SEPTEMBER 2015 S M O L I A K E T A L . 1899

Dense Network Observations of the Twin Cities Canopy-Layer Urban Heat Island*

1 BRIAN V. SMOLIAK, PETER K. SNYDER,TRACY E. TWINE,PHILLIP M. MYKLEBY, AND WILLIAM F. HERTEL Department of Soil, Water, and Climate, University of Minnesota, St. Paul, Minnesota

(Manuscript received 12 September 2014, in final form 16 May 2015)

ABSTRACT

Data from a dense urban meteorological network (UMN) are analyzed, revealing the spatial heterogeneity and temporal variability of the Twin Cities (Minneapolis–St. Paul, Minnesota) canopy-layer urban heat island (UHI). Data from individual sensors represent surface air temperature (SAT) across a variety of local climate zones within a 5000-km2 area and span the 3-yr period from 1 August 2011 to 1 August 2014. Irregularly spaced data are interpolated to a uniform 1 km 3 1 km grid using two statistical methods: 1) kriging and 2) cokriging with impervious surface area data. The cokriged SAT field exhibits lower bias and lower RMSE than does the kriged SAT field when evaluated against an independent set of observations. Maps, time series, and statistics that are based on the cokriged field are presented to describe the spatial structure and magnitude of the Twin Cities metropolitan area (TCMA) UHI on hourly, daily, and seasonal time scales. The average diurnal variation of the TCMA UHI exhibits distinct seasonal modulation wherein the daily maximum occurs by night during summer and by day during winter. Daily variations in the UHI magnitude are linked to changes in weather patterns. Seasonal variations in the UHI magnitude are discussed in terms of land– atmosphere interactions. To the extent that they more fully resolve the spatial structure of the UHI, dense UMNs are advantageous relative to limited collections of existing urban meteorological observations. Dense UMNs are thus capable of providing valuable information for UHI monitoring and for implementing and evaluating UHI mitigation efforts.

1. Introduction London, United Kingdom, by Howard (1833), and later confirmed by Chandler (1965). The basic physics that The systematic manipulation and replacement of accounts for the existence of UHIs was deduced by natural environments with built environments in and Howard (1833) in the early nineteenth century and was around cities affect surface climate such that urban areas formalized in the late twentieth century (Nunez and are warmer than their rural environs. This phenomenon, Oke 1977; Goward 1981; Oke 1982); grounded in an known as the urban heat island (UHI), appears at the energetic basis, the underlying physics implies that surface and throughout the lower atmosphere, from the UHIs are present anywhere that humans have urbanized canopy layer to the boundary layer (Rao 1972; Oke the land surface in a manner that perturbs the surface 1976). Evidence suggesting the existence of UHIs in energy budget from its natural background. Given the surface air temperature (SAT) was first documented for dramatic land-use changes that are associated with the construction of modern cities, it is unsurprising that UHIs are pervasive. UHIs have been documented in * Supplemental information related to this paper is available at cities around the globe (Arnfield 2003; Peng et al. 2012; the Journals Online website: http://dx.doi.org/10.1175/JAMC-D- Hertel 2014). 14-0239.s1. 1 For all the gains in descriptive and theoretical Current affiliation: The Climate Corporation, Seattle, Washington. knowledge of UHIs, challenges and uncertainties re- main. A paucity of urban temperature observations limits the quantification of canopy-layer UHI magnitude Corresponding author address: Peter K. Snyder, Dept. of Soil, Water, and Climate, University of Minnesota, 439 Borlaug Hall, to simple frameworks such as the difference between 1991 Upper Buford Circle, St. Paul, MN 55108. averages of two or more stations separated into urban E-mail: [email protected] and rural categories (Hage 1972; Ackerman 1985;

DOI: 10.1175/JAMC-D-14-0239.1

Ó 2015 American Meteorological Society Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1900 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54

Basara et al. 2008). Urban–rural classification schemes urban ecosystem stresses (Baker et al. 2002). UHIs are have dominated the UHI literature of the past century, superimposed upon naturally occurring regional climate providing the best available observational estimates of variability (Lowry 1977) as well as large-scale warming canopy-layer UHI magnitude. Stewart’s (2011) system- caused by increasing concentrations of greenhouse gases atic review of UHI studies has suggested that there are in Earth’s atmosphere (Grimmond 2007). It follows that often serious deficiencies in station metadata, leading to distinguishing among local impacts of the UHI, regional ambiguous urban–rural classifications and potentially climate variability, and global warming requires urban inaccurate canopy-layer UHI estimates. Furthermore, observations that are capable of resolving all three urban–rural classification schemes often reduce the rich sources of local climate variability. Here we demon- heterogeneity of the urban environment into a single strate how observations from a dense UMN may be used number dominated by the thermal source areas around a to facilitate the separation of the UHI from background small number of stations that may or may not ade- regional climate variability. quately represent the urban form (Oke 2006). Mobile Distinguishing between time-varying changes in the sensing platforms have, in some cases, temporarily re- UHI and global warming is a larger challenge that is solved this issue by providing skeletal snapshot views of beyond the scope of this study. At the scale of individual the UHI along roadways throughout cities and their cities and their surrounding metropolitan areas, natural surrounding rural areas (Oke 1995). In this sense, mo- variability partially obscures long-term trends, regard- bile traverses tend to trade increased spatial information less of their cause (Wallace 1996). The data record used for decreased temporal information (Oke 1973). Co- in this study is not long enough to warrant investigating operative observing networks coordinated by govern- secular trends in the local UHI. In the future, UMNs ment agencies may be capable of resolving the UHI sustained over decades with thoroughly documented (Hausfather et al. 2013) but are susceptible to data in- metadata may be capable of separating time-varying homogeneity (Wu et al. 2005). Dedicated and highly changes in the UHI from the local interplay of regional controlled urban meteorological networks (UMN) climate variability and global warming. composed of fixed meteorological sensors are a rela- tively recent development and are capable of providing improved spatiotemporal observations of canopy-layer 2. Study area, network details, and gridding UHIs (Grimmond 2006; Muller et al. 2013a). procedure This study describes results that are based on data a. Twin Cities metropolitan area and its UHI from a dense UMN in the Twin Cities metropolitan area (TCMA), encompassing the cities of Minneapolis and The TCMA is composed of Minneapolis, St. Paul, and St. Paul, Minnesota. It complements and advances pre- their surrounding communities in east-central Minne- vious observational studies of UHIs by overcoming most sota, just west of the Minnesota–Wisconsin border of the limitations inherent in the sparse networks that (Fig. 1a). The study area, indicated by the dashed box in are used to characterize most urban areas (World Fig. 1b, is located in the Dfb Köppen–Geiger climate Meteorological Organization 2008; Muller et al. 2013a). zone (snow and fully humid warm summer; Kottek et al. In this paper, we 1) describe the essential features of the 2006) and covers 5000 km2 within the seven counties of network; 2) present improved estimates of the TCMA the TCMA (Anoka, Carver, Dakota, Hennepin, Ram- canopy-layer UHI, including its diurnal, seasonal, and sey, Scott, and Washington). It is traversed by the Mis- nonseasonal variability; 3) relate its spatial and temporal sissippi, Minnesota, and St. Croix River valleys, above variability to human and natural factors such as mete- which the elevation varies gradually around an average orological conditions and land use; and 4) discuss im- of 275 m above sea level. The landscape is interspersed plications for network design, UHI monitoring, and with approximately 900 lakes, of which the largest is mitigation that are pertinent to investigators and Lake Minnetonka on the western periphery of the agencies seeking to replicate our efforts in other cities. TCMA. A variety of land cover and land uses exist in the Accurate quantification of the canopy-layer UHI TCMA: high-density urban development in the core magnitude is important because of its direct impacts on communities of Minneapolis, St. Paul, and Bloomington, human activity and the compounding effect of climate Minnesota; medium- to low-density urban development variability at larger scales. UHIs have a variety of eco- in the surrounding suburban communities; and agricul- nomic, social, and environmental consequences; they ture, undeveloped wetlands, forests, and prairie in its are associated with increased energy consumption rural environs. (Santamouris et al. 2001), increased heat stress (Kovats The TCMA UHI has been studied over several de- and Hajat 2008), decreased air quality (Stone 2005), and cades. Winkler et al. (1981) was one of the earliest

