Constructing climate change scenarios of urban heat ...

Environment and Planning B: Planning and Design 2008, volume 35, pages 902 ^ 919

doi:10.1068/b33066t

Constructing climate change scenarios of urban heat island intensity and air quality Robert L Wilby

Department of Geography, Lancaster University, Farrer Avenue, Lancaster LA1 4YQ, England; e-mail: [email protected] Received 2 June 2006; in revised form 1 March 2007

Abstract. As the global population becomes increasingly urbanised, so interest has grown in the potential climate change impacts on city infrastructure, services, and environmental quality. However, urban areas are only beginning to be represented explicitly in the land-surface schemes of dynamical climate models through modified energy and moisture budgets. This paper summarises recent evidence of urban impacts on climate and vice versa. The technique of statistical downscaling is then introduced through exemplar studies of London's future urban heat island and peak ozone concentrations. Projections of both indices are derived from atmospheric variables supplied by four general circulation models, driven by a medium-high (A2) emissions scenario for the 2050s. The results show further intensification of the nocturnal heat island and higher ozone concentrations that are most pronounced in summer. These changes reflect sensitivity to variations in regional climate alone, so omit other factors such as changes in land use, emissions, climate feedbacks, or synergies between air quality and heat islands. Nonetheless, the downscaled scenarios are consistent with an emerging picture of increasing risks to human health in urban areas unless appropriate adaptation measures are taken.

1 Introduction Studies of the meteorology of cities are enjoying a renaissance, thanks partly to the continued urbanisation of the world's population and a growing realisation that built environments are potentially vulnerable to climate change. Extreme weather events such as Hurricane Katrina (figure 1) have stimulated much discussion about possible changes in the destructiveness of storms (Emanuel, 2005), and have exposed the fragility of critical urban infrastructures and supporting resource systems. Likewise, the devastating European summer heat wave of 2003 raised important questions about our understanding of the human health and social implications of urban thermal stress (McGregor et al, 2006). In these cases the built form and land-cover characteristics of cities were recognised as factors that exacerbated a `natural' hazard. Hence we may consider urban areas as both a driver and receptor of climate change impacts. 1.1 Impacts of urbanisation on climate

Recent progress in urban climatology was reviewed in detail by Arnfield (2003). Since Luke Howard's classic study of London's urban heat island (UHI) in the early 19th century (Howard, 1833), it has been known that urban centres can be up to several degrees warmer than the surrounding countryside. Compared with vegetated surfaces, building materials retain more solar energy during the day, and have lower rates of radiant cooling during the night. Urban areas also have lower wind speeds, less convective heat losses, and less evapotranspiration, yielding more energy for surface warming. Artificial space heating, air conditioning, transportation, cooking, and industrial processes introduce additional sources of heat into the urban canopy layer, as shown by distinct weekly cycles in UHI intensity (figure 2).

Constructing climate change scenarios

903

Figure 1. Infrared image of hurricane Katrina prior to landfall on 29 August 2005 (source: UK Met Office, http//www.metoffice.com/satpics/namerica IR.html).

Intensity (0:1 8C)

19.0

(a)

18.5 18.0 17.5 17.0

Intensity (0:1 8C)

4.5 4.0 3.5 3.0 2.5

(b)

Thu

Fri

Sat

Sun

Mon

Tues

Wed

Figure 2. Weekly variations in London's urban heat island intensity derived from temperature differences between St James's Park and Wisley, 1959 ^ 2000: taken during (a) the night (minimum temperatures) and (b) the day (maximum temeratures). The vertical bars denote the standard error of the mean estimates for each day of the week (source: Wilby, 2003b).

904

R L Wilby

Table 1. Controls of urban heat island (UHI) magnitude and structure. Adapted from Oke (1982) and Arnfield (2003). Controls UHI intensity decreases with increasing wind speed. UHI intensity decreases with increasing cloud cover. UHI intensity is greatest during anticyclonic conditions. UHI intensity is best developed in the summer or warm half of the year. UHI intensity tends to increase with increasing city size and/or population. UHI intensity is greatest at night. UHI may disappear by day or the city may be cooler than the rural environs. Rates of heating and cooling are greater at rural sites than the city.

After numerous field studies the most important determinants of UHI character and behaviour are now well understood (table 1). For example, detailed monitoring has revealed that London's UHI is a highly mobile phenomenon with the location of the thermal maximum shifting day to day in response to variations in wind direction (Graves et al, 2001). Other studies have emphasised the control exerted by underlying land-use features, such as the built-up ratio, on the pattern of mean maximum UHI intensity (Bottya¨n et al, 2005). This implies that, as the density of development and/or area of continuous built cover increases, so does the strength of the UHI. Long-term records show that London's nocturnal UHI has indeed become more intense since the 1960s (table 2), especially during spring and summer (Lee, 1992; Wilby, 2003a). Table 2. Seasonal trends in London's nocturnal and daytime urban heat island intensity (8C per decade) 1959 ^ 98 (source: Wilby, 2003a). Season

Nocturnal Tmin

Daytime Tmax

Winter Spring Summer Autumn

ÿ0:023 a 0:131** 0:120** 0:077*

ÿ0:081** 0:015 a 0:036* ÿ0:046*

a Not

significant; * significant at p 0:05; ** significant at p 0:01.

UHIs have also figured prominently in debates about global warming. Given the above trends, much discussion has understandably surrounded the influence of urbanisation (and other land-cover changes) on long-term surface-air temperature trends (Kalnay and Cai, 2003). The possible contamination of surface-air temperature records by urban heat advection is, indeed, consistent with observed decreases in diurnal temperature ranges (IPCC, 2001). However, recent analyses of trends of global annual average minimum temperatures for windy, calm, and all conditions show identical trends of 0:19 8C 0:06 8C per decade since 1950 (Parker, 2004). If significant urban effects were present they would have emerged in the record for calm days. Even where urban heat advection has been detected at individual stations, such as De Bilt in the Netherlands, the effect is of the order of approximately 10% of observed temperature rise over the last century (Brandsma et al, 2003). 1.2 Impacts of climate change on urban areas

To date, there has been relatively little research into climate change impacts on the built environment, but this situation is changing in the UK thanks to dedicated programmes (such as Building Knowledge for a Changing Climate). Previous work


