Impacts of urban expansion on relatively smaller surrounding cities during heat waves

: Urban-induced thermal stress can threaten human health, especially during heat waves (HWs). The growth of cities further exacerbates this e ﬀ ect. Here, weather research and forecasting (WRF) with an urban canopy model (UCM) is used to assess the e ﬀ ects of megacities and their growth on the thermal regime of proximal cities during heat waves. Analysis of the heat ﬂuxes shows that advection impacts cities downwind. Results indicate that as urban areas change size (50%–100% and 100–150% of their current size), the local 2 m temperature increases by 2.7 and 1.7 ◦ C, and the 2 m speciﬁc humidity decreases by 2.1 and 1.4 g kg − 1 , respectively. A small city downwind is impacted with a 0.3–0.4 ◦ C increase in 2 m temperature. Green roof is a potential mitigation strategy for these regions (i.e., beyond the megacity). With 50% green roofs in an urban area, a 0.5 ◦ C decrease in 2 m temperature and 0.6 g kg − 1 increase in speciﬁc humidity is simulated. Urbanization upwind of a megacity will contribute to regional climate change.


Introduction
Heat waves (HWs), excessively hot periods that last for several days or longer, are a key cause of weather-related mortality [1,2].Urban citizens are more vulnerable to HWs, given the additional warmth in cities relative to rural surroundings, and the greater exposure of people to these conditions given greater population densities.During heat waves, greater heat-related deaths are predicted-for example, 148 deaths in Atlanta under a +2 • C scenario [3], or 700 documented in Chicago during a 1995 heat wave [4].Epidemiological studies suggest the mortality risk increases by 3.74% during a heat wave compared to non-heat wave periods [1].
Land cover changes and land use-related waste heat emissions have significant impacts on the climate at local and regional scales by modifying the energy, water, and momentum exchanges between the surface and atmosphere.These exchanges influence air temperature, moisture, wind, and precipitation, resulting in distinct urban climates, including the well-known urban heat island (UHI) effect [5,6].Recent theoretical, numerical, and experimental studies suggest that these urban effects may be enhanced during heat waves.Moreover, Li and Bou-Zeid [7] suggest that not only do heat waves increase the ambient temperatures, but they also intensify the difference between urban and rural temperatures.As a result, the added heat stress in cities is even higher than the sum of the background urban heat island effect and the heat wave effect.This means cities are even more vulnerable to HWs than other environments.A number of past studies have documented such urban effects.For example, a high-resolution (1 km) simulation of HW events concluded that the daily mean UHI in New York city increased the temperature by 1.5 • C [8]. Analysis of Beijing flux observations found that changes in the surface energy balance under HWs are responsible for the intensification of UHIs under HWs [9,10], with the role of wind speed being important but varying between cities [11].
The effects of urbanization on temperature and precipitation extend to surrounding rural areas and nearby downwind cities [12,13].For example, numerical modelling suggests Shanghai increases the 2 m air temperature of downwind Kunshan (10 km west of Shanghai) by 0.2-0.4• C in the afternoon and by 0.4-0.6 • C in the evening [14].It has been suggested UHI effects could possibly be reduced by 1.25 • C (25% reduction) if upwind urban areas were replaced with natural vegetation [15].Given the increasing frequency of urban conglomerations, sometimes called urban archipelagos, the impact of cities beyond their boundaries needs to be explored, especially associated with HWs.
Lowry [16] provides a framework to assess the total climate impacts (M) at a station (x) as a function of background climate (C), local landscape impacts (L), and local urbanization impacts (E) for different weather elements (i) for a time period (t): Urban impacts are easily detected if C and L do not change for the same t and x-or, if no urban impacts on climate exist at t = 0 when E itd = 0.In the latter case, upwind (u) − downwind (d) differences can be determined from Here, this framework is used to assess the impact of urbanization on downwind cities.
As urban expansion may bring some negative effects, various mitigation measures have been explored, such as green roofs (GRs).Sharma et al. [17] suggest that daytime roof temperatures can be reduced by 1-3 • C in the Chicago metropolitan area, and the amount will vary linearly with increasing green roof fractions based on urbanized weather research and forecasting (uWRF) simulations.The urban heat stress could potentially be almost completely offset if green roofs were irrigated in New York City and Phoenix during heat wave periods [18].Green roofs modify wind conditions, with the siting of GRs playing an important role under windy conditions [19].
Here, we explore the impact of urban expansion on cities during HWs, using the megacity Hangzhou (Figure 1) and its effect on a downwind city Haining (Figure 1) through numerical simulations.The specific objectives are to (1) analyze the effect of the upwind megacity size on the UHI size during HW conditions, (2) assess the evolution of spatial and temporal changes of temperature and humidity from horizontal advection from the megacity (Hangzhou) on the smaller city (Haining), and (3) evaluate the downwind effectiveness of a mitigation option (GRs) implemented in the upwind megacity.
The effects of urbanization on temperature and precipitation extend to surrounding rural areas and nearby downwind cities [12,13].For example, numerical modelling suggests Shanghai increases the 2 m air temperature of downwind Kunshan (10 km west of Shanghai) by 0.2-0.4°C in the afternoon and by 0.4-0.6 °C in the evening [14].It has been suggested UHI effects could possibly be reduced by 1.25 °C (25% reduction) if upwind urban areas were replaced with natural vegetation [15].Given the increasing frequency of urban conglomerations, sometimes called urban archipelagos, the impact of cities beyond their boundaries needs to be explored, especially associated with HWs.
Lowry [16] provides a framework to assess the total climate impacts (M) at a station (x) as a function of background climate (C), local landscape impacts (L), and local urbanization impacts (E) for different weather elements (i) for a time period (t): (1) Urban impacts are easily detected if C and L do not change for the same t and x-or, if no urban impacts on climate exist at t = 0 when  = 0.In the latter case, upwind (u) − downwind (d) differences can be determined from
As urban expansion may bring some negative effects, various mitigation measures have been explored, such as green roofs (GRs).Sharma et al. [17] suggest that daytime roof temperatures can be reduced by 1-3 °C in the Chicago metropolitan area, and the amount will vary linearly with increasing green roof fractions based on urbanized weather research and forecasting (uWRF) simulations.The urban heat stress could potentially be almost completely offset if green roofs were irrigated in New York City and Phoenix during heat wave periods [18].Green roofs modify wind conditions, with the siting of GRs playing an important role under windy conditions [19].
Here, we explore the impact of urban expansion on cities during HWs, using the megacity Hangzhou (Figure 1) and its effect on a downwind city Haining (Figure 1) through numerical simulations.The specific objectives are to (1) analyze the effect of the upwind megacity size on the UHI size during HW conditions, (2) assess the evolution of spatial and temporal changes of temperature and humidity from horizontal advection from the megacity (Hangzhou) on the smaller city (Haining), and (3) evaluate the downwind effectiveness of a mitigation option (GRs) implemented in the upwind megacity.

