Assessing the impact of changes in surface cover, human behaviour and climate on energy partitioning across Greater London

Climate-sensitive urban design is an increasingly important consideration for city planners and policy makers. This study demonstrates the use of a biophysical model to assess the response of urban climate to various changes, including population growth, reduced energy use, urban development and urban greening initiatives. Model inputs are intentionally derived using only publicly available information and assumptions involved in collating the data are discussed. Results are summarised in terms of the energy partitioning which captures changes in meteorology, surface characteristics and human behaviour. The model has been recently evaluated for the region, and those ﬁ ndings are drawn upon here to discuss the model ’ s capabilities and limitations. Model simulations demonstrate how both intentional and inadvertent changes to the urban landscape can alter the urban climate. For example, the impact of population growth depends on where, and how, people are housed, and recent changes in garden composition have reduced evaporation. This study has been designed so that model output could be combined with socio-economic data in future, enabling both risk and vulnerability to be considered together.


Introduction
Growth in urban populations puts increasing pressure on city planners, policy makers and society to develop in a sustainable and resilient manner. Cities must have the capacity to mitigate the impacts of extreme weather in order to minimise damage to human health, the environment and the economy. Urban climate-related risks include, but are not limited to, thermal stress, flooding, air quality events and extreme wind (e.g. Bell et al., 2007;Chen, Hill, & Urbano, 2009;Dessai, 2002;Hsieh & Wu, 2012). In many cases the urban environment enhances these risks. For example, urban areas are known to exacerbate heat stress for the following reasons: • The relatively limited amount of vegetation reduces the opportunity for evaporation and its associated cooling effects, contributing to city temperatures that are typically a few degrees higher than in the surrounding countryside (Howard, 1833;Oke, 1982).
• Paved and built surfaces (e.g. roads, carparks, roofs) are fairly impermeable to water so rainfall is quickly routed into drainage systems and directed away from the surface, thus removing the source of moisture for evaporation Grimmond, Oke, & Steyn, 1986;Oke, 1982;Xiao, McPherson, Simpson, & Ustin, 2007).
• Buildings and roads absorb and store a large proportion of heat during the day and the release of this heat after sunset means temperatures may remain high throughout the night (e.g. Grimmond & Oke, 1999b;Kotthaus & Grimmond, 2014a;Offerle, Grimmond, & Fortuniak, 2005;Roberts, Oke, Grimmond, & Voogt, 2006). air-conditioning in warm conditions releases additional waste heat into the environment, further augmenting temperatures and exacerbating heat stress, and putting pressure on power supplies (de Munck et al., 2013;Ohashi et al., 2007;Ramamurthy, Li, & Bou-Zeid, 2015). To lower carbon emissions, one way to encourage use of public transport over private cars is to increase the cost of residents' parking permits. In London, this has led to more people paving over their front gardens so they can park off-road (GLA, 2005b). Vegetation and pervious gravel/ soil surfaces have been replaced by impervious concrete or asphalt, enhancing runoff and reducing infiltration of rain water, with the result that evaporation is restricted.
Biophysical models use information about the urban surface (e.g. land cover, building height, radiative properties such as albedo or reflectivity) and inhabitants (population density, energy use), along with past, present or predicted meteorological data to simulate components of the energy balance and/or water balance and thus inform about the environmental conditions. In addition to identifying regions with the greatest risk of exposure, models can also indicate how the risk may change over time. The physical processes represented in models enable us to understand why the risk may be greater under certain conditions and, conversely, to identify measures that may be used to reduce exposure. Models permit the advantages and disadvantages to be explored to better inform planning decisions before investments are made. For example, the impact of several urban cooling measures (including water, vegetation, high albedo surfaces and building dimensions) on pedestrian thermal comfort was assessed for a district in Toulouse that will soon undergo redevelopment (Martins et al., 2016). Potential feedbacks resulting from decisions (made by citizens or government) can be assessed. In the Toulouse study, the high albedo scenario was found to negatively impact pedestrian comfort as more radiation was reflected from walls towards pedestrians.
Understanding and managing climate-related risks in cities is of prime importance, particularly as more variable and more extreme weather is expected in future (IPCC, 2012). The focus of this study is Greater London, home to more than 8 million people (ONS, 2011) and with a daytime population (including workers and tourists) in excess of 10 million (GLA, 2013). The Greater London region is divided into 33 districts: 32 boroughs plus the City of London. For brevity, we refer to all 33 subdivisions as boroughs. There are 12 inner boroughs (plus City of London) and 20 outer boroughs (Fig. 1). Each borough is governed by a borough council (the City of London is governed by the City of London Corporation), responsible for education, provision of services and urban planning. Some planning decisions are also made by the Greater London Authority, responsible for London as a whole. The boroughs represent useful units in terms of governance and the availability of socio-economic data (such as poverty, health status and access to services) which can be used to gauge vulnerability (e.g. Wolf & McGregor, 2013). In future, borough-level risk estimates could be combined with socio-economic data in a move towards interdisciplinary modelling of cities that involves social, economic and biophysical aspects of the city system (Masson et al., 2014). This would enable adaptive or coping strategies to be targeted towards exposed areas that are most vulnerable.
Numerous indicators exist to describe thermal (dis)comfort (de Freitas & Grigorieva, 2015), usually based on temperature and often modified according to some combination of humidity, wind speed, radiation receipt or other variables in an attempt to account for the physiological and psychological effects that translate the physical air temperature to thermal comfort experienced by humans (Johansson, Thorsson, Emmanuel, & Krüger, 2014). Well-known examples include the physiological equivalent temperature (Höppe, 1999), mean radiant temperature  and universal thermal climate index (Jendritzky, de Dear, & Havenith, 2012). Mesoscale modelling studies often rely on 2 m air temperature output, sometimes combined with humidity, to represent human comfort (e.g. Theeuwes, Solcerová, & Steeneveld, 2013). Remotely sensed land surface tempera-ture (Wolf & McGregor, 2013) or urban heat island intensity (Tomlinson, Chapman, Thornes, & Baker, 2011) products are also used. Alexander, Fealy, & Mills (2016) considers the impact of urban development in terms of the surface energy balance. Certain indicators may be more or less suited to particular applications, depending on spatial scale, period of interest and data available. Microscale studies may consider differences between sunlit and shaded areas around individual buildings (e.g. Lindberg, Holmer, & Thorsson, 2008;Middel, Häb, Brazel, Martin, & Guhathakurta, 2014), whereas the computational demands of mesoscale simulations usually restrict the study period to a few days for typical grid-box sizes. A more user-friendly approach is adopted in this study to examine how London's climate responds at the local-scale to changes in meteorology, urban design and policy.
The objectives of this paper are: (i) to assess the response of the urban environment across Greater London to changes in surface characteristics, population and energy use; (ii) to explain the routes by which these changes affect the urban climate; (iii) to demonstrate a methodology which could subsequently be applied to other cities.

