Interactions between urban heat islands and heat waves

We used the Community Earth System Model (CESM) (Hurrell et al 2013) to simulate UHIs for 50 selected cities in the United States. We conducted simulations for present-day climate conditions (1972–2004) and two future Representative Concentration Pathway (RCP) scenarios (RCP 4.5 and 8.5 for 2005–2100) at a horizontal resolution of 0.9^o latitude × 1.25^o longitude. Please see the supplementary information for more details on the model simulations.

We used thirty years of modeled outputs in the current-climate run (1975−2004) and two RCP runs (2071−2100) to represent the present-day and future climatology, respectively, and to analyze UHI intensity. Because the focus of this study is HWs, we restrict our analysis to summer months (June−August). We use hourly outputs at 13:00−15:00 (around midday) and 01:00−03:00 (around midnight) local time to represent daytime and nighttime conditions, respectively. These two sets of times were selected for two reasons. First, they are relatively close to the times of daily maximum and minimum temperature for most of the cities, giving a better representation of the diurnal range of the UHI. Second, the performance of the CLM and the biophysical factorization method (described in the next section) was validated against MODIS observations around these two times (close to the MODIS-Aqua overpassing times) in Zhao et al (2014). We selected 50 cities in the United States (figure 1) representing three Köppen–Geiger climate zones: temperate climate (21 cities, eastern and southern US), continental climate (14 cities; northeastern US) and dry climate (15 cities; arid and semiarid western US). We focus on temperate and dry regions to draw out the humidity contrasts between these two distinct zones. We also selected cities in the continental climate because they experience deadly heat as well, despite residing in an annual average colder climate (Semenza et al 1996, Whitman et al 1997, Houser et al 2015).

We computed UHI_2m (ΔT_2m) and UHI_s (ΔT_s) from the variables generated by the urban and rural sub-grids in the grid cells where our selected cities are located (please see the supplementary information for more details). Although the UHI_2m and UHI_s differ in various aspects and their magnitudes are not directly comparable (Arnfield 2003), the two UHIs show consistency in terms of the linear supposition property of their biophysical contributions (Zhao et al 2017). One merit of UHI_s compared to UHI_2m is that the former has a firm and yet simple theoretical basis derived from the surface energy balance principle that underpins the biophysical factorization (described in the next section). In this study, we use this biophysical factorization method to disentangle the mechanism underlying the UHI–HW interactions.

There are multiple ways to define HWs (Perkins 2015, Perkins and Alexander 2013, Robinson 2001, Smith et al 2013). We use the definition from the US National Weather Service (NWS): three or more consecutive days of maximum temperature reaching at least 90 °F (32.2 °C). We consider that, in each grid cell (a size on the order of 100 × 100 km), its rural sub-grid represents a local background environment for the city. Therefore, for each city we use its rural 2m-height temperature (T_2m,rural) to define HWs. This is consistent with weather station observations, as weather stations are mostly located in an open landscape (World Meteorological Organization standard) similar to a rural environment. Using this definition, we divided the 30 years of daily data in each run into two groups: (i) for each city, all sets of three or more consecutive days of modeled midday T_2m,rural larger than 32.2 °C were defined as HW days, and their UHI intensities were denoted as UHI_HW, while (ii) the remaining days were considered as normal days and hence their UHIs were denoted as UHI_NM.

To ensure the robustness of our analysis to the variations in HW definition, we examine two additional HW definitions (chosen arbitrarily to some extent), one absolute and one relative. The relative definition uses a relative threshold for each city to define HWs (denoted as Def-95Q) rather than a fixed number. Specifically, based on the histogram of summer daily T_2m,rural for each grid cell in each simulation, we define all days whose T_2m,rural are larger than the 95th percentile as HW days. The absolute one relaxes the minimum number of consecutive days of high temperature (90 °F) to two days (denoted as Def-2Day). We repeated our analysis for these two HW definitions to check the consistency of patterns identified in the main analysis.

