Urban Heat Island Mitigation: A GIS-based Model for Hiroshima

The microclimate of outdoor spaces was investigated in a previous research pertaining to the Metropolitan City of Turin (Italy) considering different outdoor air temperatures registered by various WSs [5]. The UHI models presented in this work were then applied in the Hiroshima case study. The aim was to obtain a simple GIS-based model for the simulation of the hourly air temperature through the use of a linear regression.

1.png

Figure 1. Flowchart of the methodology

Figure 1 shows the methodology used to evaluate the outside air temperature. All the variables were identified on a building block territorial unit using a GIS tool (ArcGIS 10.6). The variables were then correlated with the outside air temperature, through the use of a linear regression model, in order to identify the main parameters of influence. More models were reported as functions of different numbers of variables. Finally, the hourly variation of the outside air temperature was evaluated and the territory was classified as: ‘mountain area’, ‘plain area near the sea’ and ‘plain area not near the sea’. This classification influences the air temperatures and their daily amplitude. The air temperature variations were correlated with the following variables, which were used to analyze the UHI phenomenon [6]: altitude (m_asl), distance from the sea (D_sea), albedo of the outdoor surfaces (A_NIR), normalized difference vegetation index (NDVI), land surface temperature (LST), building coverage ratio (BCR), building density (BD), building height (BH) or relative height (H/H_avg), canyon height-to-width ratio (H/W) and main orientation of the streets (MOS). These variables were calculated at a building block scale for each WS, considering all the blocks in a buffer area of 300 m from the WS. MOS was evaluated considering a variable that ranged from 0 to 1, with 0 corresponding to the North-South direction and 1 to the West-East direction. Satellite images (Landsat 7 and 8) were used to calculate the A_NIR, LST and NDVI parameters, with reference to a typical summer day. The satellite images were chosen in the same period of weather measurements (from July 20^th to September 23^rd 2013) with no clouds the sky (less than 4 %). With the hourly distribution of air temperature, it was possible to define the typical summer day. The UHI models were defined by comparing the calculated air temperatures with the measured ones and then reducing the errors. An iterative procedure was performed on excel spreadsheets in order to reduce the errors between calculated and measured data, and to optimize the error (ε), the relative error (ε_r) and the relative absolute error (|ε_r|).

2.1 The air temperature model

The main variables of influence pertaining to the outside air temperature were identified considering the correlations between the variables and the outside air temperature. The linear regression models of the air temperature were set up considering all the variables or a limited number of variables. The linear regression model was created using 19^th August 2013 at 1:49 a.m. (Eq. (1)) as the reference:

$T_{1: 49}=I+\alpha_{1} \cdot X_{1}+\alpha_{2} \cdot X_{2}+\ldots+\alpha_{n} \cdot X_{n}$ (1)

where, ‘I’ is the intercept; ‘α_n’ are coefficients used to estimate the influence of variables X on the outdoor air temperature; ‘X_n’ are the independent variables.

The ‘min-max’ method (Eq. (2)) was introduced in order to normalize the variables and to evaluate the different weights of the variables on the air temperature:

$X_{N}=\frac{1}{X_{\max }-X_{\min }} \cdot X-\frac{X_{\min }}{X_{\max }-X_{\min }}$ (2)

Therefore, the normalized variables were dimensionless and varied from 0 to 1 (X_N). The accuracy of the models was assessed with |ε_r| and ε_r.

2.2 The hourly air temperature model

The hourly trend of a typical summer day was analyzed, with Eqns. (3), (3.1), (3.2) and (3.3), as a function of the minimum, maximum and daily distribution of the air temperature. These values were mainly influenced by the altitude, the distance from the sea and the presence of vegetation and water (as characterized by the NDVI index). The territory was then classified as mountain area, plain area near the sea or plain area not near the sea.