905

has tended to focus on managing existing weather-related risks, rather than on addressing future climate change impacts. Typical measures for mitigating high temperatures include implementing heat health warning systems (eg Philadelphia, Shanghai, and Lisbon) or reducing the intensity of the UHI using cool roofs, greenspace, and water features (eg Tokyo, Basel, and Newark). However, it is generally recognised that the most important potential impacts of climate change on cities include fewer periods of extreme winter cold; an increased frequency of heat waves, air and water pollution episodes; rising sea levels and increased risk of storm surge; and changes in the timing, frequency, and severity of urban flooding associated with more intense precipitation events (IPCC, 2001; Wilby, 2007). One of the first climate change impact studies for a world city was undertaken by the London Climate Change Partnership (LCCP, 2002). Table 3 provides a summary of the main findings by key sector. Subsequent work has focused more specifically on impacts on London's biodiversity (Wilby and Perry, 2006), transport system (GLA, 2005a), and the identification of adaptation measures for developers and planners (GLA, 2005b; 2006; LCCP, 2006). Elsewhere, adaptation scenarios for addressing the UHI of Newark and Camden, New Jersey have highlighted the benefits of urban trees both for the present and for year 2020 (Rosenzweig et al, 2005; Solecki et al, 2005). Models show that urban vegetation can deliver lower energy demands for air conditioning, as well as reduced health hazards through removal of air pollutants. Climate change is expected to affect the outdoor air quality of large urban areas. This is because future weather will have a major influence on the production, transport, dispersal, and deposition of pollutants. For instance, any increase in the frequency of hot, anticyclonic weather in summer will favour the creation of more temperature inversions trapping pollutants in the near-surface layer of the atmosphere. Lee (1993) estimates that a 18C rise in summer air temperatures (a proxy for the amount of catalysing sunshine) is associated with a 14% increase in surface ozone concentrations in London. This partly explains the poor air quality that contributed to significant mortality during the European summer heat wave of 2003 (Stedman, 2004). The exceptional temperatures stimulated emissions of natural volatile organic compounds (VOCs), such as isoprene, from vegetation, leading to the highest recorded ozone concentrations experienced in the UK for over a decade. The UK government subsequently commissioned a comprehensive review of potential impacts of climate change on outdoor air quality (Defra, 2006). The Air Quality Expert Group anticipates fewer poor air quality episodes in winter under the assumption of increased ventilation associated with greater winter storminess. However, an increase is expected for summer photochemical smog episodes and ozone precursor biogenic VOCsöboth linked to rising temperatures and solar radiation. This has implications for the future selection of species used for biofuels or urban tree planting (see Stewart et al, 2003). The remainder of this paper will illustrate methods for constructing climate change scenarios of UHI intensity and air quality (ozone) using case studies for London. These indicators of climate change were chosen because of their paramount significance to human comfort and health in urban areas (DOH, 2001; Patz et al, 2005). Section 2 reviews the options for deriving scenarios, given that neither UHI nor air quality can be obtained directly from general circulation model (GCM) output. Section 3 outlines the statistical downscaling method used to develop UHI and air quality scenarios for London, described in sections 4 and 5, respectively. Finally, section 6 briefly considers the key uncertainties and priorities for future research.

906

R L Wilby

Table 3. Potential climate change impacts for London (source: LCCP, 2002). Issue

Main impacts

Higher temperatures

Intensified urban heat island, especially during summer nights. Increased demand for cooling (and thus electricity) in summer. Reduced demand for space heating in winter.

Flooding

More frequent and intense winter rainfalls, leading to riverine flooding and overwhelming of urban drainage systems. Rising sea levels, storminess, and tidal surges require more closures of the Thames Barrier.

Water resources

Heightened water demand in hot, dry summers. Reduced soil moisture and groundwater replenishment. River flows higher in winter and lower in summer. Water-quality problems in summer associated with increased water temperatures and discharges from stormwater outflows.

Health

Poorer air quality affects asthmatics and causes damage to plants and buildings. Higher mortality rates in summer due to heat stress. Lower mortality rates in winter due to reduction in cold spells.

Biodiversity

Increased competition from exotic species, spread of disease and pests, affecting both fauna and flora. Rare saltmarsh habitats threatened by sea-level rise. Increased summer droughts cause stress to wetlands and beech woodlands. Earlier springs and longer frost-free seasons affect dates of bird egg-laying, leaf emergence, and flowering of plants. Increased likelihood of building subsidence on clay soils. Increased ground movement in winter affecting underground pipes and cables. Reduced comfort and productivity of workers.

Built environment

Transport

Increased disruption to transport system from extreme weather. Higher temperatures and reduced passenger comfort on the London Underground. Damage to infrastructure through buckled rails and rutted roads. Reduction in cold-weather-related disruption.

Business and finance

Increased exposure of insurance industry to extreme weather claims. Increased cost and difficulty for households and business of obtaining flood insurance cover. Risk management may provide significant business opportunity.

Tourism and lifestyle

Increased temperatures could attract more visitors to London. High temperatures encourage residents to leave London for more frequent holidays or breaks. Outdoor living, dining, and entertainment may be more favoured. Green and open spaces will be used more intensively.

2 Methods for constructing climate change scenarios for urban areas Approaches to the generation of building-scale climate change scenarios have already been reviewed by Hacker et al (2008). Various downscaling methods are suitable for obtaining future series of primary variables, such as temperature and precipitation, at length scales finer than the typical 300 km grid resolution of GCMs. But additional work is needed to explicitly represent urban surfaces or to derive èxotic' variables such as heat island or air quality indices. Nonetheless, the required techniques may still be


907

conveniently grouped into dynamical and statistical downscaling methods. This section gives a few examples of each approach. 2.1 Dynamical downscaling

