ARTICLE IN PRESS

1352-2310/$ - se

doi:10.1016/j.at

�Correspond

Atmospheric Environment 40 (2006) 3561–3573

www.elsevier.com/locate/atmosenv

Land use planning and surface heat island formation:A parcel-based radiation flux approach

Brian Stonea,�, John M. Normanb

aCity and Regional Planning Program, College of Architecture, Georgia Institute of Technology, Atlanta, GA 30332-0155, USAbDepartment of Soil Science, University of Wisconsin-Madison, Madison, Wisconsin, 53706, USA

Received 21 April 2005; received in revised form 14 December 2005; accepted 3 January 2006

Abstract

This article presents a study of residential parcel design and surface heat island formation in a major metropolitan region

of the southeastern United States. Through the integration of high-resolution multispectral data (10m) with property tax

records for over 100,000 single-family residential parcels in the Atlanta, Georgia, metropolitan region, the influence of the

size and material composition of residential land use on an indicator of surface heat island formation is reported. In

contrast to previous work on the urban heat island, this study derives a parcel-based indicator of surface warming to

permit the impact of land use planning regulations governing the density and design of development on the excess surface

flux of heat energy to be measured. The results of this study suggest that the contribution of individual land parcels to

regional surface heat island formation could be reduced by approximately 40% through the adoption of specific land use

planning policies, such as zoning and subdivision regulations, and with no modifications to the size or albedo of the

residential structure.

r 2006 Published by Elsevier Ltd.

Keywords: Urban heat island effect; Climate change; Land use; Urban planning; Remote sensing; Surface heat island; Thermal infrared

1. Introduction

Long recognized in the physical sciences as animportant climatological phenomenon, the urbanheat island effect has only recently gained seriousattention in the fields of urban planning and publichealth. A growing interest in urban temperaturesamong planners and other social scientists can beattributed to the steadily increasing number ofannual heat-related fatalities over the past severaldecades, which now exceeds the annual mortalityrates associated with all other forms of catastrophic

e front matter r 2006 Published by Elsevier Ltd.

mosenv.2006.01.015

ing author.

weather (National Oceanic and Atmospheric Ad-ministration (NOAA), 2000). Also significant topublic health is the influence of ambient heat onregional air quality problems. Annual exceedancesof the national standard for ground level ozone, forexample, have been found to correlate morestrongly with ambient temperatures than withannual emissions of ozone precursors (Stone,2005). The potential for steadily increasing tem-peratures and air pollution in response to bothregional and global climatological mechanisms,coupled with a projected doubling of the globalurban population by 2030 (United Nations, 2001,p. 4), greatly elevates the need for planners and

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–35733562

public health officials to devise effective strategiesfor managing climate in large cities.

A key question for the field of planning concernsthe role of urban spatial structure in the formationof heat islands. While the material composition,form, and configuration of the built environment isknown to directly influence the formation of heatislands (Golany, 1996), a consensus regarding themost thermally benign pattern of land developmentis largely absent from the heat island literature. Asdemonstrated by a large number of studies, thespatial distribution of urban air temperature max-ima appears to correlate strongly with the density ofland use, resulting in the classic bell-shapedtemperature profile of the monocentric city. Yet,as other studies have shown, lower density patternsof housing are associated with higher per householdlevels of impervious cover and energy consumption,two important drivers of heat island formation(Stone, 2004; Alberti, 1999). In light of theseconflicting observations, should large, mid-latitudecities seek to accommodate future populationgrowth through medium to high density, compactforms, or through lower density, dispersed patternsof development? The answer to this question holdsimportant implications for the practice of urbanenvironmental planning.

To address this issue, this article presents theresults of a study of residential parcel design andsurface warming patterns in the Atlanta, Georgia,metropolitan region. With the aid of high resolutionthermal infrared (IR) imagery and a parcel levelgeographic information system, the thermal proper-ties of over 100,000 single-family parcels areassociated with the spatial structure and materialcomposition of residential development throughouta 900 km2 study region. In measuring the thermalproperties of individual land parcels, we develop an‘‘emissivity adjusted’’ indicator of the parcel blackbody flux to account for the influence of variablesurface emissivity and sky radiation captured in thethermal IR imagery. Furthermore, we estimate thepredevelopment radiant flux and subtract thisquantity from the observed, post-developmentradiant flux to differentiate the warming impactsof urban development from natural, backgroundsources of surface heat. Following a detaileddescription of our methods and results, the articleconcludes with a discussion of physical planningstrategies designed to mitigate the thermal impactsof existing and future residential development in theAtlanta region and elsewhere.

2. Background

In order to associate remotely sensed thermal IRdata with land use planning policies it is essentialthat a ‘‘policy-relevant’’ indicator of surface warm-ing be developed. While remote sensing technologiescan provide a valuable tool for assessing the thermalperformance of urban environments, these must beused with great caution to yield reliable informationfor land use planning. Most importantly, there is acritical need to isolate the influence of specificclasses of land use on surface warming patterns.While previous work has demonstrated a cleardistinction between the thermal properties of‘‘urban’’ and ‘‘rural’’ land cover types, each ofthese broad categories of development is composedof a number of specific land use classes that aresubject to an array of development regulations. Ifwe are to mitigate heat island formation through themodification of specific zoning and subdivisionregulations, we must be able to isolate the influenceof these individual policies on surface warming. Tobe policy-relevant, an indicator of surface warmingmust be capable of quantifying the thermal char-acteristics of distinct and legally defined classes ofland use.

