Wintertime radiation and energy budget along an urbanization gradient in Montreal, Canada


  • Onil Bergeron,

    1. Department of Natural Resource Sciences, McGill University, Ste. Anne de Bellevue, QC, Canada
    Search for more papers by this author
  • Ian B. Strachan

    Corresponding author
    1. Department of Natural Resource Sciences, McGill University, Ste. Anne de Bellevue, QC, Canada
    • Department of Natural Resource Sciences, McGill University, Macdonald Campus, 21,111 Lakeshore Road, Ste. Anne de Bellevue, QC H9X 3V9, Canada.
    Search for more papers by this author


This study reports on the radiation and energy balance of three sites (rural, suburban, and urban) located along an urbanization gradient in the Montreal, QC, region for two winters (December–March) with contrasting snow regimes. The urban and suburban sites had similar albedo which was about half that at the rural site during the snow-cover period. Temporal variability in albedo was attributable to the presence of snow on rooftops at the urban site and to a site-specific response to cloudiness at the suburban site. As compared to the suburban site, the urban site showed higher albedo inducing lower net radiation (Q*) which was compensated for by greater anthropogenic heat flux (QF), so that the urban site had highest total available energy (Q* + QF). Hourly QF estimates were a significant term in the winter energy budget analysis. QF was dominated by building heating at both urbanized sites, while vehicular traffic contributed to rush hour peaks. Daytime total available energy was mostly dissipated as sensible heat flux (QH) at the beginning of the winter season and mostly stored (ΔQS) towards the end of the winter at both urbanized sites. Daytime energy partitioning into QH and ΔQS was correlated with air temperature with no significant differences between urbanized sites. On a daily time scale, available energy was mostly stored before noon and dissipated as QH in the afternoon at both urbanized sites. Urbanized sites showed differences in diurnal variability of QH and ΔQS occurring in the afternoon and evening. Latent heat flux (QE) was low throughout winter and accounted for 10% of the total available energy during daytime at the urbanized sites. Water vapour emissions showed intra-urban differences in their response to wintertime climatic conditions. Copyright © 2010 Royal Meteorological Society

1. Introduction

The world's population is increasingly urbanized and currently about 80% of the Canadian population lives in cities. Characteristics of urban fabrics, structure, and orientation can influence the properties of the near-surface atmosphere over cities and local climates. Understanding the interaction between the energy balance and the urban surface characteristics is therefore crucial. Simultaneously, computational power has been growing, allowing meso-scale models to run at higher resolution, thus enabling the incorporation of detailed surface characteristics, including urban features, in weather prediction schemes. However, weather forecasting abilities over cities are inadequate because of current limitations of surface process models and to a restricted number of observational studies available in the literature.

The Environmental Prediction in Canadian Cities (EPiCC) Research Network ( was created to serve a threefold purpose: (1) report observations of energy balance components using (eddy covariance) flux towers for two Canadian cities with contrasting climates; (2) quantify each site's surface characteristics using remote sensing; and (3) test and couple the urban surface parameterization scheme TEB (Masson, 2000) to the SVAT model ISBA (Noilhan and Planton, 1989) to be ultimately implemented into the Canadian weather modelling and prediction system. The cities of Montreal and Vancouver were selected to allow an emphasis on winter snow and summer irrigation, respectively. This study focuses on observations made in Montreal during the winters of 2007–2008 and 2008–2009 which experienced contrasting snow regimes.

There are a number of studies reporting on wintertime urban energy budget with varying degrees of detail (Grimmond, 1992; Ruffieux, 1995; Spronken-Smith, 2002; Christen and Vogt, 2004; Offerle et al., 2005, 2006; Spronken-Smith et al., 2006; Coutts et al., 2007; Lemonsu et al., 2008; Pigeon et al., 2008; Vesala et al., 2008; Leroyer et al., 2010). However, studies on the role of snow on urban energy budget at the neighbourhood level are rare. Lemonsu et al. (2008) and Leroyer et al. (2010) showed that snow can influence energy flux partitioning during discrete snowmelt periods in Montreal; this study represents an opportunity to extend their findings to the entire wintertime period for two years under different snow regimes.

In urban flux tower studies, anthropogenic heat flux estimates are usually overlooked or are deduced from other energy balance terms. Hence, the reported energy budgets are somewhat incomplete and may lump together the variability of different components into one residual term. Sailor and Lu (2004) proposed a simple methodology to estimate the within-day anthropogenic heat flux using information from energy consumption data, vehicular traffic, and population density at spatial scales ranging from neighbourhoods to cities or regions, a technique which is particularly suitable for flux tower studies. Their methodology has been adapted given the available data to provide hourly estimates of anthropogenic heat flux for urban and suburban areas in Montreal.

Our study focuses on characterizing the effect of urbanization on wintertime radiation and energy balance in the Montreal region. The specific objectives of this study were to (1) provide hourly estimates of anthropogenic heat loss at the spatial scale of the study sites (104–106 m2) for inclusion in the energy balance; (2) describe the seasonal and diurnal variability of the radiation and energy balance components for two winters with contrasting snow regimes; and (3) determine key environmental factors or site characteristics that can help explain this variability.

2. Methods

2.1. Site description

Three flux towers were deployed in the region of Montreal, Canada, along an urbanization gradient that followed an east-to-west transect (Figure 1). Site characteristics are listed in Table I.

Figure 1.

Aerial photographs of the suburban (SUB) and urban (URB) sites. North is towards the top of the figure. Images from and Google Earth©

Table I. Site characteristics
  • a

    Urbanization is assumed to be nil at RUR.

  • b

    Oke (2006).

  • c

    Includes only trees > zH.

  • d

    133 618 inh./15.9 km2 (Ville de Montréal, 2009a).

  • e

    65 041 inh./27.1 km2 [65 041 inh./20.6 km2 when three parks (201, 288, and 159 ha) are removed from borough surface area] (Ville de Montréal, 2009a, 2009b).

  • f

    Surface cover fractions and mean building heights estimated from Quickbird image, building, and GIS information for a 1000 m radius centred on the towers at SUB and URB (Jinfei Wang, personal communication). At SUB, 1% of the surface is classified as other (open water, swimming pools, or bareground).

