Resolving the influence of local flows on urban heat amplification during heatwaves

We use the Weather Research and Forecasting (WRF) version 4.1.3 (Skamarock et al 2019) modeling system for our convection-permitting simulations. We use Sydney, Australia as an example of a typical coastal city with a mountain range inland of the city running parallel to the coast. The computational domain consists of one parent domain at 4 km resolution (240 × 400 points) and two one-way nested domains both at 800 m resolution (411 × 451 points) (figure 1). For the nested domains, one is run with the single-layer urban canopy model (SLUCM) to describe urban-scale processes (urban) and one without (non-urban). Note that the SLUCM model code was modified to include ten urban types as opposed to the default types of high, medium and low density. For the non-urban simulations, all urban areas are replaced as grasslands. The WRF physics configuration used here includes the Mellor–Yamada–Janjic boundary layer and Monin–Obukhov (Janjic Eta) similarity surface layer scheme, the Dudhia shortwave and Rapid Radiative Transfer Model longwave radiation schemes, the Noah Land Surface Model, and the WRF Double‐Moment 5-class microphysics scheme. This physics configuration has been shown to work well in this region (Evans and McCabe 2010). Due to the resolution of both parent and nested domains, no cumulus physics is required as the resolution is fine enough to resolve convective processes.

Zoom In Zoom Out Reset image size

There are 45 hybrid eta vertical levels from the surface to 50 hPa for the atmospheric model. A 10 s timestep for the outermost parent domain is used and a reduced timestep at a ratio of 3:1 for the nested domains. The lateral boundary conditions for the parent domain and sea surface temperatures are sourced from the 6 hourly 70-level Bureau of Meteorology Atmospheric high-resolution Regional Reanalysis for Australia (BARRA) available at 12 km resolution (Su et al 2019). During early 2017 southeast Australia experienced persistent hot temperatures with frequent heatwave events. A five-member initial condition ensemble is run for the period 1 January 2017 until 28 February 2017 with initialization of the individual ensemble members starting at 6 h intervals.

For the urban grid points we use the World Urban Database Access Portal Tool (WUDAPT) (Brousse et al 2016) to create an urban classification map for Sydney following the urban types described by the local climate zones (LCZs) of Stewart and Oke (2012). Using the tools developed by Brousse et al (2016) the ten urban types (table 1) are integrated into the landcover classification of WRF (figure 1). Urban areas not covered by our Sydney LCZ map that are within the nested domain were modeled as a compact low-rise urban form.

Updating the urban types within WRF requires appropriate estimation of parameters to describe key attributes of each type (table 1). Here we use the Urban Vegetation Cover and Environmental Planning Instrument—Height of Buildings datasets maintained by the New South Wales Government Department of Planning, Industry and Environment (DPIE 2008, 2019) to estimate the vegetated fraction and building height respectively for each urban type. We use the MODIS Bidirectional Reflectance Distribution Function Albedo data (Schaaf and Wang 2015) to estimate building albedo and the anthropogenic heat emission global database by Dong et al (2017) to estimate the anthropogenic heat flux. To estimate the parameters, we extract the points corresponding to each LCZ type from the respective database and calculate the median and standard deviation. The final estimates are presented in table 1 and are within the ranges defined by Stewart and Oke (2012).

Due to the high resolution of the WRF simulations, evaluation of the model skill against gridded climate data is challenging. Therefore we use observations from 11 automatic weather stations (AWS) maintained by the Australia Bureau of Meteorology that have data for the simulation period within the model domain (figure 1). These data are available at a 1 min interval and are aggregated to 10 min averages. We extract the model outputs co-located with the geographic location of the AWS sites and compare the 10 min time series. To quantify the simulation skill we use the metrics of the International Land Model Benchmarking system (Collier et al 2018) which includes metrics of the relative bias, the root mean square error and the phase skill score (see supplementary material (available online at stacks.iop.org/ERL/16/064066/mmedia)). An example of the time series comparison for Sydney Airport is provided in supplementary figure S1 with the skill scores for various variables across all AWS sites included in supplementary tables S1 and S2. Overall, the model skill is high, although precipitation events were limited over January-February 2017 there remain limitations in the timing and magnitude of precipitation events. Additional assessment of the differences between urban and non-urban outputs is provided in supplementary figures S2 and S3.