Study Area
Hangzhou (30 • 16 N, 120 • 12 E), the capital of Zhejiang Province, is located 180 km southwest of Shanghai in the Yangtze River Delta (Figure 1b).The city covers 4876 km 2 of the 16,596 km 2 total administrative area.In 2015, the population was 9 million (80% classified as urban) [21].The city experiences a subtropical climate (mean annual temperature = 16.2 • C), with daily averages ranging from 3.8 • C (winter) to 28.6 • C (summer) [22].Historically, Hangzhou was renowned for its pleasant weather [23], but in recent years there have been very hot summers (e.g., 42.8To analyze the impacts of upwind urban expansion on regional climate, two small cities near Hangzhou are studied (Figure 1b): (1) Haining (50 km northeast of Hangzhou, downwind) and ( 2) Fuyang (30 km southwest of Hangzhou, upwind).These two cities have approximately the same size, population, and climate.These us allow to consider upwind and downwind impacts of the climate (Equation 2).

Heat Wave Characteristics
The HWs are characterized using daily maximum air temperature data [24,25] for the period 1 January 1979 to 31 December 2014, using the Meehl and Tebaldi [26] HW definition selected within ExtremeFinder/Urban Multi-scale Environmental Predictor (UMEP) [27].HWs have become more frequent, especially after 2003, with events almost every year (Figure 2).Given that the 2013 HW had the longest duration (31 days) and was the hottest (40.6 • C), we analyze that period.During this HW, the average daily maximum temperature was 36.8 • C, with seven days >39 • C and three days >40 • C.

Study Area
Hangzhou (30°16′ N, 120° 12′ E), the capital of Zhejiang Province, is located 180 km southwest of Shanghai in the Yangtze River Delta (Figure 1b).The city covers 4876 km 2 of the 16,596 km 2 total administrative area.In 2015, the population was 9 million (80% classified as urban) [21].The city experiences a subtropical climate (mean annual temperature = 16.2 °C), with daily averages ranging from 3.8 °C (winter) to 28.6 °C (summer) [22].Historically, Hangzhou was renowned for its pleasant weather [23], but in recent years there have been very hot summers (e.g., 42 To analyze the impacts of upwind urban expansion on regional climate, two small cities near Hangzhou are studied (Figure 1b): (1) Haining (50 km northeast of Hangzhou, downwind) and ( 2) Fuyang (30 km southwest of Hangzhou, upwind).These two cities have approximately the same size, population, and climate.These us allow to consider upwind and downwind impacts of the climate (Equation 2).

Heat Wave Characteristics
The HWs are characterized using daily maximum air temperature data [24,25] for the period 1 January 1979 to 31 December 2014, using the Meehl and Tebaldi [26] HW definition selected within ExtremeFinder/Urban Multi-scale Environmental Predictor (UMEP) [27].HWs have become more frequent, especially after 2003, with events almost every year (Figure 2).Given that the 2013 HW had the longest duration (31 days) and was the hottest (40.6 °C), we analyze that period.During this HW, the average daily maximum temperature was 36.8 °C, with seven days >39 °C and three days >40 °C.