Latitude, Longitude (°)45.547 N, 73.592 W45.501 N,73.811 W45.328 N,74.165 W
Land useResidentialResidentialAgricultural
Urban climate zoneb357
Population density (inh. km−2)8400d2400 (3150)e
zH, mean building height (m)f7.96.4
zTT, mean tall tree height (m)c13.013.8
zM, measurement height (m)25252
W, mean street canyon width (m)25.232.2
zH/W, street aspect ratio0.320.22
Roof materialsGravelShingle
Roof orientationFlat2-facets 10–15° from horizontal 
Exterior building wall materialsBrickBrick or shingle
Surface cover fractions (%)f   
λI, impervious (pavement)44370
λP, built (roofs)27120
λV, vegetation (grass and trees)2950100

The urban site (URB) was located in the Rosemont-La-Petite-Patrie borough which is a densely populated (8400 inh. km−2) residential area with two- or three-storey row housing which translated into well defined street canyons and alley canyons (Figure 1). Urban fabrics mostly include paved streets and alleys, gravel on flat rooftops, and brick walls. Backyards and frontyards are small with sparse vegetation (mostly grass, herbs, and shrubs) and are bordered with impervious materials (paved alley, sidewalk, parking lot, or building). Sheds or garages are present in most backyards. Single deciduous trees taller than the surrounding buildings are a regular feature in SE–NW-oriented street canyons. Opposing SW–NE-oriented streets include commercial and institutional activity. The Métropolitaine Highway is about 2 km NW of the flux tower site, while a 2.9 ha park and a 63 ha park are located 150 m N and 2.3 km NE, respectively.

The suburban site (SUB) was located about 18 km WSW of URB in the Pierrefonds-Roxboro borough, a single-family residential area (3150 inh. km−2) with detached houses and well-defined street canyons (Figure 1). Swimming pools are common in backyards while sheds are few. Rooftops are mostly sloped and covered with dark shingles. Exterior walls are bricks or vinyl siding. Grass, shrubs, and tall coniferous and deciduous trees are present in both backyards and frontyards. Commercial activity is present on one N-S major street located W of the flux tower. A 20 ha cemetery is found about 500 m SE of the flux tower site and the Des Prairies river flows 1.5 km N.

The rural site (RUR) was located in Coteau-du-Lac, Quebec, an agricultural area off the Island of Montreal about 35 and 50 km SW of SUB and URB, respectively, in the upwind direction (Figure 1). Sparse housing and farm buildings are present but were outside the flux source areas. The flux tower was deployed in a field which lay fallow during the winter periods.

2.2. Instrumentation

All flux towers were equipped with similar instrumentation that was cross-checked against each other before deployment. Each flux tower had an eddy covariance system mounted at zM (Table I) that comprised a sonic anemometer–thermometer [model CSAT3, Campbell Scientific Canada Corp. (CSCC), Edmonton, AB, Canada] to measure three-dimensional wind components and air temperature as well as an infra-red gas analyzer (model LI-7500, Li-COR Biosciences, Lincoln, NE, USA) to measure H2O concentration and atmospheric pressure. The eddy covariance system sampled at 20 Hz using a datalogger (models CR5000 at URB and SUB and CR1000 at RUR, CSCC) and raw data were stored for post-processing.

Supporting meteorological variables were sampled at 0.5 Hz and averaged on a 5-min basis using a datalogger (models CR23X at all sites and CR1000 for radiation data at RUR, CSCC). A non-heated, four-component radiometer (model CNR1, Kipp & Zonen, Delft, The Netherlands) was used to measure incoming shortwave (Kin) and longwave (Lin) radiation, outgoing shortwave (Kout) and longwave (Lout) radiation, albedo (α = Kout/Kin), and net radiation (Q*) at zM. At URB, radiation measurements were made at the top of the flux tower to represent the whole neighbourhood radiation balance (URBtower) and a second instrument was located 1.5 m above and in the centre of the rooftop (15 m long, > 40 m wide considering contiguous rooftops) of the private property where the tower was installed to represent radiation balance over flat rooftops (URBroof). URBroof was located about 14 m E of URBtower and any impact of shadowing was minor owing to the open nature of the tall tower and would have been restricted to late afternoon periods. Air temperature (Tair) was monitored at zM with a PRT detector (models HMP45C-212 at URB and HMP45C at SUB and RUR, CSCC). A snow ranging sensor (model SR50, CSCC) measured snow depth (zsnow) over bare soil at RUR, in the backyard where the tower was installed at SUB, and on the rooftop of the private property where the tower was installed at URB. Supporting meteorological variables were averaged half-hourly before analysis.

At URB and SUB, an aluminium telescopic triangular lattice tower was used. The tower base was attached to the front of a 2.5 × 5 m closed trailer which was sitting in the backyard of a private property. The data logging equipment enclosed in the trailer and the base of the tower was accessible at the ground level. Eddy covariance instrumentation was mounted about 1 m upwind of the tower, relative to prevailing winds, on a boom which was installed about 1 m above the tower top section on a mast. At RUR, the eddy covariance instrumentation was mounted on a boom more than 1 m upwind, relative to prevailing winds, of the supporting tripod mast. The tripod itself was mounted beside and upwind of an electric pole for logistical reasons. Hence, for all three sites flow distortion by the tower structure was minimized.

In accordance with our safety protocol, the flux towers at URB and SUB were retracted whenever high winds (>60 km/h), wind gusts (>70 km/h), or freezing precipitation were forecasted for the next 12–24 h. Hence, our measurements exclude those more extreme events likely associated with frontal passages.

2.3. Calculation of turbulent fluxes and quality control

Latent (QE) and sensible (QH) heat fluxes were calculated as the covariance of scalars (H2O concentration and air temperature, respectively) and vertical wind velocity (Desjardins et al., 1993; Baldocchi, 2003). Fluxes were calculated from block averages over 30-min time periods. Two rotations were applied to set mean lateral and vertical wind speed to zero (Tanner and Thurtell, 1969). A density correction was applied to QE as per Webb et al. (1980). No correction was applied for spectral loss or sensor separation as these corrections typically modify fluxes by only a few percent (Christen and Vogt, 2004). No provisions were made to account for the lack of energy balance closure as it cannot be reliably determined in urban environments.