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1901

FIG. 1. (a) The seven-county TCMA in east-central Minnesota; (b) the distribution of temperature sensors that compose the UMN described in this study, rural airport observations used to define the rural background reference temperature, and urban-airport observations used to validate the spatial interpolation schemes. Major interstate highways are indicated with dark gray lines. Major lakes and rivers are indicated with light blue shading. The largest lake in the TCMA, Lake Minnetonka, is located west of the urban area and is outlined in dark blue. Percent of impervious surface area is indicated with gray shading, as based on NLCD2011 data (30-m resolution; see text for details). Spatial interpolation is performed over the entire study area, whereas data are analyzed only within the TCMA domain (bounded by a dashed box); (c) the urban, suburban, and rural areas within the TCMA. The locations of the Minneapolis and St. Paul CBDs are indicated with stars. studies to produce a quantitative characterization of the during winter days than nights, that the presence of snow TCMA UHI. Using a 10-yr data record composed of 20 increased the UHI magnitude during daytime and cooperative weather stations and the local National nighttime, and that moderate snowfall tended to di- Weather Service station, Winkler et al. (1981) estimated minish the UHI magnitude. Their network did not de- the spatial extent and average magnitude of the TCMA tect any significant relationship between the intensity of UHI for the annual mean (1.08C) and the calendar urbanization and the TCMA UHI magnitude. As part of months of January (1.18C) and July (1.28C). Todhunter an analysis of UHIs in the world’s largest cities for the (1996) used a slightly larger network of 26 cooperative period of 2006–12, Hertel (2014) analyzed the TCMA and television weather stations to describe the effect of UHI by contrasting the average of a set of hourly rural the TCMA UHI on a variety of climatological indices airport observations against concurrent observations at during 1989. Whereas Winkler et al. (1981) showed two urban airports: KMSP and St. Paul Downtown hand-analyzed maps of the TCMA UHI, Todhunter (KSTP). Hertel (2014) documented the TCMA UHI on (1996) employed ordinary kriging to analyze local cli- the basis of the annual (1.38C), wintertime (0.98C), and mate statistics on a regular grid. Malevich and Klink summertime (1.88C) averages. (2011) studied the relationship between snow and the b. Network design, installation, and administration TCMA UHI, which they defined as the difference be- tween 19 ‘‘iButton’’ sensors around the Minneapolis This study employs data from a dense UMN designed central business district (CBD) and the Minneapolis– to advance knowledge of the TCMA canopy-layer UHI St. Paul International Airport (KMSP), located 10 km to by addressing limitations intrinsic to the observation the southeast. They reported that the UHI was larger networks used in previous studies (e.g., sparse networks

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1902 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54 or a limited period of record). Installation began in June TABLE 1. LCZs represented by the UMN described in this study, of 2011. By August of 2011, the nascent network had along with the corresponding number of UMN stations associated nearly 30 sensors located throughout the urban core and with each category, in terms of their primary and secondary clas- sifications. See Stewart and Oke (2012) for LCZ details. LCZs 1, 2, the suburban and rural areas, sufficient to capture the 3, 7, 10, C, E, and F are not represented by stations that compose urban–rural temperature gradient across the study area. the UMN described in this study. As of late 2012, the UMN reached a stable size of ap- Primary class Secondary class proximately 170 sensors distributed throughout the Local climate zone (stations) (stations) 5000-km2 study area, surpassing previous TCMA UHI studies by nearly a factor of 10. 4, open high-rise 4 0 5, open midrise 3 0 The network samples individual neighborhoods in the 6, open low-rise 112 4 urban core as well as the varied natural and agricultural 8, large low-rise 0 22 environs of the surrounding periurban areas (Fig. 1). In 9, sparsely built 79 4 the central cities of Minneapolis and St. Paul, the aver- A, dense trees 1 5 age distance between a sensor and its nearest three B, scattered trees 4 19 D, low plants 5 13 neighbors is 1.7 km. In the surrounding suburban and G, water 0 20 exurban communities, the average distance between a sensor and its nearest three neighbors is 3.8 km. The average elevation of observation sites is 276 m. The stan- dard deviation of the observation sites about the average the volunteer (e.g., that their property remains mini- elevation is 18 m, and the range is 216–316 m. None of the mally affected in terms of aesthetics and function). observation sites are located in complex terrain subject Stewart and Oke (2012) introduced local climate to topographically modified flows. Only a very small zones (LCZs) as a formal manner of categorizing the fraction of the TCMA is located in the narrow upper environments in which local meteorological observa- Mississippi River valley, which by virtue of its topogra- tions are taken. Table 1 lists each LCZ and presents the phy is essentially unpopulated. number of sites in each category as of 1 September 2014 The observation sites in our network were deter- (primary LCZ class is in the first column; secondary mined through the following process: First, local citizen LCZ class is in the second column). The network volunteers were solicited from contributors to two represents a diversity of built environments in the existing local observing networks: the cooperative TCMA: from the open high, mid-, and low rises of the Community Collaborative Rain, Hail, and Snow urban core to the sparsely built and undeveloped natural (CoCoRaHS) network and the Minnesota State Cli- areas in communities surrounding Minneapolis and matology Office’s MNgage precipitation network. We St. Paul. No stations are sited in compact high-, mid-, or subsequently expanded our network of volunteers low-rise areas because, with the exception of the through our research project Internet page (http:// downtown CBDs, these built environments do not exist islands.umn.edu) and local media outlets (television in the TCMA. The same is true of lightweight low-rise and newspaper), and by establishing partnerships with areas. Heavy industry and large low-rise commercial the municipal parks departments in Minneapolis and areas are present in the TCMA but are not represented St. Paul. Drawing from this list of volunteers, we sought among the primary LCZ classes, primarily because the to balance areal extent and density, with the goal of a volunteers for our network were private citizens and not denser network in the urban core and an areal extent private organizations with property in commercial or that reached into the rural areas beyond the TCMA. industrial areas. Because there are large low-rise com- Once a volunteer is accepted into the network, we site mercial areas adjacent to a number of the sensor sites, an observing platform in consultation with the World these areas are represented among the secondary LCZ Meteorological Organization initial guidance described classes. Some of the sensors are sited in the vicinity of by Oke (2006). To minimize aliasing differing micro- lakes, which are also represented among the secondary climates into our data, we seek sites with similar ground LCZ classes. cover immediately surrounding the sensor platform Data from individual sensors in the network are (generally low plants such as grasses). Sensors are al- downloaded manually every 3–4 months. As new data ways sited away from potential lateral heat sources such are collected, they are converted from their native as large buildings, air conditioning units, exhaust vents, comma-separated-value form to network common data and large trees. Beyond these considerations, the siting form (NetCDF), processed by an automated quality- reflects a compromise between the needs of the project assurance procedure, aggregated with existing station (i.e., ground cover and sky exposure) and the desires of data, and archived.

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1903 c. Sensor type, calibration, and quality assurance change of 28C or greater relative to both the pre- ceding and subsequent 15-min observations. This Individual dataloggers (HOBO1 Pro v2 model num- check identifies physically implausible temperature ber U23-004) are deployed at each site with naturally variability. aspirated radiation shields. Each logger records air 3) In a spatial consistency check, we flag any hourly temperature measurements via an external sensor con- observations that are more than six standard de- nected to the waterproof case, as well as an internal viations (6s) from the network mean at a particular temperature measurement. The preferred deployment time point. This check identifies temperature obser- for each datalogger is on a standard observation plat- vations with pronounced, isolated microclimatic in- form, composed of a 5-cm-diameter post, a logger, a fluences as well as some additional climatological sensor, and a radiation shield. The radiation shield is outliers that are not identified by either of the first attached to the post such that the measurement is taken two checks. as close to 2 m as possible. In cases in which the standard observing platform cannot be deployed because of The first two QA procedures are applied to the raw property restrictions, the logger, external sensor, and 15-min data. The spatial consistency check is applied radiation shield are attached to a suitable object (e.g., after averaging the data to hourly intervals. Observations flagpole, light post, or fence post) as close to 2 m as flagged as outliers are recorded and removed from the possible. The dataloggers are launched in the field and dataset. Random samples of flagged values were man- configured to record observations every 15 min. ually inspected to assess the validity of the observation The entire set of 200 dataloggers was calibrated in two and the fidelity of the QA check. separate groups of 100. Each logger was given a unique identifier (ID), launched with HOBOware software, and d. Spatial interpolation with kriging and cokriging successively placed in four uniform environments to Hourly UMN data that pass all three QA procedures verify the sensor specifications: a laboratory at 208C are interpolated to a uniform 1 km 3 1 km grid using two (room temperature), a deep freezer at 2408C, a cooler geostatistical techniques: kriging and cokriging (covar- at 08C, and the University of Minnesota St. Paul campus iate kriging). The grid was determined by the smallest weather station at 328C (field conditions). Dataloggers station separation distances in the urban core. The that departed from the manufacturer specifications at prediction grid covers a domain that fully encompasses the time of calibration were withheld from deployment. the TCMA: from 44.558 to 45.358N latitude and from See the supplemental online material for detailed sensor 93.808 to 92.758W longitude (dashed line in Fig. 1). To specifications. Dataloggers exhibiting abrupt disconti- anchor the interpolation schemes at the boundaries nuities, persistent errors, or failure while deployed were of the analysis domain, we supplement our UMN ob- removed from the field, tested, and replaced with a new servations with rural airport observations at nine loca- or recalibrated datalogger. Successfully recalibrated tions surrounding the TCMA. These airports are listed dataloggers were only redeployed after being updated in Table 2 and are shown with black squares in Fig. 1.To and launched with a new ID. evaluate the kriged and cokriged temperature fields, we To ensure that our data are free from spurious sensor cross validate the gridded temperature data by calcu- readings, we employ a series of three automated data lating RMSE using a set of six TCMA airport observa- quality-assurance (QA) procedures that are consistent tions that are not used in either interpolation scheme. with those described by Durre et al. (2010): These urban airports are also listed in Table 2 and are 1) In a local all-time-record check, we flag any 15-min shown with white squares in Fig. 1. observations that fall outside the all-time records for Kriging is a standard geostatistical tool that interpolates Minneapolis–St. Paul (temperature T ,236.68Cor a distributed field of data using information inherent in the T . 42.28C). This check identifies physically im- spatial covariances (Journel and Huijbregts 1978). Toward plausible temperature observations. that end, kriging uses semivariograms to express the spatial 2) In a temporal consistency check, we flag 15-min dependence of the field around a known data point and to observations that exhibit a change of 648C or greater develop the kriging weights that form linear combinations relative to either the preceding or subsequent 15-min at prediction data points. We develop a semivariogram for observation, as well as observations that exhibit a each set of hourly canopy-layer air temperature observa- tions using an exponential model whose three parameters (nugget, range, and sill) are determined by nonlinear least 1 HOBO and HOBOware are registered trademarks of Onset squares regression (Kawashima and Ishida 1992). Because Computer Corporation in Bourne, Massachusetts. there are no dramatic differences in the background

