Investigating the climatic impact of urban planning strategies through the use of regional climate modelling: a case study for Melbourne, Australia

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 28: (2008) Published online 19 March 2008 in Wiley InterScience ( Investigating the climatic impact of urban planning strategies through the use of regional climate modelling: a case study for Melbourne, Australia Andrew M. Coutts,* Jason Beringer and Nigel J. Tapper School of Geography and Environmental Science, Monash University, Melbourne, Vic, 3800, Australia ABSTRACT: Urban planning is a useful method for improving local climate and human health in cities through purposefully modifying urban land surface characteristics. This can reduce the potential risks of elevated city temperatures due to the urban heat island (UHI). Unfortunately, simple tools are not readily available for urban planners to assess the climatological impacts of various urban development scenarios. Urban modelling could be developed into such a tool to achieve this. This study attempts to design and evaluate a suitable tool for application in Melbourne, Australia. The Air Pollution Model (TAPM) was chosen to assess the impact of a long-term urban planning strategy on local climate and the above canopy UHI in Melbourne. Improvements were made to TAPM by increasing the number of urban land-use classes in the model and creating a higher resolution land cover database focused on housing density. This modified version of TAPM showed a good performance in replicating the surface energy balance compared with an observational flux tower site in suburban Melbourne during summer. TAPM simulated a mean maximum UHI intensity of 3 4 C at2a.m.in January. A future UHI scenario was then examined (year 2030) using an urban land cover database derived from plans in the Melbourne 2030 urban planning strategy. Results highlighted specific areas where planning intervention would be particularly useful to improve local climates, namely activity centres and growth areas. The appropriateness of the use of TAPM and climate models as a tool in urban planning is also discussed. Copyright 2008 Royal Meteorological Society KEY WORDS urban planning; urban climate; climate modelling; Melbourne; surface energy balance Received 18 August 2006; Revised 1 December 2007; Accepted 10 December Introduction Unplanned and rapid urbanization in cities can often lead to negative environmental impacts, including modifications to the local urban climate. The urban heat island (UHI) phenomenon is often evident in cities whereby urban areas are warmer than surrounding rural areas. UHIs may contribute towards elevated temperatures, which can be harmful for vulnerable urban residents, particularly during summer andheatwaveepisodes (Rankin, 1959). Higher incidences of heat-related illnesses including heart disease and even mortality have been associated with elevated temperatures within urban areas. Those particularly at risk include the elderly, lowincome earners, and residents in high density, older housing stock with limited surrounding vegetation (Smoyer- Tomic et al., 2003). Fortunately, there is sufficient evidence to suggest that urban planning can be a useful method for improving local climate and human health (Jackson, 2003; Stone, 2005). In order to reduce negative climatological impacts, those involved in urban development and design must begin to incorporate climate knowledge into planning strategies. * Correspondence to: Andrew M. Coutts, School of Geography and Environmental Science, Monash University, Wellington Road, Clayton, Victoria, 3800, Australia. amcou1@student.monash.edu.au UHIs form primarily because of high thermal heat capacity and heat storage of urban surfaces, added sources of heat from anthropogenic activities, and reduced evapotranspiration (Oke, 1988). Within the urban canopy (below maximum building height), urban geometry is also important in controlling radiative exchanges between the walls and floor of urban canyons. Small sky view factors (SVF) and large height to width ratios trap radiative energy during the day and limit nocturnal cooling. This leads to the development of peak UHI intensities during the night, as rural areas are allowed to cool uninhibited. Cloud amount and wind speed are important meteorological parameters as they affect longwave cooling and ventilation, which serve as surrogate variables describing the relative roles of radiative and turbulent exchanges in and around the urban region (Morris and Simmonds, 2000). While the UHI phenomenon has been well documented in the climatological literature over the past few decades, few cities have developed comprehensive strategies to mitigate its intensity. Reasons for little consideration of climate related understanding in urban planning include a lack of knowledge, economic constraints, and communication problems (Eliasson, 2000). Added to these reasons, planning tools are not often available for planning authorities to assess the implications of projected Copyright 2008 Royal Meteorological Society

2 1944 A. M. COUTTS ET AL. land-use change on local climate (Fehrenbach et al., 2001). According to Eliasson (2000), the development of such tools based on scientific research that can be incorporated into the urban planning framework should be a challenge and focus for urban climatologists. However, one such tool that can address the issue of climate impacts of urban planning strategies, if adequately developed, is climate modelling, both local and regional. Climate modelling that uses specific treatments at the urban surface can significantly help in determining the likely impacts of large scale urbanization on local climate and UHI development, improving weather forecasts, estimating energy consumption, and aiding in urban planning (Kusaka and Kimura, 2004). Information at all urban scales (global, regional, local, and microscale) can be highly beneficial for planners, but knowledge about climate at the city (regional) and neighbourhood (local) scale is specifically relevant as planning authorities influence/regulate features at this scale, such as heights of buildings. For climate models to be a useful tool in aiding sustainable urban planning, it is important that they are correctly able to simulate the urban climate at this scale. The urban surface is highly complex and models require additional inputs, and new and improved parameterizations, to accurately