Quality control procedures applied to meteorological variables as well as radiative and turbulent fluxes involved the exclusion of data points for time periods when absolute values of mean, minimum, maximum, or standard deviation were outside variable-specific realistic ranges. Data points from known periods of instrument malfunction, calibration, or servicing were also rejected. QE and QH were further quality controlled by discarding data points corresponding to half-hours when (1) less than two-third of raw data were usable to calculate fluxes; (2) the number of points identified as spikes was greater than 1% of the record length; (3) wind was blowing from behind the sonic anemometer–thermometer and through the tower structure; (4) the difference between block average and linear detrended fluxes was greater than the block average flux; or (5) dew, frost, rain, or snow obstructed the optical path of the gas analyzer as identified by monitoring the automatic gain control (AGC) value. Spikes in raw data records were identified as three or fewer consecutive points that were at least 3.5 standard deviations away from a running mean (2-min window) with an absolute value above a variable-specific threshold (Vickers and Mahrt, 1997). No filtering based on friction velocity (u*) was applied. Meteorological and flux variables were visually inspected and remaining data points showing obvious unrealistic behaviour were excluded.

2.4. Data analysis

Measurements were performed continuously from fall 2007 to fall 2009. This study focuses on the two wintertime periods corresponding to measurements made from 1 December 2007 to 31 March 2008 and from 1 December 2008 to 31 March 2009. During the study period, the towers were in the retracted position 28% of the time at both SUB and URB. For all three sites, only data when the towers at both URB and SUB were in the extended position were used. Hence for the study period, Q*, QH, and QE were available 67, 38, and 38% of the time at RUR, 72, 52, and 52% at SUB, and 66, 39, and 38% at URB, respectively, after quality control and exclusion of retracted-tower periods.

The energy budget (terms in W m−2) of the sites can be written as follows:

equation image(1)

where QF is the anthropogenic heat flux, ΔQS the net change of heat storage, ΔQA the net-advected flux, and S any other sources or sinks. ΔQA and S are assumed to be negligible because of the careful choice of sites and measurement heights and are thus neglected. When half-hourly Q*, QE, and QH were all available, ΔQS was calculated as a residual term:

equation image(2)

Eddy covariance measurements typically suffer from the lack of energy balance closure that is about 20% over natural surfaces (Wilson et al., 2002), therefore ΔQS is likely overestimated and represents an upper limit. It is worth noting that this residual term differs slightly from previously reported residual terms as we estimated QF which is usually not available in similar published studies.

QF was defined as

equation image(3)

where QM is the heat loss from human metabolism, QV from vehicular traffic, and QB from the building sector. The methodology to estimate QF at SUB and URB is detailed in Appendix A1. QF was estimated from the period 2007–2009 although only QB varied for one year to the next, QM and QV were assumed constant between years. QF was also assumed to be negligible at RUR and was set at zero.

The stability parameter (ζ was calculated as

equation image(4)

where zd is the zero-displacement height taken as 0.7 zH at URB and SUB. At RUR, zd was set equal to zsnow because flux measurements were made close to the surface and no canopy was present in wintertime. Obukhov length (L) was computed as per Monteith and Unsworht (1990).

A clearness index was calculated as follows:

equation image(5)

where G0 is the solar irradiance at the top of the atmosphere calculated as per Duffie and Beckman (2006) and was used to quantify cloud cover and thus characterize light regime (diffuse vs direct light).

3. Results and discussion

3.1. Environmental conditions during the study period

The study period includes two winters with above-normal temperatures and very different precipitation regimes (Table II). Winter 2007–2008 was generally colder than 2008–2009. Snowfall was above normal throughout winter 2007–2008 and the snow-cover period lasted from December to March (3 December 2007 to 8 April 2008 at RUR; Table III). In contrast in 2008–2009, snowfall was above normal only for the first half of the winter and liquid precipitation dominated during the second half, thus the snow-cover period was mostly restricted to January and February (10 December 2008 to 19 March 2009 at RUR; Table III).

Table II. Monthly mean air temperature, total snowfall, and total precipitation measured at the Montreal Trudeau airport (Environment Canada, 2010)
Mean air temperature ( °C)− 3.1− 2.0− 3.40.3
Total snowfall (cm)112.856.688.077.8
Total precipitation (mm)119.898.8107.2121.6
Mean air temperature ( °C)− 2.1− 8.9− 2.33.7
Total snowfall (cm)97.071.734.82.6
Total precipitation (mm)151.467.685.047.6
1971–2000 Normals
Mean air temperature ( °C)− 6.3− 10.2− 8.4− 2.3
Total snowfall (cm)48.352.543.336.0
Total precipitation (mm)81.378.361.573.6
Table III. Monthly daytime (night-time) means of environmental variables
YearMonthSiteTair ( °C)Kin (W m−2)Kout (W m−2)Lin (W m−2)Lout (W m−2)Q* (W m−2)αzsnow (m)
  1. Daytime (night-time) corresponds to periods when Kin ≥ (<)5 W m−2. Daytime α includes data between 10h00 and 14h00 LST. URB α corresponds to URBtowerα. zsnow was measured over bare soil at RUR, in a backyard at SUB, and on a rooftop at URB.

2007DecemberRUR− 8.4 (−9.9)151122250 (248)275 (266)3 (−19)0.820.26
  SUB− 7.9 (−8.8)15159248 (245)280 (272)59 (−29)0.420.31
  URB− 7.5 (−8.5)14264246 (244)284 (277)39 (−35)0.450.30
2008JanuaryRUR− 7.1 (−9.4)204146238 (235)280 (266)16 (−32)0.710.13
  SUB− 6.8 (−8.3)20966230 (222)282 (269)91 (−49)0.340.27
  URB− 6.5 (−7.7)19169228 (223)289 (276)61 (−57)0.360.07
 FebruaryRUR− 5.9 (−7.1)218173258 (265)286 (278)17 (−14)0.810.25
  SUB− 6.2 (−7.0)26699238 (246)286 (278)118 (−33)0.400.54
  URB− 6.0 (−6.7)240101239 (246)292 (283)85 (−39)0.440.17
 MarchRUR− 3.2 (−5.2)367276250 (254)296 (281)45 (−29)0.760.41
  SUB− 1.5 (−3.1)332114256 (254)308 (292)166 (−40)0.370.83
  URB− 2.1 (−3.1)303107255 (254)311 (295)139 (−44)0.390.19
 DecemberRUR− 8.6 (−9.0)13889250 (251)278 (273)22 (−23)0.640.17
  SUB− 8.2 (−8.1)14542250 (251)281 (277)73 (−28)0.320.20
  URB− 7.7 (−7.5)13041243 (245)286 (281)45 (−38)0.360.09
2009JanuaryRUR− 12.4 (−13.7)209175223 (227)258 (249)0 (−23)0.840.41
  SUB− 12.3 (−12.7)21177218 (221)260 (254)91 (−35)0.400.47
  URB− 11.9 (−11.8)19585217 (221)268 (262)60 (−43)0.440.17
 FebruaryRUR− 6.8 (−8.7)265197239 (240)283 (270)23 (−31)0.750.36
  SUB− 6.2 (−6.9)25886236 (239)287 (277)122 (−40)0.360.50
  URB− 5.5 (−6.2)22969239 (235)298 (283)101 (−51)0.300.10
 MarchRUR− 0.2 (−2.5)35596249 (246)316 (297)193 (−53)0.280.08
  SUB0.3 (−1.7)36166250 (246)323 (299)222 (−55)0.200.39
  URB2.2 (0.7)32042261 (259)346 (313)192 (−57)0.130.01