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1904 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54

TABLE 2. Airport observing stations (here, ID indicates station model, a linear variogram fit is applied. A range cutoff of identifier) used to define the background rural reference SAT and 25 km is applied when producing the kriged fields, ob- as boundary points in the interpolation schemes described in the tained by averaging the range parameter over all time text, as well as the TCMA stations used for the purpose of vali- dation. Hourly data for each station were obtained from NOAA points. The range parameter represents the distance be- NCDC using Climate Data Online (https://www.ncdc.noaa.gov/ yond which the autocorrelation is essentially zero. Thus, cdo-web/). the range cutoff restricts the kriged temperature at a given point to nearby observations within a range of 25 km. Name ID Lat Lon Elev Cokriging leverages covariance between a sparsely Background rural reference/boundary stations sampled prediction variable and a well-sampled sec- Airlake KLVN 44.6288N 93.2288W 292 m Cambridge KCBG 45.5598N 93.2658W 288 m ondary variable to interpolate the former with greater Glencoe KGYL 44.7568N 94.0818W 302 m accuracy than may be possible with kriging alone (Ishida L O Simenstad KOEO 45.3088N 92.6908W 275 m and Kawashima 1993). In our application, we seek to Maple Lake KMGG 45.2368N 93.9868W 313 m predict canopy-layer air temperature on a uniform 8 8 New Richmond KRNH 45.150 N 92.533 W 304 m 1km 3 1 km grid from a set of irregularly distributed Princeton KPNM 45.5608N 93.6088W 298 m Red Wing KRGK 44.5898N 92.4858W 238 m temperature observations and a field of gridded imper- Stanton KSYN 44.4768N 93.0168W 280 m vious surface area (ISA) data. As an extension of the Validation stations kriging procedure, cokriging uses a cross semivariogram Anoka County Blaine KANE 45.1508N 93.2178W 278 m to represent the spatial dependence of the two variables. 8 8 Crystal KMIC 45.062 N 93.351 W 262 m Impervious surfaces such as pavements do not permit Flying Cloud KFCM 44.8328N 93.4718W 276 m Minneapolis–St. Paul KMSP 44.8838N 93.2298W 266 m infiltration of rainwater and have pronounced thermal South St. Paul KSGS 44.8608N 93.0338W 250 m mass relative to unsaturated soil and naturally vegetated St. Paul downtown KSTP 44.9328N 93.0508W 213 m surfaces. When insolation is present, these differences contribute to contrasting thermal storage and radiative cooling rates, which may lead to the development of a climatic data over the TCMA owing to elevation or land– UHI under certain conditions. For example, Zhang et al. sea contrasts, we employ so-called simple kriging, which (2011) reported a statistically significant positive corre- assumes a stationary mean. The kriging procedure begins lation between imperviousness and the strength of the by estimating the semivariogram using all available data canopy-layer UHI magnitude in their study of Detroit, for a given time point. The three semivariogram param- Michigan, during the summer of 2008. Schatz and eters are estimated for each hour between 2000 central Kucharik (2014) also reported a significant correlation standard time (CST; UTC 2 6h)31July2011and0400 between imperviousness and the magnitude of the UHI; CST 1 August 2014. In cases of a singular semivariogram they also found that statistical models that are based on

FIG. 2. An example of the application of kriging and cokriging with impervious surface area data to (a) daily-mean SAT data (8C) observed by the UMN network on 16 May 2012, (b) the kriged SAT field and urban-airport observations, with size corresponding to the absolute error of the interpolated value as indicated below the panel, and (c) the cokriged SAT field and urban-airport observations, with size corresponding to the absolute error of the interpolated value.

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1905

TABLE 3. Bias (gridded minus observed; 8C) and RMSE (8C) of gridded daily-mean temperatures near six TCMA urban-airport stations withheld from the kriging and cokriging interpolation schemes (see Table 2 for locations).

Bias RMSE Name ID Kriging Cokriging Kriging Cokriging Anoka County Blaine KANE 20.94 20.63 1.26 0.95 Crystal KMIC 20.16 0.05 0.49 0.17 Flying Cloud KFCM 20.47 20.34 0.65 0.38 Minneapolis–St. Paul KMSP 20.41 20.09 0.62 0.22 South St. Paul KSGS 21.04 20.79 1.18 0.86 St. Paul downtown KSTP 20.32 20.07 0.63 0.33 Avg 20.56 20.31 0.80 0.49 imperviousness outperformed models that are based on A secondary concern is the time required to produce the normalized difference vegetation index, a measure the field estimated with cokriging. As the volume of ISA of vegetation abundance. Our study utilizes a set of data increases, so does the computation time. In the 30-m-resolution ISA data from the National Land Cover study area, many of the 500-m ISA values are zero, in Database 2011 (NLCD2011; Jin et al. 2013). To capture particular over rural areas. To balance the sampling of the imperviousness of the thermal source region sur- the ISA field with manageable computation time, we rounding prediction points, we regrid the 30-m-resolution randomly selected an equal number of sampling points ISA data to 500-m resolution. Depending on meteoro- from each quintile of the ISA data. This approach re- logical conditions and surface character, the thermal duces the dimensionality of the 500-m ISA field from source area of a standard 2-m temperature observa- 34 000 grid points to 4000 grid points. We tested the tion may extend from tens of meters to greater than sensitivity of our results to the choice of the number of 1km(Kljun et al. 2002; Stewart and Oke 2012). Our 500-m ISA grid points included in the cokriging scheme results were not sensitive to the choice of ISA averaging and found little sensitivity, in particular when more than radius for values of less than 750 m. Following the ap- 2500 grid points were included. The same 4000 ISA grid proach to kriging, the temperature estimates produced points were used in each cokriging estimate. by cokriging at each prediction point are based on To validate the interpolated SAT fields, we use hourly temperature observations and 500-m ISA data within a SAT observations from the six urban Twin Cities airport range of 25 km. stations listed in Table 2 to calculate bias and RMSE

FIG. 3. (a) The spatial distribution of the daily-mean TCMA UHI averaged over the 3-yr study period from August 2011 to August 2014, expressed as anomalies (8C) from the back- ground rural reference SAT. The zero line is indicated with a dashed line. Major highways are indicated with gray lines. The Minneapolis CBD is indicated with a black circle. The St. Paul CBD is indicated with a black triangle. The Minneapolis–St. Paul International Airport is in- dicated with a black square. County boundaries are indicated with black lines. (b) The PDF corresponding to the SAT anomalies in (a).

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1906 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54

FIG. 4. Diurnal variations in the TCMA UHI (8C) averaged across the 3-yr study period from 1 August 2011 to 1 August 2014 over (left) urban and (right) suburban areas, as indicated in Fig. 1c, for (a),(b) all calendar months, (c),(d) sum- mertime (June–September), and (e),(f) wintertime (December–March). The area average is indicated with a solid line, and the spatial 5th–95th percentile range (nearest-rank method) is indicated with shading. Times correspond to CST. relative to the closest grid point. These urban-airport UMN-based gridded fields and the observations from observations were not included in the interpolation the airport automated surface observing system (ASOS) schemes. Thus they represent independent data that because of the differing manner in which the sensors may be used to evaluate kriging and cokriging. housings are ventilated. The UMN temperature sensors The distribution of daily-mean SAT observed at are naturally aspirated by wind; the airport ASOS TCMA UMN sensor sites on 16 May 2012 is shown in temperature sensors are mechanically aspirated by fans. Fig. 2a. The corresponding kriged daily-mean SAT field The average bias is only slightly larger than the reported is shown in Fig. 2b, along with station observations from standard error for the HOBO sensors (60.28C), which six urban airports withheld from interpolation. The lends confidence to the gridding approach. The valida- kriged field captures the general pattern of urban tion statistics associated with the cokriged SAT field are warmth but lacks definition where observations are clearly superior to the kriged SAT field; therefore we sparse, in particular at the edge of the study domain. By will hereinafter only refer to results that are based on contrast, the field cokriged with ISA data exhibits the cokriged SAT field. Other approaches to validation, greater fidelity to observations, both in terms of the such as vehicle traverses, could help to verify the in- pattern of urban warmth and in comparison with the stantaneous structure and magnitude of the TCMA urban-airport validation stations. Bias and RMSE are UHI, but they are limited in their ability to observe the calculated at the six grid points collocated with the UHI continuously over its temporal evolution and thus urban-airport validation sites and are shown in Table 3 remain beyond the scope of this study. along with the average over all six locations. Both e. Metadata metrics exhibit smaller values at all six validation locations for the cokriged field when compared with In addition to the network and sensor details that have the kriged field. Some bias is expected between the been described in this section, extensive metadata are

