To quickly calculate 33 morphological parameters in 160 points and 10 sites, an automatic process combining the python script and Geographical Information System (GIS) model was introduced. These parameters were then correlated with UHI data 24,44,45. In general, the study was conducted by the following steps:
⑴ conducted field measurements to collect microclimate data and morphological information in 10 selected sites;
⑵ built GIS model of study sites by ArcGIS and calculated urban morphological parameters using python script and geoprocessing functions;
⑶ conducted Pearson correlation and multilinear regression analysis between UHI and urban morphological parameters. proposed empirical models to indicate their relationships.
3.1. Site identification
In terms of site selection for microclimate measurements, the influence of geographic condition, seasonal and weather variance, artificial heat sources shall be well controlled 46. Geographic location determines the regional climate background and the distances from cool resources, including sea surfaces and vegetated mountains. For seasonal and weather influence, the climatic condition (temperature, humidity, solar radiation, precipitation, wind, and cloud cover) changes between different seasons and even different days, which could impose significant effects on UHI patterns. To reduce the geographic impact, study sites were carefully selected in the adjacent area on the northern side of Hong Kong Island. Meanwhile, to minimize the seasonal and climatic impact, the measurements were conducted on clear sunny days in the summertime. Originally there were 12 sites with typical urban form investigated, but only 10 sites are analyzed in this study (Fig. 2a). In the preliminary analysis, it was found that the data collected from Sung Hing Lane site (SHL) and Lockhart Road site (LhR) may not be representative due to the difference of measurement time at SHL and land use type at LhR. The other 10 sites were all residential blocks and measured in the months between July and early September, which were typical summer months in Hong Kong. Thus, SHL and LhR are excluded from the present analysis.
In urban climate research, it is important to define the site boundary according to research scales. In this study, it was decided based on a review of different kinds of research on urban morphology and environmental performance, covering main research methods (remote sensing, simulation, and field measurement). In a study of the cooling effect of green space based on remote sensing technology, a size of 240 m × 240 m upon satellite images was employed by Kong et al. to calculate landscape metrics 47. In a physical simulation study on urban bioclimatic design strategies, Sad de Assis and Barros Frota built a scaled 500 m × 500 m model representing an area of 25 hectares 48. In a field measurement research project, Krüger and Givoni measured the temperature of seven areas in Curitiba, Brazil, each with a diameter of 250m 49. Santamouris 2, Edussuriya et al. adopted the area size of 200m × 200m and investigated 20 different urban residential areas to study the relationship between urban morphology and air quality in Hong Kong 24. Based on the six generic urban forms proposed by Leslie Martin and others to address land-use characteristics, Ratti et al. reassessed the environmental performance of these forms using computer analysis techniques, adopting an area size of 250m × 250m 50. In the research on the relationships between building energy consumption and urban texture, urban areas with dimensions of 400m × 400m were adopted to extract building form data by using a computer-based image processing technique 51. Based on the aforementioned studies, a size of 400 m × 400 m was proposed for the research sites in the present study considering the following aspects:
The cooling effect of green space may extend beyond the site, as far as the park’s width 52.
According to the technical circular of Hong Kong Air Ventilation Assessment 53 the surrounding area of up to a perpendicular distance of 2H (highest building in the site) from the project boundary must be included for investigating wind performance. Since building height in the measurement area could normally reach 100m, a 400m × 400m rectangular area was proper for this study.
3.2. Microclimate measurements and field survey
The microclimate measurements in these 10 sites were conducted during three summers from 2010 to 2012. HOBO weather stations were used to record the climate data, with all the sensors installed at the heights ranging from 1.2 to 2 m. There were 160 points measured in total, with 81 points located inside the parks and 79 points on the surrounding streets. The control point in each site was continuously measured and the other points were measured by mobile equipment (Fig. 2b). Microclimate measurements were conducted for 3 days in each site. In each day, the measurements at all points were carried out once every two hours from 1 pm to 9 pm. The microclimate data collected at 3 pm and 9 pm are analyzed in this paper, representing the hottest period during daytime and early nighttime. More details on the field measurements and climatic data were presented in the previous paper 19.
Urban building information and land use data were collected through on-site surveys combined with the information from GoogleMaps, and GeoInfo Map provided by Survey and Mapping Office/ Lands Department, HKSAR. Fish-eye lens pictures at all measurement points were taken to calculate SVFs. The urban morphological information including terrain, building layout, greenery, and measurement points was consolidated into the ArcGIS models.
3.3. Parameter calculation by ArcGIS
By collecting the morphological data of all the sites, 10 GIS models with building form, area, height, and volume information were built in ArcGIS. A systematic process was established to calculate the design parameters at both site and point level:
Firstly, geoprocessing functions such as “Buffer”, “Intersect” etc. were used to select the buildings to be included in the calculation according to a certain site or point.
Then, Python scripts were written to calculate specific parameters such as DENSITY and HV using either mathematical function built-in Python or Data processing tools in ArcGIS.
Finally, the model function in ArcGIS was used to combine these steps to calculate the parameters proposed in Sect. 2.
Figure 3 shows the schematic workflow configuration to calculate aspect ratio for each point in the “Model” interface and the lines created for getting the 1-degree resolution in aspect ratio (AR) calculation.
3.4. Data analysis
Pearson correlation and multiple linear regression analysis were conducted to quantify the impact of each parameter and to develop empirical models that explain the UHI variation. In the regression analysis, dependent variables are the UHI intensities at 3 pm and 9 pm (UHI_3pm and UHI_9pm), while independent variables are the morphological parameters identified in the last section. For dependent variables, UHI is defined as the difference between the measured air temperature at each point and the data from the rural area weather station (Ta Kwu Ling weather station). Ta Kwu Ling weather station has been widely used by researchers in Hong Kong as a rural site 54,55.
To identify important urban morphological parameters, Pearson correlation was conducted between UHI_3pm and UHI_9pm with all parameters. Only those parameters proven to be statistically correlated with UHI intensities will be considered in the multiple linear regression.
Multicollinearity problems may exist in such a large quantity of independent variables. This problem was carefully dealt with through two main steps. Firstly, Pearson correlation was carried out to examine the correlation coefficient between every two design parameters. A coefficient higher than 0.80 indicates high collinearity, which means these two parameters cannot exist in one model. Secondly, based on the first step, models with a different combination of independent variables can be generated. Some scholars have suggested a formal detection of tolerance for these models to avoid multicollinearity 55. In this study, a tolerance of less than 0.10 was considered as a collinearity problem. The final models were obtained by removing the independent variables with a tolerance lower than 0.1 and P-value higher than 0.05.