Simulation Configuration
To investigate this, we used the weather research and forecasting model (WRF) [28], coupled with the single layer urban canopy model (UCM).The single-layer UCM (urban canopy model)
To explore the impact of changing the roof materials, the Princeton roof model (PROM) [34] within the Princeton urban canopy model (PUCM) was used, as it allows the roof facet materials [2,[35][36][37][38] to be changed, while otherwise retaining the same physics as the UCM.
The Hangzhou area is modeled using three nested domains (horizontal grid resolutions of 9, 3, and 1 km; Figure 1) and IGBP (International Geosphere-Biosphere Programme) -Modified, MODIS (moderate resolution imaging spectroradiometer) 20-category land use [39].The outer domain (d01) covers most of eastern China; d02 includes Shanghai and most of Zhejiang, as well as a small part of Jiangsu and Anhui provinces; and d03 covers Hangzhou, most of Shaoxing, Jiaxing, and Huzhou, as well as part of the Zhoushan region.The analysis presented is for d03.One-way nesting with 85 sigma levels vertically below the upper boundary of 100 hPa are used in the simulations.The model is run only for part of the HW period (00:00 UTC 30 June 2013 to 00:00 UTC 7 July 2013), with the first 40 hours being model spin-up prior to the analysis periods beginning at 00:00 LST 2 July 2013.
NCEP (National Centers for Environmental Prediction) FNL (final) operational global analysis data [40] are used for both the initial and the boundary conditions.The model physical parameterization schemes chosen are (1) two-dimensional (2D) Smagorinsky [41], the Smagorinsky [42] scheme for horizontal diffusion; (2) the Mellor-Yamada-Janjić planetary boundary layer (PBL) scheme [43,44]; (3) the unified Noah land-surface model; and (4) the rapid radiative transfer model [45] for longwave radiation and the Dudhia [46] scheme for shortwave radiation.The cumulus parameterization option is off for all domains, as even the largest grid size (9 km) is less than 10 km [47], and there was no precipitation during the simulation period.Only the roof parameters differ between the UCM and PUCM/PROM simulations (Table 1, [2,29]).The latter allows GRs to be simulated.

Numerical Experimental Design
To assess the impact of expansion of the Hangzhou urban area, four different urban extents (UE) are simulated (Figure 3): 1.
UE 1.0 : current urban extent (IGBP-Modified MODIS 20-category data [48]); 2. UE 1.5 : a 1 km outward expansion in both the east-west and north-south directions (water body grids remain unchanged), creating an urban area 152.3% of the present-day extent; 3. UE 0.5 : an urban contraction in both directions, to become 57% of the current extent (area is replaced with crops, water areas unchanged); 4. UE 0.0 : all urban grids (Figure 3) changed to crops.
The expansion process is similar to the actual urbanization of Hangzhou [24], and some general conclusions are obtained under this type of urban growth.
To consider the potential of green roofs to mitigate HW effects, the GR are assumed to have an initial soil moisture of 0.3 m 3 m −3 (~50% saturation, i.e., very well irrigated), to ensure that evaporation [48] is considered over three areal extents: 1.
GR d03 : GRs added to all roofs in the d03 urban grids (Figure 4a) 2.
GR HZ : GRs added to all roofs in the Hangzhou urban grids only (Figure 4b) 3.
Atmosphere 2018, 9, x FOR PEER REVIEW 5 of 17 To consider the potential of green roofs to mitigate HW effects, the GR are assumed to have an initial soil moisture of 0.3 m 3 m −3 (~50% saturation, i.e., very well irrigated), to ensure that evaporation [48] is considered over three areal extents: 1. GRd03: GRs added to all roofs in the d03 urban grids (Figure 4a) 2. GRHZ: GRs added to all roofs in the Hangzhou urban grids only (Figure 4b) 3. GRNO: no GRs (Figure 4c).

Evaluation of Model Performance
Near-surface observations at 13 China Meteorological Administration (CMA) meteorological stations in d03 (Figure 1b) over 48 h (00:00 4 July 2013 to 00:00 6 July 2013 UTC) during this HW were used to assess the performance of WRF-UCM (or WRF-PUCM/PROM with no GR).The correlation coefficients of 2 m air temperatures are mostly greater than 0.9 (Figure 5a), but the 2 m temperature   To consider the potential of green roofs to mitigate HW effects, the GR are assumed to have an initial soil moisture of 0.3 m 3 m −3 (~50% saturation, i.e., very well irrigated), to ensure that evaporation [48] is considered over three areal extents: 1. GRd03: GRs added to all roofs in the d03 urban grids (Figure 4a) 2. GRHZ: GRs added to all roofs in the Hangzhou urban grids only (Figure 4b) 3. GRNO: no GRs (Figure 4c).

Evaluation of Model Performance
Near-surface observations at 13 China Meteorological Administration (CMA) meteorological stations in d03 (Figure 1b) over 48 h (00:00 4 July 2013 to 00:00 6 July 2013 UTC) during this HW were used to assess the performance of WRF-UCM (or WRF-PUCM/PROM with no GR).The correlation coefficients of 2 m air temperatures are mostly greater than 0.9 (Figure 5a), but the 2 m temperature

