The microscale cooling effects of water sensitive urban design and irrigation in a suburban environment

Abstract

Prolonged drought has threatened traditional potable urban water supplies in Australian cities, reducing capability to adapt to climate change and mitigate against extreme. Integrated urban water management (IUWM) approaches, such as water sensitive urban design (WSUD), reduce the reliance on centralised potable water supply systems and provide a means for retaining water in the urban environment through stormwater harvesting and reuse. This study examines the potential for WSUD to provide cooling benefits and reduce human exposure and heat stress and thermal discomfort. A high-resolution observational field campaign, measuring surface level microclimate variables and remotely sensed land surface characteristics, was conducted in a mixed residential suburb containing WSUD in Adelaide, South Australia. Clear evidence was found that WSUD features and irrigation can reduce surface temperature (T _s) and air temperature (T _a) and improve human thermal comfort (HTC) in urban environments. The average 3 pm T _a near water bodies was found to be up to 1.8 °C cooler than the domain maximum. Cooling was broadly observed in the area 50 m downwind of lakes and wetlands. Design and placement of water bodies were found to affect their cooling effectiveness. HTC was improved by proximity to WSUD features, but shading and ventilation were also effective at improving thermal comfort. This study demonstrates that WSUD can be used to cool urban microclimates, while simultaneously achieving other environmental benefits, such as improved stream ecology and flood mitigation.

Appendices

Appendix

Appendix A: Corrections for air temperature AWS measurement heights

To correct for the height difference between 1.5- and 3-m AWS sites, T _a data from the bicycle transects were used. The bicycle transects measured T _a at approximately 1.5 m and were compared with nearby fixed AWS observations to derive a height difference correction factor. Bicycle transect observations (T _b ) within 25 m of fixed stations (T _{A W S} ) were used. Using a regression linear model, correction factors were derived for four periods that reflect the heating/cooling structure of the diurnal cycle: 12 am–8am (slower cooling), 9am–11am (rapid heating), 12pm–4pm (slower heating), and 5pm–11pm (faster cooling). The eight AWS at 1.5 m were corrected to 3 m using these linear relationships:

• 12 am–8 am: T _{A W S} = 0.94T _b + 0.47 [r ² = 0.97]

• 9 am–11 am: T _{A W S} = 1.05T _b − 2.44 [r ² = 0.96]

• 12 pm–4 pm: T _{A W S} = 0.84T _b + 3.9 [r ² = 0.94]

• 5 pm–11 pm: T _{A W S} = 0.98T _b − 0.02 [r ² = 0.98]

We corrected all 1.5-m sites (sites 23–30, Fig. 3) to 3 m, using these relationships, to minimise the amount of sites that were altered. These relationships are not considered universal and only apply to the Mawson Lakes dataset.

Appendix B: Area of influence estimate: land surface influences on UCL air temperature

As outlined in Section 2.3.1, we conducted a simple analysis to ascertain the general area of influence for T _a at AWS sites. The circular buffers that we tested are shown visually in Fig. 11. For methodological simplicity, three different configurations of circular buffers were tested with different positioning of the circle relative to the source. The simplest configuration does not account for wind direction and is always centred on the AWS (circular 1) (Fig. 11a). The circular 2 configuration was set up so that the area of influence is placed 100% upwind of the point (Fig. 11b), and circular 3 is set so that 25% of the circle radius is downwind of the source (Fig. 11c). Wind direction was defined using the Parafield Airport weather station (Bureau of Meteorology). Hourly average wind direction was grouped into eight cardinal directions to define the placement of the buffers relative to the source (circulars 2 and 3).