The UK Climate Impacts Programme (UKCIP02) scenarios are a good example of dynamical downscaling, being derived from a regional climate model (RCM) with a spatial resolution of around 50 km (Hulme et al, 2002). However, the UKCIP02 scenarios are indicative of climate change for vegetated surfaces, even where grid boxes cover large urban areas such as London and Manchester. This is a major limitation because built areas exert significant influences on local surface energy and moisture budgets, and would be expected to respond differently to the radiative forcing of increased greenhouse gas concentrations. As has already been shown, artificial heat sources exert additional influence on urban temperature regimes (figure 2). As part of the BETWIXT (Built EnvironmenT: Weather scenarios for investigation of Impacts and eXTremes) project, the UK Hadley Centre developed a new landsurface `tiling' scheme to represent subgrid surface heterogeneity in a GCM (Betts and Best, 2004). Separate energy and water budgets were simulated for different land surfaces, including urban areas (figure 3). Tiles representing built areas incorporate physical characteristics such as heat storage in buildings, partitioning between sensible and latent heat fluxes, frictional drag on the atmosphere, and direct anthropogenic heat sources. Several experiments were then conducted, considering presence/absence of urban areas, the extent of artificial heat sources (equivalent to 0, 20, or 60 Wmÿ2 ), under current and doubled atmospheric CO2 concentrations. The various simulations enabled preliminary analyses of the effects of land-surface feedbacks, climate change, and anthropogenic heat sources on UHIs. Experimental results for New York signal fewer occurrences of near neutral daytime heat islands and a greater frequency of more intense nocturnal UHIs. H1

H2

H3

H4

T2 T1

T3

T4

Figure 3. Representation of the new land surface scheme implemented in the Hadley Centre regional climate model, with explicit `tiled' representation of the temperatures T and heat fluxes H on different landscape types, including urban areas (source: Betts and Best, 2004).

Dynamical downscaling has also been used to investigate the effects of climate change on continental-scale air quality. For example, Leung and Gustafson (2005) used the Penn State/NCAR mesoscale model (MM5) to investigate potential changes in atmospheric conditions related to air quality over the US, parts of Canada, and Mexico. The results show that, during summer, Texas experiences marked warming, increased downward solar radiation, less frequent rainfall, and more frequent stagnation of air massesöall of which favour an increase in ozone concentrations by the 2050s. Comparable changes emerge for central and southern Europe, owing to decreased precipitation scavenging of oxidised nitrogen compounds, increased temperatures favouring biogenic precursors of ozone, and increased radiation driving the photochemical

908

R L Wilby

production of ozone (Langner et al, 2005). Rising peak ambient ozone concentrations are, in turn, expected to translate into increased morbidity and mortality. For instance, MM5 and the Community Multiscale Air Quality atmospheric chemistry model showed a median 4.5% increase in ozone-related acute mortality across the thirty-one county New York metropolitan region by the 2050s (Knowlton et al, 2004). 2.2 Statistical downscaling

As shown above, dynamical downscaling can provide useful insights into the physical mechanisms affecting future heat stress and ambient air quality in urban areas. However, such modelling is very resource intensive and so is normally undertaken for a single driving GCM, small number of emission scenarios, and/or a few decades in the future. The alternative strategy involves statistically downscaling climate and air quality indices of interest for individual sites. As explained by Hacker et al (2008), statistical downscaling uses empirical relationships between large-scale atmospheric driving variables (such as pressure fields) and local-scale weather (such as maximum temperature) to estimate how the smaller-scale properties will change in the future, given changes in the larger-scale climate. For a detailed description of various statistical downscaling methods and their assumptions see Wilby et al (2004). A key advantage of statistical downscaling over dynamical downscaling is that a large number of uncertainties affecting future scenarios can be evaluated, such as different heat island and air quality scenarios emerging from different driving GCMs and/or emission scenarios. For example, Hayhoe et al (2004) scaled station-scale temperatures by several GCM temperature projections to investigate future extreme heat and heat-related mortality in four cities (Los Angeles, Sacramento, Fresno, and Shasta Dam). The annual number of days classified as heatwave conditions increased under all simulations. Using known relationships between extreme heat and summer excess mortality for the Los Angeles metropolitan area, heat-related deaths were shown to increase by up to seven times by the 2090s, even with acclimatization. A further benefit of statistical downscaling is the ability to construct scenarios for indices that are not directly available from either GCMs or RCMs. For example, in a previous paper (Wilby, 2003a) I used information on regional airflows, pressure, and humidity supplied by a GCM to investigate future changes in the intensity of London's UHI under two emission scenarios. The results suggest further intensification of the mean nocturnal UHI, by as much as 1 8C in summer by the 2080s, even when excluding possible changes in anthropogenic heat sources, population density, and infrastructure. Although this might seem a modest amount of urban warming it is in addition to the present average ( 2 8C), and peak ( 10 8C) UHI intensity. Furthermore, there would be a disproportionate increase in the number of nights with peak intensity events, and this is all before the regional warming projected by the UKCIP02 scenarios is factored into the absolute temperatures experienced in the centre of London. This early work is extended in following sections by considering UHI behaviour under a wider range of GCMs, and by exploring associated changes in peak ozone concentrations. 3 The Statistical DownScaling Model (SDSM) As in the previous case study, climate change scenarios for London were generated via the SDSM version 3.1 (Wilby et al, 2002). SDSM is best described as a hybrid of regression-based and stochastic weather generator downscaling methods, because daily atmospheric circulation patterns and moisture variables are used to condition local-scale predictands at target sites (in this case UHI and air quality indices). Unexplained variance is reproduced by adding stochastically resampled residuals from the predictor ^ predictand regression relationships. This enables the generation


909

Table 4. Examples of Statistical DownScaling Model applications to climate change impact assessment. Study

Source

Precipitation and temperature scenarios for water balance modelling of Lake Ziway, Ethiopia Precipitation and temperature scenarios for snowpack modelling in the Colorado Rockies, US Evaluation of links between synoptic climatology and glacier mass balances, Norway Comparison of statistical and dynamical downscaling methods for estimating future probabilities of extreme weather events, UK Modelling the effects of climate and land-use change on flood risk in the Alzette basin, Luxembourg Evaluation of daily precipitation and temperature simulations for different downscaling techniques in a Quebec watershed, Canada Future mass balance and meltwater yields from glaciers in the Alps, Switzerland Continuous simulation of daily river flows for future flood risk estimation in selected catchments, UK Evaluation of daily precipitation simulations for different downscaling techniques, China Scenarios of wet season precipitation occurrence and amounts for the Anti Atlas, Morocco Transient scenarios of nitrogen concentrations and evaluation of adaptation options for a lowland chalk stream, UK