Zoom In Zoom Out Reset image size

Total UHI_s (ΔT_s) can be approximated by linear supposition of various biophysical contributions: urban-rural changes in radiation balance, aerodynamic resistance, Bowen ratio, heat storage and anthropogenic heat (Zhao et al 2014). More details on the calculation can be found in the supplementary information. We quantified each biophysical contribution to ΔT_s and compared them between normal and HW days. These evaluations were performed using 30 years of midday and midnight data from the three simulations (current: 1975–2004; RCP 4.5 and RCP 8.5: 2071–2100) over all selected cities. These analyses allow us to contrast the biophysical drivers between normal and HW days across climate conditions and scenarios.

Zoom In Zoom Out Reset image size

We find strong diurnal and climatological signals in the interactions between UHI_2m and HWs (figure 2 and table 1). During daytime, UHI_2m is 0.4 ± 0.05 K (mean ± 1 standard error) or 80% higher during HW days than during normal days for cities in the temperate climate zone (humid eastern United States) under current climate (figure 2(a) and 3(a)). UHI_2m for these cities are significantly enhanced during HWs, indicating a significant synergistic effect between UHI_2m and HWs. This is consistent with previous findings in Baltimore/Washington metropolitan area in the US (Li and Bou-Zeid 2013) and Beijing in China (Li et al 2015). It is interesting that this synergistic effect is diminishing in future warmer climates. Near the end of this century, the average UHI_2m for cities in this region is projected to be 0.3 ± 0.08 K and 0.4 ± 0.09 K lower during HWs than during normal days under RCP 4.5 and RCP 8.5, respectively (figure 3(a) and 2(c)). This indicates negative feedbacks between UHI_2m and HWs in a future warmer world. Although our analysis and previous studies (Oleson 2012, Zhao et al 2017) have shown that different RCP scenarios have different impacts on the UHI, we find only slight difference in UHI_2m–HW interactions between the two climate scenarios (figure 3(a)). The reason might lie in the calculation of the synergistic effects of HW and UHI. A warming scenario affects the UHI intensity in both normal and HW days, and thus part of the impacts are cancelled when the difference between UHI_2m,HW and UHI_2m,NM is calculated for a given warming scenario.

An opposite pattern occurs in the dry region. We find only a slight (~0.06 K) higher UHI_2m during HWs compared to normal days in the present-day climate during daytime (figure 2(a) and 3(a)). This difference is statistically insignificant (p value > 0.38), indicating an indiscernible synergistic effect for cities in this region under current conditions. However, unlike those in the temperate region, the cities in the dry group show a stronger synergistic effect in future scenarios (figure 3(a) and 2(c)). Near the end of this century, the average HW exacerbation of UHI_2m is projected to increase to 0.2 ± 0.06 K and 0.3 ± 0.05 K under RCP 4.5 and RCP 8.5, respectively. These numbers are statistically significant (p value < 0.01).

The diurnal pattern is asymmetric as well. At night, despite significant UHI_2m–HW synergistic effects across climate zones and scenarios, demonstrated by solid circles in figures 2(b) and (d) and consistent higher red bars than blue bars in figure 3(c), there are no discernible impacts of local background climate on the effects. In the present-day climate, the nighttime UHI_2m are intensified during HWs compared to normal conditions in temperate as well as dry region. Specifically, the average nighttime UHI_2m is 0.6 ± 0.2 K and 0.4 ± 0.1 K larger during HWs than during normal conditions in the temperate and dry region, respectively. These numbers are projected to increase to 1.1 ± 0.1 K (temperate) and 1.0 ± 0.0 K (dry) under RCP 4.5, and 1.2 ± 0.2 K (temperate) and 1.2 ± 0.1 K (dry) under RCP 8.5. These results indicate that the GHG warming scenarios exacerbate the nighttime UHI_2m—HW synergistic effects and that these effects remain minimally impacted by the background local climate in future scenarios.

We also analyzed the interactions between UHI_s and HWs. This indicator is important since the radiant temperature is a key component of thermal comfort to urban residents. Moreover, it is closely related to the surface energy balance and hence quite informative as a metric of the city-scale microclimate for the purpose of comparing different cities across diverse climate conditions.