simulate the urban climate (Zehnder, 2002). In particular, the high heat storage of urban landscapes associated with high thermal admittance and radiation trapping, as well as the added sources of anthropogenic heat, need to be incorporated. Tools like satellite imagery (such as MODIS) or databases of urban land-use and land classification (LULC), now provide finer spatial resolution of the high heterogeneity of urban characteristics (albedo, emissivity or heights of buildings) across cities as input databases for models (Dandou et al., 2005; Jin and Shepherd, 2005). While accuracy in modelling the urban climate is of prime importance, features such as ease of use and short running time should also be important factors, as urban planners require tools that incorporate such features. Recent work in regional scale modelling has seen the development of a number of urban models of varying degrees of complexity based on two types of parameterization schemes. The first type of scheme involves simple modifications to existing land surface schemes by modifying or fabricating the parameters of the land surface to broadly behave like an urban surface, such as increasing roughness lengths and decreasing albedo (Atkinson, 2003). One simple parameterization scheme developed by Grimmond and Oke (2002) is called the Localscale Urban Meteorological Parameterization Scheme (LUMPS). Using net all-wave radiation, simple information on surface cover and standard weather observations, turbulent and storage heat fluxes can be calculated through a series of linked equations. The equations include the Objective Hysteresis Model (OHM), which uses net all-wave radiation and the surface properties of the site to calculate heat storage (Grimmond et al., Taha (1999) used a bulk parameterization approach to better incorporate heat storage and more explicitly account for urban canopy layer fluxes, which also included the OHM. Similarly, Dandou et al. (2005) made modifications to the thermal part of the fifthgeneration Penn state/ncar Mesoscale Model (MM5) that incorporated the OHM. The model also included anthropogenic heating, while modifications were also made to the dynamical part of the model resulting in acceleration to the diffusion processes during unstable conditions. The second type of parameterization scheme involves the inclusion of a separate urban canopy scheme to the land surface model by incorporating parameters to represent canyon geometry and interactions between the walls, rooftops, and roads. A number of variations on this approach have been developed. Some characteristics of these included using the drag force approach to represent the dynamic and turbulent effects of buildings and vegetation while the thermal modifications of the surface involve a 3D urban canopy (Dupont et al., 2004; Martilli et al., 2002). This approach calculates the surface temperature of each surface type by taking into account the interactions of shadowing and radiation trapping effects. Single level urban canopy models have also been developed and incorporated into atmospheric models where the canopy model simulates turbulent fluxes into the atmosphere at the base of the atmospheric model, parameterizing both the surface and the roughness sub-layer (Kusaka and Kimura, 2004; Masson, 2000). The Town Energy Balance model (TEB) is one such scheme and has been shown to simulate the surface energy balance and climate well compared with observations (Lemonsu et al., 2004; Masson et al., 2002). A good summary of urban modelling approaches and developments can be found in Dandou et al. (2005). As a result of such modifications and developments, the ability of climate models to simulate the urban climate has improved, as has their appropriateness as a tool that may aid urban planning. For instance, Taha (1999) modelled effects of increased albedo for all the LULC types in Atlanta (increasing residential albedo from0.16to0.29etc.)andshowedadecreaseintheair temperature of about 0.5 C. Atkinson (2003) found that in London during the day, the albedo, anthropogenic heat, emissivity, SVF, thermal inertia and surface resistance to evaporation (SRE) all aided the formation of an UHI to varying amounts of between 0.2 and 0.8 C. SRE was the most important factor increasing the UHI intensity during the day, while the roughness length decreased intensity. At night, the roughness length, emissivity, SVF andsreaideduhiformationby C, but the largest effect (2 C) came from anthropogenic heating (Atkinson, 2003). This kind of information is highly valuable to urban planners in developing policies for reducing negative climatic impacts to protect vulnerable urban dwellers from the risk of exposure to elevated heat conditions. Given the growing knowledge and capacity of urban climate modelling, this study attempts to investigate the role of climate modelling as a tool for use in urban

3 URBAN PLANNING AND CLIMATE IN MELBOURNE 1945 planning and to design and evaluate a suitable tool for Melbourne, Australia. Through the use of a regional scale model, The Air Pollution Model (TAPM), the possible climatic impacts of a long-term urban planning strategy for Melbourne, were assessed. Future planning directions of the strategy aim at encouraging a more compact city by clustering and increasing the amount of housing in established urban areas. Continued urbanization following existing development patterns is likely to lead to an intensification of the UHI (Coutts et al., 2007b). Using a modified version of TAPM, we aimed to model the regional climate of Melbourne and its subsequent UHI. Modifications included an improved urban surface parameterization and an improved land cover input database. Results will highlight to urban planners that the UHI is an issue that needs to be addressed and identify specific areas/regions where planning intervention may be required. As well as assessing the impact of the urban planning strategy, we will comment on the use of regional scale modelling as a tool in urban planning. 