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1907

FIG. 5. The spatial distribution of the TCMA UHI (8C) averaged over two seasonally varying diurnal periods: (a) nighttime, defined as the period from sunset to sunrise, and (b) daytime, defined as the period from sunrise to sunset; (c) PDFs corresponding to the patterns in (a) and (b). The display is as in Fig. 3. provided in the online supplemental material, following the nine rural airport observations used as rural boundary the standardized UMN metadata protocol introduced conditions in the interpolation schemes described in sec- by Muller et al. (2013b). tion 2. These municipal airports are located outside the urban core, well beyond the influence of the TCMA UHI. They are unambiguously rural in character, situated over 3. Results low plants away from the influence of built structures. They also encircle the TCMA, which provides a spatially a. Defining the UHI representative average of its rural environs. We do not The canopy-layer UHI has typically been defined as the specifically account for the passage of mesoscale bound- difference between urban and rural screen-level (2 m) aries through our study area, but our area-average rural temperature observations (Oke 1982). Despite broad background SAT and choice to consider time averages of agreement on this fundamental concept, the UHI litera- six or more hours reduce the likelihood that these features ture is complicated by a diversity of station classification unduly influence our results. schemes and UHI magnitude definitions (Stewart 2011). To investigate the spatial characteristics of the TCMA Because of a paucity of unambiguously urban obser- UHI, we define and analyze a time-varying SAT vations in major municipalities, studies have often relied anomaly field by subtracting the rural background SAT on as few as two stations (one urban, one rural) to es- from the cokriged SAT field described in section 2. timate the UHI (e.g., Ackerman 1985; Brazel et al. 2000; Therefore the UHI is present wherever local anomalies Fortuniak et al. 2006). Where a greater number of urban are greater than zero. The dataset has hourly sampling observations exist, investigators have begun to define and extends continuously over the 3-yr period from 1 ‘‘urban’’ and ‘‘rural’’ on the basis of averages of two or August 2011 to 1 August 2014. This allows us to resolve more stations (e.g., Kim and Baik 2005; Basara et al. the diurnal evolution of the canopy-layer UHI. 2008). Hawkins et al. (2004) pointed out that the spatial Rather than classifying the UMN sensor sites as urban variability of temperature within rural areas could in- or rural and presenting UHI magnitude estimates that are fluence estimates of UHI magnitude. Fortuniak et al. based on the difference of their means, we instead report (2006) noted that their initial estimates of UHI magni- two spatial statistics derived from the spatial UHI distri- tude were affected by the passage of mesoscale bound- bution: the area average (m) and the 95th percentile (P95). aries such as fronts and thunderstorm outflows. Thus, The area average is comparable to studies that estimate they filtered days and nights when these features were the UHI with the average difference between many urban known to be present in the vicinity of the two stations and rural stations, whereas the 95th percentile is compa- used in their study. rable to studies that estimate the UHI by the peak dif- To overcome the aforementioned challenges associated ference between urban and rural stations, often defined as with defining the UHI and estimating its magnitude, we the UHImax. The area average and 95th percentile are define the TCMA rural background SAT as an average of calculated for both urban and suburban areas.

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1908 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54

The study area is classified into urban, suburban, and rural areas in Fig. 1c. These definitions are based on the average ISA within local municipalities. The urban area is defined as contiguous communities with greater than 40% area-average ISA, including the Twin Cities of Minneapolis and St. Paul, as well as their older, more developed first-ring suburbs. The suburban area is de- fined as contiguous nonurban communities with greater than 20% area-average ISA, including second-ring suburbs and several exurban communities that con- tinue to expand the spatial extent of the TCMA. The remainder of the study area is defined as rural, although it should be noted that a small degree of urbanization is present in towns that surround the TCMA. b. Spatial structure of the time-average TCMA UHI The study-mean, daily-mean TCMA UHI shown in Fig. 3a extends across much of the study area but is concentrated in the most developed portions of the re- gion. Local maxima are apparent over the Minneapolis CBD, the St. Paul CBD, and the KMSP airport. In the suburban area, the UHI magnitude is weaker by a factor of 2–3 and appears to be more heterogeneous than in the urban core. Beyond the suburbs, the UHI pattern tapers off, except along narrow corridors that appear to follow the development along the major interstate highways that bisect the TCMA. The PDF corresponding to the pattern in Fig. 3a that is shown in Fig. 3b is positively skewed with a maximum FIG. 6. (a) The daily-mean background rural reference SAT time 8 around zero and a secondary maximum around 0.88C. The series ( C; black line) and low-pass-filtered time series [blue line; based on a Lanczos filter with 25 weights and a low-frequency primary maximum represents the outer extent of the UHI, cutoff of 0.0111 (90 days)]; (b) nighttime-mean (from sunset to and the secondary maximum represents the beginning of sunrise) TCMA UHI time series (8C) averaged over the urban area the inner-core UHI. These respective features also cor- in Fig. 1c, and low-pass-filtered time series as in (a); (c) daytime- respond to what Runnalls and Oke (2000) referred to as mean (from sunrise to sunset) TCMA UHI (8C) averaged as in (b), the UHI ‘‘cliff’’ at the rural–urban edge and the start of and low-pass-filtered time series as in (a). UHI ‘‘peaks’’ in areas of significant urban development. analysis represents (,1 km) and influence too small of a c. Diurnal and seasonal variation portion of the TCMA to affect the area average. On average, the UHI is present throughout the entire The TCMA experiences seasonal variations in the day, but it varies in strength according to time of day, length of day, which have attendant impacts on the diurnal time of year, and location (urban TCMA in Figs. 4a,c,e; evolution of the UHI (see Fig. S1 in the supplemental suburban TCMA in Figs. 4b,d,f). Averaged over the en- online material). Runnalls and Oke (2000) and Fortuniak tire calendar year, the TCMA UHI peaks at midnight, et al. (2006) normalized time to parse out UHI changes with a secondary maximum at midday and minima around independent of season. The current study aims to examine sunrise and sunset (Figs. 4a,b). This behavior is consistent seasonal changes in the UHI, including those driven by with our understanding of the energetic basis of the UHI; changes in length of day; therefore, daytime is defined as that is, it varies diurnally in accordance with the mesoscale the period from sunrise to sunset, and nighttime is defined differences in urban and rural heating and cooling rates as the period from sunset to sunrise. largely attributable to differences in radiative and ther- In the TCMA urban core, the nighttime-mean UHI is modynamic properties (Oke 1982). Microscale differ- stronger than the daytime-mean UHI (Figs. 5a,b), ences in heating and cooling rates (e.g., building shading, whereas in the suburban periphery the nighttime- and cold-air pooling) also explain local UHI magnitudes but daytime-mean UHI patterns are of comparable magni- occur on spatial scales that are smaller than our gridded tude. In other words, the nighttime UHI is characterized

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1909

FIG. 7. The spatial distribution of the TCMA UHI (8C) averaged over nighttime and daytime, disaggregated by season: (a),(b) summertime (June–September) and (c) PDFs corresponding to the patterns in (a) and (b); (d),(e) wintertime (December–March) and (f) PDFs corresponding to the patterns in (d) and (e). The display is as in Fig. 3. by a stronger rural–urban SAT gradient than the day- midnight (Figs. 4c,d). In the cold season, the TCMA time UHI. This is also evident in the corresponding UHI is strongest during the day, peaking around noon spatial PDFs shown in Fig. 5c. The nighttime UHI PDF (Figs. 4e,f). The warm-season behavior is consistent with is similar to the daily-mean PDF shown in Fig. 3b, with previous studies of UHI diurnal variability in mid- a mode near 08C, a secondary maximum around 18C, latitude cities (Kłysik and Fortuniak 1999; Runnalls and and a small set of extreme values. In contrast, the day- Oke 2000; Kim and Baik 2005; Fortuniak et al. 2006; time UHI PDF has a mode near 0.258C, followed by a Basara et al. 2008), and the cold-season behavior agrees mostly uniform distribution that abruptly drops off above with the results of Malevich and Klink (2011), who 1.258C. We examine these differences in greater detail in used a smaller UMN to study the relationship between section 3. the TCMA UHI and snow. When the data are averaged only over the summer Different TCMA UHI patterns are expected between months (June–September) and winter months (December– the warm and cold seasons given the large seasonality of March), their corresponding diurnal evolutions suggest the mean climate in the north-central United States that the small differences between the average night- (Fig. 6a). During the study period, for example, daily- time and daytime UHI magnitude in the annual mean mean SAT varied by nearly 508C from winter to sum- are an artifact of averaging nearly equal and opposite mer. The nighttime-mean UHI time series shown in seasonal behavior. In the warm season, the TCMA UHI Fig. 6b varies on multiple time scales. On daily to weekly is strongest at night, peaking in the hours preceding time scales, the nighttime-mean UHI is largest in the