3.2. Albedo

During the snow-cover period, α was twice as high at RUR (0.70–0.85) than SUB and URBtower (0.30–0.45), while URBroofα was similar to RUR (Table III, Figure 2(a) and (b)). In mid-February 2009 and also in early January 2008, URBtower showed days with markedly decreased α (Figure 2(a) and (b)). This decrease was related to the absence of snow on rooftops (Figure 2(c) and (d)) and low URBroofα (Figure 2(a) and (b)). Simple calculations based on the decrease in URB albedo at the tower and rooftop levels between snowy-roof and snow-free roof conditions suggest that rooftops accounted for 40% of the flux tower radiative footprint.

Figure 2.

Daily average (n > 3) of (a, b) daytime (10h00 to 14h00 LST) albedo (α) and (c, d) whole day snow depth (zsnow). α was measured on a roof top (URBroof) and at the top of the flux tower at URB (URBtower). zsnow was measured over bare soil at RUR, in a backyard at SUB, and on a rooftop at URB

The effect of rooftop snow on the radiation balance was studied by comparing two 4-day periods with similar air temperature (mean diurnal variation between − 5 and − 10 °C) and wind regimes (SW–NW winds). The first period (snowy rooftops at URB) was 4–8 December 2007 with mean zsnow = 0.35 m on URB rooftop. The second period (snow-free rooftops at URB) was 13–17 February 2009 with mean zsnow = 0.03 m on URB rooftop. Note that rooftops were mostly snow covered at SUB during both periods.

Under snowy-roof conditions, mid-day α was lowest at SUB at 0.41 and slightly greater for URBtower at 0.48, while α was 0.83 for URBroof, similar to RUR at 0.82 (Figure 3(a)). Under snow-free roof conditions, α was lowest for URBroof, with a mid-day value of 0.12 (Figure 3(b)). Consequently, the composite URBtower albedo was slightly lower than SUB, with mid-day values of 0.17 and 0.24, respectively (Figure 3(b)). Intra-urban variability in albedo during winter time can thus be significantly impacted by the presence of snow on rooftops, which in turn can differ with rooftop characteristics within urban areas as they did in this study (Table I). For example, most (>85%) buildings at URB were constructed before 1980, whereas about half of all the buildings were constructed after 1980 at SUB, suggesting that the thermal efficiency of rooftops is likely much lower at URB which could contribute to intra-urban snow cover and albedo variability.

Figure 3.

Diurnal ensemble average (1 < n < 5) of (a, b) daytime (Kin > 5 W m−2) albedo (α), (c, d) outgoing longwave radiation (Lout) and (e, f) net radiation (Q*) for two 4-day wintertime periods when snow was and was not present on URB rooftops. See text for details

SUB and URBtower showed similar α values when rooftops were covered with snow but SUB was slightly lower than URBtower overall (Table III) and was clearly lower on some days (Figure 2(a) and (b)). The difference between SUB and URBtower albedo observed here was inconsistent with surface cover fractions, structure, and geometry of the sites (Table I). For example, SUB had greater vegetation cover fraction (mostly grass) that was basically snow covered, which would have promoted higher albedo. Also, URB had well defined street canyons and alley canyons (row buildings, greater aspect ratio zH/W, Table I) that were expected to induce lower albedo than SUB (Kondo et al., 2001; Fortuniak, 2008). However, the SUB:URBtowerα ratio was positively correlated with the clearness index kT (Figure 4(a)). This result indicates that both sites had very similar α values under overcast conditions (diffuse light dominated), while URBtower showed α up to 28% greater than SUB under clear sky conditions (direct light dominated). The SUB:URBtowerα ratio was also negatively correlated with SUB α (Figure 4(b)), while no significant relationship was found with URBtowerα, thus indicating that the between-site difference in albedo was because of the variability observed at SUB. At SUB, the buildings are detached and surrounded by relatively large snow-covered areas in winter. Under direct light, buildings shadow those snow-covered areas, inducing a decrease in the albedo of those surfaces which translates into a significant albedo decrease at the neighbourhood level. The same effect could have taken place around numerous tall coniferous trees and on the different facets of sloped roofs. This shadowing effect likely induces a milder albedo decrease in street and alleys canyons at URB because the underlying surfaces are darker and include a greater proportion of walls, which do not accumulate snow, and ploughed streets or alleys. Also, flat rooftops, generally bordered by edges of limited height, are not prone to shadowing effects.

Figure 4.

Daily ratio of URB over SUB daytime (10h00 to 14h00 LST) albedo (α) against (a) the daytime clearness index kT and (b) SUB α for days when mean daily zsnow > 0.1 m and α> 0.3 at URB. See text for details

3.3. Net radiation

In accordance with the between-site albedo differences observed during the snow-cover period, daytime Q* was much higher at SUB and URB than RUR and was generally 20–30 W m−2 higher at SUB than URB over the course of the winter (Table III). The monthly mid-day (10h00 to 14h00 LST) difference in Q* between SUB and URB was larger than the compensating greater QF at URB (Table V). Night-time Q* was consistently 2–10 W m−2 higher at SUB than URB because of generally lower Lout at SUB (Table III). Q* showed the smallest diurnal amplitude at RUR and highest amplitude at SUB while it was almost similar, yet slightly lower, at URB (Figure 5(a) and (e)). Q* was higher at SUB than URB at all times except in the early morning when there was no difference between sites. The afternoon decline in Q* began about 1 h earlier at URB than SUB. These results show that the albedo differences observed between urban areas induced significant intra-urban variability in net radiation during wintertime over Montreal, both seasonally and diurnally.

Figure 5.