Model description
This study uses the Surface Urban Energy and Water balance Scheme (SUEWS), which has already been evaluated against observational datasets at two urban sites in this region (Ward, Kotthaus, Järvi, & Grimmond, 2016). SUEWS considers the urban surface comprised of seven surface types (paved surfaces, buildings, evergreen trees and shrubs, deciduous trees and shrubs, grass, bare soil and open water) with a single-layer soil store beneath each surface (except water). The exchange of energy at the surface is written (Oke, 1987): (1) Q* is net all-wave radiation; Q F is anthropogenic heat flux, i.e. the additional energy supplied through human activities. These inputs heat the air (Q H , turbulent sensible heat flux), evaporate water (Q E , turbulent latent heat flux) or are stored in (and later released from) the urban volume (ΔQ S , net storage heat flux). The storage heat flux is calculated using the Objective Hysteresis Model (OHM, ). Evaporation is calculated using an adapted Penman-Monteith equation  with surface conductance formulated after Jarvis (1976), Ward et al. (2016a). A running water balance is calculated at each time-step, providing soil moisture, runoff and surface wetness. Further details can be found in Christen (2011), Järvi et al. (2014) and Ward et al. (2016a). One of the advantages of SUEWS is its simplicity. High-performance computing is unnecessary, even when running the model for multiple years or multiple areas. Required inputs include information about the surface characteristics (e.g. land cover, building height, albedo, emissivity) and human behaviour (energy use, water use, population density), along with basic meteorological data: incoming shortwave or solar radiation (K ↓ ), air temperature (T air ), relative humidity (RH), barometric pressure (p), wind speed (U) and precipitation (P). The versatility of the model allows additional input information to be accepted if available (Lindberg, Grimmond, Onomura, & Järvi, 2015), otherwise recommended values should provide a reasonable approximation in many cases (Ward, Järvi, Onomura, & Lindberg, & Grimmond, 2016). Key site-specific information may need to be derived from other sources (Section 3.1).
The model runs in this study were performed using SUEWS v2016a (Ward et al., 2016b). Here, we focus on the modelled energy fluxes (Eq. (1)). Results are presented in terms of the energy partitioning using the median midday (1100-1400) Bowen ratio (Q H /Q E ), β MM . Several studies have concluded that the balance between evaporation and sensible heat flux is a key factor influencing elevated urban temperatures and contributing to the urban heat island (Fischer, Seneviratne, Lüthi, & Schär, 2007;Li and Bou-Zeid, 2013;Li et al., 2015). The less energy used for evaporation, the more energy is available for heating the atmosphere and augmenting air temperatures, i.e. greater risk of thermal stress. Since β MM includes Q H and Q E , it will respond to changes in the surface energy partitioning which result from meteorological conditions (including temperature and humidity), surface water availability, additional energy supplied by anthropogenic activities, land cover characteristics, properties of surface materials and state of vegetation. This approach can therefore be used to summarise the impact of a range of urban development or climate mitigation strategies on urban climate.

Model evaluation
Model output should always be interpreted with care, especially when applied to a region or conditions that differ from those for which the model has been tested. The year-round performance of SUEWS was recently evaluated using observational datasets collected in central London, UK, and suburban Swindon (a large town located about 120 km to the west of London) during the period 2011-2013. We are therefore able to present this application of the model across Greater London with a thorough understanding of its performance. To set the model results in context, values from the observed datasets are provided where available. These are denoted 'OBS URB' for the dense urban central London site and 'OBS SUB' for the suburban residential Swindon site. The UK evaluation demonstrated good performance in summer, particularly for the suburban site, but highlighted two issues. Firstly, there is insufficient information to enable the OHM coefficients used to calculate the storage heat flux to account for building volume or density of building materials, with the result that ΔQ S is underestimated and Q H overestimated at URB. Secondly, evaporation from open water is likely overestimated, contributing to a modelled Q E that is too high. The reader is referred to the full evaluation paper for detailed discussion (Ward et al., 2016a).
In addition to the UK evaluation, SUEWS has also been evaluated against observations at suburban sites in Los Angeles and Vancouver, North America (Järvi et al., 2011); for residential areas of cold-climate cities in Helsinki and Montreal (Järvi et al., 2014); also at a city-centre site in Helsinki (Karsisto et al., 2015); and at an urban and suburban site in Dublin (Alexander, Mills, & Fealy, 2015). These studies demonstrate that SUEWS generally reproduces reasonable results, particularly for suburban areas during summer months. In dense urban areas the large magnitude and considerable spatial variability of Q F presents a challenge to both modelling and measurement, so uncertainties are largest in these areas. In general, as for the UK sites, ΔQ S appears to be underestimated for dense urban areas (Järvi et al., 2014). As Q E is often small in city centres it can be difficult to measure and model accurately (Karsisto et al., 2015;Kotthaus & Grimmond, 2014a). These are fairly typical issues encountered by modellers and urban climatologists, as opposed to particular limitations of SUEWS.

Methodology
An objective here is to demonstrate a methodology for locations lacking detailed information. Thus, the model runs have been carried out using readily available information. One advantage of SUEWS is its relatively undemanding input requirements, but appropriate meteorological forcing data, representative surface cover and reasonable estimates of energy use are important.
The flexibility of SUEWS allows the user to define model grids (e.g. to reflect data availability and areas of interest). The focus of this work is at borough-level, but different areas can be specified depending on the application. Each borough is treated as a separate model grid and, at this coarse scale, there is no interaction between grids. The approximation has been made that all boroughs experience the same meteorological forcing, but surface characteristics must be specified individually for each borough.
To give an idea of the summertime behaviour under different conditions, data for the month of July in two consecutive years (2011,2012) with very different rainfall patterns are analysed. 2011 was a dry year (602 mm rainfall, 76% of the 1981-2010 Normal for southern England), whereas 2012 was very wet (1047 mm rainfall, 132% of the Normal) (Met Office, 2016). SUEWS is run from 1 January for the two years of interest, allowing several months' spin-up prior to the July study periods. The model setup and parameters are essentially the same as for the evaluation in central London (Ward et al., 2016a), except for the meteorological forcing data and site-specific input.

Preparation of input information
3.1.1. Meteorological forcing data The required meteorological forcing is provided by the WFDEI global data product (Weedon et al., 2014), available at 0.5°× 0.5°r esolution at 3-h intervals for 1979-2012, downscaled to the model time-step (5-min). Differences in incoming shortwave radiation, rainfall and relative humidity result in small differences when using WFDEI instead of observed data: Q* is overestimated but the impact on the heat fluxes is reduced by compensating effects (see Appendix A).