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1910 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54 cold season, as exemplified by the daily peaks in the TABLE 4. Time-average TCMA UHI magnitude (8C), quantified area-average time series in Fig. 6b. On seasonal time in terms of the area average m and spatial P95 (nearest-rank scales, however, the nighttime-mean UHI is largest in method) across urban and suburban areas, disaggregated by di- urnal and seasonal period as indicated, on the basis of the cokriged the warm season, as exemplified by the time series after hourly SAT anomaly field. Anomalies are calculated relative to the being filtered by a low-pass Lanczos filter with 25 rural background SAT (see text for details). Nighttime is defined as weights and a low-frequency cutoff of 0.0111 (90 days; the period from sunset to sunrise; daytime is defined as the period Duchon 1979). The daytime-mean UHI time series from sunrise to sunset. Summertime is defined as the calendar shown in Fig. 6c exhibits less variability than the months June–September (JJAS). Wintertime is defined as the calendar months December–March (DJFM). nighttime-mean time series, both on daily and seasonal time scales. These contrasts between the nighttime and Summertime Wintertime daytime UHI magnitude become clearer when seasonal (JJAS) (DJFM) averages are compared with seasonal extremes. Night Day Night Day The summertime and wintertime average nighttime Urban m 1.50 0.90 0.88 0.99 and daytime UHI patterns are shown in Fig. 7 along Suburban m 0.96 0.56 0.51 0.71 with their corresponding spatial PDFs. UHI magni- Urban P95 1.99 1.20 1.19 1.19 tudes (m and P95) associated with these maps are listed Suburban P95 1.38 0.88 0.87 1.01 in Table 4, disaggregated by diurnal and seasonal period. During the summer, the UHI is strongest at night over the than in the summertime (Fig. 8c), particularly in the ur- urban core, with magnitudes approaching 28C locally in ban core. Another difference between the time-average and around the KMSP airport and the Minneapolis–St. and extreme UHIs is evident in the contrast between the Paul CBDs. During the day, the summertime UHI is more daytime UHI strength in the summer and winter. In the spatially uniform. Hotspots are still evident, but horizontal time average, there is little difference between the sum- SAT gradients are much weaker during the day than at mertime and wintertime daytime UHI; in the extreme night. During the winter, the UHI is of comparable mag- cases, however, the wintertime daytime UHI (Fig. 8d)is nitude during the day and night, both in the urban core 50% larger than the summertime daytime UHI (Fig. 8b). and the suburban periphery. In fact, some of the largest Taken together, the pronounced difference in UHI differences between the nighttime and daytime winter- magnitude and patterns between the time-average and time UHI are seen in the rural TCMA, where the UHI extreme cases suggests that different processes modu- anomalies are largely negative at night and are near zero late the daily UHI variability and the seasonal UHI or slightly positive during the day. This may be a mani- variability. festation of the contrast between strong radiative cooling by snow-covered surfaces at night and modest radiative e. Modulation by regional meteorological conditions heating by low-albedo pavements, structures, and bare vegetation. The preceding results indicate that the TCMA UHI exhibits differing patterns of variability on daily and d. Extremes seasonal time scales. Previous studies have demon- As discussed in section 3b, the nighttime and daytime strated the pronounced influence of weather on the UHI time series shown in Figs. 6b and 6c exhibit pro- formation and strength of UHIs around the world (e.g., nounced high-frequency daily variability apart from their Ackerman 1985; Runnalls and Oke 2000; Morris and low-frequency seasonal modulation. The time-average Simmonds 2000; Morris et al. 2001; Kim and Baik 2005; maps shown in Fig. 7 illustrate the low-frequency seasonal Yow 2007). Following these analyses, we examined the patterns. To examine the patterns of high-frequency meteorological conditions concurrent with the top 30 daily extremes, we composited the 30 warmest nights extreme nighttime and daytime UHI events during our and days in each season on the basis of the time series in study period. Table 6 lists the number of events with Figs. 6b and 6c. Maps are shown in Fig. 8, and the asso- particular categorical sky cover, wind speed, tempera- ciated spatial statistics are shown in Table 5. ture, and dewpoint conditions. Regardless of season, Like the seasonal contrasts that are evident in the pronounced nighttime UHI events are strongly favored time-average UHI patterns (Fig. 7), the nighttime ex- under clear or scattered sky cover conditions with light treme UHIs (Fig. 8a) are much stronger than the day- or no winds. Furthermore, these nighttime events are time extreme UHIs (Fig. 8b). Unlike the seasonal generally associated with warm conditions during the contrasts that are evident in the time-average UHI summertime and very cold conditions during the win- patterns, the nighttime extreme UHIs are stronger and tertime. The former are indicative of the residual day- more spatially concentrated in the wintertime (Fig. 8a) time heat retained in the urban environment during an

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1911

FIG. 8. The spatial distribution of the TCMA UHI (8C), averaged over the top 30 events in each diurnal [(left) nighttime and (center) daytime] and seasonal [(a),(b) summertime and (d),(e) wintertime] period, as defined by the magnitude of the urban time series shown in Figs. 6b and 6c. (c),(f) The PDFs corresponding to the patterns in (a), (b), (d), and (e), as indicated. The display is as in Fig. 3. extreme warm-season UHI event. The latter are con- urban core elevated and maintain any residual UHI sistent with enhanced anthropogenic heat flux to the from the previous night. Another possibility is that at- urban environment during the wintertime (Sailor and mospheric and/or soil moisture act to influence the Lu 2004). Intense daytime UHI events do not show a thermal admittance of the surface in rural areas, help- particular tendency toward sky cover or wind condi- ing the surface to retain heat at night and keep cool tions, but they appear to be associated with warmer during the day (Runnalls and Oke 2000). conditions that are indicative of greater energy flux into To further understand the signals implied by the statis- the urban land–atmosphere system. During the summer, tics in Table 6, we also examined composite difference there is some indication that extreme nighttime UHI maps that represent the change in UHI magnitude under events occur during dry conditions, when large vapor opposing sky cover and wind speed conditions (Fig. 9). For pressure deficits could conceivably drive enhanced ra- example, on calm nights during summer, clear skies con- diative cooling in the rural areas. In contrast, Table 6 tribute to a TCMA UHI that is 0.58–18C more intense than suggests that extreme daytime UHI events occur under when overcast skies are present (Fig. 9a). On clear nights moist conditions during the warm season. Increased at- during summer, the UHI is more than 18C stronger when mospheric moisture would tend to make the lower at- calm or light winds prevail across the TCMA than when 2 mosphere more emissive during the day, absorbing and skies are clear and winds are greater than 5 ms 1 (Fig. 9b). reemitting outgoing longwave radiation from the sur- Local maxima are evident over the Minneapolis and face, which would tend to keep temperatures in the St. Paul CBDs in the summertime patterns (Figs. 9a,b), as

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1912 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54