Diurnal ensemble average of energy balance components. Note the different scale used for Q*. Hourly bins (n > 30) are shown for clarity

The presence of snow on rooftops influencing the sites' albedo significantly impacted the radiation budget of the studied urban areas. Under snowy-roof conditions, URBroof showed no diurnal difference in Q* as compared with RUR (Figure 3(e)). SUB and URBtower showed markedly higher daytime Q* than RUR with SUB being up to 50 W m−2 higher than URBtower at noon (Figure 3(e)). Under snow-free roof conditions, SUB, URBtower, and URBroof showed daytime Q* up to 200 W m−2 higher than RUR. The absence of snow on rooftops at URB induced significant daytime Lout increase from rooftop surfaces which increased daytime Lout at the neighbourhood level (Figure 3(d)). No such effect was observed at SUB where the roofs remained snow covered during both periods. Despite higher daytime Lout at URBtower under snow-free roof conditions, mid-day Q* was similar between URBtower and SUB, suggesting that the decrease in α due to the absence of snow on rooftops at URB compensated for the lower α at SUB under snowy-roof conditions.

3.4. Anthropogenic heat flux

Total daily QF was about three times higher at URB than SUB (Table IV). Consequently, the total available energy (Q*+ QF) was highest at URB despite higher Q* being observed at SUB on a full daily cycle. Also, total daily QF was of the same magnitude as QH at URB and accounted for up to half of QH at SUB (Table IV). Daytime Q*/(Q*+ QF) ranged from 0.91 to 0.98 and from 0.71 to 0.93 at SUB and URB, respectively (Table V). These results show that QF is a significant term of the energy balance over Montreal residential areas in winter, especially in December when Kin and Q* are lowest, and can account for up to 9–29% of daytime total available energy.

Table IV. Mean wintertime daily totals (MJ m−2 day−1)
  1. For each site, only half-hours when all variables were available were used.

RUR− 1.140.00− 0.740.45− 0.85
RUR0.430.00− 0.150.97− 0.39
Table V. Daytime (10h00 to 14h00 LST) monthly means of Q* and QF and ratios of Q*, QH, QE, and ΔQS over (Q* + QF)
YearMonthSiteQ* (W m−2)QF (W m−2)Q*/ (Q* + QF)QH/ (Q* + QF)QE/ (Q* + QF)ΔQS/ (Q* + QF)
  1. For each site, only half-hours when all variables were available were used.

2007DecemberRUR1201.00− 0.831.000.83

Figure 6 shows QF and its components for SUB and URB as determined for the period 1 December 2007 to 31 March 2008. QF ranged from 7 to 13 W m−2 and from 25 to 45 W m−2 at SUB and URB, respectively. Peak hours were around 7h00–8h00 and 18h00–19h00 LST at both sites when QV reached its maximum. QB dominated QF, while QM was negligible at both sites. Our QF estimates for residential areas in Montreal are lower than those reported by Summers (1965 in Klysik, 1996), Oke (1997), and Taha (1997). However, these latter estimates included denser parts of the city (downtown business district) and areas enclosing industrial activities, which were excluded from our estimates. Also, QB was estimated using energy consumption data from warm winters (Table II), inducing lower estimations of QB and QF relative to normal or below-normal years. Nevertheless, the diurnal trends described here are consistent with other studies (Ichinose et al., 1999; Sailor and Lu, 2004; Offerle et al., 2005, Pigeon et al., 2007).

Figure 6.

Diurnal ensemble average of wintertime (1 December 2007 to 31 March 2008) anthropogenic heat flux components at the (a) suburban and (b) urban sites (see Appendix A1 for details)

QF between-year variability was only due to QB through (1) year-specific annual energy consumption totals and (2) temperature regimes which determined the distribution of space heating annual totals within each year. As a consequence, the QF between-year variability was negligible at SUB and rather small at URB (Table IV) and hourly estimates for the period 1 December 2008 to 31 March 2009 were within 1 W m−2 of those in Figure 6.

These estimates represent a first attempt to quantify the diurnal and seasonal variability of QF over Montreal. However, their uncertainty is unknown and sources of bias are expected. The greatest source of error is likely the population density estimates which were taken at the borough level to match the available information on population displacements. These estimates are expected to be lower than the population density at the neighbourhood level (e.g. 1 km radius around the flux tower) as this latter scale does not include mid- to large-sized parks. Hence, QF estimates may suffer from an underestimation. Also, because population density was fixed, any seasonality is not represented; this source of uncertainty is expected to have more of an impact on summer QV. In calculating QB, residential households and commercial/institutional floorspace comprising the study sites were assumed to be similar to the province-wide average. As the province-wide average is biased towards Montreal as this region represents about half the provincial population, this source of uncertainty is likely minor. Despite these minor limitations, the QF estimates reported here represent the most detailed information on heat loss from human activity available to date for Montreal.

3.5. Sensible heat flux

Total daily QH was higher at URB than SUB in both years (Table IV), which is in accordance with surface cover fractions (Table I). Total daily QH was higher in 2008–2009 than in 2007–2008 at all sites, which is consistent with total Q*. Overall, monthly means of daytime QH/(Q*+ QF) tended to decrease from about 0.5 to about 0.4 throughout the winter at both SUB and URB (Table V). Daytime QH/(Q*+ QF) was generally higher at URB than SUB, especially in March 2009 when snow cover disappeared earlier at URB (Table III).

At the daily time scale, daytime ratios of QH/(Q*+ QF) were negatively correlated with Tair which explained 20, 40, and 37% of QH/(Q*+ QF) variability at RUR, SUB, and URB, respectively (Figure 7(a)). At RUR, QH/(Q*+ QF) had the same sign as Tair and was around 0 when Tair = 0 °C. This was expected as the sign of QH in winter over a crop field covered with snow depends on the temperature gradient between the air and the snowpack, the latter being around 0 °C. QH/(Q*+ QF) was consistently lower at RUR than SUB and URB at any temperature. A coincidental regression test (Zar, 1984) indicated that regressions of QH/(Q*+ QF) with Tair were not significantly different between SUB and URB [F = 2.69, degrees of freedom = (2; 218)]. It is worth noting that the equilibrium temperature of QH/(Q*+ QF) [temperature at which QH/(Q*+ QF) would change sign] lies between 16 and 24 °C, which corresponds to typical indoor temperatures in Montreal. This result suggests that the indoor–outdoor temperature gradient might help explain the temporal variability of daily daytime QH/(Q*+ QF) during the winter time over temperate cities through the regulation of heat conduction from indoor to outdoor air. Assuming that the variability in outdoor air temperature is compensated for by building heating (QB) to keep indoor temperature, and thus the amount of energy stored in the indoor air volume, constant, an increase in the indoor–outdoor temperature gradient would increase heat conduction from indoor to outdoor and thus increase QH. However, a more detailed study on the interaction of these energy balance terms would be needed to shed light on this phenomenon.