Surface cover fractions
Neighbourhood Statistics (NeSS, 2014) provides land cover data for each borough based on the Generalised Land Use Database (GLUD). The most recent 2005 dataset is used here. The following additional information is used to convert the GLUD classes (domestic buildings, non-domestic buildings, road, path, rail, domestic gardens, greenspace, water, other land uses, unclassified land) to the SUEWS land cover classes (Section 2.1): (i) A breakdown of land cover types within gardens for 2006-8 (Smith et al., 2011) (ii) Total tree cover in London is 20% (GLA, 2005a) (iii) It is assumed 20% (80%) of trees are evergreen (deciduous).
Thus, domestic gardens are split up into land cover types using (i); the total proportion of trees (ii) in greenspace plus gardens is used to deduce the proportion of greenspace that is grass; road, path and rail are counted as paved surfaces; other land uses and unclassified land are counted as bare soil surfaces.
Surface cover varies widely across London (Table 1, Fig. 2). City of London has the highest proportion (70%) of impervious surfaces (45% buidings, 25% paved surfaces), followed by Kensington and Chelsea, Islington, and Westminster (60%). These are also the least vegetated boroughs (City of London has only 5% vegetation), along with Tower Hamlets which has the highest proportion of water (22%). The outer boroughs are much more vegetated, with agricultural land towards the edge of the city, open greenspace (parks and commons) and large gardens. Bromley and Havering comprise over 70% vegetation. The spatial variability within boroughs tends to be most pronounced in central London, for example Westminster contains two large parks (Hyde Park and Regent's Park) but is very densely built-up outside these areas. The outer boroughs are predominantly suburban. Substantial water bodies include the River Thames which runs through the centre of London, as well as other open water (lakes, reservoirs, docks) in Tower Hamlets, Waltham Forest and Enfield, for example. The boroughs range in size from 150 km 2 (Bromley) to 12 km 2 (Kensington and Chelsea) (City of London is 3.3 km 2 ).
The observation site URB is located in a very densely built up area of Westminster borough, close to the boundary with City of London (Fig. 1). The proportion of paved and building surfaces at the URB site (81%) is greater than the borough averages. The proportion of vegetation (5%) is similar to City of London and the proximity to the river means 14% of the source area is made up of open water. The URB site thus represents amongst the most densely built up, least vegetated, and highly populated areas of London. The observation site SUB is more similar to the average values, with 45% vegetation and 49% impervious surfaces. For more details about the observation sites see the original references for these datasets (Kotthaus & Grimmond, 2014a, 2014bWard, Evans, & Grimmond, 2013).

Surface characteristics
Characteristics for each surface type are, in this application, assumed to be the same for all boroughs (Table 2). In reality, the variety of building materials will lead to differences in surface properties at the neighbourhood scale or below. The values selected are representative of the surfaces in general (e.g. buildings in European cities) and are the same as those used for the model evaluation.
Building height is estimated at 6.5 m for outer boroughs, 10.0 m for inner boroughs, 20.0 m for Westminster and 25.0 m for City of London (based on Lindberg & Grimmond (2011)). Tree heights are estimated as function of building height (Lindberg & Grimmond, 2011). Roughness length and displacement height are calculated within SUEWS using the rule-of-thumb (Grimmond & Oke, 1999a). Other required characteristics are assumed to be the same across London as public quantitative data are unavailable. Ward et al. (2016a) shows that surface cover and energy use play important roles in surface energy partitioning, whereas building height and small differences in meteorology generally have little impact on model output.

Anthropogenic energy use
The anthropogenic heat flux Q F can either be provided as input time-series data or calculated within SUEWS. Here, Q F is calculated within SUEWS to enable values to respond to changes in temperature, population or energy-use characteristics for different scenarios.
SUEWS calculates Q F based on heating degree days (HDD), cooling degree days (CDD) and population density, ρ pop , (Järvi et al., 2011;Sailor & Vasireddy, 2006): where a F0,1,2 are coefficients related to non-temperature dependent energy use, building cooling and building heating, respectively. These coefficients can be specified individually for weekdays and weekends and for each model grid. Alternatively generic coefficients could be used for particular land use types (e.g. residential, commerical). Population data from the 2011 census (ONS, 2011) are used here. The average of the workday and resident (night-time) population is used as some boroughs have a low resident population but very large workday population, particularly in central London (Table 1, Fig. 3a). The same a F0,1,2 coefficients are used for all boroughs with separate values for weekdays and weekends. The coefficients were informed by the GreaterQF model (Iamarino, Beevers, & Grimmond, 2012) (Appendix B). Compared to GreaterQF, this approach underestimates Q F by about 30 W m −2 (≈20%) for City of London and by about 10 W m −2 for Kensington and Chelsea and Westminster, and is within about 3 W m −2 for most other boroughs. Modelled Q F varies considerably between boroughs (Fig. 3b), with mean annual values for 2012 between 3.9 W m −2 (Bromley) and 111.7 W m −2 (City of London).
In reality, population varies through the day and workers and residents have different energy-use profiles (e.g. Hamilton et al., 2009;Ichinose et al., 1999). SUEWS' simple approach provides an estimate of Q F from readily available information, but output from more complex models could be included if appropriate. Accurate estimates of Q F are more important for densely populated areas (where Q F is large relative to other terms in the energy balance (Eq. (1))) than suburban areas where a rough estimate of 5-10 W m −2 may be sufficient.
External water use is assumed negligible as residents in the UK tend to irrigate far less extensively than in climates with hotter drier summers (Alexander et al., 2015;Ward et al., 2016a).  alter the available energy, modifying sensible and latent heat fluxes. Thus, changes to the urban system associated with urban development or mitigation measures are communicated to the model by modifying the input information. Model output from these 'alternative scenarios'

Model scenarios
are then compared to the 'base run' (input data as described in Section 3.1). Several example scenarios based on real situations have been developed to illustrate the versatility of this approach (Table 3)   Dotted lines indicate possible linkages: an increase in population density may require additional building which changes the surface cover and/or building height; in the current setup Q F does not impact ΔQ S , but T air indirectly impacts ΔQ S via selection of the summer or winter OHM coefficients a 1,2,3 . Surface cover fraction for each surface i is denoted f i . All other notation is defined in the text. Inadvertent climate modification by residents' decisions about their gardens is considered in scenario 1, based on Smith et al. (2011) which describes the actual change in garden composition that occurred between 1998-9 and 2006-8. Scenario 2 considers the Mayor of London's plans for increased tree cover, based on published reports from the Greater London Authority (GLA, 2011). Scenarios 1 and 2 therefore involve changes to the surface cover fractions, f i . In scenario 3, a 50% reduction in vehicle use is simulated, mimicking similar measures to improve air quality in other cities (e.g. Zhang, Lin Lawell, & Umanskaya, 2017). In scenario 4, an increase in population density is considered, first with no change to the landscape (ρ pop increases directly affecting Q F , indirectly affecting the available energy), then with associated changes in building density (increased building fraction or building height, directly affecting ΔQ S and/or Q E ). The required change in building density is estimated using input information for the base run (Appendix C). Scenario 5 combines several measures for climate-sensitive development (reducing energy consumption, increasing vegetation). Finally, the impact of meteorological conditions is explored by running the model with forcing data from 2011 and 2012. However, underestimation of Q F (Section 3.1.4) is thought to result in a slight underestimation in β MM for City of London.