In this section of the paper, we discuss threecriteria that must be satisfied to derive a policy-relevant measure of heat island formation. Theseinclude: (1) the measurement of surface warming atthe level of the individual land parcel; (2) theadoption of a flux-based rather than a temperature-based indicator to capture the effects of land areaon warming; and (3) the adjustment of the radiantflux to more closely correspond to the sensible fluxof heat.

2.1. Measuring thermal performance at the parcel

level

In assessing the surface thermal properties ofurban and rural land covers, many studies haveadopted a unit of analysis consistent with theresolving dimension of a satellite radiometer, suchas 1.1 km or 120m (e.g., Roth et al., 1989; Nichol,1996). An important limitation of adopting such aspatially uniform unit of analysis, however, is that itfails to conform to the irregularly shaped bound-aries of individual land parcels, the unit of area atwhich land use is controlled. For example, in denseand well-mixed urban districts, a spatially uniformunit of analysis is likely to capture both commercial

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–3573 3563

and residential land uses, confounding any attemptto isolate the thermal properties of a single land useclass. Such an outcome may be categorized as aspecial instance of mixed pixel error—mixed, that is,in terms of land use class rather than land covertype.

The limitation of this type of mixed pixel error forplanning analysis is that it obscures the underlyingpolicy drivers of environmental performance. If weare interested in evaluating the environmentalimpacts of a particular planning policy, such asmaximum impervious area regulations, we mustfirst identify those zones subject to the policy. Forexample, within a square kilometer of urban landthere is likely to be a number of unique land useclasses, each with its own zoning regulationsgoverning the permissible area of impervious coverper parcel. In measuring the environmental char-acteristics of an area of land subject to multiplezoning regulations, it is difficult to assess the relativeinfluence of any single policy on environmentalperformance. For the purposes of planning re-search, a more methodologically sound approachentails the measurement of surface warming at adimension compatible with land use regulation,such as a zoning district or individual land parcel.As adopted herein, a parcel-based approach tosurface warming analysis permits parcel-specificland use policies, such as zoning regulations,subdivision regulations, and building codes, to beassociated with thermal performance at the samespatial dimension.

2.2. Accounting for the effects of land area on

surface warming

While parcel-based measures of surface warmingare needed to associate heat island formation withland use policies, the adoption of a spatiallyirregular unit of analysis complicates the use ofsurface temperature as an indicator of warming. Asnoted by Price (1979) over 25 years ago, tempera-ture-based measurements, such as the urban–ruraltemperature differential, often fail to reflect themagnitude of heat islands created by urban areas ofdifferent sizes. Price states, ‘‘[f]or investigation ofurban heating, the peak temperature is less sig-nificant than a summation (area integral) of theexcess power radiated as a result of the surfacetemperature elevation’’ (p. 1557). In other words, ameasure of the excess flux of thermal energy from acity provides a more accurate indicator of the total

magnitude of warming within an urbanized areathan does the urban–rural temperature differential.

We believe the magnitude of surface warmingbetween urban and rural zones (defined herein asthe excess flux of sensible heat energy) to be morerelevant to urban planning policy than heat islandintensity (the urban–rural temperature differential).The reason for this is that the area-integrated flux ofheat energy emitted from a city or parcel provides amore accurate indicator of the volume of the nearsurface atmosphere influenced by elevated surfacetemperatures than does the surface temperaturedifferential. In this sense, two cities of similarclimate and development type but dissimilar landarea are likely to exhibit similar average surfacetemperatures but different fluxes of energy. As thearea of a city expands, the volume of the heat island‘‘dome’’—a volume subject to elevated air tempera-tures and enhanced pollution formation—expandsas well. Measures of heat island intensity acrosscities may not reflect a differential in the volume ofheat island domes. While smaller in scale, the sameis true when comparing two parcels of differentarea.

The significance of heat island magnitude toplanning policy is clearly illustrated through Price’sanalysis of heat island formation in a number ofNew England cities. Through an analysis of thermalIR imagery collected by the Heat Capacity MappingMission satellite, the thermal impacts of city sizewere found to vary substantially based on theindicator of thermal performance selected. Asdetailed in Table 1, the urban–rural temperaturedifferential and a measure of the excess radiant flux,the ‘‘excess power radiated,’’ were found to varysignificantly based on city size. For example, whilethe urban–rural surface temperature differentialsfor Syracuse, New York, and Waterbury, Connecti-cut, were found to be equivalent, the quantity ofexcess radiant energy emitted from Syracuse wasfound to be over five times greater than that fromWaterbury.

Defined by Price as the excess radiant energy (inmegawatts) emitted from a city as a product of theurban–rural temperature differential and urbanland area, the excess power radiated measureaccounts for both the intensity of warming andthe area of land impacted by elevated surfacetemperatures. As the land area of Syracuse is manytimes greater than that of Waterbury, the total fluxof excess radiant energy across this area is greater aswell, increasing the spatial magnitude of warming.

ARTICLE IN PRESS

Table 1

Population, area, and thermal characteristics for ten New England Cities (adapted from Price (1979))

City 1970 population

(� 1000)

Area (km2) Population

density (km2)

DT (1C) Excess power

radiated (mW)

Excess power

radiated per

capita (kW)

New York, NY 7895 547.0 14,433 17.0 40,000 5.07

Syracuse, NY 197 6.5 30,307 10.9 417 2.12

Providence, RI 179 48.2 3714 13.2 3190 17.82

Worcester, MA 177 5.1 34,706 11.5 327 1.85

Hartford, CT 158 26.2 6031 15.0 1770 11.2

Bridgeport, CT 157 14.2 11,056 12.1 954 6.08

New Haven, CT 138 5.8 23,793 11.2 372 2.70

Albany, NY 116 3.7 31,351 10.3 227 1.96

Stamford, CT 109 3.2 34,063 11.2 202 1.85

Waterbury, CT 108 1.2 90,000 10.9 74 0.69

Source: Price (1979, Tables 1 and 2).