Figure 7.

Ratios of (a) mean daytime (10h00 to 14h00 LST) QH and (b) ΔQS over (Q*+ QF) as a function of air temperature (Tair)

QH showed a marked diurnal trend that demonstrated the departure of SUB and URB from RUR (Figure 5(b) and (f)). Similar wintertime trends have been reported for other temperate cities highlighting the effect of urbanization on sensible heat loss (Grimmond, 1992; Offerle et al., 2006; Vesala et al., 2008). Although similar trends were observed, QH peaked about 1 h later at URB than SUB (Figure 5). Consequently, URB showed greater QH/(Q*+ QF) than SUB after noon (Figure 8(a) and (d)). Daytime QH/(Q*+ QF) increased throughout the day at both sites with a greater increase in the afternoon. In the early morning, no marked between-site difference in QH was apparent but URB showed lower QH/(Q*+ QF) than SUB which was mostly because of higher QF (Figure 6).

Figure 8.

Ratios of (a, d) QH, (b, e) QE, and (c, f) ΔQS over (Q*+ QF) for periods when (Q*+ QF)> 0 W m−2. Hourly bins (n > 30) are shown for clarity

QH decreased to a slightly positive and negative value in the evening at URB and SUB, respectively (Figure 5(b) and (f)). Christen and Vogt (2004) reported a similar night-time difference between the urban and suburban sites for a summertime period and related this difference to more frequent non-stable conditions of the near-surface air layers at the denser sites. Similarly, URB showed less frequent stable conditions than SUB during the study period (Figure 9), hence, the stability regime could also affect intra-urban differences in night-time QH over Montreal during winter time.

Figure 9.

Histogram of near-surface stability. Stable conditions correspond to 10 > ζ> 0.1, neutral to 0.1 > ζ> − 0.1, unstable to − 0.1 > ζ> − 0.5, and strong unstable to − 0.5 > ζ> − 100. n = 1313 co-incident half-hours

3.6. Latent heat flux

Total daily QE was on average higher in 2008–2009 at both RUR and SUB, corresponding to the winter with most amount of liquid water readily available for evaporation from rainfall (Table I) and snowmelt (Table III, Figure 2) and sublimation from snow. In contrast at URB, total daily QE was higher in 2007–2008 than in 2008–2009 when the long snow-cover period promoted evaporation through a constant supply of snow to melt from rooftops and streets. Hence, Montreal showed intra-urban differences in terms of the water vapour emission response to wintertime climatic conditions.

Total daily QE averaged over both winters was 0.71, 1.08, and 1.20 MJ m−2 day−1 at RUR, SUB, and URB, respectively (Table IV), which is slightly lower than the wintertime values reported by Christen and Vogt (2004) for a denser temperate European city. The daily QE totals reported from our sites are also much lower than those measured over a site exposed to much milder winter conditions with frequent rainfall in Vancouver, Canada (Grimmond, 1992). Our results show that water vapour emissions in wintertime were 51–69% higher over urban areas relative to the rural baseline. The main sources of water vapour include combustion from vehicular traffic and heating fuels as well as snow sublimation/evaporation from different surfaces, especially rooftops and roads. The increase in water vapour emissions with urbanization is consistent with the between-site differences in the distribution of those sources as indicated by QV, QB (Figure 6), λP, and λI (Table I).

The monthly mean daytime QE/(Q*+ QF) was generally constant around 0.1 at both SUB and URB (Table V). Also, QE showed minimal diurnal variation at all three sites (Figure 5(c) and (g)) and QE/(Q*+ QF) was very similar between SUB and URB throughout daytime hours at values below 0.2 (Figure 8(b) and (e)). A significant relationship between daily daytime QE/(Q*+ QF) and any environmental variable could not be found at any site (data not shown). However, most high QE/(Q*+ QF) values (>0.2) were associated with values of the stability index (ζ) between − 0.1 and 0.1, i.e. neutral conditions (Figure 10). Hence, high QE/(Q*+ QF) values appeared to be restricted to specific meteorological events occurring in neutral stability conditions, suggesting that QE/(Q*+ QF) is basically constant within and between days over snow covered, urbanized sites in winter, at least under unstable conditions.

Figure 10.

Ratio of mean daytime (10h00 to 14h00 LST) QE over (Q*+ QF) as a function of the stability parameter ζ. Vertical bars enclose neutral conditions (0.1 > ζ> − 0.1)

3.7. Net change of heat storage

Mean total daily ΔQS was positive at both SUB and URB on a wintertime basis (December–March; Table IV) although mean total daily ΔQS was negative from October to January and positive the rest of the time (not shown). Other studies reported negative net change of heat storage during wintertime (Christen and Vogt, 2004; Offerle et al., 2005; Offerle et al., 2006) but these studies did not calculate QF separately from the residual term ΔQS. Between-year variability of ΔQS was consistent with that of Q*, while between-site variability for the winter as a whole was mostly due to that of QE (Table IV). These results suggest that intra-urban differences in ΔQS during the winter depend on net radiation and (lack of) liquid water available to evaporate (see previous section).

Overall, the monthly mean daytime ΔQS/(Q*+ QF) generally increased from about 0.35 to about 0.45–0.5 (Table V). Daily daytime ΔQS/(Q*+ QF) was positively correlated with Tair at all sites and Tair explained 13, 15, and 26% of ΔQS/(Q*+ QF) variability (Figure 7(b)). ΔQS/(Q*+ QF) was consistently higher at RUR than SUB and URB for any given temperature which is in accordance with the correlation found between QH/(Q*+ QF) and Tair. A coincidental regression test (Zar, 1984) indicated that regressions of ΔQS/(Q*+ QF) with Tair were not significantly different between SUB and URB [F = 0.05, degrees of freedom = (2; 215)]. Hence, the temporal variability of the net change of heat storage was correlated to air temperature with no sensitivity with respect to intra-urban difference in surface characteristics.

