vegetation types to suburban Etotal, an important step in evaluating decisions related to the seasonal management of urban water and energy budgets.

[49] Despite the good agreement we found between component-based and measured Etotal in the suburban ecosystem we studied, differences in species composition among cities make it complex to extrapolate our component measurements of evapotranspiration to other urban and suburban areas, particularly in more southern latitudes. For example, the turfgrass site in this study represented only cool season turfgrass species, as warm season species are not typically planted in the climate zone associated with Minnesota. Yet cool and warm season turfgrasses have large differences in evapotranspiration, with cool season species using up to 45% more water during the growing season compared to warm season species [Feldhake et al., 1983; Zhang et al., 2007]. Cities located in warmer climates with higher frequencies of warm season turfgrasses will need to consider these species' differences in water use when scaling up and predicting urban Etotal.

5. Conclusions

[50] We quantified seasonal variations in evapotranspiration from a suburban landscape in the Upper Midwest region of the United States, with peak rates > 3 mm d−1 in summer. The good agreement between our component-based and measured Etotal, with an overall imbalance of 3%, suggests that the major plant functional types captured most of the spatial and temporal variability required to quantify urban Etotal. Component-based estimates overestimated measured Etotal in spring (20%) and fall (10%) and underestimated measured Etotal in summer (11%), likely due to errors associated with the irrigated turfgrass model and variability from microclimate effects. We found turfgrass was the major contributor to annual Etotal from both recreational and residential land uses types (87% and 64%, respectively) due to a high fractional cover (74% and 34%, respectively) and higher average rates of water use per cover area in midsummer than deciduous broadleaf or evergreen needleleaf trees (3.0, 1.4, and 2.3 mm d−1, respectively). The maximum contribution to Etotal from nonirrigated turfgrass occurred in spring and fall, irrigated turfgrass in midsummer, deciduous broadleaf trees in late summer, and evergreen needleleaf trees in early spring and late fall. Turfgrass-dominated recreational land use areas had higher average annual Etotal compared to deciduous tree-dominated residential areas (467 and 324 mm yr−1, respectively), as well as an altered seasonal pattern of Etotal with higher fluxes in spring and fall, and during a midsummer drought. These results suggest that a plant functional type approach could be used to estimate and compare Etotal among different cities and to evaluate the effects of changes in land use, vegetation composition, or management practices on urban water and energy budgets.


vapor pressure deficit, kPa.


net radiation, W m−2.


incoming solar radiation, W m−2.


below canopy solar radiation, W m−2.


ground heat flux, W m−2.


latent heat flux, W m−2.


sensible heat flux, W m−2.


leaf area index, m2 m−2.


ecosystem evapotranspiration, mm time−1.


evapotranspiration from trees, mm time−1.


evapotranspiration from turfgrass, mm time−1.


evaporation from water, mm time−1.


potential evapotranspiration, mm time−1.


tree transpiration, mm time−1.


interception loss from tree canopies, mm time−1.


evapotranspiration from evergreen needleleaf trees, mm time−1.


evapotranspiration from deciduous broadleaf trees, mm time−1.


evapotranspiration from nonirrigated turfgrass, mm time−1.


evapotranspiration from irrigated turfgrass, mm time−1.


evapotranspiration from turfgrass under evergreen needleleaf trees, mm time−1.


evapotranspiration from turfgrass under deciduous broadleaf trees, mm time−1.


[51] We thank ∼350 homeowners in Saint Paul, Roseville, Falcon Heights, and Lauderdale, the City of Lauderdale, and the University of Minnesota for granting us permission to conduct this research on their property; Brian Horgan for advice and access to the turfgrass site at the University of Minnesota Turfgrass Research, Outreach, and Education Center; Larry Oberg for technical advice and access to the KUOM broadcast tower; Dave Ruschy for data from the University of Minnesota climate station; Natascha Kljun for sharing her footprint model code; and Matthias Zeeman for sharing his footprint mapping script. We are grateful for advice from Tim Griffis, Sarah Hobbie, Rebecca Montgomery, and Tracy Twine, and for assistance with field measurements, software development, and image processing from Ahmed Balogun, Ben Freeman, Vikram Gowreesunker, Cheyne Hadley, Vicki Kalkirtz, Rebecca Koetter, Mark McPhail, Julia Rauchfuss, Scott Shatto, and Yana Sorokin. This work was supported by an Interdisciplinary Doctoral Fellowship to E.B.P. from the University of Minnesota Institute on the Environment and a grant to J.P.M. from NASA Earth Science Division (NNG04GN80G).