Evaluation of Model Performance
Near-surface observations at 13 China Meteorological Administration (CMA) meteorological stations in d03 (Figure 1b) over 48 h (00:00 4 July 2013 to 00:00 6 July 2013 UTC) during this HW were used to assess the performance of WRF-UCM (or WRF-PUCM/PROM with no GR).The correlation coefficients of 2 m air temperatures are mostly greater than 0.9 (Figure 5a), but the 2 m temperature is underestimated (mean bias error (MBE) = −1.03• C).This is consistent with previous findings [49,50].This leads to overestimation of 2 m relative humidity (MBE = 14.54%), but generally reasonable correlation coefficients (>0.8).The 10-m wind speed correlation coefficients are mostly between 0.4 to 0.8.The wind direction hit rate (HR, [51]) for within 30 • of observed values is 45%.Unfortunately, WRF-PROM with larger GR extents cannot be evaluated, as there are currently insufficient GRs in the area to evaluate with meteorological observation sites available.Prior to this, WRF-PUCM/PROM evaluations at both Tsinghua University in Beijing, China and Princeton University in New Jersey, United States, found that PROM is able to capture the diurnal cycle of roof temperatures and the soil moisture dynamics of green roofs with high accuracy.[38].
Of interest are the sites in Hangzhou (#10) and in the downwind city Haining (#8).For both there is very good performance (correlation coefficients are >0.95 and >0.9, respectively) for both 2 m air temperature and 2 m relative humidity, with consistent performance across the diurnal cycle as well.
Unfortunately, WRF-PROM with larger GR extents cannot be evaluated, as there are currently insufficient GRs in the area to evaluate with meteorological observation sites available.Prior to this, WRF-PUCM/PROM evaluations at both Tsinghua University in Beijing, China and Princeton University in New Jersey, United States, found that PROM is able to capture the diurnal cycle of roof temperatures and the soil moisture dynamics of green roofs with high accuracy.[38].
Of interest are the sites in Hangzhou (#10) and in the downwind city Haining (#8).For both there is very good performance (correlation coefficients are >0.95 and >0.9, respectively) for both 2 m air temperature and 2 m relative humidity, with consistent performance across the diurnal cycle as well.[52] plots are for hourly data with correlation coefficients (polar axis), with normalized standard deviation (horizontal axis) and normalized root mean square error (RMSE) (internal circular axes).The overall, cross-site, mean bias error (MBE) and RMSE are also indicated.

Impacts of the Upwind Megacity Size
To examine the impacts of increasing city size during an HW period (Figure 6), the variables analyzed are the surface energy balance fluxes (storage heat flux, turbulent sensible, and latent heat fluxes) and the resulting impacts from the surface energy balance flux changes (i.e., surface temperature, 2 m air temperature, 2 m specific humidity, and 10 m wind speed).The variables are analyzed spatially, using transects that are perpendicular to the predominantly southwesterly wind direction (Figure 3c) (background wind, Appendix A).The spatial means (5 km) are determined from the 1 km grids for 3 h time periods.This gives eight time series for each variable investigated.
With the expansion of the urban area (UE 1.5 -UE 1.0 difference), the storage heat flux has an enhanced diurnal cycle, with larger positive values during the day and larger negative values at night.As expected, there are larger storage heat fluxes in the urban area.Similarly, the turbulent sensible heat fluxes increase and latent heat fluxes decrease with the reduction in vegetation and therefore soil moisture in the urban areas.
Atmosphere 2018, 9, x FOR PEER REVIEW 7 of 17 To examine the impacts of increasing city size during an HW period (Figure 6), the variables analyzed are the surface energy balance fluxes (storage heat flux, turbulent sensible, and latent heat fluxes) and the resulting impacts from the surface energy balance flux changes (i.e., surface temperature, 2 m air temperature, 2 m specific humidity, and 10 m wind speed).The variables are analyzed spatially, using transects that are perpendicular to the predominantly southwesterly wind direction (Figure 3c) (background wind, Appendix A).The spatial means (5 km) are determined from the 1 km grids for 3 h time periods.This gives eight time series for each variable investigated.