TABLE 5. Composite-average TCMA UHI magnitude (8C), as TABLE 6. Number of UHI events with indicated meteorological based on the top 30 daily TCMA UHI events (as defined by the conditions among the top 30 TCMA UHI events in each diurnal magnitude of the area-average urban time series shown in Figs. 6b and seasonal category (as defined by the magnitude of the urban and 6c), quantified in terms of m and spatial P95 (nearest-rank time series in Figs. 6b and 6c) for sky cover S, wind speed W, method) across urban and suburban areas, disaggregated by di- temperature T, and dewpoint D. Meteorological conditions are urnal and seasonal period. based on nighttime and daytime averages of observations at KMSP. Summertime Wintertime (JJAS) (DJFM) Summertime Wintertime Night Day Night Day (JJAS) (DJFM) Night Day Night Day Urban m 2.99 1.85 3.01 2.58 Suburban m 2.13 1.36 2.09 2.08 Overcast (S $ 70%) 0 2 5 5 Urban P95 3.76 2.26 3.82 2.90 Scattered (30% , S , 70%) 8 15 13 12 Suburban P95 2.79 1.81 2.75 2.48 Clear (S # 30%) 22 13 12 13 2 Windy (W $ 5ms 1) 0302 2 2 Breezy (2 m s 1 , W , 5ms 1) 7 22 14 15 well as over Lake Minnetonka in the western TCMA 2 Calm (W # 2ms 1) 23 5 16 13 (indicated with a dark blue outline). On the basis of the $ 8 relative magnitudes of the summertime (Figs. 9a,b)and Hot (T 25 C) 1 21 0 0 Warm (158C # T , 258C) 22 9 0 2 wintertime patterns (Figs. 9c,d), the influence of clouds Mild (58C # T , 158C) 7024 and winds on the UHI is slightly greater in winter than in Cool (258C , T , 58C) 0 0 9 14 summer. The composite difference patterns in Fig. 9 Cold (T # 258C) 0 0 19 10 bear a strong resemblance to the extreme UHI patterns Moist (D $ 158C) 6 17 — — shown in Fig. 8. This confirmatory evidence further Comfortable (108C , D , 158C) 12 7 — — supports the pronounced influence of weather variabil- Dry (D # 108C) 12 6 — — ity on the formation and strength of the UHI on the daily time scale. The patterns also appear to be similar to the The summertime pattern in Fig. 10a contrasts strongly average UHI pattern shown in Fig. 3, which suggests with the wintertime pattern in Fig. 10b, wherein the that the same processes that drive the long-term average frozen lake has no discernable influence on the local UHI also control the daily variability of the UHI. UHI. During winter, the daytime UHI is larger than the nighttime UHI everywhere except a small area that is f. Modulation by land–atmosphere interactions centered just east of the Minneapolis CBD. Over the The pronounced seasonal variability of the TCMA suburban TCMA the daytime and nighttime UHIs are canopy-layer UHI that is evident in the statistics in Table 4, roughly equivalent, and over the rural areas the daytime the diurnal curves in Fig. 4, and the patterns in Fig. 7 implies UHI is larger than the nighttime UHI, suggestive of the major changes in the urban energy balance throughout the strongly emissive, often snow-covered surface. Unlike course of the year. In particular, these results reveal diurnal the lake, which freezes seasonally, the CBD areas re- and seasonal variations in the UHI magnitude. These dif- main exposed throughout the year and are a concen- ferences are highlighted in Fig. 10, which shows differences trated source of anthropogenic heating. Hence, the between nighttime and daytime UHI magnitude in the urban core exhibits a stronger nighttime UHI through- summer (Fig. 10a)andwinter(Fig. 10b) and differences out the year. Conversely, the suburban TCMA experi- between the summertime and wintertime UHI during the ences a larger daytime UHI only during the winter. night (Fig. 10c)andday(Fig. 10d). The summertime pat- The pattern shown in Fig. 10c indicates that the tern shown in Fig. 10a is characterized by positive anoma- nighttime UHI is larger in summer than in winter lies throughout almost the entire domain, testifying to the throughout the TCMA, except for the far southeastern dominance of the nighttime UHI. Two local maxima are portion of the domain. This area is unambiguously evident: one centered just east of the Minneapolis CBD, rural, which suggests the possibility that on average the and one centered over the eastern end of Lake Minne- UHI is advected southeastward during winter. This tonka. The maximum in the western TCMA is likely a re- is consistent with the prevailing northwesterly winds sult of both the cooling influence of the large lake during during that season. Further analysis is necessary to the day and its warming influence during the night. The confirm this adevection, however, possibly following anomaly is not centered directly over the lake but instead is the deployment of additional sensors to that sector of displaced eastward, consistent with the prevailing south- the TCMA. westerly winds during that season. Further study is neces- The daytime pattern shown in Fig. 10d also supports sary to confirm the spatial extent of the lake’s influence. the advection hypothesis, where wintertime UHI magnitudes

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1913

FIG. 9. The spatial distribution of composite UHI differences (8C) for opposing regional meteorological conditions, as indicated, with composite size in parentheses: (a) for warm-season nights, clear and calm (68) minus cloudy and calm (10); (b) for warm-season nights, clear and calm (68) minus clear and windy (13); (c) the PDFs corresponding to the patterns in (a) and (b); (d) for cold-season nights, clear and calm (27) minus cloudy and calm (17); (e) for cold-season nights, clear and calm (27) minus clear and windy (19), (f) the PDFs corresponding to the patterns in (d) and (e). See Table 6 for meteorological-category definitions. Lake Minnetonka is outlined in dark blue as in Fig. 1; otherwise, the display is as in Fig. 3. are larger than summertime UHI magnitudes in the in time and space. For example, whereas previous studies southeastern TCMA. Seasonal differences in the day- presented monthly-mean or annual statistics (Winkler time UHI are small throughout the remainder of the et al. 1981; Baker et al. 1985; Todhunter 1996), the results TCMA domain. presented here are based on hourly data averaged over nighttime and daytime. Malevich and Klink (2011) used a much smaller UMN 4. Discussion and conclusions restricted to a 6-km radius around the Minneapolis CBD in The foregoing results confirm, refine, and extend the their study of the relationship between the TCMA UHI limited set of previous studies on the TCMA canopy- and the presence of snow; they found that the wintertime layer UHI performed over the past several decades UHI is larger during daytime than at night. Our results (Winkler et al. 1981; Baker et al. 1985; Todhunter 1996; support this finding and demonstrate that the daytime Malevich and Klink 2011). They represent an un- peak is also present in the suburban periphery sur- precedented level of spatiotemporal detail for the study rounding the urban core. Similar to Malevich and region, facilitated by a UMN that exceeds previous Klink (2011), we hypothesize that the pattern and in- TCMA network sampling by nearly an order of magnitude tensity of the daytime, wintertime UHI and wintertime

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1914 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54

FIG. 10. The spatial distribution of TCMA UHI differences (8C) between nighttime and daytime periods, disaggregated by season [(a) summertime; (b) wintertime], and UHI differences between summertime and wintertime, disaggregated by diurnal periods [(c) nighttime; (d) daytime]. Lake Minnetonka is outlined in dark blue as in Fig. 1; otherwise the display is as in Fig. 3. differences between daytime and nighttime UHIs are and urban areas, which may be associated with the ab- due to the influence of snow, which limits the thermal sorption of insolation by low-albedo impervious sur- mass of suburban regions by covering dark, impervious faces and dirty snow in the developed portions of the surfaces with its high-albedo, low-conductivity, highly TCMA. Outside the TCMA, small positive UHI emissive surface (Warren 1982; Baker et al. 1991). Be- anomalies are present in the seasonal average and sea- cause of these properties, snow reflects a large amount sonal extremes, which also may be related to absorption of insolation during the day and provides an efficient of insolation, in this case by non-snow-covered surfaces surface for radiative cooling at night. Municipal snow or structures above the snow surface. Notwithstanding removal is ubiquitous in the developed portions of the these new insights, additional years of data and further TCMA, often reexposing dark, impervious surfaces that analysis are necessary to test the snow hypothesis may be warmed more than the surrounding snow- rigorously. covered surfaces (Yuan and Bauer 2007). Further- In addition to contributing to advances with regard to more, urban activity tends to darken snow over the the TCMA UHI, the dense UMN employed for the course of several days following accumulation (Ho and purposes of this study also offers generalized insights for Valeo 2005). The average (Fig. 7e) and extreme (Fig. 8e) the UHI literature. By representing thermal source re- daytime UHI patterns are very uniform across suburban gions across a variety of LCZs (Table 1), the dense