Abraham (2006) AGCI (2006) Fealy (2004) Goodess et al (2003; 2007) Guangul (2003) Khan et al (2006) MacDonald (2004) Reynard et al (2004) Wetterhall et al (2007) Wilby (2005) Whitehead et al (2006)

of multiple realisations with slightly different time series attributes, but the same overall statistical properties. An advantage over conventional time series adjustment (`change factor' or `morphing') methods is that new temporal sequences of (extreme or more persistent) events can be generated under future climate forcing (Diaz-Nieto and Wilby, 2005). A further strength of SDSM is that scenarios can be produced in seconds using a desktop computer. As noted before, this means that it is feasible to evaluate a range of GCMs and/or emission scenarios ö both important sources of uncertainty affecting climate change assessments. The tool is also extremely versatile as is evidenced by the breadth of recent applications (table 4). 3.1 Large-scale climate information

Atmospheric predictor variables used to calibrate SDSM were obtained from the National Center for Environmental Prediction (NCEP) reanalysis. Future climate change scenarios were downscaled from four GCMs: the Canadian Centre for Climate Modelling and Analysis model (CGCM2); the Commonwealth Scientific and Industrial Research Organisation model (CSIRO Mk2); the Max-Planck-Institut for Meteorology model (ECHAM4); and the Hadley Centre's coupled ocean/atmosphere climate model (HadCM3). The archive of NCEP and GCM output contains twenty-nine daily predictors (describing atmospheric circulation, thickness, and moisture content at the surface, 850 hPa and 500 hPa levels), for nine grid boxes covering the British Isles, for the period 1961 ^ 2100 (figure 4). The present study uses predictors taken from the grid boxes centred on eastern (EE) and southwest (SW) England under the A2 (medium-high emissions) scenario of the IPCC Special Report on Emission Scenarios.

910

R L Wilby

60 8N

57 8N

54 8N

51 8N

48 8N 9 8W

6 8W

3 8W

08

3 8E

Figure 4. General circulation model grid boxes employed by the statistical downscaling model for sites across the UK. 3.2 Application of SDSM

Downscaling using SDSM involved two main steps. First, statistical relationships must be established between the variables of interest (ie daily UHI intensity and ozone concentrations) and large-scale indices of regional weather over the target region obtained from the NCEP reanalysis for the current climate. Table 5 shows the correlation coefficient for significant predictor ^ predictand relationships. These illustrate the strength of association between the variables for August. Second, the empirical predictor ^ predictand relationships for the observed climate were used to downscale ensembles of the same local variables for the future climate, using data supplied by four GCMs driven by the A2 emission scenario for the period 1961 ^ 2100. Table 5. Large-scale atmospheric predictor variables (from the National Center for Environmental Prediction reanalysis) used to downscale daily urban heat island (UHI) and air quality for London. The correlation coefficients (r) are given for the strength of association between variables in August. All variables included in the downscaling model are significant at the p 0:01 level. Predictand

Predictors

r

Nocturnal UHI 1961 ± 90

near-surface airflow strength near-surface westerly wind component near-surface vorticity 850 hPa geopotential height near-surface relative humidity

ÿ0.40 ÿ0.36 ÿ0.41 0.48 ÿ0.41

Maximum fifteen-minute ozone concentration 1992 ± 2000

mean sea-level pressure near-surface relative humidity near-surface westerly wind component mean regional temperature at 2 m

0.17 ÿ0.55 ÿ0.21 0.60


911

4 Case study 1: nocturnal UHI intensity As in a previous paper (Wilby, 2003a), the strength of London's UHI was represented by the temperature gradient between St James's Park (urban station) and Wisley in Surrey (a rural reference station 30 km from the city centre). The nocturnal UHI is given by the difference in the daily minimum temperature between the two stations, and is on average strongest in August (2:2 8C) and weakest in January (1:1 8C). However, these averages conceal considerable day-to-day variability. For example, the annual mean nocturnal heat island intensity for the period 1961 ^ 90 was 1:8 8C, but daily values ranged from 10:0 8C (on 14 January 1982) to ÿ8:9 8C (on 31 May 1970), with 5% of days having an intensity of 5 8C or more. Following a previous paper (Wilby, 2003a), SDSM was calibrated using 1961 ^ 90 daily minimum temperature differences between St James's and Wisley, and NCEP climate variables centred on the EE grid-box (figure 4). Significant correlations were found between the intensity of the heat island and several regional climate indices (most notably strength of airflow, vorticity, and near-surface relative humidity) (table 5). This combination of predictors explains up to 40% of the daily variability in the nocturnal heat island intensity in August. As expected, UHI intensity is greater under clear sky conditions (as implied by low humidity) and low wind speeds (under anticyclonic weather). Interestingly, the nocturnal heat island intensity is only weakly correlated with regional temperatures, suggesting that future intensification of London's heat island would be largely independent of projected temperature changes. The limitations of SDSM in reproducing London's present and projected UHI have been discussed before (Wilby, 2003a). The nocturnal model is known to have a cool bias ( 0:2 8C), and to underestimate the observed frequency of very intense episodes. Nonetheless, despite the coarse resolution of the input data, and the absence of thermal storage effects or anthropogenic heat sources, the model does emulate the time-varying UHI of notable heatwaves, such as the summer of 1995 (figure 5). Even so, in common with all downscaling methods (statistical and dynamical), the quality of future projections depends on the stationarity of the calibrated predictor ^ predictand relationships and their accompanying parameterisation. With the above points in mind, SDSM was used to downscale the UHI using largescale predictor variables supplied by the four GCMs under present and future (2050s) greenhouse gas forcing. Perfect prognosis results derived from NCEP fully capture the observed seasonal variability in the mean strength of the UHI, but consistently underestimate the frequency of the most intense episodes (figure 6). These outcomes 80

Intensity (0:1 8C)

60 40 20 0 ÿ20 ÿ40 1 April 1995

1 June 1995

1 August 1995

1 October 1995

Figure 5. Observed (grey lines) and downscaled (black lines) nocturnal urban heat island during the heatwave of 1995. These data were not used in model calibration. Thin lines denote daily values; thick lines are seven-day weighted averages (source: Wilby, 2003a).