With the expansion of the urban area (UE1.5-UE1.0difference), the storage heat flux has an enhanced diurnal cycle, with larger positive values during the day and larger negative values at night.As expected, there are larger storage heat fluxes in the urban area.Similarly, the turbulent sensible heat fluxes increase and latent heat fluxes decrease with the reduction in vegetation and therefore soil moisture in the urban areas.Daytime surface temperature differences are warmer, but 2 m air temperature decreases, with even larger changes at night, as urban areas absorb more heat during the day and release it at night.This results in an increased 2 m nocturnal air temperature.Urban development also influences humidity, with a more evident "dry island" as the modeled city grows.A decrease of 3.5 g kg −1 in the daytime of the 2 m specific humidity is simulated between UE0.0 and UE1.0 (Figure 6a).As the 2 m temperature increases, the specific humidity is reduced, especially during daytime when the temperature is relatively high.Wind speed is also affected by urbanization.The urban expansion is associated with larger nocturnal wind speeds.This is consistent with Kang and Lenschow's [53] findings (WRF-LES (large eddy simulation) simulations) of surface heterogeneity causing larger winds perpendicular to the mesoscale wind direction.Similarly, in the Hangzhou simulation, high pressure influencing the region creates a continuous background wind that is enhanced by the urbaninduced thermal difference.The strongest wind speed difference occurs at night, associated with the greater thermal heterogeneity.Daytime surface temperature differences are warmer, but 2 m air temperature decreases, with even larger changes at night, as urban areas absorb more heat during the day and release it at night.This results in an increased 2 m nocturnal air temperature.Urban development also influences humidity, with a more evident "dry island" as the modeled city grows.A decrease of 3.5 g kg −1 in the daytime of the 2 m specific humidity is simulated between UE 0.0 and UE 1.0 (Figure 6a).As the 2 m temperature increases, the specific humidity is reduced, especially during daytime when the temperature is relatively high.Wind speed is also affected by urbanization.The urban expansion is associated with larger nocturnal wind speeds.This is consistent with Kang and Lenschow's [53] findings (WRF-LES (large eddy simulation) simulations) of surface heterogeneity causing larger winds perpendicular to the mesoscale wind direction.Similarly, in the Hangzhou simulation, high pressure influencing the region creates a continuous background wind that is enhanced by the urban-induced thermal difference.The strongest wind speed difference occurs at night, associated with the greater thermal heterogeneity.
The influence of land use changes is consistent between variables along the cross-section (Figure 6a), with the unchanged water bodies evident.Comparing Figure 6a,b, the trend for the three land use scenarios is similar through the day.However, the location of the changes varies, from all of Hangzhou (Figure 6a: urban expansion, UE 1.0 − UE 0.0 ) to mainly on the edge of Hangzhou (Figure 6b: UE 1.0 − UE 0.5 ).This is as expected with the land use changes simulated.Note that the variables in the center of Hangzhou are almost unaffected (Figure 6b,c), suggesting that the impact of urban expansion on surrounding cities is not obvious in the direction perpendicular to the background wind.
Potential temperatures within the planetary boundary layer (PBL) change between the land use scenarios (Figure 7), with larger differences during the day than at night.The maximum near surface difference (0.2 • C) is at 13:00.The changes in potential temperatures are consistent with PBL height changes, indicating that urban expansion modifies the thermal regime throughout the PBL.A deeper PBL can result in enhanced heating because of entrainment.A decrease of 0.1 in potential temperature is simulated at 23:00 near the top of the PBL (~200 m agl WRF determined), and is attributed to the temperature inversion in the entrainment zone.
Atmosphere 2018, 9, x FOR PEER REVIEW 8 of 17 The influence of land use changes is consistent between variables along the cross-section (Figure 6a), with the unchanged water bodies evident.Comparing Figures 6a and 6b, the trend for the three land use scenarios is similar through the day.However, the location of the changes varies, from all of Hangzhou (Figure 6a: urban expansion, UE1.0 − UE0.0) to mainly on the edge of Hangzhou (Figure 6b: UE1.0 − UE0.5).This is as expected with the land use changes simulated.Note that the variables in the center of Hangzhou are almost unaffected (Figure 6b,c), suggesting that the impact of urban expansion on surrounding cities is not obvious in the direction perpendicular to the background wind.
Potential temperatures within the planetary boundary layer (PBL) change between the land use scenarios (Figure 7), with larger differences during the day than at night.The maximum near surface difference (0.2 °C) is at 13:00.The changes in potential temperatures are consistent with PBL height changes, indicating that urban expansion modifies the thermal regime throughout the PBL.A deeper PBL can result in enhanced heating because of entrainment.A decrease of 0.1 °C in potential temperature is simulated at 23:00 near the top of the PBL (~200 m agl WRF determined), and is attributed to the temperature inversion in the entrainment zone.[44] and Janjić [45] planetary boundary layer (PBL) scheme.