An air temperature hourly-distribution factor, ‘f(t)’, was identified for the typical summer day and for each area to reduce |ε_r| and ε_r between the measured and calculated hourly air temperatures:

$T_{h}=T_{\min }+\Delta T^{\cdot} f(t)$ (3)

where,

$T_{\min }=I+\alpha_{A I t} \cdot A l t+\alpha_{D i s t} \cdot D i s t+\alpha_{N D V l} \cdot N D V I$ (3.1)

$\Delta T=I+\alpha_{A l t} \cdot A l t+\alpha_{D i s t} \cdot D i s t+\alpha_{N D V I} \cdot N D V I$ (3.2)

$f(t)=\frac{T_{h}-T_{\min }}{T_{\max }-T_{\min }}$ (3.3)

2.3 The thermal comfort indexes

The microclimate is affected by the local urban morphology [7] and several parameters, such as the H/W ratio, the sky view facto (SVF), the main orientation of the streets (MOS) or the albedo of outdoor surfaces, are used to describe the urban context, [8]. It has been found, from a literature review, that indexes used to assess comfort can be classified into three categories: energy balance models (i.e. Physiologically Equivalent Temperature ‘PET’; Predicted Mean Vote ‘PMV’; Perceived Temperature ‘PT’), empirical indices (i.e. Actual Sensation Vote ‘ASV’; Thermal Sensation Vote ‘TSV’) and indices based on linear equations (i.e. Apparent Temperature ‘AT’; Cooling Power Index ‘PE’; Wind Chill Temperature ‘WCT’) [9, 10, 11].

Among the various indicators for calculating thermal comfort, those that depend only on temperature T_air (°C), relative humidity RH (%), vapour pressure vp (hPa) and velocity v (m/s) of the outdoor air have been selected. In this work the following indicators were used with the relative correlations [9]:

• Apparent Temperature ‘AT’ is an equivalent perceived temperature, caused by the combined effects of air temperature, relative humidity and wind speed:

$A T=-2.7+1.04 \cdot T_{a i r}+\frac{2 \cdot v p}{10}-0.65 \cdot v$ (4)

• Discomfort Index ‘DI’ is used to quantify the effective temperature combining the effect of temperature, humidity and air movement on the sensation of heat or cold perceived by the human body:

$D I=T_{a i r}-0.55 \cdot(1-0.01 \cdot R H) \cdot\left(T_{a i r}-14.5\right)$ (5)

• Normal Effective Temperature ‘NET’ is the effective temperature felt by the human organism for certain values of meteorological parameters such as air temperature, relative humidity of air, and wind speed:

$\begin{aligned} N E T=37 &-\frac{37-\tau_{\text {air }}}{0.68-0.0014 \cdot R H+\frac{1}{1.76+1.4 \cdot V^{-10.75}}}-0.29 \cdot T_{\text {air }} \\ &(1-0.01 \cdot R H) \end{aligned}$ (6)

• Humidex ‘H’ was created to quantify and the degree of risk to the human body in the event of heat and excessive moisture (in cooling season); the simplified formula is the following:

$H=T_{a i r}+\frac{5}{9} \cdot(v p-10)$ (7)

• Heat Index ‘HI’, also known as an apparent temperature, is the perceived temperature by the human body when relative humidity is combined with the air temperature:

$\begin{aligned} H I=& 8.784695+1.61139411 \cdot T_{a i r}+2.338549 \\ R H-0.14611605 \cdot T_{a i r} \cdot R H-1.2308094 \cdot 10^{-2} & . \\ T_{a i r}^{2}-1.6424828 \cdot 10^{-2} \cdot R H^{2}+2.211732 \cdot 10^{-3} \\ T_{a i r}^{2} \cdot R H+7.2546 \cdot 10^{-4} \cdot T_{a i r} \cdot R H^{2}-3.582 \\ 10^{-6} \cdot T_{s i r}^{2} \cdot R H^{2} \end{aligned}$ (8)

• Relative stain index ‘RSI’ is used to describe the thermal comfort of a standard pedestrian under specific environmental conditions:

$R S I=\frac{\left(T_{a i r}-21\right)}{(58-v p)}$ (9)

The results shown below are divided into 4 sections. The first part is dedicated to the linear regression model used to create the air temperature model for the considered typical summer day (August 19^th 2013). The results of the hourly air temperature distribution model are presented in the second part, where the weather stations located in the mountains area, plain area not near the sea and plain area near the sea are distinguished. In the third part, assessments of the urban heat island intensity were made using the UHI-driven indicators (Q1 and Q2) and land-cover-driven indicators (Q3) [13]; the heatwaves and cold waves for the years 2007 and 2013 were also analyzed. Outdoor thermal comfort indexes have been calculated in the last section comparing the results on seven weather stations.