912

R L Wilby

CGCM2

ECHAM4

Episodes (days)

Mean intensity (0:1 8C)

18 14

OBS

NCEP

4

2

0

10

Change in episodes (days)

6 4 2 0 ÿ2

(b)

HadCM3

6

22

(a) Change in mean (0:1 8C)

CSIRO

26

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

2 1 0 ÿ1


Figure 6. London's observed and downscaled nocturnal urban heat island for 1961 ^ 90 (a) and changes projected for the 2050s (b). The T-bars denote the upper limit of the 95% confidence band due to the variance unexplained by the statistical downscaling model. Intense nocturnal heat island episodes (right column) are arbitrarily defined as nights with temperature differences (between the urban and rural reference stations) exceeding 4 8C. For a description of the models, see text.

are repeated by the four GCMs, although CGCM2 tends to overestimate UHI intensity between August and November, whereas ECHAM4 and HadCM3 underplay mean intensities between July and August. All SDSM ^ GCM combinations underestimate the frequency of intense episodes, owing to the recognised tendency of SDSM to reproduce means at the expense of extreme events, combined with known biases in the GCM output driving the downscaling (see Wilby and Harris, 2006). Nonetheless, for most months, observations fall within the 95% confidence ranges of the GCM-downscaling model uncertainty. The four GCMs project intensification of the nocturnal UHI by the 2050s between May and October, but show a mixed response during the rest of the year (figure 6). The most rapid warming is shown by ECHAM4 in August (a further 0:5 8C by the 2050s)ö approximately half the historic rate observed in summer since the 1960s (table 2). Interestingly, ECHAM4 also shows weakening of the mean UHI intensity in winter, a pattern that is mirrored to a lesser extent by CSIRO. All GCMs suggest more frequent intense UHI episodes, with, on average, seven such events in August, compared with five presently. Although there are subtle differences between the GCM projections, overall the scenarios are qualitatively similar. All point to continued intensification of London's nocturnal UHI and a greater frequency of intense heat island episodes in summer. These changes are set against a background of more persistent and intense heatwaves over much of Europe and the USA signalled by other studies (eg Meehl and Tebaldi, 2004). Because SDSM makes an instantaneous connection between synoptic conditions and the UHI, compounding effects from long-duration heatwaves are underestimated in the present analysis. However, even a conservative outlook supports the view that there will be an increase in the risk of morbidity and mortality due to more extreme future heatwaves combined with more intense UHIs (Patz et al, 2005; Rooney et al, 1998).


913

Summer anomaly (days)

5 Case study 2: ozone pollution episodes As noted previously, atmospheric circulation patterns are a major determinant of ambient air quality and pollution episodes (eg Gardner and Dorling, 2000; McGregor and Bamzelis, 1995; O'Hare and Wilby, 1995), and, hence, the health of urban populations (eg Anderson et al, 1996; Stedman, 2004). Future air pollution concentrations in London will reflect local and regional patterns of emissions, as well as the frequency of high-pressure systems over southeast England. Whereas vigorous westerly airflows favour the dispersal of pollutants, stagnant anticyclonic weather provides ideal conditions for in situ pollution episodes. Whilst it is beyond the scope of this study to model the complex interactions between pollutant emissions, photochemistry, transport, and dispersal, it is possible to speculate about the future frequency of `pollution-favouring' weather patterns. For example, figure 7 shows changes in the frequency of highpressure systems centred on the EE grid box projected by HadCM3 under the A2 and B2 (medium-low) emissions scenarios (LCCP, 2002). Under A2 emissions there is a projected increase in the frequency of stagnant air masses favouring pollution episodes of over four days per summer by the 2080s, compared with 1961 ^ 90. Changes under the B2 emissions lead to an increase of more than two days per summer by 2080s. However, both plots show considerable interannual variability in the frequency of summer pollution episodes. The above analysis is extended by investigating the influence of regional-scale atmospheric predictor variables on site-specific ozone concentrations. Multivariate linear and nonlinear regression analyses have been used before to disaggregate longterm trends in urban ozone concentrations into climatic and nonclimatic components (eg Gardner and Dorling, 2000; Wise and Comrie, 2005). In this case SDSM was calibrated using 1992 ^ 2000 daily maximum fifteen-minute ozone concentrations at Russell Square in London, and NCEP predictor variables centred on the SW gridbox but covering SE and EE (figure 4). Peak ozone concentrations were found to be associated most strongly with mean sea level pressure, near-surface westerly wind speeds, relative humidity, and temperature (table 5). These predictor variables explain up to 58% of the daily variability in ozone at the site. Despite a large difference in spatial and temporal scales between the predictors and predictand, the downscaling yields a good approximation of the overall distribution of peak ozone concentrations at the site [figure 8(a)]. Closer inspection of the daily concentrations during the 1995 summer pollution episode reveals that SDSM faithfully captures the timing of individual peaks but, as in the case of the UHI, underestimates the severity of the very highest concentrations [figure 8(b)]. The unexplained variance may be due to missing factors, such as the release of natural VOCs during intense heatwaves, accumulative effects, or chemical reactions. 20 15 10 5 0 ÿ5 ÿ10 ÿ15 1961 1981 2001 2021 2041 2061 2081

(a)

1961 1981 2001 2021 2041 2061 2081

(b)

Figure 7. Projected change in the number of summer weather patterns centred on the EE grid box (see figure 4) favouring pollution episodes in London under A2 (medium-high) emissions (a) and B2 (medium-low) emissions (b), with respect to the 1961 ^ 90 average (source: LCCP, 2002). Annual data are shown in black, with trendlines in grey.

914

R L Wilby

0.20 0.18 0.16 Probability

0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00

0

Daily peak concentration (ppb)

(a)

20

40 60 Concentration (ppb)

80

100

140 120

Observed

100

SDSM

80 60 40 20 0 01/95

02/95 03/95

04/95 05/95

06/95 07/95

08/95

09/95

10/95

11/95

12/95 01/96

(b) Figure 8. Observed (black line) and downscaled (grey line) (a) distributions of daily maximum fifteen-minute ozone concentrations for 1992 ^ 2000, and (b) time series of the daily peak concentrations during the 1995 episode at Russell Square, London. SDSM: the statistical downsizing model.