Note: The population reported for New York, NY, is for the metropolitan area; all other populations are reported for the municipality.

We have corrected a typographical error in the original table, in which the unit of excess power radiated was reported as kW rather than as

mW, and we have computed the fields of population density and excess power radiated per capita based on the data reported by Price.

B. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–35733564

Thus, while the temperature differential of these twocities is equivalent, the total land area in Syracuseexposed to elevated surface temperatures, and, byextension, the total volume of the near surfaceatmosphere, is many times greater.

In the context of planning research, area-inte-grated measures of heat island magnitude areneeded to illuminate the spatial drivers of regionalclimate change. The significance of heat islandmagnitude to planning policy is clear from Price’sanalysis of population size and surface warming inSyracuse and Waterbury. While the population ofSyracuse is about 80% greater than that of Water-bury, its excess power radiated is over 500% greater.Why might this be the case? Once we account forany intra-regional variation in climate or land covertypes, an important structural distinction betweenthe two cities is the density of development. With agross population density of about 900 persons perhectare, Waterbury’s population density in 1970was almost three times that of Syracuse. As a result,the average quantity of land developed to support asingle Syracuse resident was three times as great asthe per capita land requirements of a resident ofWaterbury, potentially translating into a greatermagnitude of surface warming per resident. For theten largest cities examined by Price and presented inTable 1, a significant negative correlation was foundbetween population density and the excess powerradiated per capita (r ¼ �0:65; po:001). Theseresults support the hypothesis that, when control-ling for population size and climate, lower density

cities in New England are characterized by a greatersurface heat island magnitude than higher densitycities.

2.3. Approximating the sensible heat flux

While aerial and satellite radiometers provide themost viable means of obtaining continuous surfacethermal data across a large study region, a principallimitation of remote sensing studies of the urbanheat island effect is a failure to account for thevariable distribution of surface emissivity. Given thewide array of surface material types found in urbanareas, ranging from heavily canopied areas to theimpervious materials of driveways and rooftops,radiant flux densities recorded by thermal sensorsare prone to systematically underestimate thermo-dynamic temperatures from the heterogeneoussurfaces of urban environments. As a result, thesurface radiant flux density must be adjusted tomore closely approximate the black body fluxdensity, which is a better indicator of sensible heat.

The radiant flux density is a reasonable choice forrepresenting heating of the atmosphere by the urbansurface for three reasons: (1) it contributes a portionof the heating itself; (2) it can be expressed as afunction of the surface-to-air temperature differ-ence, making it roughly proportional to the sensibleheat flux density; and (3) it can be measured at theparcel level through the use of remote sensing. Twomeasures of the surface radiant flux are commonlyused: (1) the apparent radiant flux density, which is

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–3573 3565

measured directly by remote sensing instruments,and (2) the black body radiant flux density, which isadjusted for the surface emissivity and represents amaximum emitted flux density proportional to thethermodynamic temperature of the surface. Theblack body radiant flux density is more closelyrelated to the thermodynamic temperature than theapparent radiant flux density and thus is likely to bemore highly correlated to the surface sensible heatflux density. Therefore, we need to adjust themeasurements of apparent radiant flux density froma remote sensing instrument using the surfaceemissivity. To do so, we employ the followingequation:

UBB ¼Umeasured � ½ð1� eÞLSKY�

e, (1)

where UBB is the black body flux density, Umeasured

the apparent flux density, e the parcel compositeemissivity, LSKY the emitted sky radiation.

The emissivity value associated with a particularparcel is based on the proportion of the parceloccupied by impervious and vegetative materials.As demonstrated by Carlson et al. (1995), thefraction of vegetative cover (fc) for a particular landparcel may be estimated through the derivation of anormalized difference vegetation index (NDVI).Once computed, the parcel composite emissivitymay be estimated through the following equation:

ec ¼ evðf cÞ þ eið1� f cÞ, (2)

where ec is the parcel composite emissivity, ev theemissivity of vegetative component, fc the fractionof vegetative cover, ei the emissivity of imperviouscomponent.

As a final step, the quantity of radiant energyattributable to predevelopment land characteris-tics—the rural ‘‘background’’ radiation—must besubtracted from the measured parcel black bodyflux. As illustrated by Price’s measure of the excesspower radiated, in deriving an indicator of theurban heat island effect, we are interested in thethermal differential between developed and undeve-loped land features. When measuring heat islandintensity, this variable is quantified as the tempera-ture differential between developed and undeve-loped zones. When measuring heat islandmagnitude, this variable becomes the flux differen-tial between developed and undeveloped zones. Tocalculate this flux differential at the parcel level, wefirst calculate the black body flux density for adeveloped parcel based on the methods described

above, and then subtract from this quantity anestimate of the predevelopment black body fluxdensity that would result were the same parceloccupied by a natural land cover, such as forestcanopy cover. We then multiply this flux densitydifferential times the parcel area to estimate theparcel net black body flux. Eq. (3) demonstrates thiscalculation:

NFBB ¼ AparcelðUBB urban parcel �UBB rural parcelÞ,

(3)

where NFBB is the parcel net black body flux, Aparcel

the parcel area, UBBurban parcel the black body fluxdensity of post development parcel, UBB rural parcel

the black body flux density of predevelopmentparcel.