Impacts on the Downwind Regions
The simulations show that urbanization of Hangzhou has a negative impact (i.e., warming), not only on Hangzhou but also the surrounding area, especially downwind of Haining (Figure 1b).The impact on Haining varies between UE0.5, UE1.0 and UE1.5 (Figure 8).Following Lowry's [16] methodology (Equations 1 and 2), analyzing changes in Fuyang (a small city upwind of Hangzhou, Figure 1b) is useful to rule out other influences.In Fuyang, the urban expansion in Hangzhou results in negligible changes to storage, as well as turbulent-sensible and latent heat fluxes.However, differences in temperature, humidity, and wind speed from urban expansion are much more evident in Haining (downwind of Hanghzou) than in Fuyang, highlighting the impact of urbanization downwind.The 2 m temperature changes were mainly at night, while specific humidity changed mainly during the day.This is consistent with the temporal pattern (Figure 6), supporting the explanation that the cause is the expansion of Hangzhou.[44] and Janjić [45] planetary boundary layer (PBL) scheme.

Impacts on the Downwind Regions
The simulations show that urbanization of Hangzhou has a negative impact (i.e., warming), not only on Hangzhou but also the surrounding area, especially downwind of Haining (Figure 1b).The impact on Haining varies between UE 0.5 , UE 1.0 and UE 1.5 (Figure 8).Following Lowry's [16] methodology (Equations 1 and 2), analyzing changes in Fuyang (a small city upwind of Hangzhou, Figure 1b) is useful to rule out other influences.In Fuyang, the urban expansion in Hangzhou results in negligible changes to storage, as well as turbulent-sensible and latent heat fluxes.However, differences in temperature, humidity, and wind speed from urban expansion are much more evident in Haining (downwind of Hanghzou) than in Fuyang, highlighting the impact of urbanization downwind.The 2 m temperature changes were mainly at night, while specific humidity changed mainly during the day.This is consistent with the temporal pattern (Figure 6), supporting the explanation that the cause is the expansion of Hangzhou.The vertical cross-section (Figure 2, transect AB) of the difference in 2 m air temperatures between the UE0.5 and UE1.0 scenarios has a clear nocturnal increase of more than 1 °C in the rural region that has become urban (Figure 9).This impact extends into areas where land use has not changed, nearly 30 km upwind and more than 50 km downwind.The maximum temperature difference downwind (0.36 °C) is almost twice of the difference at an equivalent distance upwind (0.19 °C).The spatial extent of the influence upwind and the size of the temperature difference are both larger at night, which is consistent with the diurnal cycle of air temperature change in Hangzhou.However, the extent of influence downwind is large during the day because of stronger winds (i.e., a secondary impact in the downwind regions).The result, between UE1.0 and UE1.5, also gets an approximate conclusion.The vertical cross-section (Figure 2, transect AB) of the difference in 2 m air temperatures between the UE 0.5 and UE 1.0 scenarios has a clear nocturnal increase of more than 1 • C in the rural region that has become urban (Figure 9).This impact extends into areas where land use has not changed, nearly 30 km upwind and more than 50 km downwind.The maximum temperature difference downwind (0.36 • C) is almost twice of the difference at an equivalent distance upwind (0.19 • C).The spatial extent of the influence upwind and the size of the temperature difference are both larger at night, which is consistent with the diurnal cycle of air temperature change in Hangzhou.However, the extent of influence downwind is large during the day because of stronger winds (i.e., a secondary impact in the downwind regions).The result, between UE 1.0 and UE 1.5 , also gets an approximate conclusion.Turbulent sensible heat flux increases by more than 160 W m −2 in the daytime in rural regions that are urbanized (Figure 10).This is consistent with the changes in air temperature (Figure 9).However, the sensible heat flux difference in the downwind urban area, where the 2 m air temperature increases, is close to zero, suggesting that it is not driving the 2 m temperature change downwind.Similarly, small changes occur in the storage and latent heat fluxes (Figure 8).This supports that advection is the critical influence with regard to relative humidity changing with the increasing temperature.Turbulent sensible heat flux increases by more than 160 W m −2 in the daytime in rural regions that are urbanized (Figure 10).This is consistent with the changes in air temperature (Figure 9).However, the sensible heat flux difference in the downwind urban area, where the 2 m air temperature increases, is close to zero, suggesting that it is not driving the 2 m temperature change downwind.Similarly, small changes occur in the storage and latent heat fluxes (Figure 8).This supports that advection is the critical influence with regard to relative humidity changing with the increasing temperature.

Can these Urban Expansion Impacts Be Mitigated with Green Roofs?
Green roofs (GRs) are one potential strategy to mitigate temperature effects in and downwind of a city.Analysis of the difference between simulations when all the urban roofs in d03 are assumed to be green (GRd03) and when none are green (GRNO) (Section 2.5) shows a reduction in sensible heat flux (maximum 55 W m −2 ) and an increase in latent heat flux (maximum 85 W m −2 ).For both fluxes, the largest difference occurs in the daytime.Associated with these changes in turbulent fluxes are small reductions in surface temperature during the day and night of 1.5 °C and 0.5 °C, respectively.This is consistent with the sensible heat flux changes.The effect of GRs is to reduce the amplitude of the diurnal cycle of the storage heat flux.This is attributed to an increase in heat storage capacity in urban areas by green roofs (especially while wet).Thus, in these simulations GRs do mitigate UHI effects, and could reduce surface temperature in cities during the daytime (if sufficient water was available).The size of the changes is potentially large enough to have an impact-for example, Anderson [1] suggests that a 0.5 °C decrease in 2 m air temperature (Figure 11e) decreased mortality risk by 2.25% during a HW in the United States.Other impacts simulated include an increase of 0.6 g kg −1 , 2 m-specific humidity, as well as a decrease of 0.4 m s −1 , 10 m wind speed during the daytime.The reduction of wind speed may improve thermal comfort with less advected heat into the urban area.However, the higher humidity and lower wind speed may not be more pleasant.