As with the nocturnal UHI, SDSM was used to downscale peak ozone concentrations at the site using predictor variables supplied by the four GCMs for present (1992 ^ 2000) and future (2050s) climate conditions. As before, comparison of observed concentrations with results derived from NCEP predictors provides a means of assessing biases due to the downscaling technique per se, as well as the information content of the atmospheric predictor variables. Differences between NCEP and the GCMs are due only to biases in each model's climatology under present greenhouse gas concentrations. Overall, downscaling using NCEP predictors faithfully captures seasonal variations in mean ozone concentrations, but underestimates the 95th percentile concentration in summer, most notably in July and August (figure 9). Downscaling using predictors supplied by ECHAM4 and HadCM3 tends to overestimate mean concentrations in summer, but compensates for biases in the peak concentrations, giving a close match to observations, especially for ECHAM4. In contrast, downscaling from CGCM2 and CSIRO consistently underestimates both mean and peak ozone concentrations in summer months. All four GCMs project increased ozone concentrations throughout much of the year for the 2050s relative to present conditions (figure 9). The largest increases in the mean occur in July and August under the ECHAM4 predictors, and in the 95th percentile concentrations in July under HadCM3 output. Absolute changes in the


CSIRO

ECHAM4 95th percentile (ppb)

40 30 20 10

(a)

12 8 4 2 ÿ4 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

HadCM3

OBS

NCEP

70 50 30 10

Change in 95th percentile (ppb)

Change in mean concentration (ppb)

Mean concentration (ppb)

CGCM2

915

25 20 15 10 5 0 ÿ5


(b)

Figure 9. Observed (black line) and downscaled (grey line, NCEP) maximum fifteen-minute ozone concentrations at Russell Square, London for 1992 ^ 2000 (a) and changes projected for the 2050s (b). For a description of the models, see text.

mean maximum fifteen-minute concentration in July were up to 12 ppb, which is comparable to results of dynamical downscaling of ozone in New York City for the 2050s under A2 emissions (Knowlton et al, 2004). The range of uncertainty due to the choice of GCM was 8 ^ 13% for the mean, and 3 ^ 5% for the 95th percentile concentration. However, it is stressed that the projected changes and model uncertainty reflect only the influence of future atmospheric conditions, thereby excluding other important factors such as future emissions of ozone precursors. In addition, the downscaling does not take into account major feedbacks, such as ozone depletion by the increased water vapour content of a warmer atmosphere, or increased stratospheric influx of ozone to the troposphere (Stevenson et al, 2006). 6 Concluding remarks This paper has demonstrated some applications of statistical downscaling to climate impact assessments for urban areas. The case studies illustrate potential changes in UHI intensity and ambient air quality in isolation (although some commonality is implied by the shared downscaling predictor variables shown in table 5). Ideally, the various impacts of climate change should be mutually consistent to give plausible narratives of future environmental conditions in built areas. This includes broader considerations such as changes in urban energy consumption, emission controls, growth for the urban area, or future land-use conversions (as in Solecki and Oliveri, 2004). The London case studies show responses to instantaneous, large-scale weather forcing alone; no account is taken of the potential synergy between the heat island and ozone concentrations, or of possible feedback between climate change and atmospheric chemistry (see Defra, 2006). There is clearly scope for refinement of these analyses. The latest SDSM version 4.1 incorporates new routines for handling serially correlated data and for frequency

916

R L Wilby

t maximum 15-min ozone concentration (ppb)

estimation of extreme events. The former enables more sophisticated treatment of the time series of downscaled indices (beyond the simple use of lagged predictor variables in the current version). This could improve the representation of extremes that depend on cumulative effects, such as thermal storage in the urban fabric, or the build up of pollutant concentrations in the boundary layer. For example, summer ozone concentrations are known to be positively correlated (r 0:78) with the concentration of the preceding day (figure 10). Alternatively, greater attention could be paid to the choice of downscaling predictor variables, by compositing training data into low, medium, and high-intensity episodes (rather than assuming a common set of predictors and model weights for all events). 100

80

60

40

20

0

0

20 40 60 80 100 t ÿ 1 maximum 15-min ozone concentration (ppb)

Figure 10. Evidence of serial correlation in daily maximum fifteen-minute ozone concentrations (ppb) observed in August at Russell Square, London. Points lying above the line signal increasing concentrations between consecutive days.

Given these opportunities, statistical downscaling remains a valuable option for the rapid production of local-scale climate change information that cannot be obtained directly from GCMs or dynamical downscaling. The low computational demand of SDSM also enables the evaluation of key uncertainties affecting future scenarios, such as the host GCM. This is consistent with wider efforts to undertake multimodelensemble (eg Pryor et al, 2005; Stevenson et al, 2006) or perturbed-physics-ensemble experiments (eg Murphy et al, 2004), leading to the development of probabilistic climate change information for impact assessment (eg Benestad, 2004; New et al, 2007; Wilby and Harris, 2006). However, new downscaling frameworks will be required to translate coarse resolution (space and time) probabilistic information into local climate change scenarios for impact assessment, not least for urban locations. The investigation of London's nocturnal heat island shows qualitatively similar results regardless of GCM. All models point to further strengthening of the UHI and increases in the frequency of intense summer episodes owing to projected changes in weather patterns over southern England. By the 2050s mean nocturnal temperatures in August could be 3 8C warmer in London than surrounding nonurban landscapes (which are themselves projected to warm by several degrees). Given that the SDSM projections are conservative, and that peak intensity episodes are expected to warm by even more, maximum urban ^ rural differences could exceed 10 8C. There is also broad agreement amongst the GCMs that future weather patterns will lead to a deterioration in London's air quality, particularly during midsummer. Building designers and spatial