2. Methods 2.1. The urban planning strategy for Melbourne, Australia The city of Melbourne, Australia is a rapidly growing city with an anticipated population increase of up to 1 million people by the year 2030, requiring the development of approximately new households (Department of Sustainability and Environment, 2002). In 2002, the Victorian Government introduced a planning strategy to accommodate this growth titled Melbourne The strategy seeks to achieve a more compact city through the development of activity centres (built up centres for business, shopping, working and leisure with forms of higher density housing) and the establishment of an urban growth boundary (Figure 1) (Department of Sustainability and Environment, 2002). The anticipated development of a more compact city, if not planned in an informed manner, could lead to an exacerbated UHI intensity. This will be compounded by increased hot weather and hazardous climatic conditions through global warming (IPCC, 2007), which will impact especially on vulnerable urban dwellers. Melbourne already shows an UHI signature, with a 20-year mean maximum UHI of 2.68 C found at 6 a.m. (Morris et al., 2001). During summer, anti-cyclonic events often bring warm and dry North to Northeast airflow over Melbourne, and can send temperatures in excess of 35 C during the day, while mean early morning (6 a.m.) UHI intensity during these conditions has been observed at 3.56 C (Morris and Simmonds, 2000). These are mean UHI intensities, suggesting that under optimal conditions (clear skies and low winds) UHI intensity can be much higher. An automobile transect across Melbourne in 1992 showed a peak UHI intensity as high as 7.1 C in the central business district (CBD) during the evening (9 p.m.) (Torok et al., 2001). Unplanned and hasty urban development could compromise the overall goal of the Melbourne 2030 strategy, which aims to achieve a liveable, attractive and prosperous city. Cities that are low density and reliant on private car transport and strong zoning that separates housing, employment and services are unsustainable. Rather, a sustainable city is often described in the urban design literature as compact, high density urban form and supported by a comprehensive transport network, which emphasizes connectivity and mixed use developments at critical nodes (intersecting transport routes) (Mills, 2005). However, this city model can encourage UHI development and compromise green-space, potentially threatening the environmental quality of the city (Pauleit et al., 2005). Melbourne 2030 aims for a sustainable city and the planning strategy provided a good opportunity to investigate the use of regional climate modelling in assessing urban climate modifications resulting from land-use and planning policies. Our approach consisted of two scenarios: (A) a simulation of the current urban climate and UHI intensity in Melbourne and (B) a year 2030 scenario of increased urbanization based on the Melbourne 2030 key directions to investigate likely future changes to urban climate The air pollution model (TAPM) and urban modifications Selecting an appropriate model as a tool for urban climate impact assessment and use by urban planners is likely to depend on a number of parameters. The accuracy of the model must be sufficient to robustly simulate the urban climate yet not overly complex, computationally expensive, and should be user-friendly. Dandou et al. (2005) suggested that despite the simplicity of their bulk urban parameterization scheme, improvements in results were comparable with that produced by the complex canopy scheme of Martilli et al. (2002). The ease of use is likely to be important, and inputs of surface characteristics into the model should be simply described and readily available, such as through easily obtainable data on types of surface cover, vegetation cover, albedo, mean building height, anthropogenic heating, and dwelling density. TAPM (Hurley, 2005) was selected for this study as benefits included the ability to conduct year-long simulations; the ability to run simulations without surface observational inputs; the ease of a PC-based interface for use in Windows operating systems; user-defined surface cover databases; and a range of methods for analysing outputs. Therefore, TAPM has the potential to be adopted as an urban planning tool. The meteorological component of TAPM is an incompressible, non-hydrostatic, primitive equation meteorological model with a terrain following vertical coordinate for 3D simulations and a 3D nestable, eularian grid (Hurley, 2005). As described in Hurley (2005), the prognostic meteorological component solves approximations to the fundamental fluid dynamics equations, and rather than requiring site specific observations, flows such as sea breezes and terrain flows are predicted against

4 1946 A. M. COUTTS ET AL. Figure 1. The Melbourne 2030 compact city with the location of activity centres and the urban growth boundary (Department of Sustainability and Environment, 2003) ( State of Victoria, Department of Sustainability and Environment, 2003). This figure is available in colour online at a background of larger scale meteorology provided by global synoptic analysis. The vertical fluxes are represented by a gradient diffusion approach, including a counter-gradient term for heat flux from turbulence terms determined by solving equations for turbulent kinetic energy and eddy dissipation rate. TAPM includes a soilvegetation-atmosphere transfer (SVAT) scheme, which is used at the model surface, and a radiative flux parameterization at both the model surface and at upper levels in the atmosphere. For urban land surfaces in TAPM, temperature and specific humidity are calculated depending on the fraction of urban surface cover following similar approaches for non-urban surfaces, except that the surface properties (albedo, thermal conductivity) are given appropriate urban values. The anthropogenic heat flux is also included in the surface flux equations (Hurley, 2005). A number of validation studies and evaluations have been conducted on TAPM, including one in Melbourne (Hurley et al., 2003). For the period July 1997 to June 1998, model verification was completed using eight monitoring sites across Melbourne. Results showed that the 10- m winds and screen level temperature were predicted very well with a low average Root Mean Square Error (RMSE) and a high index of agreement (IOA) (Hurley et al., 2003). However, TAPM only incorporated a