Can these Urban Expansion Impacts Be Mitigated with Green Roofs?
Green roofs (GRs) are one potential strategy to mitigate temperature effects in and downwind of a city.Analysis of the difference between simulations when all the urban roofs in d03 are assumed to be green (GR d03 ) and when none are green (GR NO ) (Section 2.5) shows a reduction in sensible heat flux (maximum 55 W m −2 ) and an increase in latent heat flux (maximum 85 W m −2 ).For both fluxes, the largest difference occurs in the daytime.Associated with these changes in turbulent fluxes are small reductions in surface temperature during the day and night of 1.5 • C and 0.5 • C, respectively.This is consistent with the sensible heat flux changes.The effect of GRs is to reduce the amplitude of the diurnal cycle of the storage heat flux.This is attributed to an increase in heat storage capacity in urban areas by green roofs (especially while wet).Thus, in these simulations GRs do mitigate UHI effects, and could reduce surface temperature in cities during the daytime (if sufficient water was available).The size of the changes is potentially large enough to have an impact-for example, Anderson [1] suggests that a 0.5 • C decrease in 2 m air temperature (Figure 11e) decreased mortality risk by 2.25% during a HW in the United States.Other impacts simulated include an increase of 0.6 g kg −1 , 2 m-specific humidity, as well as a decrease of 0.4 m s −1 , 10 m wind speed during the daytime.The reduction of wind speed may improve thermal comfort with less advected heat into the urban area.However, the higher humidity and lower wind speed may not be more pleasant.6, but between GRd03 (green roofs added to all roofs in the d03 urban grids) and GRNO (no green roofs) scenarios (Section 2.5).
The results also indicate that GRs modify the potential temperature and PBL height (Figure 12).The GRs in upwind cities (GRHZ scenario, Figure 12a) result in a reduction of potential temperature (0.2 °C) and a decrease in the daytime PBL height (~50 m).Effects such as this are important to explore in the context of air quality.The GRs could alleviate the thermal effects of urban expansion on the downwind cities.The maximum potential temperature difference in Haining was 0.25 °C (GRHZ and GRNO scenarios, see Figure 12b), which is greater than the change in Hangzhou itself.As the timing of maximum differences vary (Hangzhou 13:00, Haining 17:00), this suggests the mitigation in Haining is due to advection.The GRs decrease the wind speed in downwind cities.
The green roofs greatly reduce the air temperature of Hangzhou and the downwind city Haining, and can be used to mitigate thermal effects of regional urban expansion.This result is expected to have important implications for planning scenarios and decisions regarding where mitigation strategies have maximum effects, both locally and regionally (downwind).6, but between GR d03 (green roofs added to all roofs in the d03 urban grids) and GR NO (no green roofs) scenarios (Section 2.5).
The results also indicate that GRs modify the potential temperature and PBL height (Figure 12).The GRs in upwind cities (GR HZ scenario, Figure 12a) result in a reduction of potential temperature (0.2 • C) and a decrease in the daytime PBL height (~50 m).Effects such as this are important to explore in the context of air quality.The GRs could alleviate the thermal effects of urban expansion on the downwind cities.The maximum potential temperature difference in Haining was 0.25 • C (GR HZ and GR NO scenarios, see Figure 12b), which is greater than the change in Hangzhou itself.As the timing of maximum differences vary (Hangzhou 13:00, Haining 17:00), this suggests the mitigation in Haining is due to advection.The GRs decrease the wind speed in downwind cities.
The green roofs greatly reduce the air temperature of Hangzhou and the downwind city Haining, and can be used to mitigate thermal effects of regional urban expansion.This result is expected to have important implications for planning scenarios and decisions regarding where mitigation strategies have maximum effects, both locally and regionally (downwind).

Concluding Remarks
This study investigates the impacts of upwind urbanization on the UHI effects of downwind cities, by conducting numerical simulations in the Yangtze River Delta Region for the heat wave conditions of 2013.Four scenarios of Hangzhou's extent are modelled.From analysis of these results, the following conclusions are drawn: • Without the urban surface of Hangzhou, the 2 m temperature is 5 °C lower at night, and the 2 m specific humidity is 3.5 g kg −1 higher during the daytime, compared to the current urban extent (UE1.0).

•
With increasing urban expansion, an increase of nighttime 2 m air temperature of 2.7 °C for UE0.5 to UE1.0, and 1.7 °C for UE1.0 to UE1.5, as well as a 2.1 g kg -1 and 1.4 g kg -1 decrease, respectively, of daytime 2 m specific humidity are predicted.

•
Greater heat flux heterogeneity caused by urban areas leads to an increase of 1.0 and 0.8 m s -1 for 10 m wind speed in Hangzhou.

•
Comparison of the conditions for a small city upwind (Fuyang) and downwind (Haining) of the megacity (Hangzhou) indicates large impacts from upwind urban expansion on regional climate.

Concluding Remarks
This study investigates the impacts of upwind urbanization on the UHI effects of downwind cities, by conducting numerical simulations in the Yangtze River Delta Region for the heat wave conditions of 2013.Four scenarios of Hangzhou's extent are modelled.From analysis of these results, the following conclusions are drawn:

•
Without the urban surface of Hangzhou, the 2 m temperature is 5 • C lower at night, and the 2 m specific humidity is 3.5 g kg −1 higher during the daytime, compared to the current urban extent (UE 1.0 ).

•
With increasing urban expansion, an increase of nighttime 2 m air temperature of 2.7 • C for UE 0.5 to UE 1.0 , and 1.7 • C for UE 1.0 to UE 1.5 , as well as a 2.1 g kg −1 and 1.4 g kg −1 decrease, respectively, of daytime 2 m specific humidity are predicted.

•
Greater heat flux heterogeneity caused by urban areas leads to an increase of 1.0 and 0.8 m s −1 for 10 m wind speed in Hangzhou.

•
Comparison of the conditions for a small city upwind (Fuyang) and downwind (Haining) of the megacity (Hangzhou) indicates large impacts from upwind urban expansion on regional climate.

•
Given the predominant southwesterly winds, the urbanization in Hangzhou increases 2 m air temperature in Haining by about 0.3 • C between UE 0.5 and UE 1.0 , and 0.4 • C between UE 1.0 and UE 1.5 , while the increases are about 0.05 • C and 0.1 • C, respectively, for Fuyang.

•
The strongest and widest warming effect appears at 0700 LST, and the weakest effect occurs in the afternoon.Similar results are predicted for 2 m specific humidity and 10 m wind speed.
The driving heat flux changes in Haining are less than 5 W m −2 , but vary by hundreds of W m −2 in Hangzhou.This suggests that it is advection that give rise to the atmospheric changes in Haining.
One mitigation strategy (green roofs) is examined in the upwind megacity.From the simulations, the following conclusions are drawn:

•
Green roof coverage of 50% in the d03 area could reduce the daytime 2 m air temperature in Hangzhou by 0.5 • C, and increase the 2 m specific humidity by 0.6 g kg −1 .Such changes effectively alleviate the UHI effect and the "dry island" effect within Hangzhou.