917

planners will need to factor such information into their long-term strategies to help safeguard the future health and comfort of city dwellers. Acknowledgements. This study develops earlier work on urban heat islands supported by the London Climate Change Partnership (LCCP). The author thanks the two anonymous referees, as well as Richard Betts and Martin Best for supplying information on the Met Office subgrid tiling scheme. References Abraham L Z, 2006,``Climate change impact on Lake Ziway watershed water availability, Ethiopia'', unpublished MSc thesis, Institute for Technology in the Tropics, University of Applied Science, Cologne AGCI, 2006 Climate Change and Aspen: An Assessment of Impacts and Potential Responses Aspen Global Change Institute, 100 East Francis St, Aspen, Colorado Anderson H R, Ponce de Leon A, Bland J M, 1996, `Àir pollution and daily mortality in London: 1987 ^ 92'' British Medical Journal 312 665 ^ 669 Arnfield A J, 2003, ``Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island'' International Journal of Climatology 23 1 ^ 26 Benestad R E, 2004, ``Tentative probabilistic temperature scenarios for northern Europe'' Tellus 56A 89 ^ 101 Betts R, Best M, 2004, ``Simulating climate change in urban areas: interactions between radiative forcing, landscape effects and heat sources'', BETWIXT technical briefing note 3, Met Office, Exeter Bottya¨n Z, Kircsi A, Szegedi S, Unger J, 2005, ``The relationship between built-up areas and the spatial development of the mean maximum urban heat island in Debrecen, Hungary'' International Journal of Climatology 25 405 ^ 418 Brandsma T, Ko«nnen G P, Wessels H R A, 2003, `Èmpirical estimation of the effect of urban heat advection on the temperature series of De Bilt (The Netherlands)'' International Journal of Climatology 23 829 ^ 845 Defra, 2006 Air Quality and Climate Change: A UK Perspective report of the Air Quality Expert Group to the UK Department of Environment, Food and Rural Affairs, http://www.defra.gov.uk/ environment/airquality/publications/airqual-climatechange/ Diaz-Nieto J, Wilby R L, 2005, `À comparison of statistical downscaling and climate change factor methods: impacts on low flows in the River Thames, United Kingdom'' Climate Change 69 245 ^ 268 DOH, 2001 Health Effects of Climate Change in the UK report of the UK Department of Health, London Emanuel K, 2005, `Ìncreasing destructiveness of tropical cyclones over the past 30 years'' Nature 436 686 ^ 688 Fealy R, 2004 An Assessment of the Relationship Between Glacier Mass Balance and Synoptic Climate in Norway: Likely Future Implications of Climate Change unpublished PhD thesis, Department of Geography, University of Ireland, Maynooth Gardner M W, Dorling S R, 2000, ``Meteorologically adjusted trends in UK daily maximum surface ozone concentrations'' Atmospheric Environment 34 171 ^ 176 GLA, 2005a The Impacts of Climate Change on London's Transport Systems Greater London Authority, London GLA, 2005b Adapting to Climate Change: A Checklist for Development Greater London Authority, London GLA, 2006 London's Urban Heat Island: A Summary for Decision Makers Greater London Authority, London Goodess C, Osbord T, Hulme M, 2003, ``The identification and evaluation of suitable scenario development methods for the estimation of future probabilities of extreme weather events'', Tyndall Centre for Climate Change Research, technial report 4, http://www.tyndall.ac.uk/ research/theme3/final reports/it1 16.pdf Goodess C M, Anagnostopoulo C, Bardossy A, Frei C, Harpham C, Haylock M R, Hundecha Y, Maheras P, Ribalaygua J, Schmidli J, Schmith T, Tolika K, Tomozeiu R, Wilby R L, 2007, `Àn intercomparison of statistical downscaling methods for Europe and European regions ö assessing their performance with respect to extreme temperature and precipitation events'' Climate Change, in press Graves H M, Watkins R, Westbury P, Littlefair P J, 2001 Cooling Buildings in London (BRE Press, London)

918

R L Wilby

Guangul S G, 2003 Modelling the Effect of Climate and Land-use Changes on Hydrological Processes: An Integrated GIS and Distributed Modelling Approach PhD thesis, Department of Hydrology and Hydraulic Engineering, Vrije Universiteit, Brussels Hacker J N, Belcher S E, Goodess C M, Holmes M J, Roaf S, 2008,``Building scale scenarios: climate change and their applications to the design of climatically sensitive buildings'', submitted to Environment and Planning B: Planning and Design Hayhoe K, Cayan D, Field C B, Frumhoff P C, Maurer E P, Miller N L, Moser S C, Schneider S H, Cahill K N, Cleland E E, Dale L, Drapek R Hanemann R M, Kalkstein L S, Leniham J, Lunch C K, Neilson R P, Sheridan S C,Verville J H, 2004, `Èmission pathways, climate change and impacts on California'' Proceedings of the National Academy of Sciences 101 12422 ^ 12427 Howard L, 1833 The Climate of London (Harvey and Darton, London) Hulme M, Jenkins G J, Lu X, Turnpenny J R, Mitchell T D, Jones R G, Lowe J, Murphy J M, Hassell D, Boorman P, McDonald R, Hill S, 2002, ``Climate change scenarios for the UK: the UKCIP02 scientific report'', Tyndall Centre for Climate Change Research, School of Environmental Sciences, University of East Anglia, Norwich IPCC, 2001 Climate Change 2001: The Scientific Basis contribution of working group I to the third assessment report of the Intergovernmental Panel on Climate Change (Cambridge University Press, Cambridge) Kalnay E, Cai M, 2003, `Ìmpacts of urbanization and land-use change on climate'' Nature 423 528 ^ 531 Khan M S, Coulibaly P, Dibike Y, 2006, `Ùncertainty analysis of statistical downscaling methods'' Journal of Hydrology 319 357 ^ 382 Knowlton K, Rosenthal J E, Hogrefe C, Lynn B, Gaffin S, Goldberg R, Rosenzweig C, Civerolo K, Ku J, Kinney P L, 2004, `Àssessing ozone-related health impacts under a changing climate'' Environmental Health Perspectives 112 1557 ^ 1563 Langner J, Bergstro«m R, Foltescu V, 2005, `Ìmpact of climate change on surface ozone and deposition of sulphur and nitrogen in Europe'' Atmospheric Environment 39 1129 ^ 1141 LCCP, 2002 A Climate Change Impacts in London Evaluation Study final technical report, Entec UK Ltd, London Climate Change Partnership, Greater London Authority, London LCCP, 2006 Adapting to Climate Change: Lessons for London London Climate Change Partnership, Greater London Authority, London Lee D O, 1992, `Ùrban warming? An analysis of recent trends in London's heat island'' Weather 47 50 ^ 56 Lee D O, 1993, ``Climate change and air quality in London'' Geography 78 77 ^ 79 Leung L R, Gustafson W I, 2005, ``Potential regional climate change and implications to US air quality'' Geophysical Research Letters 32 L16711, doi: 10.1029/2005GLO22911 MacDonald O, 2004 Coupling Glacier Mass Balance and Meltwater Yields in the European Alps with Future Climate Change: Downscaling from Integrations of the HadCM Model unpublished PhD thesis, Department of Geography, University of Salford, Salford McGregor G R, Bamzelis D, 1995, ``Synoptic typing and its applications to the investigation of weather air pollution relationships, Birmingham, UK'' Theoretical and Applied Meteorology 51 223 ^ 236 McGregor G R, Pelling M, Wolf T, 2006, `À report to the Environment Agency on the social impacts of heat waves'', Department of Geography, King's College London Meehl G A, Tebaldi C, 2004, ``More intense, more frequent, and longer lasting heatwaves in the 21st century'' Science 305 994 ^ 997 Murphy J M, Sexton D M H, Barnett D N, Jones G S,Webb M J, Collins M, Stainforth D A, 2004, ``Quantification of modelling uncertainties in a large ensemble of climate change simulations'' Nature 430 768 ^ 772 New M, Dessai S, Lopez A, Wilby R L, 2007, ``Challenges in using probabilistic climate change information for impacts and adaptation decision-making: an example from the water sector'' Philosophical Transactions of the Royal Society forthcoming O'Hare G P, Wilby R L, 1995, `Òzone pollution in the United Kingdom: an analysis using Lamb circulation types'' Geographical Journal 161 1 ^ 20 Oke T R, 1982, ``The energetic basis of the urban heat island'' Quarterly Journal of the Royal Meteorological Society 108 1 ^ 24 Parker D E, 2004, ``Large-scale warming is not urban'' Nature 432 290 Patz J A, Campbell-Lendrum D, Holloway T, Foley J A, 2005, `Ìmpact of regional climate change on human health'' Nature 438 310 ^ 317