4.1 The air temperature model

In order to identify the main variables of influence on the air temperature, the correlations between the variables and air temperature were evaluated (Figure 9). The altitude shows a negative correlation because the air temperature decreases as the altitude increases. The presence of vegetation and water also reduces the air temperature and NDVI therefore has a negative correlation, while positive correlations can be observed for BD, BCR, H/H_avg and H/W.

Only variables that were not dependent on each other were used for the higher correlation factor. As BD and BCR are dependent variables, only BCR was used for the model; BH and H/H_avg are also dependent, so only H/H_avg was used for the model. The linear regression models of the air temperature are presented hereafter, where models with non-normalized variables are distinguished (Figure 10a): a linear regression model with all the non-normalized variables (Eq. 11); a linear regression model with the non-normalized variables but without LST (Eq. 12); and with normalized variables (Figure 10b): a linear regression model with all the normalized variables (Eq. 13); a linear regression model with the normalized variables but without LST (Eq. 14); a linear regression model with the normalized variables but without LST and NDVI (Eq. 15); a linear regression model with the normalized variables but without ANIR and LST (Eq. 16).

The best results, with the highest R² coefficient of determination, were provided by Eq. 11 (all the non-normalized variables) and Eqns. 13 and 14 (all the normalized variables and all the normalized variables without LST). Table 4 reports the R² values that show to what extent the variations in air temperature can be explained, by the regression model, as functions of the selected variables (Eqns. 12 and 14(Eqns. 12 and 14 without LST, Eq. 15 without LST and NDVI, and Eq. 16 without LST and ANIR).

The measured value of the relative error |ε_r|, which is the ratio of the absolute error, between the measured and calculated values of the air temperatures, was used to describes the accuracy of the models; low values can be observed for all the linear regression models and they tend to increase slightly when some variables, such as LST and NDVI, are excluded. Moreover, NDVI and A_NIR are dependent variables, with a correlation coefficient of 0.76, and the weight of the variables should therefore be negative in the models, but the A_NIR tends to be positive due to the presence of the NDVI (compensatory effect); in addition, for this case study, A_NIR is quite constant, with an average value of 0.22 and a low standard deviation of 0.04.

The models were then applied to the Hiroshima territory through the use of the GIS tool. Figure 11 shows the air temperature simulated for August 19^th 2013 at 1:49 a.m. using the model with all the normalized variables (Eq. 13). The air temperature is higher in urban areas than in the peripheral plain and mountain areas, where the temperature is mitigated by the altitude, the presence of vegetation and a lower buildings density.

9.png

Figure 9. Correlations between the variables and the air temperature

10.png

Figure 10. Air temperature models with (a) all the non-normalized variables and (b) all the normalized variables

Table 4. Eqns. 11-16: Coefficients, relative error |εr| and coefficient of determination R2 for air temperature models

 The range of variability of the different variables multiplied by the relative coefficients α is indicated in Figure 12. The main variables of influence (the non-normalized variables are in green, while the normalized variables are in blue) are: the altitude, the presence of vegetation, the characteristics of the outdoor surfaces (A_NIR), the buildings density (BCR) and the relative building height (H/H_avg). LST is present in two equations (Eq. 11 and 13); the A_NIR coefficient becomes negative when NDVI is not included in the model (Eq. 15). The distance from the sea and the altitude are uncontrollable variables, and in order to improve the microclimatic conditions and, to mitigate the air temperature, it is therefore necessary to intervene on the other variables. For example, in newly built areas (where there is an increase in BCR and H/H_avg, and consequently an increase in air temperature), the share of green areas can be improved (NDVI is inversely proportional to the air temperature) to compensate for the UHI effect. Eqns. 15 and 16 confirm the correlation between A_NIR and NDVI; NDVI in Eq. 15 is inversely proportional to T_air, and the same relationship may be observed for A_NIR in Eq. 16 without NDVI.