•
Green roofs in Hangzhou can relieve the thermal stress in the downwind city Haining by decreasing the potential temperature by 0.25 • C.

Figure 2 .
Figure 2. Heat waves in Hangzhou (1 January 1980 to 31 December 2014) by year: (a) duration and (b) days (yellow boxes) and their maximum temperature.See Section 2.2 for methods.

Figure 2 .
Figure 2. Heat waves in Hangzhou (1 January 1980 to 31 December 2014) by year: (a) duration and (b) days (yellow boxes) and their maximum temperature.See Section 2.2 for methods.

Figure 3 .
Figure 3. IGBP-modified MODIS 20-category land use [48] in (a) d03 (Figure 1a) with the location of the area modified in the four scenarios (Section 2.5) (white rectangle, detail in (b)); cross-section analyzed (yellow rectangle, detail in (c) of modified urban extent); and the vertical cross-section (yellow line AB, see Figures 9 and 10); (b) modified (Section 2.5) land use; and (c) detail of area analyzed for the four scenarios (in Figure 6 and 11).

Figure 3 .
Figure 3. IGBP-modified MODIS 20-category land use [48] in (a) d03 (Figure 1a) with the location of the area modified in the four scenarios (Section 2.5) (white rectangle, detail in (b)); cross-section analyzed (yellow rectangle, detail in (c) of modified urban extent); and the vertical cross-section (yellow line AB, see Figures 9 and 10); (b) modified (Section 2.5) land use; and (c) detail of area analyzed for the four scenarios (in Figures 6 and 11).
Figure 3. IGBP-modified MODIS 20-category land use [48] in (a) d03 (Figure 1a) with the location of the area modified in the four scenarios (Section 2.5) (white rectangle, detail in (b)); cross-section analyzed (yellow rectangle, detail in (c) of modified urban extent); and the vertical cross-section (yellow line AB, see Figures 9 and 10); (b) modified (Section 2.5) land use; and (c) detail of area analyzed for the four scenarios (in Figures 6 and 11).

Figure 3 .
Figure 3. IGBP-modified MODIS 20-category land use [48] in (a) d03 (Figure 1a) with the location of the area modified in the four scenarios (Section 2.5) (white rectangle, detail in (b)); cross-section analyzed (yellow rectangle, detail in (c) of modified urban extent); and the vertical cross-section (yellow line AB, see Figures 9 and 10); (b) modified (Section 2.5) land use; and (c) detail of area analyzed for the four scenarios (in Figure 6 and 11).

Figure 4 .
Figure 4. Land use with GRs implemented in (a) GR d03 all d03 urban grid roofs, (b) GR HZ roofs in Hangzhou urban grids, (c) GR NO no green roofs.

Figure 10 .
Figure 10.The same as Figure 9, but with surface turbulent sensible heat flux (W m −2 ).The values between -5 W m −2 to 5 W m -2 are masked.

Figure 10 .
Figure 10.The same as Figure 9, but with surface turbulent sensible heat flux (W m −2 ).The values between -5 W m −2 to 5 W m −2 are masked.

Figure 11 .
Figure11.The same as Figure6, but between GRd03 (green roofs added to all roofs in the d03 urban grids) and GRNO (no green roofs) scenarios (Section 2.5).

Figure 11 .
Figure11.The same as Figure6, but between GR d03 (green roofs added to all roofs in the d03 urban grids) and GR NO (no green roofs) scenarios (Section 2.5).

Figure 12 .
Figure 12.The same as Figure 7, but for GRNO and GRHZ (all green roofs in Hangzhou urban grids) scenarios, in (a) Hangzhou and (b) Haining (see Figure 1b).

•
Given the predominant southwesterly winds, the urbanization in Hangzhou increases 2 m air temperature in Haining by about 0.3 °C between UE0.5 and UE1.0, and 0.4 °C between UE1.0 and UE1.5, while the increases are about 0.05 °C and 0.1 °C, respectively, for Fuyang.• The strongest and widest warming effect appears at 0700 LST, and the weakest effect occurs in the afternoon.Similar results are predicted for 2 m specific humidity and 10 m wind speed.The driving heat flux changes in Haining are less than 5 W m -2 , but vary by hundreds of W m −2 in Hangzhou.This suggests that it is advection that give rise to the atmospheric changes in Haining.One mitigation strategy (green roofs) is examined in the upwind megacity.From the simulations, the following conclusions are drawn: • Green roof coverage of 50% in the d03 area could reduce the daytime 2 m air temperature in Hangzhou by 0.5 °C, and increase the 2 m specific humidity by 0.6 g kg −1 .Such changes effectively alleviate the UHI effect and the "dry island" effect within Hangzhou.• Green roofs in Hangzhou can relieve the thermal stress in the downwind city Haining by decreasing the potential temperature by 0.25 °C.• The strongest mitigation effects appear at 1300 LST in Hangzhou and 1600 LST in Haining.

Figure 12 .
Figure12.The same as Figure7, but for GR NO and GR HZ (all green roofs in Hangzhou urban grids) scenarios, in (a) Hangzhou and (b) Haining (see Figure1b).