Fig. 11

A representation of the buffers used to define the area of influence in this study: a centred on the fixed AWS (circular 1), b 100% of buffer upwind of the fixed AWS (circular 2), c circular buffer centred a distance of 75% of radius distance in the upwind direction (circular 3), and d example of a Horst and Weil (1992) approach with 10–90% isopleths shown. Horst and Weil (1992) approach was calculated at hourly timesteps. The location of the sensor (yellow dot) and the wind direction (easterly) are indicated

The analytical solution from Horst and Weil (1992) was also used in this research (e.g. Figure 11d). This model was designed for scalar sources influencing a micrometeorological flux station, and not a concentration source area. However, Horst (1999) states that the upwind extent of a source area associated with concentration-profile flux is similar to that of the source area for flux measurements. Further, this analytic solution is designed to be used above the roughness sublayer (i.e. not in the UCL); nevertheless, it was tested for comparison with the circular buffer approach. Site-specific observation heights were used, and roughness lengths were defined after (Wieringa 1993) (values between 0.25–0.5 chosen). Assumed values for Monin-Obukhov length (day = -140 and night = 80) and friction velocity (day = 0.3 and night = 0.15) were defined from Peña et al. (2010). Sensitivity testing suggested that the Horst and Weil (1992) model is not highly sensitive to Monin-Obukhov length and friction velocity for a low observation height. When averaging T _s for the Horst and Weil method, we took a distance-from-source weighted average over 10–90% isopleths. The Horst and Weil source area was calculated for the same time period over which the T _s image was captured (2–3 h) using the average wind speed and direction from the Parfield Airport AWS.

To calculate the correlation statistics (as in Fig. 12), for all area of influence configurations, the T _s from the thermal images (Section 2.2.1) was compared with the measured in situ T _a. A separate analysis was conducted for the nocturnal (15th February) and daytime (16th February) thermal images. For each buffer type/size, the T _s was averaged and compared with AWS T _a. The T _a was averaged over the appropriate time interval corresponding to the thermal image (approximately 2–3 h). For all circular buffers, T _s was averaged using the same configuration at all AWS sites, and for each buffer configuration, a linear regression model between T _a and T _s was fitted (Fig. 12). As such, with the exception of the Horst and Weil method, the correlation coefficients in Fig. 12 describe how well each buffer configuration (Fig. 11), when applied uniformly to all AWS sites (i.e. the same buffer for all sites), can describe the variability between T _s and T _a across the suburb (the limitations of this approach are outlined below).

Fig. 12

The coefficient of determination (r ²) that were calculated for different buffer configurations. The r ² values were calculated using T _s (independent variable) and T _a (dependent variable) for day (top) and night (bottom) cases. On the basis of these correlations, the best daytime footprint was judged to be the 50-m-diameter circular 2 configuration and the best nighttime footprint to be the 25-m circular 1 configuration

The analysis of the T _a-T _s relationships yielded the following best estimates of the source area for an UCL AWS in Mawson Lakes:

• day: 50-m (or less)-diameter circle upwind of the source (circular 2 or circular 3);

• night: 25-m-diameter circle centred on the source (circular 1).

For both the day and night cases, the general trend was for the correlation to increase as circular buffer size decreased (Fig. 12). This pattern was also found by Schwarz et al. (2012) who observed that the correlation decreased as the buffer size increased, as a greater proportion of unrepresentative surfaces were included in the buffer. However, during the day, the different buffer sizes returned more similar results than the nighttime example. This similarity represents the effects of advection and atmospheric mixing, which are more apparent during the day. Overall, for both day and night, the smaller buffers with a diameter of 50 m (or less) effectively covered the canyon floor that surrounded the site. It is thought that the adjacent canyon surfaces were most representative of the surfaces that influenced the microclimate T _a at these UCL sites, when considering T _s from a planar view.