In summary, to be useful in planning research, anindicator of the urban heat island effect must bespatially compatible with the legal dimensions ofland use control, and must accurately approximatethe quantity of energy contributed by a single landparcel to elevated temperatures within cities. Thepreceding discussion outlines a conceptual basis forderiving such an indicator. What follows in theremainder of the paper is an application of thisconceptual approach to the Atlanta, Georgia,metropolitan region and a presentation of policyinsights that may be gleaned from this analysis.

3. Methods

In May of 1997, the National Aeronautic andSpace Administration (NASA) obtained high reso-lution multispectral imagery over the Atlanta,Georgia, metropolitan region for the purpose ofinvestigating the influence of land use on urban heatisland formation. Recorded from an aerial platformwith the Advanced Thermal and Land ApplicationsSensor (ATLAS) at a ground resolution of 10m, 15channels of multispectral data were obtained overthe Atlanta study region by NASA and madeavailable to a range of research efforts knowncollectively as ‘‘Project Atlanta.’’ The exceptionallyhigh spatial resolution of this data permits thethermal characteristics of high-density urban parcelsto be reliably measured and integrated into a parcellevel geographic information system. With the aidof the Project Atlanta data, three principal stepswere carried out to develop an integrated surfacewarming and land use database for the Atlantaregion. These included: (1) data processing, (2)construction of a parcel-based land information

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–35733566

system, and (3) derivation of the parcel ‘‘net blackbody flux’’ measure.

3.1. Data processing

Prior to the analysis, a number of steps wereperformed to atmospherically correct and spatiallyrectify the ATLAS data. Collected under mostlyclear skies and calm conditions in May 1997, themultispectral imagery required corrections for posi-tional accuracy of the sensor, atmospheric radiance,and georectification. To correct for atmosphericradiance, temperature profiles recorded by radio-sondes launched during the period of data collec-tion were used in concert with the MODTRAN3model to account for the effects of water vaporand other atmospheric constituents on atmosphericradiance. Surface measurements recorded withhand held radiometers were employed in validatingthe ATLAS observations, and positional datarecorded continuously during the flights was usedto correct the data for positional changes during theperiod of data collection. Following these atmo-spheric corrections, each of the eleven flightlinesrecorded by NASA was georeferenced to GeorgiaState Plane coordinates and merged into a singleimage mosaic. Of the 15 channels of multispectral data obtained from NASA, three were usedin this research: a channel in the red (0.63–0.69 mm),near infra-red (0.76–0.90 mm), and thermal IR(9.6–10.2 mm) regions of the electromagnetic spec-trum.

3.2. Deriving the parcel black body flux

The first step in estimating the parcel net blackbody flux requires that the thermal characteristics ofindividual land parcels be measured. To do so,parcel boundary information for the approximately104,000 single-family parcels located in the studyregion was obtained from the City of Atlanta andFulton County offices of tax assessment. Byregistering the multispectral imagery and parcelgeography data layers to the same coordinatesystem, it was possible to spatially overlay the twodata sets in a geographic information system. Onceintegrated, the average apparent radiant flux densityof each parcel was derived through a zonalsummary function. Given the bandwidth of thethermal IR channel, the apparent radiant fluxdensity (Umeasured) for a parcel can be derivedthrough an application of Planck’s model for a

black body radiator:

Umeasured ¼

Z 10:2 mm

9:6 mm

C1

pl5½expðC2=lTÞ�dl, (4)

where C1 ¼ 3:7404� 108Wm4m�2, C2 ¼ 14; 387, T

is temperature in Kelvin.In addition to the apparent radiant flux density,

information on the composite parcel emissivity andemitted sky radiation were required to apply Eq. (1),presented above. Employing the method of Carlsonet al. (1995), the fraction of vegetative cover foreach land parcel was estimated through the deriva-tion of a NDVI with ATLAS bands in the red andnear infra-red spectral regions. Based on publisheddata for vegetative and impervious materials char-acteristic of residential zones, an emissivity value of0.97 was assigned to the vegetative component ofeach parcel and a value of 0.90 was assigned to theimpervious component (Jensen 2000; Oke, 1987).The composite parcel emissivity was then calculatedwith Eq. (2). Radiosondes launched during theperiod of data collections recorded an average valueof 2.26Wm�2 for the flux density of emitted skyradiation in the 9.6–10.2 mm wavelength band.

As a final step, the parcel net black body flux wascalculated by subtracting an estimate of thepredevelopment flux density from the post develop-ment flux density, and by then multiplying thisquantity times the parcel area in square meters toderive an estimate of the area-integrated flux ofexcess thermal energy at the parcel level (Eq. (3)). Itis important to note that, due to the narrowbandwidth of the thermal channel used for thisresearch, 9.6–10.2 mm, the net black body fluxindicator derived for this study does not representthe complete flux of excess radiant energy from thesurface. Rather, it serves as an indicator of surfacewarming that should scale closely with the excessblackbody radiant flux across the full spectrum ofwavelengths.

3.3. Constructing a parcel design database

In addition to an indicator of surface warming,information on the physical design attributes ofsingle-family parcels was needed to associate heatisland formation with variable residential land usepatterns. We are interested in residential land usefor several reasons: (1) the majority of the developedland area in the Atlanta region is occupied by thisland use class; (2) residential land use tends to

ARTICLE IN PRESS

Table 2

Parcel design attributes

Variable Definition Hypothesized significance

Area of impervious cover The area of the parcel occupied by the residential

structure, ancillary buildings, or paving materials

in m2.

Positive: increments in impervious cover are

associated with an increase the net black body flux.

Area of lawn and

landscaping

The area of the parcel occupied by lawn and

other forms of surface vegetation in m2.