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1915

TABLE 7. Time-average bias of station-based UHI estimates TABLE 8. Composite-average bias of station-based UHI esti- using two TCMA airport locations [KMSP (urban) and KFCM mates using KMSP and KFCM relative to network-based UHI (suburban)] relative to network-based UHI estimates using m and estimates using m and spatial P95 (nearest-rank method) over ur- spatial P95 (nearest-rank method) over urban and suburban areas. ban and suburban areas, on the basis of the top 30 daily TCMA Positive values indicate that, on average, the airport location UHI events (as defined by the magnitude of the area-average urban overestimates the UHI; negative values indicate that, on average, time series shown in Figs. 6b and 6c). Positive values indicate that, the airport location underestimates the UHI. during extreme UHI events, the airport location tends to over- estimate the UHI; negative values indicate that, during extreme Summertime Wintertime UHI events, the airport location tends to underestimate the UHI. (JJAS) (DJFM) Night Day Night Day Summertime Wintertime (JJAS) (DJFM) KMSP urban m 0.40 0.32 0.32 0.26 Night Day Night Day KFCM suburban m 0.44 0.23 0.24 0.15 KMSP urban P95 20.22 20.07 20.13 20.07 KMSP urban m 0.50 0.32 0.38 0.33 KFCM suburban P95 20.22 20.24 20.38 20.36 KFCM suburban m 0.62 0.18 0.33 0.16 KMSP urban P95 20.34 20.08 20.28 20.14 KFCM suburban P95 20.14 20.26 20.50 20.48 network captures much more of the heterogeneous urban– suburban environment than would a sparser network. To briefly illustrate the impact of increased resolution of dense UMNs for UHI monitoring, we did not investigate offered by a dense UMN, we compare the UMN-based the optimal network size for describing the bulk features of statistics shown in Tables 4 and 5 with identical statistics the TCMA UHI, let alone local UHIs in general. that were calculated using only the grid cells corre- Pigeon et al. (2006) designed a UMN optimized sponding to the locations of two TCMA urban airports: for UHI monitoring by strategically distributing 20 KMSP in Bloomington and Flying Cloud Airport (KFCM) sensors to sample the leading EOF of the summertime in Eden Prairie, Minnesota. Bias statistics based on sea- surface air potential temperature and specific humidity sonal averages are shown in Table 7; corresponding bias fields output from a 250-m-resolution mesoscale at- statistics that are based on seasonal extremes are shown mospheric model simulation centered over the city of in Table 8. The bias of airport-based seasonal average Marseille in France. They showed that the leading UHI magnitude tends to be positive for the area average, EOF of the nighttime temperature field corresponds to close to zero for the urban spatial 95th percentile, and the UHI and that the leading EOF of the daytime negative for the suburban spatial 95th percentile. This temperature field corresponds to the sea-breeze circu- suggests that TCMA airports will tend to overestimate lation. They validated their network design by demon- the area-average UHI, that KMSP is a good proxy for the strating that the same EOF structures could be obtained spatially extreme UHI magnitude, and that KFCM un- using the entire field, only the 20 grid points nearest to derestimates the suburban spatial extreme UHI magni- the sensor locations, and the sensor observations them- tude. The same behavior is also observed with respect to selves. Thus, Pigeon et al. (2006) established that a me- the bias of seasonal extremes (Table 8). soscale numerical model could capture the large-scale Depending on where urban observations are located features of the Marseille UHI, which suggests that well- in relation to the large-scale UHI pattern, individual calibrated numerical models may be viewed as an alter- stations may or may not be representative of the area- native approach to quantifying the large-scale features average UHI or spatially extreme UHI magnitudes. Our and area-average magnitude of UHIs. Similar to our use of a UMN underscores the notion that no one urban study, however, Pigeon et al. (2006) did not address how location is likely to be comprehensively representative of many UMN sensors are necessary to adequately describe the entire metropolitan area; that is, the mesoscale features the UHI. Future work could establish a relation among of urban climate must be characterized by multiple stations. city size, land surface heterogeneity, and optimal net- In cities without UMNs, the canopy-layer UHI is generally work density. monitored using existing, limited urban observations. UMNs, such as the one described here, are ideal for Vehicle-based traverses offer instantaneous snapshots but deployment in other cities while emerging observation may not be suitable for sustained observations. Our results platforms mature because they are immediately available, indicate that this small network approach comes at the risk are straightforward to control following published guide- of biased estimates of the local UHI, particularly if the lines, and offer flexibility in spatial and temporal sampling. distribution of station LCZ classifications do not correspond Technologies like cellphone-battery-derived temperature to the distribution of LCZs represented within the city. Al- (Overeem et al. 2013) and remotely sensed land skin though the results presented here demonstrate the efficacy temperature (Jin 2012) are undoubtedly promising but

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC 1916 JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY VOLUME 54 require substantial assumptions to derive estimates of Durre, I., M. J. Menne, B. E. Gleason, T. G. Houston, and R. S. canopy-layer air temperature from their raw observa- Vose, 2010: Comprehensive automated quality assurance of tions. The spatial and temporal resolution afforded by daily surface observations. J. App. Meteor. Climatol., 49, 1615– 1633, doi:10.1175/2010JAMC2375.1. UMNs surpasses that of existing networks. This increase Fortuniak, K., K. Kłysik, and J. Wibig, 2006: Urban–rural contrasts in information content facilitates the identification of of meteorological parameters in Łódz. Theor. Appl. Climatol., spatial structures at finescales relevant in urban areas. For 84, 91–101, doi:10.1007/s00704-005-0147-y. example, UMNs could inform mitigation and adaptation Goward, S. N., 1981: Thermal behavior of urban landscapes and the efforts within cities by singling out particular locations for urban heat island. Phys. Geography, 2, 19–33. Grimmond, C. S. B., 2006: Progress in measuring and observing the investment such as hot spots and hot corridors. UMNs can urban atmosphere. Theor. Appl. Climatol., 84, 3–22, doi:10.1007/ also provide critical data for evaluating the efficacy of s00704-005-0140-5. mitigation strategies. Last, UMNs that leverage volun- ——, 2007: Urbanization and global environmental change: Local teers are capable of strengthening links among citizens, effects of urban warming. Geogr. J., 173, 83–88, doi:10.1111/ businesses, municipal governments, and researchers, j.1475-4959.2007.232_3.x. Hage, K. D., 1972: Nocturnal temperatures in Edmonton, Alberta. broadening the exposure and impact of urban climato- J. Appl. Meteor., 11, 123–129, doi:10.1175/1520-0450(1972)011,0123: logical research. NTIEA.2.0.CO;2. Hausfather, Z., M. J. Menne, C. N. Williams, T. Masters, Acknowledgments. This research was funded by the R. Broberg, and D. Jones, 2013: Quantifying the effect of University of Minnesota Institute on the Environment urbanization on U.S. Historical Climatology Network under Discovery Grant DG-0015-11 and the U.S. Na- temperature records. J. Geophys. Res. Atmos., 118, 481– tional Science Foundation under Grant 1231325. We also 494, doi:10.1029/2012JD018509. Hawkins,T.W.,A.J.Brazel,W.L.Stefanov,W.Bigler,and acknowledge the support and cooperation of the many E. M. Saffell, 2004: The role of rural variability in urban citizens, institutions, and municipalities who have pro- heat island determination for Phoenix, Arizona. J. Appl. vided ongoing access to their property for the tempera- Meteor., 43, 476–486, doi:10.1175/1520-0450(2004)043,0476: ture observations that are associated with this project. TRORVI.2.0.CO;2. Hertel, W. F., 2014: A multi-city analysis of the natural and hu- REFERENCES man drivers of the urban heat island. M.S. thesis, De- partment of Soil, Water, and Climate, University of Ackerman, B., 1985: Temporal march of the Chicago heat Minnesota, 90 pp. [Available online at https://conservancy. island. J. Climate Appl. Meteor., 24, 547–554, doi:10.1175/ umn.edu/bitstream/handle/11299/166776/Hertel_umn_0130M_ 1520-0450(1985)024,0547:TMOTCH.2.0.CO;2. 15243.pdf?sequence51&isAllowed5y.] Arnfield, A. J., 2003: Two decades of urban climate research: A Ho, C. L. I., and C. Valeo, 2005: Observations of urban snow review of turbulence, exchanges of energy and water, and the properties in Calgary, Canada. Hydrol. Processes, 19, 459–473, urban heat island. Int. J. Climatol., 23, 1–26, doi:10.1002/ doi:10.1002/hyp.5544. joc.859. Howard, L., 1833: The Climate of London. Dalton, 291 pp. Baker, D. G., E. L. Kuehnast, and J. A. Zandlo, 1985: Climate of Ishida, T., and S. Kawashima, 1993: Use of cokriging for air tem- Minnesota. Part XV: Normal temperatures (1951–1980) and perature. Theor. Appl. Climatol., 47, 147–157, doi:10.1007/ their application. University of Minnesota Agricultural Ex- BF00867447. periment Station Tech. Bull. AD-SB-2777-1985, 70 pp. Jin, M. S., 2012: Developing an index to measure urban heat [Available online at https://conservancy.umn.edu/bitstream/ island effect using satellite land skin temperature and land handle/11299/122927/SB566.pdf?sequence51&isAllowed5y.] cover observations. J. Climate, 25, 6193–6201, doi:10.1175/ ——, R. H. Skaggs, and D. L. Ruschy, 1991: Snow depth required to JCLI-D-11-00509.1. masktheunderlyingsurface.J. Appl. Meteor., 30, 387–392, Jin, S., L. Yang, P. Danielson, C. Homer, J. Fry, and G. Xian, doi:10.1175/1520-0450(1991)030,0387:SDRTMT.2.0.CO;2. 2013: A comprehensive change detection method for up- Baker, L. A., A. J. Braze, N. Selover, C. Martin, N. McIntyre, F. R. dating the National Land Cover Database to circa 2011. Steiner, A. Nelson, and L. Musacchio, 2002: Urbanization and Remote Sens. Environ., 132, 159–175, doi:10.1016/ warming of Phoenix (Arizona, USA): Impacts, feedbacks and j.rse.2013.01.012. mitigation. Urban Ecosyst., 6, 183–203, doi:10.1023/ Journel, A. G., and C. J. Huijbregts, 1978: Mining Geostatistics. A:1026101528700. Academic Press, 600 pp. Basara, J. B., P. K. Hall Jr., A. J. Schroeder, B. G. Illston, and K. L. Kawashima, S., and T. Ishida, 1992: Effects of regional tempera- Nemunaitis, 2008: Diurnal cycle of the Oklahoma City urban ture, wind speed and soil wetness on spatial structure of sur- heat island. J. Geophys. Res., 113, D20109, doi:10.1029/ face air temperature. Theor. Appl. Climatol., 46, 153–161, 2008JD010311. doi:10.1007/BF00866095. Brazel, A. J., N. Selover, R. Vose, and G. Heisler, 2000: The tale of Kim, Y.-H., and J.-J. Baik, 2005: Spatial and temporal structure of two climates—Baltimore and Phoenix urban LTER sites. the urban heat island in Seoul. J. Appl. Meteor., 44, 591–605, Climate Res., 15, 123–135, doi:10.3354/cr015123. doi:10.1175/JAM2226.1. Chandler, T. J., 1965: The Climate of London. Hutchinson, 292 pp. Kljun, N., M. W. Rotach, and H. P. Schmid, 2002: A three- Duchon, C. E., 1979: Lanczos filtering in one and two di- dimensional backward Lagrangian footprint model for a mensions. J. Appl. Meteor., 18, 1016–1022, doi:10.1175/ wide range of boundary-layer stratifications. Bound.-Layer 1520-0450(1979)018,1016:LFIOAT.2.0.CO;2. Meteor., 103, 205–226, doi:10.1023/A:1014556300021.