5 URBAN PLANNING AND CLIMATE IN MELBOURNE 1947 single homogeneous urban surface class for the entire urban region with single values for land surface parameters such as albedo and thermal conductivity derived from the literature. The surface energy balance is simulated in TAPM and considered fundamental to an understanding of boundary layer climates and is basic to an understanding of such features as thermodynamic behaviour of air and surface temperature and humidity, and the dynamics of local airflow (Oke, 1988). Hence, models need to be able to replicate the partitioning of available energy adequately in order to robustly simulate the resultant climate. The urban surface energy balance is given by (Oke, 1988): Q + Q F = Q H + Q E + Q S where Q is net radiation, Q F is anthropogenic heating, Q H is sensible heat flux, Q E is the latent heat flux and Q S is the storage heat flux. TAPM was modified to improve simulations of urban environments by incorporating four urban land surface types (low, medium and high density, and CBD) replacing the existing single urban surface. Surface parameters in TAPM for the medium density surface type were specified using information from an observational site located in suburban Melbourne (Coutts et al., 2007b). The site was located in Preston, north of Melbourne ( , ) in a sprawling, moderately developed housing area consisting largely of detached dwellings, typical of the Melbourne urban landscape (Figure 2). Site characterization showed a plan impervious surface cover of 62%, which included a plan building area of 45% and had a mean height to width ratio of Surface characteristics observed at the site included fraction of urban and vegetation cover (determined from aerial photography); mean building height; anthropogenic heat flux (determined using population, energy, and transport databases); roughness length; and albedo (Table I) and were available as model input parameters. Surface energy balance measurements from the Preston site were used to evaluate the performance of the model for simulated medium density housing in Preston. Observations were taken from instruments mounted on a tall tower at a height of 40 m using the eddy correlation technique (Baldocchi et al., 1988) to measure local scale fluxes ( m) of sensible and latent Figure 2. The medium density observational study site in Preston, located north of Melbourne CBD. This figure is available in colour online at Table I. The original urban surface characteristics from Preston are given along with the values assigned in the model for each level of urban density: fraction of urban cover σ u ;albedoα u ; anthropogenic heat flux A u (W.m 2 ); thermal conductivity k u (W.m 1.K 1 ); roughness length z 0u (m); building height z H (m); fraction of non-urban area covered by vegetation σ f ; leaf area index LAI; minimum stomatal resistance rsi (s 1.m 1 ). σ u α u A u k u z ou z H σ f LAI rsi Observed (Preston) TAPM (3.0.2) default Urban (low) Urban (medium) Urban (high) Urban (CBD)

6 1948 A. M. COUTTS ET AL. heat and momentum. Radiation sensors measured each component of the radiation balance, giving net radiation. The storage heat flux was calculated as a residual of the energy balance equation. Observations of temperature, vapour pressure, wind speed, and friction velocity were also conducted. Model outputs matching the location of the Preston site were compared with the observed surface energy balance and meteorological parameters and provided information to evaluate how well TAPM replicated the surface energy balance and simulated the local urban climate at the observational site. This site was one of three urban flux sites operating at various times and locations in Melbourne during (Coutts et al., 2007b). The urban land surface characteristics for Urban (medium) (Table I) were set to be very similar to the medium density observational site described (Preston). The low, high, and CBD urban land surface characteristics were then assigned with reference to the values of Urban (medium) (Table I) using expert knowledge of the Melbourne urban landscape from the observational campaign (Coutts et al., 2007b) and other literature values. However, in order to replicate the storage capacity of a complex 3D urban surface in a one-dimensional surface scheme the thermal conductivity was substantially increased in the model. Using the value of a component material such as concrete in bulk model parameterizations does not capture the full influence of the heterogeneous urban landscape or the effects of the urban canopy. Sugawara et al. (2001) created a thermal property parameter (combining the product of specific heat and thermal conductivity) that better represented urban surfaces. This parameter was determined to be much larger than a homogenous surface type such as asphalt and concluded that the value should be a few times larger than the component material in bulk urban models that do not deal with canyon shape explicitly (Sugawara et al., 2001). In TAPM, the thermal conductivity value for the land surface was modified in order to match the storage heat flux results from TAPM with the observational results at the medium density site in Preston during January. The thermal conductivity needed to be increased well above realistic values before the surface began behaving similarly to a real urban surface, identifying the importance of canyon geometry in trapping and storing energy Model configuration and database development Model scenarios for the current and year 2030 scenarios were performed for January during the Austral summer, as urban residents are exposed to higher ambient temperatures at this time. Large scale synoptic analyses were used to force the model between the periods 1997 and Synoptic scale forcing meteorology was provided from the 6-hourly Limited Area Prediction Systems (LAPS) (Puri et al., 1998) at a 0.75 grid spacing. These 8 years of January simulations were conducted so that modelled differences were due to land surface changes and not due to year-to-year climate variability. Moreover, the same forcing data were used for each experiment. The modified TAPM version was configured with three nested grids of horizontal points with the inner grid encompassing the Melbourne metropolitan area at a grid spacing of 1000 m centred at E and S. The middle and outer grid spacings were 3000 m and m, respectively, and 25 vertical grid levels were selected with the highest level at 8000 m. Other databases of terrain height (9-s DEM), sea surface temperature and soil classification data, were also used in the scenarios (Hurley, 2005). In order to run the scenarios described, relevant land surface databases were developed at a suitable resolution for input into TAPM. For the current day scenario (Scenario A) a vegetation (land-use) database was obtained that provided recent vegetation cover (1988) (Geoscience Australia, 2003). In addition, a surface database of low, medium, and high density areas, as well as the CBD, was constructed. Information on census districts for the entire Melbourne metropolitan area were collected (Australian Bureau of Statistics (ABS), 2001) and the dwelling density calculated for each district (dwellings per km 2 ). This information was converted to mean dwelling density for 0.01 decimal degree grid cells (approximately 1 km). Plans for Melbourne 2030 aim to increase the average housing density significantly from 1000 dwellings per km 2 to an average of 1500 dwellings per km 2 (Department of Sustainability and Environment, 2002). Therefore, high density areas were deemed to be greater than 15 dwellings per hectare, medium density areas between 10 and 15 dwellings per hectare and low density areas less than 10 dwellings per hectare (though greater than 1) (Figure 3). This database was overlain on the vegetation database and used as input into TAPM. The database for the Melbourne 2030 scenario (Scenario B) was based on the document s key directions as discussed earlier. Taking the current urban density database, the urban growth boundary was added and those areas not currently developed within the urban growth boundary were assigned to the low density class. The location of the proposed 26 Principal, 82 Major, and 10 Specialized activity centres were then added, by assuming that the surrounding housing for a 1-km radius would be high density (within walking distance). Housing within another 1-km radius was anticipated to increase to at least medium density while existing high density housing areas and the CBD areas remained as such (Figure 3). 3. Results and Discussion 3.1. Evaluating TAPM against urban surface energy balance observations Using the new land surface database of the current Melbourne urban landscape, TAPM was run for the month of January and compared with the observations at the medium density observational site (Preston) (Figure 4). TAPM showed a good performance in replicating the diurnal course and monthly mean surface energy balance

7 URBAN PLANNING AND CLIMATE IN MELBOURNE 1949 Figure 3. Land surface database of current density scenario A (left) and the Melbourne 2030 scenario B with the urban growth boundary (right). This figure is available in colour online at Figure 4. Comparison of the observed (dashed line) and modelled (solid line) diurnal surface energy balance (observed height 40 m and model level 50 m) for location ( , ) and corresponding grid point (043, 084) for the month of January Regression (fit) equations were Q (y = 1.032x ); Q H (y = 1.137x ); Q E (y = 0.789x ); Q S (y = 0.833x ). This figure is available in colour online at for the month of January The evaporative flux was replicated well by the model, only showing an overestimation in the afternoon. The storage heat flux was also well replicated, although some discrepancy was evident in both Q and Q H. This is caused by an overestimation of incoming solar radiation due to the inability of the model to capture cloudy skies and poor air quality (which can reflect and scatter incoming short wave radiation), so monthly averages of Q were overestimated. This extra energy led to additional partitioning into Q H.On the majority of the January days, the TAPM model performed well. The model also captured important features of urban energy balance partitioning (Figure 4). These included the hysteresis pattern in Q S, showing a peak

8 1950 A. M. COUTTS ET AL. approximately 1 2 h before the peak in Q. The peak was not as evident in the observations during this month as what was generally seen during the observational campaign (Coutts et al., 2007b). The asymmetry in Q H was also evident with the peak occurring later in the afternoon. Importantly, both Q H and Q E remained positive into the evening, supported by heat storage release from the urban fabric, and remained slightly positive throughout the night. Despite the discrepancy in Q and Q H, the model accurately simulated temperature, relative humidity, and wind speed (Figure 5). Slight discrepancies were seen in the diurnal temperature plot, with a reduced lag in temperature approaching its maximum and the nighttime temperatures were underestimated, due to underestimation of nighttime heat storage release. Table II gives the January 2004 monthly comparison of modelled meteorological variables and their associated error for the model grid point corresponding with the measurement tower location, compiled following Willmott (1981). Statistical comparisons are also given for the surface energy balance components. Changes in urban surface characteristics influence how net radiation is partitioned into each of the surface energy balance components, so the flux ratios and how they vary between density classes were of particular interest (Figure 6). Also, while the summer month (January) was of primary interest, there was also observational data available for a full year (Coutts et al., 2007b) and it was possible to see how well the model reproduced the partitioning of the urban surface energy balance seasonally (Figure 6). Therefore, TAPM was also run from August 2003 to July 2004 corresponding with the year-long observational study. Naturally, partitioning in January was good as the model parameters for the urban surface characteristics were adjusted to match this data, yet over the course of the year, the model did not capture Q S and Q H well. A reasonable replication Figure 5. Comparison of observed (left) and modelled (right) temperature, relative humidity, and wind speed (observed height 40 m, model level 50 m) for location ( , ) and corresponding grid point (043, 084) for the month of January Table II. Statistical comparison between variables for the observational location ( , ) and corresponding model grid point (043, 084) for the month of January 2004 of temperature T ( C); wind speed WS (m/s); specific humidity q (g/kg); friction velocity u (m/s); sensible heat flux Q H (W.m 2 ); latent heat flux Q E (W.m 2 ); and storage heat flux Q S (W.m 2 ). T WS q u Q Q H Q E Q S n O P s O s P CORR RMSE RMSE S RMSE U MAE d r n, number of observations; O, observations; P, predicted values; s O s P, observed and predicted standard deviations; CORR, Pearman Correlation; RMSE, Root Mean Square Error; RMSE S, Systematic RMSE; RMSE U, Unsystematic RMSE; d, Index of Agreement; r 2, Coefficient of determination (Willmott, 1981).