Positive: increments in the area of lawn and

landscaping are associated with an increase in the net

black body flux.

Percent tree canopy cover The percentage of the parcel overlaid by tree

canopy cover.

Negative: increments in tree canopy cover are

associated with reductions in the net black body flux.

Number of bedrooms The number of bedrooms in the residential

structure.

Control variable: the number of bedrooms serves as a

control for the capacity of the residential structure.

Table 3

Descriptive statistics

Variable Minimum Maximum Mean Standard

deviation

Lot size (m2) 90 4050 1360 800

Impervious area

(m2)

22 1326 208 63

Lawn area (m2) 46 3863 1152 758

Tree canopy

cover (%)

0 100 45 32

Number of

bedrooms

1 14 2.9 0.8

Composite

emissivity

0.91 0.97 0.96 0.01

Net black body

flux (W)

19 22,030 3080 2110

B. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–3573 3567

constitute the leading edge of urban expansion; and(3) residential land uses are more suitable formodest design changes than commercial or indus-trial land uses, which tend to be more strictlygoverned by their function.

Four parcel design variables were derived for thisstudy. These include the area of imperviousmaterials, such as driveways and building foot-prints, the area of lawn and landscaping, theproportion of the parcel overlaid by tree canopy,and the number of bedrooms in the residentialstructure. This final variable is used to control forresidential capacity. In short, we are interested incomparing the thermal characteristics of parcelsvarying in size and material composition but similarin the number of residents the parcel was designedto accommodate. The inclusion of this variable inour analysis ensures that differences in thermalperformance by region of the city are not attribu-table to differences in residential capacity.

Information on the area, footprint size ofresidential buildings (houses and detached build-ings), and number of bedrooms in the residentialstructure was obtained from the City of Atlanta andFulton County offices of tax assessment.

As driveway areas are not recorded in theproperty assessment process, a sample of parcelsstratified by age and size was selected to estimate thearea of these paved surfaces. The driveways ofselected parcels were then measured directly todetermine how paved areas scale with parcel sizethroughout the study region. In combination, theestimated area of driveway paving and the area ofthe building footprint constitute the imperviouscomponent of the parcel. The area of lawn andlandscaping was derived by subtracting the esti-mated impervious area from the total parcel area.

Finally, the proportion of each parcel overlaid bytree canopy was estimated from the NDVI values,discussed above. Table 2 presents the name,definition, and hypothesized significance of eachparcel design incorporated into the database.

4. Analysis and result

With the aid of the Statistical Package for theSocial Sciences (SPSS), descriptive and explanatoryanalyses were performed to assess the influence ofparcel design on the parcel net black body flux. Asdetailed in Table 3, the average single-familyresidential parcel is characterized by a three bed-room house and is approximately 1360m2 in area,of which 208m2 or 15% is occupied by theimpervious components of the house and driveway,with the remaining 85% occupied by lawn andlandscaping. Approximately, 45% of the averagesingle-family residential parcel is overlaid by tree

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–35733568

canopy cover. In general, the design of the averageresidential parcel is consistent with a medium to lowdensity pattern of development. As indicated by therange and standard deviation of the lot size variable,the data set includes the full continuum of densitylevels.

Two measures of thermal performance arereported in Table 3. As described above, the parcelemissivity is a reflection of the relative proportion ofimpervious and vegetated materials. Values of thisvariable range from 0.90, reflective of completeimpervious cover, to 0.97, reflective of a fullyvegetated parcel. The average parcel emissivity is0.957, indicating a mix of pervious and imperviousmaterials.

The average net black body flux for a single-familyresidential parcel is 3080W, or about 2.3Wm�2 inthe 9.6–10.2mm wavelength band. This findingsuggests that, on average, the conversion of a naturalland cover to a single-family residential dwellingresults in the emission of an additional 3080W to theatmosphere within the 9.6–10.2mm band, underconditions present at the time of data collection. Itis this excess surface flux that most directly capturesthe contribution of an individual land parcel toregional surface heat island formation. The largerange and standard deviation for this variablesuggests a wide distribution of parcel surfacewarming throughout the Atlanta study region.

To evaluate the graphic correlation betweenparcel area, material composition, and surfacewarming, values of increasing lot size are plottedagainst impervious area, lawn area, and the net

0

500

1000

1500

2000

2500

3000

3500

4000

0 - 500 500 - 1,000 1,000 - 2

Lot Size (sq.

Are

a o

f L

and

Co

ver

(sq

. met

ers)

Imperv Area Lawn Area

Fig. 1. Mean area of land cover and

black body flux in Fig. 1. As illustrated in thisfigure, the magnitude of surface warming scalesclosely with lot size, with the mean net black bodyflux increasing by a factor of almost six betweenthe highest and lowest density classes. While thisdescriptive evidence supports our hypothesis of anegative relationship between the density of single-family residential development and surface warm-ing, it is important to note that this simplecovariation does not account for the distributionof residential capacity or tree canopy coverthroughout the region. If larger lot sizes are alsoassociated with greater residential capacities (e.g.,four and five bedroom houses) and a less maturetree canopy cover, then the relationship between thedensity of development and thermal performancemay be attributable to incompatible design objec-tives or to the age of development in different areasof the city. In order to control for these importantinfluences, a multivariate statistical model is devel-oped and evaluated in the following section.

4.1. Explanatory analysis

In order to quantify the potential thermal benefitsof specific design-based strategies, ordinary leastsquares was employed to develop a predictive modellinking parcel design to the net black body flux. Theresulting model included the following dependentand independent variables:

,00

me

net

Y ¼ net black body flux (W),X1 ¼ area of impervious cover (m2),

0 2,000 - 4,000 > 4,000

ters)

0

1000

2000

3000

4000

5000

6000

7000

8000

Net

Bla

ck B

od

y F

lux

(W)

Net Black Body Flux

black body flux by lot size.