11.png

Figure 11. Air temperature model with all the normalized variables (Eq. 11)

12.png

Figure 12. Correlations between the variables and the air temperature

4.2 The hourly air temperature model

The outside air temperature was simulated using the equations reported in Table 5, in which the intercept ‘I’ and the weight coefficient of the ‘α’ variables (distance from the sea, altitude and presence of vegetation and water) are indicated: Eq. 17 refers to the whole territory; Eq. 18 refers to the mountain areas; Eq. 19 refers to the plain areas near the sea; Eq. 20 refers to the plain areas not near the sea.

Figure 13 show the equations used to evaluate the daily minimum air temperature, T_min, and the daily amplitude of the air temperature, Delta (t), for a hot summer day. The hourly air temperatures were then simulated, using Eq. 3, for a typical hot summer day (August 19^th 2013). In these models, the relative error |ε_r| is higher and the coefficient of determination R² is lower than in the other models, because T_air does not only depend on these variables; consequently, there is a greater dispersion of data from the average value (this trend is more evident in the case of the Delta (t) which has an average |ε_r| of 10.8 %).

13.png

Figure 13. Equations used to evaluate the: (a) minimum T_min and (b) amplitude Delta(T) of the air temperature

Table 5. Eqns. 17-20: Coefficients, relative errors |ε_r| and coefficients of determination R² for the hourly air temperature model

14.png

Figure 14. Distribution of the hourly air temperature for August 19^th 2013

15.png

Figure 15. Application of the hourly air temperature model for 6 a.m. at a building block scale (Eq. 12)

16.png

Figure 16. Application of the hourly air temperature model for 3 p.m. at a building block scale (Eq. 12)

4.3 The UHI-driven and land-cover-driven indicators

The UHI intensity (UHII) is an indicator that can be used to measure the hourly and daily amplitude and temperature gradient of the air between the urban and the surrounding rural areas. Two types of indicators can be used to evaluate the different microclimate conditions: a UHI-driven type (Q1 and Q2) and a land-cover-driven type (Q3) [13]:

Q1. The ‘magnitude’ of the UHI-driven indicator is equal to the maximum temperature minus the average daily air temperature, where daytime is distinguished from nighttime;

Q2. The ‘range’ of the UHI-driven indicator is equal to the difference between the maximum and the minimum daily air temperatures, and daytime is distinguished from nighttime;

Q3. The ‘urban-rural’ land cover-driven indicator, which describes the difference between the hourly air temperatures in the urban and surrounding areas.

These three indicators were calculated with the hourly data for the years 2007 and 2013 for two municipal WSs: WS 7 (at altitude of 26.7 m a.s.l.), which is located in the surrounding rural area and WS 3 (at altitude of 5.8 m a.s.l.), which is located in an urban area (Figure 5).

Figure 17 shows the trends of the indicators for the years 2007 and 2013 considering the annual average quantitative values of UHI intensity at different times. It is possible to see that the values of Q1 and Q2 remain almost stable and the values decrease from midnight to 6 a.m., then there is a slight increase and the UHI effect tends to decrease after 4 p.m. (this trend is similar for both years); Q3 decreases during the day and tends to increase after 6 p.m.

17.png

Figure 17. The UHI-driven and land-cover-driven indicators

Figure 18, Figure 19 and 20 show the moving average times series of the previous 30 days of the UHII values at 0 a.m., 6 a.m., 10 a.m. and 2 p.m. Different trends may be observed: Q1 and Q2 have lower daytime values in winter than nighttime values, while the daytime values in summer are higher than the nighttime values. Moreover, the land-cover-driven indicator Q3 has higher values in winter due to the higher thermal excursion that occurs during the night (this effect is due to lower air temperatures); Q3 is higher in the daytime in summer due to higher solar irradiation. The urban station always has smaller values than the suburban station because T_air is influenced by the urban density, with higher air temperature values. In fact, WS 3 has a BCR of 0.33 m²/m² and a BD of 3.05 m³/m², which are higher than WS 7 (BCR = 0.23 m²/m² and BD = 1.64 m³/m²).