We find that UHI_s and UHI_2m generally show consistent spatiotemporal patterns in their interaction with HWs, despite differing in magnitude. Figures 3(b) and (d) demonstrate the UHI_s—HW interactions for cities in the two regions during daytime and nighttime under current and future climates. In the daytime, the average UHI_s for temperate cities is 2.8 ± 0.8 K or 78% higher during HWs than during normal days in the present-day climate (figure 3(b)), indicating a significant synergistic effect. The magnitude of the effect is much larger than that estimated using UHI_2m. This is as expected because UHI_s is generally by nature larger than UHI_2m (Nichol et al 2009, Roth et al 1989). Consistent with the UHI_2m-HW interaction, the daytime synergistic effect in this region is projected to reverse in future warmer climates. Near the end of this century, the average UHI_s is 0.7 ± 0.3 K and 1.0 ± 0.4 K lower during HWs than during normal days under the RCP 4.5 and RCP 8.5, respectively (figure 3(b)).

Cities in the dry region do not show any daytime synergistic effects under current climate. The average daytime UHI_s is marginally lower during HWs than during normal days for the period 1975–2004 (figure 3(b)). Similar with the UHI_2m-HW interaction in this region, the synergistic effect develops in future warming scenarios. The UHI_s is 0.7 ± 0.2 K larger during HWs than during normal days near the end of this century under RCP 8.5 (figure 3(b)). At night, the synergistic effect is spatially consistent across climate regions. Under current climate, the difference between UHI_s,HW and UHI_s,NM is 1.8 ± 0.3 K and 1.0 ± 0.2 K for cities in the temperate and dry region respectively. These numbers are projected to be 1.4 ± 0.2 K and 1.7 ± 0.1 K near the end of this century under RCP 8.5.

Our analysis demonstrates that the spatiotemporal patterns of UHI-HW interactions are robust to different HW definitions (table 1). Although defining HW in the other two ways (Def-95Q and Def-2Day) affects the magnitude of the HW-UHI interactions, the spatiotemporal patterns agree well with results using Def-NWS. The consistency is confirmed for UHI_2m and UHI_s (table 1), strongly supporting the robustness of the aforementioned spatiotemporal patterns of the UHI-HW interactions.

Although UHI_s and UHI_2m are fundamentally distinct concepts (Arnfield 2003), they demonstrate consistent spatiotemporal patterns in terms of their interactions with HWs (figure 3). An important advantage of UHI_s is that it has a firm theoretical basis derived from the surface energy balance principle that explains the contributions of the biophysical properties of urban surfaces (Zhao et al 2014). Therefore, utilizing this biophysical factorization method, we can disentangle the relative importance of processes that contribute to the UHI_s–HW interactions. Because of the consistency between UHI_s and UHI_2m, and the physical links between surface and air temperature, the physical mechanism behind the UHI_2m–HW interactions can be inferred from this analysis of UHI_s, but not necessarily in a direct way. For example, a large contribution of convection to the cooling of the surface (reduced UHI_s–HW) would imply increased heating of the air (increased UHI_2m–HW).

Zoom In Zoom Out Reset image size

3.4.1. Daytime

Figure 4 illustrates the biophysical drivers of daytime UHI_s during HWs and normal days in current climate and RCP 8.5. Because UHI_s–HW interactions show slight differences between RCP 8.5 and RCP 4.5 (figure 3), we present the results of biophysical factorization for RCP 8.5 only. The bars of CLM-modeled ΔT_s are shown in figure 4 for comparison as a validation of our biophysical factorization. Under current climate, the positive UHI_s–HW synergistic effect in the temperate region primarily comes from the urban-rural difference in the contribution of evaporation (term 3 in supplementary equation (4)) and enhanced anthropogenic heat release (term 5 in equation (4)) during HWs compared to normal days (figure 4(a), compare the bars during HWs and normal conditions; and table 2). The contribution from evaporation to UHI_s is increased by 1.0 ± 0.1 K during HWs compared to normal days and anthropogenic heat by 1.0 ± 0.2 K in the current climate. This elevated contribution of evaporation during HWs is consistent with observational (Li et al 2015) and modeling studies (Li and Bou-Zeid 2013). Over a water-sufficient surface, an increase of air temperature favors increasing latent heat flux rather than increasing sensible heat flux (Bateni and Entekhabi 2012). In temperate climates, cities are usually water limited due to the large fraction of impervious surfaces; whereas their surrounding rural surfaces are considered as water sufficient because of ample precipitation in this region. During HWs (i.e. temperature increases) evaporation is increased over urban surfaces but not increased as much as over rural surfaces, thus resulting in an enhanced contribution from evaporation to UHI_s compared to normal days. The opposite occurs in the dry region under the present-day climate where urban and rural surfaces are water limited. Therefore, the enhanced evaporative contribution to UHI_s is weaker for cities in this region (figure 4(c) and table 2). The enhanced anthropogenic heat release during HWs, present in temperate and dry climate region (figures 4(a) and (c) and table 2), are primarily the result of higher air-conditioning (AC) energy use to cope with the heat extremes (see supplementary information available at stacks.iop.org/ERL/13/034003/mmedia).