ΔQS diurnal variability was of similar amplitude at URB and SUB although buildings are generally smaller and sparser at SUB (Figure 5(d) and (h)). However, snow is removed from streets and rapidly melts from flat roofs at URB while it accumulates as relatively undisturbed snowpacks in front and back yards and in dense and high snowbanks along most streets at SUB. Hence, the different snow accumulation regimes can likely compensate to some extent for the lower building density in terms of heat storage. The similar ΔQS diurnal amplitude at both SUB and URB also suggests that the net change in heat storage in urban environments is independent from the total amount of heat stored as expected from building density, assuming that ΔQS estimates do not suffer from site-specific systematic errors (e.g. lack of energy balance closure).

URB and SUB showed the highest and lowest ΔQS values at mid-day and late afternoon, respectively (Figure 5(d) and (h)). Similar diurnal trends have been reported for different urban densities and climatic conditions (Grimmond, 1992; Christen and Vogt, 2004; Spronken-Smith et al., 2006; Coutts et al., 2007). However, the ΔQS morning increase and afternoon decrease occurred about 1 h earlier at URB than SUB (Figure 5(d) and (h)), while ΔQS/(Q*+ QF) was similar between SUB and URB before noon (Figure 8(c) and (f)). Except for early morning hours when (Q*+ QF)> 0 W m−2 and Q*< 0 W m−2 at URB, ΔQS/(Q*+ QF) reached a maximum of about 0.6 around 8h00 LST at both sites and decreased thereafter. After noon, ΔQS/(Q*+ QF) dropped earlier and faster at URB than SUB which is consistent with between-site difference in Q* (Figure 5(a) and (e)). URB showed higher early morning ΔQS than SUB induced by greater QF (Figure 6). These results indicate that intra-urban differences in the diurnal course of ΔQS, especially in the afternoon, are to be expected over temperate, snow-covered cities during winter time.

Measurements of QH and QE with the eddy covariance technique have been showed to suffer from systematic underestimation due to the lack of energy balance closure (Wilson et al., 2002) which appears to be because of theoretical or methodological rather than instrumental issues (Foken, 2008). Hence, estimates of ΔQS calculated as a residual term should be considered as an upper limit. It is also worth noting that the energy used for water phase changes between liquid water, snow, and ice, which can be significant during snowmelt (Leroyer et al., 2010), and water draining in the sewer system was not included in our QE measurements and might have affected our ΔQS estimates. However, this study showed that independent estimates of QF, which is a significant term of the energy budget of high latitude cities during wintertime, can be used to estimate ΔQS at fine time scale.

4. Conclusions

This study reports on the radiation and energy balances of three sites located along an urbanization gradient in the Montreal region for two wintertime (December–March) periods in 2007–2009. This study represents a first opportunity to analyse a unique two-year data set from a city with cold and snowy winters that will be used to validate and refine urban climate process models, especially regarding snow.

During the snow-cover period, the urban and suburban sites showed similar albedo (∼0.4) which was about half that at the rural site (∼0.8). The seasonal variability of the urban site's albedo was primarily affected by the presence of snow on rooftops; snow here tended to melt rapidly and rooftops represented about 40% of the radiative footprint area. The decrease in albedo due to the absence of snow on rooftops at the urban site induced greater net radiation (Q*) overall despite increased daytime longwave radiation losses at the neighbourhood level. The suburban site showed lower albedo by up to 0.05 as compared to the urban site, which resulted in generally higher daytime net radiation. The albedo ratio of the two urbanized sites varied with a sky clearness index. The geometry of the suburban site, including the distribution of tall trees, along with relatively extended snow-covered areas between buildings and trees likely induced a site-specific response in albedo to cloudiness resulting in greater variability in albedo with direct-beam radiation at the suburban site.

This study presents the first estimates of hourly anthropogenic heat flux (QF) over residential areas in Montreal. QF was 7–13 W m−2 at the suburban site and 25–45 W m−2 at the urban site. QF was dominated by heat loss from buildings, while vehicular traffic contributed to rush hour peaks at both sites. Mean daily QF was of the same magnitude as sensible heat flux (QH) at the urban site and up to half of QH at the suburban site, while daytime QF accounted for up to 9–29% of the total available energy (Q*+ QF). Higher QF compensated for lower Q* at the urban site which showed higher (Q*+ QF) overall on a daily basis. Thus QF represented a significant term of the energy budget over Montreal urban areas during wintertime.

The total available energy was mostly dissipated as sensible heat flux at the beginning of the winter season and was mostly stored towards the end of the winter at both urbanized sites. Both QH/(Q*+ QF) and ΔQS/(Q*+ QF) showed significant correlation with air temperature at all three sites and regressions were not significantly different between the urban and suburban sites. The correlations of daytime QH/(Q*+ QF) and ΔQS/(Q*+ QF) with Tair indicate that as air temperature decreases in wintertime, available energy is released as sensible heat flux more than it is stored, independently from any intra-urban variability in neighbourhood characteristics.

Water vapour emissions over Montreal in wintertime were low and accounted for a small fraction (10%) of total available energy during daytime at the urbanized sites. However, water vapour emissions were 51–69% greater at the urbanized sites than the rural baseline, which was consistent with between-site differences in expected emission sources. Water vapour fluxes show intra-urban variability in their response to between-year differences in climatic conditions.

Urbanized sites showed a distinct diurnal variability in energy balance components, except for QE. Mid-day Q* and QH were 2–5 times and about an order of magnitude greater at the urbanized sites than the rural site, respectively. ΔQS was maximal in early morning and minimal in late afternoon and showed similar diurnal amplitude at both urbanized sites. These results suggest that the net change of heat storage in urban environments is independent from the total energy stored that would be expected from building structure and density. The total available energy was mostly stored before noon and dissipated as QH in the afternoon at both urbanized sites. However, the urban site showed earlier and more rapid changes in energy partitioning between QH and ΔQS than the suburban site. The atmospheric stability regime induced intra-urban differences in night-time QH over Montreal during winter time.

These observations, along with the underlying data set, can be used to refine, validate, and calibrate urban climate models in terms of the influence of snow, anthropogenic heat flux, and the intra-urban variability in structure and composition of urban features, on the radiation and energy balance of a Canadian temperate city during winter time.


This study was funded through a grant to Drs J. Voogt and T. R. Oke and IBS by the Canadian Foundation for Climate and Atmospheric Sciences (CFCAS). Post-doctoral funding to OB was provided by CFCAS. The authors wish to thank the assistance of Mitchell Lavoie for the QF estimates, Olivier Gagnon (Environment Canada) for vehicular traffic profiles, and Jinfei Wang (University of Western Ontario) for the surface cover classification. The field and technical assistance of Eric Christensen, Kenton Ollivierre, Pierre-Luc Lizotte, Josée Thibodeau, Jean-François Aublet at McGill, and Dr Frédéric Chagnon and Bruno Harvey at Environment Canada is greatly appreciated.