The heatwaves and cold-waves were evaluated for the years 2007 and 2013 using the same weather station data (WS 3 and WS 7). The heatwaves were considered as events with temperatures over the 97.5^th percentile and cold-waves as events with temperature under the 2.5^th percentile [14]. In 2007 and 2013, there were 10 hot days a year in Hiroshima, with an average air temperature of 30.43 °C in 2007 and 31.39 °C in 2013, with heatwaves with higher air temperature than 29.8 °C for the year 2007 and 31.01 °C for the year 2013. Moreover, there were 9 cold days a year with an average air temperature of 3.36 °C in 2007 and 1.37 °C in 2013, with cold-waves with a lower air temperature than 4.24 °C for the year 2007 and 2.48 °C for the year 2013.

18.png

Figure 18. The Q1: 30 UHI-driven indicator for the moving average times series of UHI intensity at 0 a.m. (in blue), 6 a.m. (in green), 10 a.m. (in yellow) and 2 p.m. (in orange)

19.png

Figure 19. The Q2:30 UHI-driven indicator for the moving average times series of UHI intensity at time 0 a.m. (in blue), 6 a.m. (in green), 10 a.m. (in yellow) and 2 p.m. (in orange)

20.png

Figure 20. The Q3: 30 UHI-driven indicator for the moving average times series of UHI intensity at time 0 a.m. (in blue), 6 a.m. (in green), 10 a.m. (in yellow) and 2 p.m. (in orange)

Figure 21 shows the average hourly UHI variability during the summer months (June-August) for Hiroshima. The data for the summer months for the years 2007 and 2013 were used to analyze the UHI variability trend. The shaded area in Figure 21 represents the average hourly standard deviation, while the dashed lines with the sun and the moon symbols represent the approximate local sunrise (5 a.m.) and sunset (7 p.m.) times. The UHII increases between the hours 3 p.m. to 1 a.m. and decreases during the period 3 a.m. to 3 p.m.; the highest UHI value is 1.5 °C and it may be observed at 1 a.m. (average value for the years 2007 and 2013). This trend is due to the influence of coastal winds and it is similar to that of the City of Seattle, which is located near the coast [1]. In fact, wind effects, in addition to the surface characteristics, play a crucial role in influencing the UHII; as the wind speed increases, the volume of relatively cooler air arriving from the surrounding rural areas reduces the urban air temperature. These air circulations play a crucial role in reducing the horizontal temperature gradient between the urban and rural areas [1].

21.png

Figure 21. Mean daily variability of the UHII for summer months (June-August)

4.4 The thermal comfort assessment

Six indexes were used to analyze outdoor thermal comfort in summertime [9] based on linear equations depending on the available three climate variables: air temperature, relative humidity and velocity. The outdoor thermal conditions were assessed using the data from seven weather stations considering the typical summer day with hourly intervals. In general, it is possible to observe that the low distance from the sea (WS1, WS4), the high altitude (WS7), the presence of the wind (WS1, WS4, WS5, WS6 and WS7), and of green areas (WS5) significantly increase the outdoor thermal comfort.

Figure 22 shows some examples of the hourly results for three thermal comfort indexes: Apparent Temperature (AT), Humidex (H) and Heat Index (HI). Only for WS1 and WS4 warm thermal comfort conditions were recorded for H and HI indexes; these results are mainly due to the proximity to the sea, and to the presence of the wind. The worst thermal comfort conditions are observed for WS3 and WS6; in WS3 there was no wind and WS6 low green areas and high buildings density. Thermal comfort in summertime is correlated to the type of urban environment but also by the presence of wind.

22.png

Figure 22. Hourly trends of thermal comfort indicators