Variability across greater London ('base run')
Observational campaigns in urban areas have found that vegetation plays a major role in surface energy partitioning, with increasing vegetation fraction often linked with decreasing summertime Bowen ratio (e.g. Christen & Vogt, 2004;Grimmond & Oke, 2002;Offerle, Grimmond, Fortuniak, & Pawlak, 2006). Primarily, this behaviour is explained by the greater evaporation potential of vegetation and permeable soil surfaces compared to impervious built surfaces. The impervious surfaces tend to have very little capacity to store moisture and thus dry quickly after wetting with most surface water going to runoff which is rapidly removed via pipes and sewerage systems.
SUEWS reproduces this behaviour (Fig. 5a,c): β MM is seen to decrease as vegetation fraction increases. The rate of change is greatest when there is only a small proportion of vegetation. Those boroughs with little vegetation have the highest β MM : City of London, followed by Islington, Kensington and Chelsea, and Westminster (numbers 7, 19, 20 and 33 in Fig. 5a,c). At the other end of the spectrum, Bromley (5) and Havering (16) are ≥70% vegetation and have low β MM . The observed values (URB and SUB in Fig. 5a,c) demonstrate that the model is performing reasonably. Bexley (3) and Tower Hamlets (30) both have notably low β MM given their vegetation fractions because these boroughs contain a significant proportion of open water (Fig. 2), which, according to the model, can evaporate freely. Note that comparison of modelled and observed values at the URB site suggested that SUEWS probably overestimates evaporation from open water (Ward et al., 2016a), however this is a topic of ongoing research since is it is not currently clear whether the depth of the River Thames constrains its temperature and restricts evaporation, or whether considerable evaporation from the river is taking place but is not detected at the height of the sensor (Kotthaus and Grimmond, 2014b). In Figs. 6 and 7 β MM is instead plotted as a function of the proportion of vegetation plus water area.
The robustness of these results is assessed by comparing the monthly average β MM with daily values (Fig. 6a). Generally, the results  Table 3 Assumptions and changes made to the model input for the various scenarios.

Scenario
Changes to input 1 Garden composition Surface cover fractions adjusted to represent 1998-9 instead of 2006-8 garden composition. 2i Increasing tree cover (on paved) Tree cover increased by 25% in each borough with a corresponding reduction in paved surfaces. 2ii Increasing tree cover (on grass) Tree cover increased by 25% in each borough with a corresponding reduction in grass surfaces. 3 Reducing road traffic Coefficient a F0 re-derived having reduced the contribution of vehicle energy use by 50%. 4i Increasing population (no building) Population increased to 2020 projection but with no change in land cover (i.e. no new building occurs). 4ii Increasing population (with building) Population increased to 2020 projection and homes are built on available land (i.e. vegetation and bare soil surfaces). 4iii Increasing population (with building) Population increased to 2020 projection and homes are built on bare soil/unmanaged land only (i.e. no loss of vegetation). 4iv Increasing population (with building) Population increased to 2020 projection and homes built upwards only (i.e. building height increased but surface cover fractions unchanged). 5 Climate-sensitive adaptation Coefficient a F0 and a F2 adjusted to reflect reduced building energy use by 20%. Coefficient a F0 adjusted to reflect reduced vehicle energy use by 10%. Tree cover increased by 25% with a corresponding reduction in paved surfaces for inner boroughs. 25% of bare soil surfaces changed to grass for wealthy boroughs.
are representative of the daily values, although the range of values increases as β MM increases (also seen in the observations). The few points below 0 occurred on 16 July 2012 which was rainy and overcast so heat fluxes were small. The five-day period 22-26 July 2012, identified as a heat-wave in London (using the method in Li et al. (2015)), had clear-skies, no rain and mean daily temperatures of 21.5°C (cf. monthly mean daily temperature of 17.0°C measured at URB). β MM on these heat-wave days (crosses, Fig. 6) is generally higher than on other days (exceptions for Bexley and Tower Hamlets are related to evaporation from water surfaces).
To investigate the relative conditions in each borough, daily β MM values are ranked from 1 (lowest) to 33 (highest) (Fig. 6b). City of London always has the highest β MM , followed by Islington. Although most boroughs change rank with neighbouring boroughs (in terms of rank order, not geographical location) depending on meteorological, surface and sub-surface conditions, the overall ranking remains similar throughout the study period. Exceptions are, again, Bexley and Tower Hamlets. Their substantial water reserves evaporate freely on sunny days with abundant energy whereas evaporation is limited on cloudier days with lower energy input. Therefore, for these boroughs, the sunniest days are those with lower rank positons (e.g. during the heat-wave). Other boroughs with lower rank positions during heatwave days also contain large areas of water, for example Hammersmith and Fulham, Newham, Waltham Forest, Westminster. Although open water evaporation in SUEWS requires more detailed evaluation, the qualitative result that water bodies help ameliorate heat stress is in agreement with other studies (e.g. Theeuwes et al., 2013).
Results are similar for July 2011, but with more variation in daily values and less consistent rank positions (Fig. 7). Frequent rainfall in 2012 maintains a plentiful supply of moisture: the surface is often wet following rainfall and soils are moist. In 2011, less frequent rainfall allows the surfaces to dry out, reducing evaporation (except where the moisture supply is maintained by large water bodies; those boroughs have notably low rankings in Fig. 7b). This behaviour results in a wider spread of β MM values. That rainfall prompts a sharp decrease in the Bowen ratio followed by an increase as surfaces dry has been demonstrated in observational data (Kotthaus & Grimmond, 2014a;Ward et al., 2013). Differences in evaporation between years can also result from differences in Q*, but summer 2011 and 2012 were both cloudier than normal .
There is little spatial variation in Q* between boroughs (Fig. 8). Incoming shortwave (K ↓ ) and longwave (L ↓ ) radiation are identical for all boroughs as the same meteorological forcing has been assumed (Section 3.1.1). Atmospheric pollution can lead to reduced shortwave radiation receipt in city centres compared to their surroundings (Arnfield, 2003). The size of this reduction is about 10% in London (Ryder and Toumi, 2011) but can be larger for more polluted cities (Jáuregui & Luyando, 1999). Conversely, an increase in L ↓ is typical due to warmer temperatures and increased pollutant concentration over cities (Arnfield, 2003;Christen & Vogt, 2004). These effects are not simulated here but, to a certain extent, these urban-rural differences in K ↓ and L ↓ compensate for each other. The highest Q* values are thus obtained for the least vegetated boroughs with the lowest bulk albedos (≈0.12) and the lowest Q* values for the most vegetated boroughs with the highest bulk albedos (≈0.17). Compared to the day-to-day variation in midday Q*, differences between boroughs are small. This is thought to be realistic as many observational studies report little difference in Q* between urban and rural sites or multiple sites within a city (e.g. Christen & Vogt, 2004; Schmid, Cleugh, . For higher resolution studies, albedo could be adjusted to  Table 1 incorporate the effect of shading (Sailor & Fan, 2002) or to represent specific materials (e.g. slate tile versus ceramic tile roofs). Modelled Q* for all boroughs is larger than the observed values at the evaluation sites, mainly because the WFDEI K ↓ is higher than the observed values (Appendix A).
For midday in summer, these results suggest the additional energy supplied by anthropogenic activities is about 30% of Q* for City of London (45% at URB). Q F can be even more important in winter and at high latitudes. For the most densely populated boroughs, Q F acts to increase Q H as Q E is limited by the low vegetation fraction and lack of available moisture. Thus, boroughs with the highest Q H tend to be those with the highest Q F (City of London, Islington, Kensington and Chelsea, Westminster). The difference in energy partitioning between study years can be seen here: Q H generally exceeds Q E in July 2011, whereas Q E exceeds Q H for the more vegetated boroughs in July 2012 (Fig. 8).