9 URBAN PLANNING AND CLIMATE IN MELBOURNE 1951 Figure 6. Mean monthly plots of daytime fractions of Q for each energy balance component and the Bowen Ratio. CBD, HIGH, MEDIUM, and LOW correspond to each of the urban classes and OBSERVATIONS correspond to the measured data from Preston. The Bowen Ratio (Q H /Q E ) is not shown for the CBD as it was significantly higher ( 20). of observations for the evaporative fraction (Q E /Q ) was seen over the course of the year. Figure 6 also demonstrates the differences in partitioning of energy between each of the urban land surface classes in the model and shows that the influence of changing the land surface values alters energy balance partitioning. The modelled data for these urban classes were not verified against any observations. Some differences in energy partitioning over the course of the year could result from a number of uncertainties. Actual deep soil volumetric moisture contents were not available to initialize the model and we found that there was a mismatch in the seasonal course of Q E /Q between the observations and the model output (Figure 6). In the model, moisture contents were the lowest over the Austral summer months (December February). Rainfall in Melbourne during February and March 2004, however, was well below average, so it was likely that deep soil volumetric moisture contents were also below average at this time, leading to the reduced energy partitioning into Q E. More accurate values of monthly soil moisture content could improve this result. As expected, Q E /Q decreased with increasing urban density as the vegetated surface cover was replaced with greater impervious surface cover, restricting evapotranspiration. Generally, the partitioning of energy into Q E was acceptable over the course of the year and responded well to the changes in surface cover. The modelled Urban (medium) heat storage fraction ( Q S /Q ) during the summer period generally showed a slight underestimation compared with the observations, but were much improved compared with earlier versions of TAPM. The substantial increases of the values for thermal conductivity in the model were large enough to capture the significant energy storage by the 3D urban landscape. Comparing each of the densities, the amount of heat storage increased with increasing density. However, absorption of energy by the urban surface in the

10 1952 A. M. COUTTS ET AL. model appeared to saturate (reach a maximum absorptive capacity) when increasing thermal conductivity. Therefore, despite the increasing thermal conductivity with urban density, the 1D land surface in the model did not capture the full influence of the 3D canyon morphology on heat storage. During winter, the land surface scheme did not replicate the high heat absorption ( Q S /Q ) by the urban fabric that was seen in the observations (Figure 6). In most Northern Hemisphere energy balance studies, a decrease in heat storage is seen during winter, following the reduction in Q,aswellasaddedsurfacemoisture for increased Q E /Q (Grimmond, 1992), a pattern that TAPM did replicate. However, in this case, it may not be that TAPM inaccurately represented urban heat storage, but rather the uncertainty may lie in the observations. Spronken-Smith et al. (2006) found in Christchurch, New Zealand that under settled anti-cyclonic conditions, a strong inversion can occur that can severely restrict turbulent mixing and influence the above canopy flux measurements. As the observations in Melbourne calculated Q S as a residual of the eddy correlation technique, it could be that the observational results over emphasize the importance of heat storage during stable wintertime conditions and is an area that requires further study. On account of the slightly underestimated Q S /Q, the sensible heating fraction (Q H /Q ) during the summertime for the Urban (medium) density was also slightly higher than observed. The partitioning of energy was very similar for each urban density during summer, though all sites were slightly higher than the observations. This is often why above canopy temperatures are similar across an urban area during the day, as higher density sites absorb more energy and Q S /Q increases, restricting the availability for atmospheric heating, which sometimes aids Urban Cool Island (UCI) development in combination with shading (Morris and Simmonds, 2000). The Bowen ratio (Q H /Q E ) throughout the year was well replicated compared with Urban (medium) and increased with higher urban density (Figure 6). The Bowen ratio from the model results also preceded the observations again as a result of the lack of input data for the soil moisture content and the influence of this on Q E. The model was not able to accurately replicate the partitioning of energy outside of the summer months. However, as TAPM was replicating the partitioning of energy and meteorological parameters at the surface reasonably well in January, it can be used for the scenarios described with a good degree of confidence. While a crude method of parameterizing the model to behave more like an urban surface was used, and direct validation was not completed on the energy balance partitioning, the model has vastly improved on the performance of TAPM version 2.0 before the modifications were made (data not shown). Additionally, the model was only evaluated for the medium density urban class, so there may be limitations in the model s applicability to other urban density classes. The lack of an urban canopy scheme could also limit the model s capacity to accurately replicate urban heat storage across density classes. There is obvious scope for a specific urban parameterization in TAPM Modelling UHI intensity and the impact of Melbourne 2030 TAPM was configured as described in Section 2.3 and run for eight Januaries from 1997 to 2004 to provide an ensemble average for current summertime