On the basis of the correlation coefficients, the 25-m-diameter circular 1 configuration (centred at the source) was clearly the most appropriate footprint for the nighttime case (Fig. 12). Given that wind speeds tended to become lighter and more variable at night, this centred configuration (as opposed to an upwind setup) also made physical sense. The 50-m-diameter circular 2 configuration was used for daytime cases throughout this study, but arguably there was no clear “best” configuration for the day case. There was little difference in T _a-T _s correlation observed between the circular 2 and 3 configurations. However, the fact that the 12.5-, 25-, and 50-m-diameter circular 2 and 3 buffers produced better correlations than the circular 1 buffers suggests that taking into account wind direction during the day is important. Overall the results suggest the buffer should be up to 50 m in diameter and located upwind of the source.

The Horst and Weil (1992) method when applied to the day case produced similar sized area of influence to the circular buffers (see Fig. 11d) and the T _a-T _s correlations were broadly equivalent to the better performing circular configurations (Fig. 12). However, for the night case, the Horst and Weil (1992) method generated much larger estimates than the best performing circular buffers, and the correlation between T _a and T _s was lower than the circular buffer approach. Given the methodological complexity and assumptions associated with the Horst and Weil (1992) approach, it was deemed unnecessary for this study.

The larger source area at night is typical for flux and concentrations source area models. However, this contradicts the findings from the circular buffers, which implied that source area decreased at night. This may reflect the fact that traditional source area modelling methods, such as Horst and Weil (1992), are not valid for instruments within the UCL. Given these findings, and the fact that predicted source areas in this study were considerably smaller than suggested in other studies, we strongly recommended ongoing research in UCL source area modelling. Source area analysis is of critical importance for interpretation of results in microscale observational studies and is often overlooked.

It is unclear to what extent roof T _s affects T _a inside in the urban canopy. To address this, the T _s was tabulated both with and without building roofs included (Fig. 12). This analysis shows that removing roof temperatures improves the T _s-T _a, especially at night (Fig. 12). This suggests that roof T _s is not particularly important for defining the surface street-level T _a. However, roof surfaces have heterogeneity of emissivity, and the improvement of the T _s-T _a correlation without roofs may also reflect some error in the emissivity of roofing materials, as uniform emissivity values for roofs were used in this research. Without high-resolution roof emissivity data to correct T _s, it was difficult to ascertain what factor was causing this trend. Due to this uncertainty, roof surfaces were left in all land cover tabulations.

3.1 Limitations of the area of influence analysis

An important factor in the area of influence analysis, is that for circular buffers, the T _s was taken as an average over a given surface, and not as a distance-from-source weighted average (as in Fig. 11d). Therefore, the T _s from smaller circular buffers were more representative of T _a than the larger buffers, which tended to capture less representative surfaces, such as those located in adjoining urban street canyons. The use of unweighted average across the buffer surface meant that these less representative surfaces were treated equally in the calculations. However, this analysis was not attempting to calculate a flux station-type source area estimate. The analysis was attempting to capture the surfaces that are most representative of microscale T _a variability. Given that the correlation coefficient does not constantly increase as buffer size decreases (3- and 6-m-diameter circular buffers were also tested), we are confident that we are inferring a reasonable estimate of the buffer size that captures the surfaces most representative of microscale T _a variability. Nevertheless, source area analyses for UCL sensors is a very important area that requires future research. A number of factors were not captured with this method, including three-dimensional effects (e.g. canyon wall temperatures), local-to-mesoscale advective effects, and differences in characteristics of the upwind fetch. The approach used did not capture these processes, and it is expected that all these factors will have some influence on microscale T _a temperature variability, especially during the day. Nevertheless, the good correlation between T _a and T _s suggests that microscale T _a is highly influenced by the thermal characteristics of the surfaces directly adjacent to the source. Another limitation of this analysis is the assumption that the same area of influence can be applied to all the sites. Factors such as surface roughness, measurement height, and the configuration of buildings will affect the actual source area from site to site. It is acknowledged that there will be variability from site to site in actual source area size and positioning. Overall, because the Mawson Lakes site is a relatively open environment (i.e. few tall buildings), it is thought that the method presented here was an acceptable means of getting a broad understanding of the average area influencing T _a across the 27 fixed AWS sites.