Scenario 1: change in garden composition
The composition of gardens in London has changed over recent years (Smith et al., 2011). Compared to 1998-9, gardens in 2006-8 contain more buildings and paved surfaces (41% cf. 31%) and less vegetation (58% cf. 65%). For this first scenario, we assess the impact on local climate by considering the change in β MM if today's (i.e. 2006-8) garden composition was returned to that of 1998-9.
Overall, domestic gardens comprise 24% of the total area of Greater London. The decrease in vegetation fraction associated with the change in garden composition is small for many central boroughs, but for inner boroughs with appreciable amounts of garden (≈20%, e.g. Islington) the small increase in vegetation fraction causes a discernible reduction in β MM ( Fig. 9a and b). For the outer boroughs, even though there is a lot of garden space, the already higher vegetation fraction means the change in garden composition generally has a smaller impact. However, all boroughs show a reduction in β MM , indicating that the change in garden composition that has occurred since 1998-9 has altered the surface-atmosphere exchange: nowadays more energy is directed to heating and less to evaporation. Such trends are expected to continue as more people pave their front gardens for parking or opt for low maintenance gardens (GLA, 2005b). Other environmental impacts associated with replacing vegetation with paved surfaces include reduced biodiversity and increased flood risk. SUEWS suggests an increase in annual runoff of 5-15 mm due to this change in garden composition (modelled annual totals for each borough range between 376 and 553 mm in 2012).
The impact of these land cover changes on the surface energy balance can be understood using Fig. 4. Modifying the surface cover fractions changes the bulk radiative properties of the surface. Emissivity has little impact on the output, but vegetation generally has higher albedo than typical materials used for paving or building in the UK (Table 2) so the increase in bulk albedo results in lower Q*. The anthropogenic heat flux is unchanged by this scenario, but the storage heat flux is reduced (as grass stores less heat than paved and building surfaces, and because Q* is lower). The increase in vegetation fraction means more of the available energy is used in evaporation, leaving a smaller proportion for the sensible heat flux. Hence Q E increases and Q H decreases.  Table 1. Buildings and paved surfaces tend to store (and later release) a large amount of heat as their constituent materials have high heat capacities. In SUEWS, this behaviour is simulated through different OHM coefficients (a 1,2,3 ) for each surface type. However, Ward et al. (2016a) suggests the range of OHM coefficients in the literature may be insufficient to account for the wide range of urban surface types, with the result that ΔQ S may be underestimated in built-up areas. Central boroughs (City of London, Kensington and Chelsea, Islington, Westminster) have the highest ΔQ S , but ΔQ S OBS URB is much larger (Fig. 8). As previous work accounting for the three-dimensional building volume did not significantly improve the performance of OHM (and decreased performance in some cities) (Grimmond & Oke, 1999b), in the current configuration building height does not affect ΔQ S , although it is reasonable to expect larger ΔQ S with greater built volume.

Scenario 2: increase in tree cover
Recognising the benefits of urban trees (e.g. Salmond et al., 2016), the Mayor of London plans to increase London's tree cover from 20% to 25% by 2025 (GLA, 2011). Two scenarios are considered here: (i) trees replace paved surfaces; (ii) trees replace grass surfaces (Table 3). In each case, the input surface cover fractions are altered so that the tree cover increases by 25% for each borough (in line with total tree cover   Table 1) and for urban (URB) and suburban (SUB) observations. across Greater London increasing from 20% to 25%).
When trees replace paved surfaces β MM decreases across all boroughs ( Fig. 9c-f). The largest changes in vegetation fraction occur for boroughs with the largest proportion of trees, but the largest changes in β MM occur where the vegetation fraction is small. The Mayor's Street Tree Programme aims to plant new trees in the most deprived areas, including areas with few street trees (GLA, 2011). Although no borough-scale plans were available, this strategy will be more effective than suggested by scenario 2i, as the rate of change of β MM is greater where vegetation is scarce. The impact is also greater in 2011 than 2012 (mean Δβ MM = −0.042/-0.018 in 2011/2012, which corresponds to a change in Q H and Q E of around ± 4 W m −2 assuming a Q* of 400 W m −2 and β MM ≈ 1).
If trees instead replace grass surfaces (scenario 2ii) there is no change in vegetation fraction and the results are quite different (Fig. 9g-j). The presence of more trees and less grass lowers the surface conductance, which decreases Q E and allows Q H to increase (mean Δβ MM = 0.072/0.077 in 2011/2012). In 2011 the impact on the outer boroughs is less than in 2012 (Fig. 9h, j), possibly because soil moisture (rather than vegetation fraction) plays a significant role in limiting evaporation in dry years. These results highlight the importance of considering both where trees will be planted and what will they replace. Note that the types of vegetation planted/replaced are also relevant. These can be represented in the model by specifying vegetation characteristics (particularly maximum surface conductance).