ARTICLE IN PRESS

Table 4

Linear regression results for parcel net black body flux (W)

Variable B- Standardized Significance

1

OL

non

squ

to

in

B. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–3573 3569

X2 ¼ area of lawn and landscaping (m2),X3 ¼ tree canopy cover (%),X4 ¼ number of bedrooms.

coefficient coefficient

Impervious area

(m2)

0.083 0.290 o 0.001

Lawn area (m2) 0.012 0.512 o 0.001

Tree canopy cover

(%)

�24.42 �0.432 o 0.001

Number of

bedrooms

0.440 0.020 o 0.001

Summary statistics Adj. R2 F-statistic F

significance

0.55 31,866 o 0.001

Note: the square root of the net black body flux was used in this

model as a variance stabilization transformation.

With the aid of SPSS, a set of parameter estimatesand model summary statistics was generated for theapproximately 104,000 single-family residential par-cels located in the Atlanta study region. The resultsof this modeling process are presented in Table 4.1

The results of the regression analysis indicate thatthe various parcel design attributes are significantlyrelated to the parcel net black body flux. Asexpected, the area of impervious cover exhibits asignificant positive relationship with parcel warmingwhen controlling for other parcel design attributesbelieved to be associated with the dependentvariable, as indicated by the positive sign on theB-coefficient. While the use of a transformed-dependent variable complicates the interpretationof model coefficients, the standardized regressioncoefficients provide an indication of the relativestrength of the relationship between the parceldesign attributes and parcel warming. The standar-dized regression coefficient for the area of imper-vious cover indicates that, with each one standarddeviation increase in impervious area, the level ofthe transformed net black body flux variableincreases by approximately 0.29 standard devia-tions. Significantly, this relationship was found tohold when controlling for the number of bedroomsin the residential structure, suggesting that theelevated surface warming is not solely attributableto variation in residential capacity.

Perhaps surprisingly, impervious cover was notfound to be the strongest determinant of parcelwarming. With a standardized regression coefficientof 0.51, variation in the area of lawn and land-scaping was found to have the strongest associationwith the net black body flux. Specifically, anincrease of one standard deviation in pervioussurfaces is associated with an increase in thetransformed net black body flux of approximately0.51 standard deviations. This standardized effecton the dependent variable is significantly greaterthan that of the other independent variables and isbelieved to be attributable to two mechanisms.

A residual analysis conducted to evaluate the reliability of the

S model showed evidence of heteroscedasticity, suggesting

-normal variance of model residuals. The performance of a

are root transformation of the dependent variable was found

correct for this potential source of error. The results reported

Table 4 reflect this variance stabilization transformation.

First, although not as great as the impervious fluxdensity, the flux density of grass and other land-scaping materials is greater than that of thepredevelopment land cover of forest canopy. As aresult, the displacement of tree canopy by residen-tial lawn areas increases the flux of energy per unitof area. Second, the area of lawn and landscapingaccounts for a larger proportion of the averageresidential parcel than impervious materials. Thus,while the flux density of energy from lawn materialsis less than that of impervious materials in watts persquare meter, the number of square meters ofcoverage is greater—resulting, in many cases, in agreater total flux of energy from pervious materialsper parcel. As discussed in the following section, thisfinding holds important implications for design-based approaches to heat island management.

With a standardized regression coefficient of�0.43, the proportion of tree canopy cover wasalso found to be a strong predictor of parcelwarming. As discussed in the following section, thisfinding confirms the value of tree planting programsas a strategy to mitigate surface warming in cities.

Overall, a regression model incorporating thefour parcel design attributes was found to explainapproximately 55% of the variation in the squareroot of the net black body flux within the Atlantastudy region (Adj. R2

¼ 0.55). As such, this modelprovides a reasonable basis for assessing the thermalimpacts of various planning policies governing thedesign of residential parcels in the Atlanta, Georgiastudy area. However, given the complexity ofinterpreting the model coefficients with a trans-formed dependent variable, there is a need to assess

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–35733570

what magnitude of benefits may be realized fromtargeted design modifications to residential parcels.To address this limitation of the analysis, thefollowing discussion quantifies the potential benefitsof three specific planning strategies for reducingsurface heat island formation in residential zones.

5. Discussion

The results of this analysis provide compellingevidence that the size and material composition ofsingle-family residential parcels is significantlyrelated to the magnitude of surface warming in theAtlanta study region. Specifically, smaller, higherdensity parcels were found to be associated with alower net black body flux than larger, lower densityparcels when controlling for the class of land useand the number of bedrooms in the residentialstructure. While both the area of imperviousmaterials and lawn and landscaping were found tobe positively related to the parcel net black bodyflux, the area of lawn and landscaping—a strongcorrelate of parcel size—was found to be thestrongest predictor of excess surface warming.

In contrast to previous work on the urban heatisland (e.g., Hoyano, 1984), the results of this studysupport the hypothesis that lower density, dispersedpatterns of urban residential development contri-bute more surface energy to regional heat islandformation than do higher density, compact forms.While specific to the Atlanta metropolitan region,due to the uniformity of building materials anddevelopment patterns found throughout NorthAmerican cities, we believe these findings aregeneralizable to a number of large, mid-latitudecities characterized by a similar climatological andphysiographic characteristics. For such cities, thiswork illuminates at least three specific planningstrategies that may be employed to offset the excesssurface heat energy emitted from both existing andfuture residential development.