To identify the heatwave days during January/February 2017 we use observational station data from Sydney Airport (66037), as well as Observatory Hill (66062) and Richmond (67105) stations that are part of the Australian Climate Observations Reference Network—Surface Air Temperature (ACORN-SAT) maintained by the Australia Bureau of Meteorology. Although many definitions exist for heatwave identification, we use a modified version of the excess heat factor originally defined by Nairn and Fawcett (2013) that implicitly incorporates a seasonal cycle to characterize heatwaves according to the climatological conditions for a particular time of the year (Perkins and Alexander 2013). Heatwave days are identified by calculating the anomaly in the daily mean temperature (determined from the average of the daily maximum and minimum temperature) above the calendar day 90th percentile (equation (1)). By using the daily average temperature, we can account for the impact of warm nights as well as warm days during heatwave events. When the daily average temperature anomaly is positive for at least three consecutive days, these are classified as heatwave days.

$$\begin{equation}{\text{EH}}{{\text{F}}_{{\text{SIG}}}} = \frac{1}{3}\left( {{T_1} + {T_{t - 1}} + {T_{t - 2}}} \right) - {T_{90}}.\end{equation} \tag{ 1 }$$

Here T₉₀ is calculated using a 15 day window centered on the corresponding calendar day with a temperature T₁ for the standard climatological reference period of 1961–1990. This definition has been used previously to identify Australian heatwave events in Hirsch et al (2019) and Hirsch and King (2020). Applying this definition to the weather station data, we identify five heatwave events that occurred in quick succession over Sydney during January-February 2017. Using a slightly different definition based on maximum daily temperatures yielded comparable heatwave day identification.

We assess the temporal evolution of the urban gradient along an East–West transect, from coast to inland, of the temperature, moisture content, and horizontal wind speed components (figure 2). Across the simulation period the temperature gradient across the city varies with a mean value of 0.83 °C and a maximum value of 15.79 °C. The largest temperature gradients (figure 2(a)) often occur during the heatwave events when there is a westerly flow over the Blue Mountains located west of the city (figure 2(c); positive u-wind component). More specifically, at instances where there is an abrupt change to a westerly flow at the urban/rural interface, corresponding to the dates 11 January 2017, 17 January 2017, 30 January 2017 and 10 February 2017, an easterly flow at the coast is present. This contributes to the establishment of large temperature gradients due to the relatively cooler conditions present at the coast. This is also observed in available data from the weather stations (supplementary figure S4). A closer examination of the time series for the first heatwave event (supplementary figure S5) shows that there are instances where the v-wind direction is different between coastal grid cells and those further inland. More specifically on the 11 and 13 January the v-wind is Southerly at the coast and Northerly further inland, which would also contribute to the development of the large temperature gradients observed on those days.

Zoom In Zoom Out Reset image size

A recurrent feature during the simulation is the sea breeze (figure 2(c); negative u-wind component) which contributes to ventilating the city and reducing the temperature gradient (figure 2(a)). The second heatwave is the only event where a strong westerly flow does not occur at the onset but is present during the event. Instead, the south-easterly air flow contributes to delaying the transport of the warmer inland flow across the whole city to the second heatwave day. The timing of this event is sensitive to the station data used, with a one day difference in the onset timing between the inland Richmond and coastal Observatory Hill site. This is likely associated with the easterly flow that persists over the 14 and 15 January preceding the second heatwave event. This is indicative of how influential sea breezes can change the onset timing of a heatwave across a large coastal city. All heatwave terminations are coincident with a strong southerly flow (figure 2(d); positive v-wind component) that cool temperatures across the city.

Examining the radiative changes across the transect (supplementary figure S11) indicates that the downward short- and longwave radiation is generally consistent across the city. Furthermore, over the days preceding each heatwave event, the smooth time series of the downwards shortwave radiation are indicative of the clear sky conditions that develop at heatwave onset. Given the temperature increases over the same days that clear sky conditions occur prior to each heatwave, we are confident that the timing of the heatwave events are consistent with the observed days.