Scenario 3: reduction in road traffic
Scenario 3 considers a change in human behaviour, without changing the landscape, namely a 50% reduction in vehicle usage. This scenario is intended to represent strict traffic controls enforced to manage air pollution. For example, during the 2008 Beijing Olympics, vehicle use was restricted on alternate days based on odd/even licence plates and vehicles not meeting emission standards were banned completely, which meant only about 40% of vehicles were permitted to use the roads each day (Song & Wang, 2012;Yu, Wang, Zong, Li, & Lü, 2010). As well as the impact on pollutant emissions, there will be an associated reduction in heat emissions.
According to GreaterQF, vehicle use contributes about 15% of total Q F . To model a 50% reduction in road traffic, the coefficient a F0 (see Section 3.1.4) was reduced to 0.1457 and 0.1384 W m −2 (cap ha −1 ) −1 on weekdays and weekends, respectively. These values were derived using the same method as the generic London borough coefficients (Appendix B) but with the vehicle contribution halved. The reduction in Q F reduces the energy input to the surface and results in lower β MM (not shown). The impact is generally very small (mean Δβ MM = −0.012 in 2012), except for very densely populated boroughs, mainly because building energy use is so high. Policy may therefore be better directed towards reducing building energy consumption. In warmer climates, air conditioning can substantially increase Q F during summer, maintaining cooler indoor temperatures but exacerbating heat stress outside. Reducing transport use is much more important in terms of air quality (e.g. Beevers & Carslaw, 2005).

Scenario 4: population increase
Scenario 4 considers the impact of population increase. The resident population of Greater London is expected to increase from 8.2 million (ONS, 2011) to 9.1 million by 2020 (GLA, 2014). Here, the percentage projected change in resident population for each borough is applied to the current population densities (Table 1). First, the impact of the additional population alone (i.e. no new building) is considered (scenario 4i, Fig. 10a-d). There is negligible impact for boroughs where ρ pop and Q F is small (Bexley, Bromley, Havering) or where the change in population is very small (Kensington and Chelsea Δρ pop = 0.2%). For central boroughs, the increase in ρ pop caused an increase in β MM , particularly for City of London (Δβ MM = 0.155/0.253 in 2011/2012) where considerable population growth is combined with the already high population density to increase Q F (although note that part of the reason for this large change is due to the assumption that has been made in scaling the mean of day-and night-time ρ pop by the projected increase in resident ρ pop ). The additional population has a larger impact during dry weather as, when water is limited, the extra energy acts to increase Q H rather than Q E .
A more realistic scenario also increases the proportion of buildings in order to provide the growing population with somewhere to live and work. The increase in buildings required to match the population growth was calculated from the relation between building fraction and ρ pop fitted to the current values for each borough (Appendix C). In scenario 4ii, the land cover fractions are adjusted to account for this increase in building fraction (the proportion of vegetation and bare soil is reduced accordingly; it is assumed existing paved, building and water surfaces are not available to be built on). In scenario 4iii, the building fraction is increased as for scenario 4ii, but with a corresponding decrease in bare soil surfaces only (i.e. development of unmanaged land rather than removing vegetation). In scenario 4iv, the land cover fractions are unchanged but building height is increased to provide the same new building volume as for scenario 4ii and iii.
Results for scenario 4iv (not shown) are very similar to scenario 4i (within 0.5%) except for Tower Hamlets (Δρ pop = 32%) where β MM increases from 0.78 to 0.81 (scenario 4i) to 0.82 (scenario 4iv). Equivalent values for 2011 are 0.75 increasing to 0.79 and 0.82. These results suggest that building height has a much smaller impact than building extent (assuming roughness length and displacement height scale with mean building height). Note that the parameters within SUEWS were not changed to account for shading or radiation trapping within the urban canyon, both of which increase with building height, reducing daytime Q* but increasing night-time Q* (e.g. Mavrogianni et al., 2011).
If building occurs horizontally rather than vertically, the associated loss of vegetation increases β MM (scenario 4ii, Fig. 10e-h), particularly for the inner boroughs and more so for drier years than wetter years. If vegetated areas are protected from development and new buildings are constructed only on bare soil, the impact is generally reduced (scenario 4iii, Fig. 10i-l). This reduction is greater in 2012 than 2011, which again, is probably related to soil moisture availability. The lack of rainfall in 2011 means soil moisture stores are low. When rain does fall, it can directly infiltrate the soil beneath pervious surfaces but not beneath impervious surfaces (SUEWS assumes buildings and paved surfaces are totally impervious). A reduction in pervious surfaces (bare soil and vegetation) reduces infiltration and increases runoff, so the soil receives less water. Therefore, both scenario 4ii and 4iii have a larger impact in 2011 than 2012, as in 2012 soil moisture is higher and less of a limiting factor. Thus, the proportion of vegetation is the important factor in 2012, whilst the proportion of pervious surfaces is important in 2011.

Scenario 5: cimate-sensitive adaptation
Scenario 5 combines several aspects from previous scenarios to assess the potential impact of adapting London in a climate-sensitive way (Table 3) including: reduced anthropogenic energy emissions brought about through improved energy efficiency of buildings and improved access to public transport (see Appendix B); and increased vegetation cover brought about through tree planting in inner boroughs (i.e. focusing on the least vegetated boroughs) and replacing bare soil with grass in wealthy boroughs (assuming wealthy boroughs have the means to develop unmanaged land into green spaces). Wealthy boroughs were identified as those with the top 20% highest median household income (GLA, 2012/3): Kensington and Chelsea, City of London, Westminster, Richmond upon Thames, Camden, Wandsworth.
Scenario 5 results in a reduction in β MM across all boroughs, but the largest changes occur for the inner boroughs (where the proportion of vegetation increases and the considerable anthropogenic heat flux decreases) and wealthy boroughs (where the proportion of vegetation increases) (Fig. 11). The impact is greater in 2011 than 2012. Boroughs classified as both inner boroughs and wealthy boroughs experience the largest change: for Camden, City of London, Kensington and Chelsea, Wandsworth and Westminster Δβ MM exceeds -0.10 (Fig. 12). The decrease in energy use for the outer boroughs (where Q F is small) has a small impact in 2012 (Δβ MM < −0.01 for many boroughs) but has a greater effect in 2011. There would also be a reduction in carbon emissions associated with the decrease in fuel combustion for transport and home heating.
All the scenarios presented here produce a change in the same direction for all boroughs. However, alternative development pathways can be envisaged where the situation improves in some boroughs but worsens in others. For example, population growth accompanied by extensive building in some areas, compared to greening initiatives in other boroughs where there is less pressure on space. Input data can be easily adapted to represent the relevant changes; the demonstrations here are intended to be examples. Fig. 10. As for Fig. 9 but for scenario 4i-iii (Table 3). These scenarios assume an increase in population with (4i) no additional building; (4ii) an increase building fraction, with buildings replacing bare soil and vegetation; and (4iii) an increase in building fraction, with buildings replacing bare soil only. See text for details.