conditions and then again for the 2030 planning scenario. The current scenario (A) showed a mean nighttime (0200) UHI of approximately 3 4 C in the CBD, reducing as distance from the CBD increased (Figure 7(a)). Variability was high with anomalous warmer and cooler areas seen across the metropolitan area corresponding with urban density class. The modelled UHI intensity was similar in range to those previously observed in Melbourne (Morris and Simmonds, 2000; Morris et al., 2001). During the day (1400) the current Scenario (A) screen level UHI was not as intense as at 0200, but still showed an UHI intensity of 1 2 C, with temperatures being more uniform across the region (Figure 7(b)). The CBD was not warmer than the surrounding urban area. Temperatures away from the coast to the north and east of Port Phillip Bay showed higher values as a result of mesoscale airflows and a regional sea breeze. During the night, the lower wind speeds reduced the influence of the regional flows and the urban density more strongly controlled the development of the UHI. The modelled UHI also varied with synoptic conditions that supported maximum UHI development under conditions of anti-cyclonic highs centred just east of Melbourne, low wind speeds and cloudless skies (Morris and Simmonds, 2000). The Melbourne 2030 scenario (B) revealed a slightly modified UHI pattern from the current scenario (A) (Figure 7(c) and (d)). While the maximum intensity of the UHI did not increase, the areal extent of elevated temperatures expanded. The nighttime UHI reduced in its spatial variability, becoming more uniform across the urban area similar to that of the daytime UHI. Analysing the difference in screen level temperature between the current and Melbourne 2030 scenarios allows specific areas of significant warming to be identified and is what urban planners are most interested in. The extent of change in the UHI resulting from planning strategies shows areas that are particularly vulnerable. This information can be used for improved planning decisions. The greatest temperature increases during nighttime maximum UHI intensity (Figure 8) were seen in areas where development replaced pasture land and in new activity centres. Temperatures in other areas of Melbourne also appeared to respond significantly to increases in housing density especially along the edge of the current urban-rural boundary. Some of these areas are located along transport links and growth areas designated for concentrated expansion as outlined in Melbourne While these areas are likely to show the greatest increase in temperatures in 2030, temperatures were only seen to

11 URBAN PLANNING AND CLIMATE IN MELBOURNE 1953 Figure 7. Spatial variability in mean screen level temperatures for the Melbourne area at 02 : 00 (2 a.m.) and 14 : 00 (2 p.m.) h for each scenario: A Current development at 02 : 00 (a) and 14 : 00 (b); B Melbourne 2030 planned development at 02 : 00 (c) and 14 : 00 (d). This figure is available in colour online at increase to levels currently seen within the CBD. Initiatives that can help reduce temperature increases can be more easily incorporated into newly developing regions, rather than in existing urban development. Therefore, these growth areas and new or minimally developed existing activity centres could provide excellent opportunities for UHI mitigation strategies to be put in place. During the day, some portions of Melbourne to the west and north showed elevated temperatures following the planned development (Figure 9). Interestingly, during the day a large fraction of the urban area, mainly where development increases from low to higher densities, actually showed a very slight decrease (largely insignificant) in temperature due to the increased heat storage limiting the amount of energy available for atmospheric heating and reducing temperatures. The areas of greatest temperature increase were the planned growth areas where development will replace existing natural landscapes. While it may seem that these mean temperatures are not high, during periods in summer of extreme heat, temperatures can be much higher. While higher nighttime temperatures from restricted nocturnal cooling in urban areas may not seem like a significant problem, extended periods of warmer temperatures can limit nighttime recovery from daily heat stress. Inland activity centres that do not feel the effects of the cooling sea breeze would especially benefit from UHI mitigation measures. The planned increase in urban density through the establishment of an urban growth boundary and the development of activity centres in Melbourne will likely lead to a more intense UHI during the night, while during the day this is less significant. Coutts et al. (2007b) in their observational study in Melbourne found that during the summer across three urban sites of varying urban density, all sites showed a mean daytime Bowen ratio of over 2 and the daily Bowen ratio was sometimes

12 1954 A. M. COUTTS ET AL. Figure 8. Change in mean nighttime (02 : 00) screen level temperature change from the current urban development, to that proposed by the Melbourne 2030 planning strategy. Areas within the contours are statistically significant at the 95% confidence level. This figure is available in colour online at Figure 9. Change in mean daytime (14 : 00) screen level temperature from the current urban development, to that proposed by the Melbourne 2030 planning strategy. Areas within the contours are statistically significant at the 95% confidence level. This figure is available in colour online at