Most of the heatwaves commence with heat transport into the region from the west to north-west of Sydney with cooler conditions over the adjacent ocean surface (figure 3). This confirms how large gradients can establish over the city by the contrasting temperatures between the continental interior and the adjacent ocean. For the first heatwave, surface wind speeds over the adjacent ocean are weak compared to subsequent heatwave events where a strong southward flow persists. The contrasting wind directions and temperature gradient across the city (figure 2(c)) are attributed to how the large-scale flow influences heat dispersion. For example, the third heatwave commences with the convergence of strong surface wind (figure 3(d)) which results in strong updrafts close to the coastline (figure 3(m)). For the second heatwave, the southward flow extends over the city to inhibit the transport of dry air from the west into the city, although the urban footprint of Sydney is still evident with a temperature of 28 °C–30 °C compared to 24 °C–26 °C over the surrounding vegetated areas (figure 3(b)). The moisture contrast between the ocean and the land is anticipated however for the 3rd and 4th heatwaves the moisture content increases along the coastline coincident with the convergence of the continental westerly flow and oceanic southward flow. Examining the impact of the urban environment for heatwave onset (supplementary figure S12) shows that the city contributes to hotter onset for the first three events, drier surface conditions for all events, and can induce intense thermal updrafts.

Zoom In Zoom Out Reset image size

To evaluate the UHI contribution we compare the WRF output over January–February 2017 between simulations where the urban parameterization is activated against those where all urban areas are represented as grasslands (hereafter referred to as non-urban simulations). The UHI is a persistent feature during the simulations with a mean value of 1 °C over the city that extends west over the Blue Mountains (∼0.5 °C) up to a depth of 1 km into the atmosphere from the surface (figure 4(a)). This is accompanied by a drier atmospheric profile (∼0.003 kg kg⁻¹) with a comparable extent to the temperature amplification (figure 4(b)). The drier atmosphere is attributed to a decrease in moisture supply from the land surface as the contribution from evapotranspiration is significantly reduced when the urban parameterization is activated due to the smaller vegetation fractions.

Zoom In Zoom Out Reset image size

There is considerable variability in the strength and extent of the UHI as determined by the prevailing airflow across the region (not shown). In particular, there is further extension over the Blue Mountains of the UHI when the airflow is easterly, and when the airflow is westerly, the UHI temporarily extends over the adjacent ocean. Over January–February 2017, the UHI has a diurnal cycle with an average daytime UHI of 0.65 °C and an average nighttime UHI of 1.58 °C across the city.

Due to the reduced efficacy of sea breezes to ventilate the whole city during the heatwave events (figure 2) the cumulative heat attributed to the urban environment over the five heatwave events, calculated as the sum of the mean daily temperature difference divided by the number of heatwave days, increases to more than 10 °C over the city and 3 °C further west of the city over the Blue Mountains (supplementary figure S13). Again, the UHI has a larger impact at night, with an average night-time cumulative heat of 8.01 °C over the city during the five heatwave events, compared to a daytime cumulative heat of 3.89 °C.

To examine whether the UHI effect varies between the urban types we compare the diurnal variations in temperature, humidity and wind speed (figure 5). By considering the difference of the urban from the non-urban simulations, we can account for the effect of distance from the coast as some urban types are predominantly coastal (e.g. compact high-rise) compared to others. For air temperature (figure 5(a)), dew-point temperature (figure 5(b)) and relative humidity (figure 5(c)), the heavy industry, large low-rise and sparsely built urban classes show a greater diurnal variability compared to the compact and open urban classes. The compact urban classes and open high-rise are consistently warmer than non-urban areas at all times of the diurnal cycle, whereas the remaining urban classes are predominantly warmer than non-urban types at night. (figure 5(a)). For all urban classes the relative humidity is lower at all times, consistent with figure 4(b), however the moisture deficit is larger at night. The wind speed (figure 5(d)) is increased over most urban areas at night and decreased during the day between 0900 and 1600 for all urban classes. The diurnal variation can be attributed to how the urban environment establishes the strength of the temperature gradient across the city. More specifically, as more heat is retained by the thermal mass of the urban environment, large thermal contrasts between the city and adjacent ocean contribute to the strength of both land and sea breezes. Therefore, as a result of the larger thermal gradient at night between urban and non-urban simulations, irrespective of the urban class, higher wind speeds occur.

Zoom In Zoom Out Reset image size