Summary and conclusions
Numerous policy decisions must be made without knowledge of their outcome. Scenario modelling can help inform about likely (or unintended) outcomes, or can be used to compare likely impacts of one decision versus another. Since all models are approximations of reality they must be interpreted at such, bearing in mind their imperfect representation of physical processes and the assumptions involved in specifying model parameters and required input information. Estimates of these can be made from the literature, datasets or other models and sensitivity studies can be used to assess uncertainties. The decisions involved in implementing environmental measures are significant. For instance, increasing the number of trees could have a different outcome (positive or negative) depending on where and how they are planted and on the species selected. The impact of population growth depends on where and how people will be housed: increasing building height was found to have a much smaller impact than increasing building extent due to the important role of vegetated surfaces in moderating the urban climate. Compact cities may also offer advantages in terms of reduced transport demand (Mills, 2007). The results show that changes in land cover can have major impacts on urban energy partitioning, with the biggest changes tending to occur in the least vegetated areas where heat stress is greatest. In recent years, trends in garden composition (i.e. decisions taken by residents) have resulted in more Fig. 11. As for Fig. 9 but for scenario 5. Scenario 5 combines aspects of the previous scenarios to assess the potential impact of adapting London in a climate-sensitive way (see text for details). of the available energy being used to heat the atmosphere. Human behaviour has a considerable impact on the environment, particularly in very densely populated areas. However, it should not be overlooked that uncertainties are also greatest for such areas, where fluxes can vary dramatically over short temporal and spatial scales and observations are scarce. This paper demonstrates how the biophysical SUEWS model can be used to assess the response of the urban climate across a city to various urban development and climate mitigation measures. SUEWS has the capability to use more detailed information when available, for example through consultation with city planners. We describe the approaches taken to convert readily available datasets to the inputs required by the model, with the intention that analyses could be performed at other sites using a similar methodology. In terms of urban design, this study highlights the following: • (Well-watered) vegetation is a key control on urban energy partitioning and important for moderating elevated urban temperatures • Consequently, building upwards has a smaller impact on the urban energy balance than building on vegetated areas • Adding vegetation has greatest impact in sparsely vegetated areas • A range of meteorological conditions should be considered when assessing the impact of scenarios.
This study overcomes two limitations often encountered when assessing impacts of urban planning scenarios: few models respond to the range of controls assessed here (e.g. surface cover, population density, energy use, building properties); and many studies focus only on a few days due to high computational demands. Furthermore, SUEWS has also been evaluated against energy balance observations in a range of cities. Although model results must be interpreted with care, especially when pertaining to different conditions than those for which the model has been evaluated, ongoing evaluation work will extend the range of conditions and locations for which the model behaviour can be understood. Model development currently underway should improve the capabilities and accuracy of SUEWS, particularly in terms of the storage and anthropogenic heat flux.
This study focuses mainly on the risk of heat stress (with some consideration of hydrology through the water balance, although the water balance has not yet been thoroughly evaluated in the UK). There are also many other environmental factors to consider, such as air quality, noise pollution and biodiversity (beyond the scope of the current study), as well as economic and social factors. To analyse biophysical exposure risk in conjunction with social vulnerability , these results could be combined with socio-economic data.
The SUEWS model is available to download from http://micromet. reading.ac.uk.
The WFDEI dataset (WATCH Forcing Data methodology applied to ERA-Interim data) (Weedon et al., 2011(Weedon et al., , 2014 is used to provide the meteorological data required for SUEWS: incoming shortwave radiation (K ↓ ), air temperature (T air ), relative humidity (RH), barometric pressure (p), wind speed (U) and precipitation (P). Continuous, long-term observational datasets of these variables at sub-daily resolution are extremely rare due to the difficulties of siting instrumentation in urban environments and the workload involved in maintaining high-quality measurements over an extended period. The WFDEI dataset has global coverage (0.5°× 0.5°resolution) and is available at 3-h intervals for the years 1979-2012, allowing analysis of past weather events.
Although Greater London covers about 4 half-degree grid boxes, WFDEI data extracted for the central location of the URB site (51.51°N 0.12°W) are used, i.e. each borough uses the same meteorological forcing data. Differences between the 4 grid boxes were minor; here the single URB grid was chosen to simplify the comparison with observed data.
SUEWS requires input data at 60 min (which are linearly downscaled at the outset to the time-step of the model, here 5 min). Therefore, all variables are first linearly interpolated from the 3-h WDFEI data to 60 min, except 3-h rainfall which is downscaled directly to 5 min accounting for the intermittency of wet periods and conserving the total mass in each 3-h period (Table A1).
Using WFDEI data in place of observations (usually at 1-h resolution) can introduce errors (i) through interpolation of 3-h to 1-h data and (ii) as a result of differences between what the WFDEI and observed dataset represent, namely the height (see Table A1 for WFDEI; 49.6 m for URB), elevation (78 m for WFDEI; 10.7 m for URB) and land use.
To investigate (i), observations from URB (1 h) are averaged to 3 h to represent the resolution of the WFDEI data according to Table A1. That is, K ↓ is the average of the previous 3 h; T air , RH, p and U are instantaneous values (approximated here by the mean for the current hour and for the following hour). This 3-h dataset is then downscaled to 1 h in the same way as for the WFDEI dataset and compared to the original 1-h observations (Fig. A1). The assumption that the variables change linearly in time results in some scatter, particularly for K ↓ which can change dramatically over hourly time scales. The diurnal cycle of K ↓ is also slightly widened and flattened (not shown). The impact on the other variables is small and generally makes no appreciable difference to the model output.
More substantial differences are found between WFDEI and OBS datasets (Fig. A2). K ↓OBS is smaller than K ↓W1h (mean values for daytime are 195.7 W m −2 (at URB) and 220.2 W m −2 in 2012). To some extent, this is expected due to attenuation by aerosols over central London (see Section 4.1), but comparison between WFDEI and OBS data for the SUB site also revealed overestimation by WFDEI. Other differences which may be partly attributable to the urban setting of the observations include slightly warmer observed temperatures at night and lower observed relative humidity. The difference in elevation and measure-

Table A1
Details of WFDEI variables and conventions (Weedon et al., 2014) and the adjustments used to prepare 5-min input data for SUEWS. CRU precipitation totals were used.    Figure A1 but for 3-h WFDEI data interpolated to 1 h ('W1h') versus observed data ('OBS'). N rain is the number of hourly data points with rain > 0 mm.