Figures 4(b) and (d) explain the future trends of UHI_s–HW interactions under RCP 8.5. For cities in the temperate region, the synergistic effect is diminished in future scenarios. This is because the enhancement of evaporative contribution to the urban-rural difference during HWs compared to normal days is diminished near the end of this century under RCP 8.5 (figure 4(b) and table 2). The reason lies in the projected increase of precipitation in this region in future warmer climates by CESM (supplementary figure S1(a)). The increase of precipitation, to some extent, turns cities into water-sufficient surfaces assuming that future cities have the same impervious fractions (see section 3.6), so that evaporation over cities can increase as much as over their surrounding rural surfaces when HWs come in.

The situation is relatively different in the dry region. The synergistic effect becomes stronger in future warmer climates, mainly a result of the significantly elevated anthropogenic heat release during HWs. The source of the increase in anthropogenic heat is again primarily from higher AC energy consumption to cope with the extreme heat during HWs. The evaporative contribution to ΔT_s is increased during normal and HW days under future warmer climates compared to the current climate. The reason again lies in the projected increase of precipitation in future scenarios (figure S1(b)). In the current climate, cities and their surrounding rural areas largely lack water in this region. In future warmer climates, the water availability of rural surfaces is increased due to the increase in precipitation. The current version of CESM captures the dynamic feedbacks of vegetation to a changing climate state via its biogeochemical component (Oleson et al 2010). The moisturized soil, combined with the CO₂ fertilization effect and increasing air temperature, helps vegetation in rural areas develop rapidly in future climates. Rural surfaces in the dry region are no longer 'dry' in the future in the model. Therefore, the total rural evapotranspiration is generally increased in this region in the future. The water availability of urban surfaces, however, is much less increased than rural surfaces by the increased precipitation because of the lack of permeable surfaces in cities. Thus, the urban surfaces will remain water-limited to some extent. The lack of surface evaporative cooling over urban surfaces compared to over rural surfaces is intensified in future warmer climates. Therefore, the evaporative contribution to ΔT_s is increased during normal and HW days by the end of century under RCP 8.5 compared to current climate.

Zoom In Zoom Out Reset image size

3.4.2. Nighttime

We note four limitations of this study. First, urban land use is not dynamically represented in the current version of CESM. In other words, the urban fraction and geometry in every grid cell are fixed over time. Given that the nighttime UHI shows dependence on the morphological aspects of the city (Oke 1982, Oleson et al 2008), nighttime UHI is likely to increase with urbanization in the future, unless the mitigation of UHI is explicitly taken into account in urban planning. Therefore, the simulated nighttime UHI in the two RCP scenarios in this study are likely lower bounds of the potential nighttime heat stress for growing cities in future climates. Second, although CESM can simulate the dynamic response of building energy use to the environment, its parameterization scheme in the current version is still primitive. The total anthropogenic heat in the model includes only fluxes related to heating and AC, but not traffic-related heat fluxes, potentially causing a slight underestimate in anthropogenic heat. Third, local-scale processes such as advection between adjacent urban and rural land in the same grid cell are not resolved in the CESM. Therefore, impacts of changes in wind speed on the synergistic effects cannot be simulated. According to Li et al (2016), an increase in urban wind speed from 0.1 to 10 m s^–1 during daytime under HWs would increase UHI by less than 0.3 K. Last, our results are based on simulations from one ESM under two RCP scenarios rather than multi-model ensembles. CESM is currently the only ESM that has a sufficiently detailed physically-based urban land parameterization in CMIP5 (Phase 5 of the Coupled Model Inter-comparison Project) models. If other ESMs or future scenarios show a different trend in precipitation, the outcomes could be different.