A1. Appendix

A1.1. Estimating anthropogenic heat flux

Hourly anthropogenic heat flux (QF) was estimated using an approach based on Sailor and Lu (2004) and adapted to the available data for the urban and suburban sites. The general approach was to estimate each term separately (human metabolism, vehicular traffic, and building sector) by calculating hourly per reference unit (person, vehicle, household, or floor space m2) heat release and multiplying by hourly number of units per surface area (population, vehicle, household, or floor space m2 density) for the corresponding site. The reference unit used here differed somewhat from that of Sailor and Lu (2004), who strictly used a per capita approach, given the available data for the Montreal region. Daylight savings time was accounted for when appropriate.

A1.2. Hourly population densities

Hourly population densities (ρpop) were estimated separately for weekdays and weekends. Night-time (18h00 to 6h00 on weekdays, 18h00 to 9h00 on weekends) ρpop was calculated as the total number of inhabitant in the borough from 2006 census data over total borough surface area (Ville de Montréal, 2009a). At SUB, three large parks (201, 288, and 159 ha; Ville de Montréal, 2009b) were removed from borough surface area as these are considered as natural reserves more than urban parks. Daytime ρpop was calculated using borough's total population after accounting for work, school, shopping, and/or leisure displacements into, out of, and within each borough according to the Agence métropolitaine de transport's Fall 2003 origin-destination survey (AMT, 2009a). All displacements were considered for weekdays and only shopping and leisure displacements for weekends. As per Sailor and Lu (2004), transition periods (7h00 and 17h00 on weekdays and 10h00 and 17h00 on weekends) were linearly interpolated.

A1.3. Heat from human metabolism (QM)

QM was defined as follows:

equation image(A1)

where pcHM is the per capita rate of heat from human metabolism (W person−1) and ρpop is hourly population density (person m−2). pcHM was set to 175 and 75 W person−1 during daytime and night-time, respectively (Sailor and Lu, 2004). QM was linearly interpolated during transition periods (6h00 and 22h00 on weekdays and 9h00 and 22h00 on weekends based on QV profiles).

A1.4. Heat from vehicular traffic (QV)

Hourly QV was calculated as follows:

equation image(A2)

where pvDVD is the per vehicle daily vehicle distance (km vehicle−1 day−1), NV the number of vehicle per person (vehicles person−1), Ft the fraction of daily traffic per hour (day hour−1), and EV the amount of energy released per vehicle per distance travelled (J km−1). pvDVD was calculated as the total vehicle kilometres travelled (aggregating vehicle types) for all 1 × 1 km cells comprising road segments with traffic count estimates around the flux towers, divided by traffic counts for one segment along each road with data according to the Agence métropolitaine de transport's 1998 origin-destination survey (AMT, 2009b). NV was taken from 2003 origin-destination survey (AMT, 2009a). For weekdays, all vehicle classes (car, light truck, and heavy truck) were included, while heavy trucks were excluded for weekend estimates. Weekday Ft was computed using simulated displacements along a simplified road network within an approximately 3 km radius of each site for a weekday in January 2009 (Olivier Gagnon, personal communication) based on vehicle counts by class from the 2003 origin-destination survey (AMT, 2009a)). Weekend Ft was taken from Noriega et al. (2006). EV was calculated as follows:

equation image(A3)

where NHC is the net heat of combustion of gasoline (45 × 106 J kg−1), ρfuel the optimal fuel density of gasoline (0.75 kg l−1), and FE the fuel economy (l km−1). NHC and ρfuel were taken from Sailor and Lu (2004). FE was taken as a weighted average over vehicle MOBILE 6 classes (EPA, 2009) for each study site.

A1.5. Heat from the building sector (QB)

QB was estimated from energy consumption information assuming all the energy consumed was ultimately lost as heat. Annual energy consumption data by province were available publicly for the period 1990–2007 from the Natural Resource Canada's Office of Energy Efficiency (OEE, 2010). These data included energy efficiency (consumption per unit of space) stratified by sector, energy sources, end-use, and house or activity type but lacked information on intra-annual variability. In temperate regions, climate and people's behavior determine the temporal variability of energy demand for space heating and appliance use (Sailor et al., 1998; Yao and Steemers, 2005; Hamilton et al., 2009). Thus QB was divided into two end-uses:

equation image(A4)

where QBh is the heat lost from space heating and QBa the appliance use or non-space heating. QBh is temperature-dependant and includes all energy sources except the fraction of electricity consumption that corresponds to QBa which is temperature-independent. To estimate QB around our sites, only the residential and commercial/institutional sectors were relevant. House types within the residential sector were treated separately, while no distinction was made between activity types for the commercial/institutional sector. Energy efficiency was expressed on a per household basis and on a per floorspace m2 basis for the residential and commercial/institutional sectors, respectively.

Annual energy efficiency for the period 2000–2007 was extrapolated to 2008 and 2009 using a relationship with mean wintertime (January–March and December) air temperature at the Montreal Trudeau airport (Environment Canada, 2010) and with time for space heating and non-space heating, respectively. Annual QBh and QBa were calculated by multiplying energy efficiency by number of household (residential) or by floorspace m2 (commercial/institutional) per ground surface area estimated from a GIS database along 14 census tracks and 13 dissemination areas within a radius of about 1 km around the urban and suburban sites, respectively. Commercial/institutional floor space m2 was estimated as building dimensions estimated from a GIS database multiplied by number of floors assessed visually from aerial photographs (Bing Maps, 2009).

Annual QBh and QBa were divided into hourly profiles following different methodologies. For QBh, hourly air temperature data measured at the Montreal Trudeau airport for the period 2007–2009 (Environment Canada, 2010) was used to calculate a heating degree-hour index (HDHi) using a reference temperature of 18 °C following Sailor and Vasireddy's (2006) work. Fractional HDHi was obtained on an annual basis by dividing hourly HDHi by the sum of all HDHi for each year and was then multiplied by annual QBh to get hourly QBh. For QBa, energy consumption was assumed constant for one day to the next within each year and daily QBa was obtained by dividing annual QBa by the number of days of each year. Separate diurnal fractional profiles for domestic and non-domestic uses were taken from Hamilton et al. (2009) and were multiplied by daily QBa to get hourly QBa.