Variable
H.C. Ward, C.S.B. Grimmond Landscape and Urban Planning 165 (2017) 142-161 ment height contribute to small differences between the datasets, evident in the offset between p OBS and p W1h ; the influence of height differences and urban roughness on wind speed is more complicated. However, the impact of these differences in T air , p and U on the model output is very small and no attempt is made to adjust the WFDEI values. For locations in mountainous areas, it is advisable to check whether the elevation of the grid square is comparable to that of the study site, or if adjustments may be required Kokkonen et al., 2017). A major difference occurs in the precipitation data, with the WFDEI dataset considerably wetter than OBS at both URB and SUB. In 2012, annual rainfall according to WFDEI (858 mm) is slightly larger than the observed total (821 mm at URB). According to OBS, rainfall occurs for 11% of the hourly periods; for the interpolated WFDEI dataset the value is 25%. The difference is also apparent in the original 3-h WFDEI dataset (18% compared to 29% of 3-h periods have rain). This increased frequency and volume of rainfall leads to an overestimation of modelled evaporation.
Out of the meteorological driving variables, SUEWS is most sensitive to P, K ↓ and RH (as also found by Kokkonen et al. (2017)). For the sites and years considered here, using WFDEI data instead of observed forcing leads to an overestimation in Q* and hence ΔQ S , but the change in Q H and Q E is less significant because the effects of overestimated P and K ↓ are partly compensated for by the overestimation of RH. For URB in 2012, using WFDEI data instead of observed forcing leads to an overestimation of about 20 W m −2 in Q* compared to observations; the effect on Q H and Q E is smaller (< 7 W m −2 and < 5 W m −2 respectively).

Appendix B. Calculation of coefficients for anthropogenic heat flux model
Coefficients for the simple anthropogenic heat flux model (Eq. (2)) have been derived for several locations: North American cities (Sailor & Vasireddy, 2006), suburbs of Vancouver (Järvi et al., 2011), cold-climate cities of Helsinki and Montreal (Järvi et al., 2014), and at the URB and SUB sites Ward et al. (2016a). These coefficients differ, in part due to variation in building standards (i.e. insulation), energy use (e.g. air conditioning) and land use (residential, commercial, institutional). To calculate appropriate coefficients for use across London, the GreaterQF model (Iamarino et al., 2012) is used. As the most recent GreaterQF data are for 2008, the WFDEI air temperature for 2008 are used to calculate heating and cooling degree days, with a base human comfort temperature of 18.2°C (Sailor & Vasireddy, 2006). Mean daily Q F is compared with mean daily temperature, and the coefficients a F0,1,2 fitted for each borough (Fig. B1). As the data suggest minimal use of air conditioning (i.e. little increase in Q F at high temperatures), in accordance with expectations for the UK, a F1 is set to zero. The population density is taken as the mean of resident and workday values. Using resident/workday population densities alone gave much larger/smaller a F0 and a F2 values for inner boroughs, particularly those with contrasting day and night populations, whereas the mean gave a smaller range of a F0 and a F2 values and fewer extreme values.
Values of a F0 and a F2 are consistently higher for weekdays than weekends. a F0 ranged from 0.1134/0.1073 W m −2 (cap ha −1 ) −1 on weekdays/weekends to 0.2281/0.2119 W m −2 (cap ha −1 ) −1 . a F2 ranged from 0.0038/0.0034 W m −2 K −1 (cap ha −1 ) −1 on weekdays/ weekends to 0.0076/0.0069 W m −2 K −1 (cap ha −1 ) −1 . In both cases, the lowest values are for Hackney and the highest for Hillingdon. No substantial difference is found between inner and outer boroughs, so the mean of the individual borough values is used for all model grids (Table B1). These values are slightly bigger than obtained for SUB and much smaller than for URB (Ward, Kotthaus et al., 2016), which is unsurprising given the characteristics of the observation sites relative to the boroughs. Note, Ward et al. (2016a) used Westminster borough population density (204.58 ha −1 ) for URB. As the URB site is particularly densely built up, the borough average ρ pop is likely an underestimation for this area which would explain the very large a F0 values obtained.
Building energy use is the greatest source of Q F , contributing an average of 79% of the total, less in spring/summer more in autumn/ winter (Fig. B2). Values range between 67% (Havering) and 92% (City of London). The vehicle contribution ranges from 6% (City of London) to 28% (Havering) with an average contribution of 15%. Human metabolism contributes between 2% (City of London) and 7% (Havering), with an average of 6%. Although building energy use varies substantially with temperature, the overall composition of Q F is fairly constant throughout the year and Q F is much lower in summer than winter.
For scenario 5, a 10% reduction in vehicle use is approximated by reducing a F0 by 0.10 × 0.15 = 1.5%. A 20% reduction in building energy use is approximated by reducing a F0 and a F2 by 0.20 × 0.79 = 15.8%. The resultant adjusted coefficients for scenario 5 are a F0 = 0.1346/0.1260 W m −2 (cap ha −1 ) −1 and a F2 = 0.0045/ 0.0041 W m −2 K −1 (cap ha −1 ) −1 for weekdays/weekends. For scenario 3, a F0 was reduced to 0.1457/0.1384 W m −2 (cap ha −1 ) −1 for weekdays/weekends. These values were derived by fitting to GreaterQF data with the vehicle contribution halved. If instead, the approximation had been to reduce a F0 by 0.5 × 15% = 7.5%, slightly larger values of 0.1505 and 0.1409 W m −2 (cap ha −1 ) −1 would have been obtained; this difference is mainly due to the variation in building energy use with temperature.

Appendix C. Estimating change in building density with population growth
Population growth for scenario 4 is based on GLA (2014) projections. To house this additional population, it is assumed that an increase in building density is required. For scenarios 4ii-iii additional buildings are constructed on available land (i.e. vegetated and bare soil surfaces). To estimate the change in the surface cover fraction of buildings required, data from the base run are used to derive a relation between the fraction of buildings and population density (Fig. C1). From this relation (fitted to the solid circles in Fig. C1), the change in modelled building fraction between the current and projected population is calculated and added to the current building fraction to give the new building fraction (empty circles).
The non-linear equation accounts for the changing relation between building fraction and population density: as population density increases buildings are taller and living space decreases. However, this is a simplistic approach. The relation should not be used for very low population densities (below 10.5 ha −1 it returns f buildings < 0). The fit is based on the mean of day-and night-time populations, so includes workers and residents, and the fraction of all buildings, so includes residential, commercial, industrial and institutional buildings. Therefore, the estimated change in f buildings includes new homes, places of work and services in an approximate sense, but these are not explicitly accounted for (nor their varying composition between boroughs). Furthermore, only the change in building fraction has been estimated; no additional roads or gardens are considered.
Instead of building horizontally, scenario 4iv considers the case where land cover fractions are unchanged but the new buildings are constructed vertically (i.e. new floors added to existing building areas) so mean building height increases. To estimate the new mean building height, the current mean building height is multiplied by the ratio of and a F2 [W m −2 K −1 (cap ha −1 ) −1 ] are given separately for weekdays and weekends; a F1 was set to zero. new to current f buildings , giving the same change in building volume as before. In reality, of course, a mixture of horizontal and vertical development is likely.