13 URBAN PLANNING AND CLIMATE IN MELBOURNE 1955 in excess of 5. The increase in Bowen ratios with increasing urban density modelled in this study were not found in the observational results as evaporative fluxes were very similar across all housing densities despite varying vegetation cover. This was a result of poor moisture availability in response to water restrictions when observations were conducted (Coutts et al., 2007b). Therefore during the summer, the entire Melbourne region experienced warm, dry and hence unfavourable climatic conditions. Adoption of the Melbourne 2030 strategy is not likely to increase Bowen ratios across the city significantly, but there will be an extension of warm and dry conditions over longer periods of the day as well as an extension of the seasonal exposure to unfavourable conditions along with an increased spatial extent, especially if water restrictions remain tight. In this study, the effect of heat trapping and storage in the urban environment was replicated in the simulations by significantly increasing the thermal conductivity. The 3D nature and complexity of the urban landscape was not explicitly included in the model, unlike new urban canopy schemes. As a result, the model could not deliver within-canopy temperatures, which could possibly be greater than the modelled temperatures in this study. Modelled screen level temperatures were also slightly underestimated during the evening and night due to an underestimation of the slow release of heat stored in the urban fabric due to complex canyon morphology (including walls). Finally, the Melbourne 2030 scenario (B) only accounts for climatic impacts from land cover change. Mean global temperatures over the last 100 years ( ; 100-year linear trend) have increased by 0.74 C (±0.18 C) largely as a result of carbon dioxide (CO 2 ) emissions (IPCC, 2007), so projected global temperature rises (0.2 C per decade for the next two decades (IPCC, 2007)) coupled with heating from further urban development will lead to further increases in urban temperatures. Also, the frequency of extreme warm days and nights has increased since 1961 (Plummer et al., 1999). Urban areas themselves are a significant source of CO 2 mostly from vehicles emissions with local annual emissions from urban Melbourne as high as 84.9 t CO 2 ha 1 y 1 (Coutts et al., 2007a). Urban planning measures such as energy-efficient buildings and increased public transport use would help contribute to combating greenhouse gas emissions. 4. Conclusions Simulations of the changes in climate resulting from the proposed land cover changes identified in the directions of the Melbourne 2030 plan showed that continued increases in density would result in an increased intensity of the nighttime Melbourne UHI. Growth areas and particular activity centres were predicted to have the greatest temperature increases. During the day, the impact of changes in urban development was not seen to increase the peak daytime temperature due to increased storage limiting the amount of sensible heating of the atmosphere. Yet, existing urban climates during summer days can already be unfavourable with high Bowen ratios regularly observed across varying densities of the city (Coutts et al., 2007b). These results demonstrate the utility of regional scale climate modelling as a tool for climate impact assessment and show the ability to determine likely climate modifications from simple land-use changes based on planning directions. The use of TAPM for the Melbourne urban landscape was adequate for January, and identified that continued urban development in Melbourne could lead to higher diurnal exposure to warmer temperatures. Modelling results such as these are an excellent way to present and convey information and issues to environmental planners. Planning in urban areas to ameliorate, and limit the development of degraded local climates has been known for decades (Aron, 1984; Oke, 1984; Oke, 2005), yet policy development in this area is still lacking despite calls for improvements. The concept of sustainable settlements is recognized within Melbourne 2030 with initiatives such as those under the direction of A greener city, including reducing the impacts of storm-water on bays and catchments, and the management of water resources (Department of Sustainability and Environment, 2002). Melbourne 2030 currently notes concern for issues such as global warming and a livable city but an assessment of the impact of a more compact city on climate had not been undertaken. Our analysis should persuade the development of new policies for UHI mitigation by planners. This work may be opportune since the Melbourne 2030 plan is due for review in Some initiatives already exist that aid in reducing UHI intensity include energyefficient buildings and encouraging a shift in travel from private vehicles to public transport, which will reduce anthropogenic heat emissions. While this is good, a comprehensive UHI mitigation strategy for Melbourne is required and it is hoped that this study will prompt the Melbourne 2030 planning group to act and encourage the implementation of UHI mitigation initiatives. It would be a great opportunity for the Victorian Government, who wish to lead by example in environmental management (Department of Sustainability and Environment, 2002). Regional scale modelling of urban climate is a powerful tool and the use of TAPM as a model for use in urban planning has both benefits and shortfalls. As TAPM is now set up for Melbourne, further summertime scenarios could be conducted to investigate the potential of mitigation strategies such as alterations in surface albedo or the effect of increasing vegetation cover. Also, any type of urban spatial configuration at the neighbourhood scale could also be easily modelled. This study has demonstrated the potential for TAPM to become a rigorous model for use in urban planning. However, much improvement is still required before it could be commonly used. Operating the model for other Australian or international cities may not be feasible without some modification of surface parameters (requiring local field observations) or development of new parameterizations.