5.1. Reducing lawn areas

The results of our analysis suggest that the mosteffective design-oriented strategy for mitigating thethermal impacts of new housing development is areduction in the area of the residential lawn. For theaverage parcel, a reduction in the lawn area of 25%,holding other design attributes constant, would beassociated with a 13% reduction in the parcel netblack body flux. By serving to increase the density

of land use, a reduction in the lawn area of newresidential development would offset enhanced sur-face warming through at least two mechanisms.First, by reducing the average lot size of develop-ment at the urban periphery, the area of rural landconverted from forest to urban land uses is reduced,constraining the zone of enhanced warming. Sec-ond, by limiting the total zone of urban expansion,higher density development limits the extension ofroadways and other thermally intensive infrastruc-ture, also serving to preserve rural landscapes.

A reduction in the area of residential lawns maybe achieved through a number of planning policytools. The most straightforward approach entailsthe inclusion of maximum lot size restrictions inmunicipal zoning codes. Consistent with many largecities, Atlanta’s zoning code imposes a minimumstandard for residential lot sizes by land use classbut no maximum, and thus fails to effectivelycontrol the area of land conversions at the urbanperiphery. Alternatively, cities may adopt a largeminimum lot size (e.g., greater than 40 acres) orrestrictions on land uses in future developmentzones beyond the periphery of present day devel-opment, a policy known as an ‘‘urban growthboundary.’’ By limiting parcel subdivision andrestricting residential classes of development in therural hinterland, municipalities and regional gov-ernments can create clearly demarcated boundarybetween urban and rural zones, serving to promotehigher densities in the established urban core. Afinal policy employed to encourage smaller lot sizesis the scaling of development impact fees with lotsize. If sufficiently graduated, impact fees for largerlot sizes can provide an effective deterrent todeveloping expansive residential lawn areas.

5.2. Decreasing impervious cover through

reconfiguring the parcel

The finding of a positive relationship between thearea of impervious cover and the parcel net blackbody flux confirms the well-established role ofmineral-based building and paving materials inurban heat island formation. The most widelyadvocated approach to mitigating the thermalimpacts of impervious materials is through theapplication of high albedo surface treatments toincrease surface reflectivity—an approach that iscost-effective and applicable to both new andexisting development. For new construction, areduction in the total area of impervious cover

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–3573 3571

provides an additional tool for offsetting an increasein surface warming. Our findings suggest that a 25%reduction in the impervious cover of an averagesingle-family parcel is associated with a 16%reduction in the net black body flux. Whencombined with an equivalent reduction in lawnarea, parcel warming was found to be 28% lower.

The area of parcel impervious cover may bereduced in a number of ways with no net loss in theliving space of residential structures. For newdevelopment, the use of multistory constructionprovides a straightforward approach to minimizingbuilding footprint areas.2 Likewise, a revision ofzoning and subdivision regulations to shorten frontyard setbacks and narrow lot frontages cansignificantly reduce driveway areas and the area ofstreet paving required to service the parcel. For bothnew and existing development, the replacement ofdriveway paving at the time of routine resurfacingwith driveway runners (narrow strips of paving forvehicle tires) can reduce the area of parking surfacesby as much as one-third. Research has demon-strated that modest revisions to zoning and sub-division regulations, with no change in the size ordesign of residential structures, can achieve a 30%reduction in the impervious cover of residential lots(Stone 2004).

5.3. Shading existing development through tree

planting

While the potential to reduce the thermal impactsof new growth is significant, it is important to notethat changes to a city’s land development regula-tions will have only a limited influence on existingdevelopment. As the vast majority of the Atlantaregion’s 2025 built area is already in place, changesin future peripheral development will serve to offsetcontinued growth rather than abate present dayheat island formation. In light of this observation,the most effective design-oriented strategies forreducing total regional warming must address bothnew and existing development. As noted above, onestrategy for doing so would entail the replacementof traditional driveways with driveway runners atthe time of routine resurfacing. A second critical

2It should be noted that, while reducing the area of the

impervious footprint, multistory construction can increase the

potential for radiative warming through increasing urban canyon

geometries. As our analysis is limited to single-family residential

zones, we assume any enhanced warming through increased

canyon geometries to be minimal.

strategy for existing development entails the plant-ing of trees. As indicated by our analysis, for theaverage single-family parcel, an increase in treecanopy cover from 45% to 60% reduces the parcelnet black body flux by 14%. For trees strategicallyplanted along roadways and in proximity to houses,the thermal benefits are likely to be greater.

Cities can promote residential tree plantingthrough the development and enforcement ofmunicipal tree ordinances. Such ordinances canregulate the number, size, species, and placement oftrees along streets and within private parcels. Whilemost cities have incorporated some type of treeprotection ordinance into their municipal codes, theenforcement of these policies is often lax due to theinadequate funding of city arborists. However, inlight of the substantial costs of energy consumptionand air pollution associated with heat islandformation, the cost effectiveness of urban treeplanting is likely to enhance the appeal of thesestrategies for growing metropolitan regions overtime.