919

Pryor S C, Schoof J T, Barhelmie R J, 2005, ``Climate change impacts on wind speeds and wind energy density in northern Europe: empirical downscaling of multiple AOGCMs'' Climate Research 29 183 ^ 198 Reynard N, Crooks S,Wilby R L, Kay A, 2004, ``Climate change and flood frequency in the UK'', in Proceedings of the 39th Defra Flood and Coastal Management Conference Flood Management Division, Defra, London; available from the author Rooney C, McMichael A J, Kovats R S, Coleman M, 1998,`Èxcess mortality in England and Wales, and in Greater London, during the 1995 heatwave'' Journal of Epidemiology and Community Health 52 482 ^ 486 Rosenzweig C, Solecki W D, Parshall L, Chopping M, Pope G, Goldberg R, 2005, ``Characterizing the urban heat island in current and future climates in New Jersey'' Environmental Hazards 6 51 ^ 62 Solecki W D, Oliveri C, 2004, ``Downscaling climate change scenarios in an urban land use change model'' Journal of Environmental Management 72 105 ^ 115 Solecki W D, Rosenzweig C, Parshall L, Popec G, Clark M, Cox J, Wiencke M, 2005, ``Mitigation of the heat island effect in urban New Jersey'' Environmental Hazards 6 39 ^ 49 Stedman J R, 2004, ``The predicted number of air pollution related deaths in the UK during the August 2003 heatwave'' Atmospheric Environment 38 1083 ^ 1085 Stevenson D S, Dentener F J, Schultz M G, Ellingsen K, van NoijeT P C,Wild O, Zeng G, Amann M, Atherton C S, Bell N, Bergmann D J, Bey I, Butler T, Cofala J, Collins W J, Derwent R G, Doherty R M, Drevet J, Eskes H J, Fiore A M, Gauss M, Hauglustaine D A, Horowitz L W, Isaksen I S A, Krol M C, Lamarque J-F, Lawrence M G, Montanaro V, Mu«ller J-F, Pitari G, Prather M J, Pyle J A, Rast S, Rodriguez J M, Sanderson M G, Savage N H, Shindell D T, Strahan S E, Sudo K, Szopa S, 2006, ``Multimodel ensemble simulations of present-day and near-future tropospheric ozone'' Journal of Geophysical Research 111 D08301, doi:10.1029/ 2005JD006338 Stewart H E, Hewitt C N, Bunce R G H, Steinbrecher R, Smiatek G, Schoenemeyer T, 2003,`À highly spatially and temporally resolved inventory for biogenic isoprene and monoterpene emissions: model description and application to Great Britain'' Journal of Geophysical Research 108 D20, 4644, doi: 10.1029/2002JD002694 Wetterhall F, Ba¨rdossy A, Chen D, Halldin S, Xu C, 2007, ``Daily precipitation-downscaling techniques in three Chinese regions'' Water Resources Research 42 W11423, doi:10.1029/ 2005WR004573 Whitehead P G,Wilby R L, Butterfield D,Wade A J, 2006,`Ìmpacts of climate change on nitrogen in a lowland chalk stream: an appraisal of adaptation strategies'' Science of the Total Environment 365 260 ^ 273 Wilby R L, 2003a, ``Past and projected trends in London's urban heat island'' Weather 58 251 ^ 260 Wilby R L, 2003b, ``Weekly warming'' Weather 58 446 ^ 447 Wilby R L, 2005, ``Constructing wet season precipitation scenarios for a site in the Anti Atlas Mountains, Morocco'', presented at the Conference on Optimising Land and Water Resources in Arid Environments, Agadir, Morocco; available from the author Wilby R L, 2007, `À review of climate change impacts on the built environment'' Built Environment 33 31 ^ 45 Wilby R L, Harris I, 2006, `À framework for assessing uncertainties in climate change impacts: low flow scenarios for the River Thames, UK'' Water Resources Research 42 W02419, doi: 10.1029/2005WR004065 Wilby R L, Perry G L W, 2006, ``Climate change, biodiversity and the urban environment: a critical review based on London, UK'' Progress in Physical Geography 30 73 ^ 98 Wilby R L, Dawson C W, Barrow E M, 2002, ``SDSMöa decision support tool for the assessment of regional climate change impacts'' Environmental and Modelling Software 17 145 ^ 157 Wilby R L, Charles S, Mearns L O, Whetton P, Zorito E, Timbal B, 2004, ``Guidelines for use of climate scenarios developed from statistical downscaling methods'', IPCC Task Group on Data and Scenario Support for Impacts and Climate Analysis (TGCIA), http://ipcc-ddc.cru.uea.ac.uk/ guidelines/StatDown Guide.pdf Wise E K, Comrie A C, 2005, ``Meteorologically adjusted urban air quality trends in the southwestern United States'' Atmospheric Environment 39 2969 ^ 2980

ß 2008 Pion and its Licensors

Conditions of use. This article may be downloaded from the E&P website for personal research by members of subscribing organisations. This PDF may not be placed on any website (or other online distribution system) without permission of the publisher.