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC SEPTEMBER 2015 S M O L I A K E T A L . 1917

Kłysik, K., and K. Fortuniak, 1999: Temporal and spatial charac- Rao, P. K., 1972: Remote sensing of ‘‘urban heat islands’’ teristics of the urban heat island of Łódz, Poland. Atmos. from an environment satellite. Bull. Amer. Meteor. Soc., 53, Environ., 33, 3885–3895, doi:10.1016/S1352-2310(99)00131-4. 647–648. Kottek, M., J. Grieser, C. Beck, B. Rudolf, and F. Rubel, 2006: Runnalls, K. E., and T. R. Oke, 2000: Dynamics and controls World map of the Köppen-Geiger climate classifica- of the near-surface heat island of Vancouver, British tion updated. Meteor. Z., 15, 259–263, doi:10.1127/ Columbia. Phys. Geog., 21, 283–304, doi:10.1080/ 0941-2948/2006/0130. 02723646.2000.10642711. Kovats, R. S., and S. Hajat, 2008: Heat stress and public health: Sailor, D. J., and L. Lu, 2004: A top-down methodology for de- A critical review. Annu. Rev. Public Health, 29, 41–55, veloping diurnal and seasonal anthropogenic heating profiles doi:10.1146/annurev.publhealth.29.020907.090843. for urban areas. Atmos. Environ., 38, 2737–2748, doi:10.1016/ Lowry, W., 1977: Empirical estimation of urban effects on climate: j.atmosenv.2004.01.034. A problem analysis. J. Appl. Meteor., 16, 129–135, doi:10.1175/ Santamouris, M., N. Papanikolaou, I. Livada, I. Koronakis, 1520-0450(1977)016,0129:EEOUEO.2.0.CO;2. C. Georgakis, A. Argiriou, and D. N. Assimakopoulos, 2001: Malevich,S.B.,andK.Klink,2011: Relationships between On the impact of urban climate on the energy consumption snow and the wintertime Minneapolis urban heat island. of buildings. Sol. Energy, 70, 201–216, doi:10.1016/ J. Appl. Meteor. Climatol., 50, 1884–1894, doi:10.1175/ S0038-092X(00)00095-5. JAMC-D-11-05.1. Schatz, J., and C. J. Kucharik, 2014: Seasonality of the urban heat Morris, C. J. G., and I. Simmonds, 2000: Associations between island effect in Madison, Wisconsin. J. Appl. Meteor. Clima- varying magnitudes of the urban heat island and the tol., 53, 2371–2386, doi:10.1175/JAMC-D-14-0107.1. synoptic climatology in Melbourne, Australia. Int. J. Cli- Stewart, I. D., 2011: A systematic review and scientific critique of matol., 20, 1931–1954, doi:10.1002/1097-0088(200012)20: methodology in modern urban heat island literature. Int. 15,1931::AID-JOC578.3.0.CO;2-D. J. Climatol., 31, 200–217, doi:10.1002/joc.2141. ——, ——, and N. Plummer, 2001: Quantification of the influ- ——, and T. R. Oke, 2012: Local climate zones for urban tem- ences of wind and cloud on the nocturnal urban heat island perature studies. Bull. Amer. Meteor. Soc., 93, 1879–1900, doi:10.1175/BAMS-D-11-00019.1. of a large city. J. Appl. Meteor., 40, 169–182, doi:10.1175/ Stone, B., 2005: Urban heat and air pollution: An emerging role for 1520-0450(2001)040,0169:QOTIOW.2.0.CO;2. planners in the climate change debate. J. Amer. Plann. Assoc., Muller, C. L., L. Chapman, C. S. B. Grimmond, D. T. Young, and 71, 13–25, doi:10.1080/01944360508976402. X. Cai, 2013a: Sensors and the city: A review of urban mete- Todhunter, P. E., 1996: Environmental indices for the Twin Cities orological networks. Int. J. Climatol., 33, 1585–1600, Metropolitan Area (Minnesota, USA) urban heat island— doi:10.1002/joc.3678. 1989. Climate Res., 6, 59–69, doi:10.3354/cr006059. ——, ——, ——, ——, and X.-M. Cai, 2013b: Toward a standard- Wallace, J. M., 1996: Observed climatic variability: Spatial struc- ized metadata protocol for urban meteorological networks. ture. Decadal Climate Variability: Dynamics and Pre- Bull. Amer. Meteor. Soc., 94, 1161–1185, doi:10.1175/ dictability, D. L. T. Anderson and J. Willebrand, Eds., BAMS-D-12-00096.1. Springer, 31–81. Nunez, M., and T. R. Oke, 1977: The energy balance of an Warren, S. G., 1982: Optical properties of snow. Rev. Geophys., 20, urban canyon. J. Appl. Meteor., 16, 11–19, doi:10.1175/ 67–89, doi:10.1029/RG020i001p00067. , . 1520-0450(1977)016 0011:TEBOAU 2.0.CO;2. Winkler, J. A., R. H. Skaggs, and D. G. Baker, 1981: Effect of Oke, T. R., 1973: City size and the urban heat island. Atmos. En- temperature adjustments on the Minneapolis-St. Paul urban viron., 7, 769–779, doi:10.1016/0004-6981(73)90140-6. heat island. J. Appl. Meteor., 20, 1295–1300, doi:10.1175/ ——, 1976: The distinction between canopy and boundary-layer 1520-0450(1981)020,1295:EOTAOT.2.0.CO;2. urban heat islands. Atmosphere, 14, 268–277. World Meteorological Organization, 2008: Guide to meteorologi- ——, 1982: The energetic basis of the urban heat island. Quart. cal instruments and methods of observation. 7th ed. In- J. Roy. Meteor. Soc., 108, 1–24. struments and Observing Methods Rep. 8, 681 pp. [Available ——, 1995: The heat island of the urban boundary layer: Charac- online at https://www.wmo.int/pages/prog/gcos/documents/ teristics, causes, and effects. Wind Climate in Cities,J.E. gruanmanuals/CIMO/CIMO_Guide-7th_Edition-2008.pdf.] Cermak et al., Eds., Kluwer Academic, 81–107. Wu, H., K. G. Hubbard, and J. You, 2005: Some concerns when ——, 2006: Initial guidance to obtain representative meteoro- using data from the cooperative weather station networks: A logical observations at urban sites. World Meteorological Nebraska case study. J. Atmos. Oceanic Technol., 22, 592–602, Organization IOM Rep. 81, WMO/TD 1250, 51 pp. [Available doi:10.1175/JTECH1733.1. online at https://www.wmo.int/pages/prog/www/IMOP/publications/ Yow, D. M., 2007: Urban heat islands: Observations, impacts, and IOM-81/IOM-81-UrbanMetObs.pdf.] adaptation. Geogr. Compass, 1, 1227–1251, doi:10.1111/ Overeem, A., J. C. R. Robinson, H. Leijnse, G. J. Steeneveld, j.1749-8198.2007.00063.x. B. K. P. Horn, and R. Uijlenhoet, 2013: Crowdsourcing urban Yuan, F., and M. E. Bauer, 2007: Comparison of impervious air temperatures from smartphone battery temperatures. surface area and normalized difference vegetation index as Geophys. Res. Lett., 40, 4081–4085, doi:10.1002/grl.50786. indicators of surface urban heat island effects in Landsat Peng, S., and Coauthors, 2012: Surface urban heat island across 419 imagery. Remote Sens. Environ., 106, 375–386, doi:10.1016/ global big cities. Environ. Sci. Technol., 46, 696–703, j.rse.2006.09.003. doi:10.1021/es2030438. Zhang, K., E. M. Oswald, D. G. Brown, S. J. Brines, C. J. Gronlund, Pigeon, G., A. Lemonsu, N. Long, J. Barrié, V. Masson, and J. L. White-Newsome, R. B. Rood, and M. S. O’Neill, 2011: P. Durand, 2006: Urban thermodynamic island in a coastal city Geostatistical exploration of spatial variation of summertime analysed from an optimized surface network. Bound.-Layer temperatures in the Detroit metropolitan region. Environ. Meteor., 120, 315–351, doi:10.1007/s10546-006-9050-z. Res., 111, 1046–1053, doi:10.1016/j.envres.2011.08.012.

Unauthenticated | Downloaded 09/24/21 10:03 PM UTC