6. Conclusions

The results of this study indicate that, whenmeasured as an enhanced flux of surface energy andadjusted for residential capacity, the magnitude ofsurface warming associated with single-family re-sidential development in the Atlanta, Georgia,metropolitan region is greatest in low density areasand is sensitive to the size and material compositionof residential development. We believe the results ofthis study make two important contributions to theheat island literature. First, the development of aparcel-based measure of surface heat island magni-tude permits the thermal impacts of developmentactivities to be measured at a spatial dimensioncompatible with legal land use controls. As a policy-relevant measure of heat island formation, variationin the parcel net black body flux provides a muchneeded quantitative basis for evaluating the influ-ence of specific municipal land development regula-tions such as zoning regulations, subdivisionregulations, and building codes, which controlparcel design. As a result, in addition to enhancingthe surface albedo of existing development, plannerscan modify the land development regulationsgoverning the underlying structure of cities toenhance the thermal performance of new develop-ment at the urban periphery and redevelopment inthe established core. In this sense, policy-relevant

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–35733572

measures of environmental performance enablepolicy makers to be both reactive and proactive inplanning for urban growth and environmentalchange.

Second, this study assesses the magnitude ofbenefits associated with parcel design changes in amajor metropolitan area of the United States. Basedon our findings, a 25% reduction in the area ofimpervious and residential lawn space for theaverage single-family parcel, combined with anincrease in average tree canopy cover from 45% to60%, would reduce the parcel net black body fluxby approximately 40%. It is important to note thatthese benefits may be achieved with no changes tothe size, capacity, or material design of theresidential structure. When combined with strate-gies to increase the surface reflectivity of paving androofing materials, additional cooling benefits maybe realized. Future work will focus on the full rangeof benefits that may be derived from integratedland use, tree planting, and albedo enhancementstrategies across an expanded array of land usetypes.

Before concluding, it is important to highlight afew limitations of our analysis. First and mostimportantly, this analysis considers only a singleclass of land use, residential, which, while significantto urban growth for the reasons noted above, is notfully representative of the urban landscape. Byfocusing on single-family residential development,we have made no attempt to assess the impacts ofdense canyon geometries that can intensify warmingin zones characterized by tightly clustered mediumto high rise buildings. Furthermore, the results ofour analysis are highly responsive to the naturalland cover of the Atlanta study region, which is thatof a mixed deciduous forest canopy. For more aridregions characterized by non-forested pre-develop-ment land covers, the divergence between pre- andpost development fluxes of energy are likely tobe much less pronounced. And, finally, while weassume a reasonably close correspondence betweenthe surface and near surface canopy heat islands inAtlanta, averaged over time, we have made noattempt to link parcel based development patternsto air temperatures directly, and thus our resultsmay be presumed to apply to the surface heat islandonly. A more complete coupling of the surface andnear-surface canopy heat islands will require ex-tensive data on regional air circulation patterns andwaste heat emissions. Given our limited focus on therole of residential land use patterns in the enhance-

ment of surface radiant heat fluxes, we have notattempted to model these more complex interactionsherein.

In closing, we believe heat island abatement willprove to be one of the most effective means for citiesto adapt to the process of climate change over time.While the drivers of the greenhouse effect are globalin nature, and are largely insensitive to the policyactions of individual cities or regions, regional scaleclimate change in the form of heat island formationcan be abated over the near to medium termthrough comprehensive land use controls, buildingdesign, and energy conservation at the metropolitanscale. As a result, municipal decision makersconcerned about the implications of climate changeon human and environmental health can take stepsin the near term to mitigate the thermal impacts ofurban land use on surface temperatures, ambienttemperatures, and air quality. In formulating suchlocal scale strategies, it is imperative that land useplanners have a more complete understanding of therole of land use policies in heat island formation. Tothis end, this work is intended to inform a morecomprehensive approach to climate change manage-ment in cities.

References

Alberti, M., 1999. Urban patterns and environmental perfor-

mance: what do we know? Journal of Planning Education and

Research 19, 151–163.

Carlson, T., Gillies, R., Schmugge, T., 1995. An interpretation of

methodologies for indirect measurement of soil water content.

Agricultural and Forest Meteorology 77 (3–4), 191–205.

Golany, G., 1996. Urban design morphology and thermal

performance. Atmospheric Environment 30, 455–465.

Hoyano, A., 1984. Relationships between the type of residential

area and the aspects of surface temperature and solar

reflectance (based on digital image analysis using air-

borne multispectral scanner data). Energy and Buildings 7,

159–173.

Jensen, J., 2000. Remote Sensing of the Environment: An Earth

Resource Perspective. Prentice-Hall, Upper Saddle River.

National Oceanic and Atmospheric Administration (NOAA),

2000. Summer forecast—hot! NOAA, CDC work to save lives

from potential deadly heat (News Online Story’ #455).

Retrieved 10 June, 1003, from http://wTA’wnoaanews.noaa.

gov/stories/s445.htm

Nichol, J., 1996. High-resolution surface temperature patterns

related to urban morphology in a tropical city: a satellite-

based study. Journal of Applied Meteorology 135, 135–146.

Oke, T., 1987. Boundary Layer Climates. Routledge, New York,

NY.

Price, J., 1979. Notes and correspondence: assessment of the

urban heat island effect through the use of satellite data.

Monthly Weather Review 107, 1554–1557.

ARTICLE IN PRESSB. Stone, J.M. Norman / Atmospheric Environment 40 (2006) 3561–3573 3573

Roth, M., Oke, T., Emery, W., 1989. Satellite-derived urban heat

islands from three coastal cities and the utilization of such

data in urban climatology. International Journal of Remote

Sensing 10, 1699–1720.

Stone, B., 2004. Paving over paradise: how land use regulations

promote residential imperviousness. Landscape and Urban

Planning 69, 101–113.

Stone, B., 2005. Urban heat and air pollution: an emerging role

for planners in the climate change debate. Journal of the

American Planning Association 71, 13–25.

United Nations Department of Economic and Social

Affairs Population Division, 2001. World Urbanization

Prospects: The 1999 Revision. United Nations, Washington,

DC.