The empirical analysis includes the trend of green space change and the impact of green space changes on air pollution and the microclimate. Thus, the above mentioned variables and hypotheses will be defined in the following section.

#### 2.3.1. Green Space Change Analysis

Analysis of green space change is achieved through such indicators as landscape ecology metrics. Landscape ecology metrics includes three levels: patch level, class level and landscape level.

The patch level is the sum of the grids, and is measured by calculating the characteristics of each patch (such as shape index, edge contrast index,

etc.). The class level is the sum of a group of the same category of patches, and is indicated by calculating the characteristics of all types of classes (such as class area, core area, percentage of landscape,

etc.). The landscape level is the sum of all patches or classes in the region, and is indicated by measuring the characteristics of all kinds of classes (such as Shannon’s diversity index, relative patch richness,

etc.) [

45].

When examining the green-space change in the Taipei Metropolitan Area from 1995 to 2007, we used landscape ecology metrics of the class level for analysis. Since the indicators of landscape ecology metrics are numerous and complex, an explanation of some indicators requires repetition. Therefore, 14 front indicators of landscape ecology metrics were selected and analyzed for the purpose of research, such as Percentage of Landscape (PLAND), Number of Patches (NP), Patch Density (PD), Mean Patch Area (AREA_MN), Area-weighted Mean Patch Area (Area_AM), Largest Patch Index (LPI), Mean Shape Index (MSI), Area-weighted Mean Shape Index (AWMSI), Mean Nearest Neighbor Distance (ENN_MN), Area-weighted Mean Nearest Neighbor Distance (ENN_AM), Percentage of Like Adjacencies (PLADJ), Splitting Index (SPLIT), Radius of Gyration (GYRATE_MN), Area-weighted Radius of Gyration (GYRATE_AM), Clumpiness Index (CLUMPY), Aggregation Index (AI). The formula, units and methodology employed in measuring these 14 indicators are explained in

appendix, Table A1.

#### 2.3.2. Impact of Green Space Change

(1). Definition of latent variables and of observed variables.

The purpose of analyzing the impact of green space change is to learn whether green space changes will affect the microclimate changes and air pollution changes, and if so, the degree of that influence. Therefore, we have provided the definitions of the perspective, the latent variables and the observed variables in

Table 1.

The perspective includes three parts: “change of green space,” “change of air pollution” and “change of microclimate.”

The “change of green space” perspective includes six latent variables: “change of landscape,” “change of fragmentation,” “change of aggregation,” “change of area,” “change of proximity” and “change of largest patch percentage.” With the exception of the observed variable of “changed area of maintaining and switching to green space,” the other observed variables for measuring each latent variable are the changed rates of the landscape ecology metrics index.

The “change of air pollution” includes one latent variable: “change of air pollution emission.” The observed variables for measuring the latent variable are the changed rates of different air pollutants, such as sulfur dioxide, nitrogen oxide, airborne particulate, carbon dioxide, nitric oxide and nitrogen dioxide.

The “change of microclimate” includes two latent variables: “change of rainfall type” and “change of temperature.” The observed variable for measuring the “change of temperature” latent variable is the “ change of mean annual temperature,” and the observed variables for measuring the “change of rainfall type” latent variable are the “ change of mean annual rainfall,” “change of light rainy days,” “ change of torrential rainy days,” and “change of non-rainy days”. According to Taiwan's Climate Change Science Report [

48] and the rainfall classification of the Central Weather Bureau, a standard of 0.1 mm ≤ daily precipitation <1.0 mm is defined as “light rainy day,” a standard of daily precipitation ≥50.0 mm is defined as a “torrential rainy day,” and a standard of daily precipitation <0.1 mm is defined as a “non-rainy day”.

**Table 1.**
The variables of PLS model.

**Table 1.**
The variables of PLS model.
Perspective | Latent Variables | Latent Variables Code | Observed Variables | Observed Variables Code |
---|

change of green space | change of Landscape | CL | | cPLAND cWAERA |

change of fragmentation | CF | change of NP change of PD change of SPLIT
| cNP cPD cSPLIT |

change of aggregation | CA | change of PLADJ change of CLUMPY change of AI
| cPLADJ cCLUMPY cAI |

change of area | CR | change of AREA-MN change of AREA-AM
| cAREA-MN cAREA-AM |

change of proximity | CN | change of ENN-MN change of ENN-AM
| cENN-MN cENN-AM |

change of largest patch percentage | CP | | cLPI |

change of air pollution | change of air pollution emission | CAP | change of SO_{2 }emission change of NO_{x} emission change of PM emission change of CO_{2} emission change of NO emission change of NO_{2} emission
| SO_{2} NO_{x} PM CO_{2} NO NO_{2} |

change of microclimate | change of rainfall type | CRT | change of mean annual rainfall change of light rainy day change of torrential rainy day change of non-rainy day
| Rain lrd brd nrd |

change of temperature | CST | | Temp |

(2). Set of hypothetical relationship

The PLS model constructed for this study includes an outer model and an inner model. In the hypothetical relationship of the outer model, with the exception of the relationships between the “changes of rainfall type” the latent variables and the “change of mean annual rainfall,” the “change of light rainy days” observed variables were negative, and the other relationships of latent variables and observed variables were positive. The hypothetical relationships of the inner model include the impact of green space change on air pollution change and the microclimate change, and the impact of air pollution change on the microclimate change (

Table 2).

**Table 2.**
Hypothetical relationship of latent variables in PLS model.

**Table 2.**
Hypothetical relationship of latent variables in PLS model.
Endogenous latent variables/exogenous latent variables | Change of Landscape | Change of fragmentation | Change of aggregation | Change of area | Change of proximity | Change of largest patch percentage | Change of air pollution emission |
---|

change of air pollution emission | − | + | − | − | + | − | none |

change of rainfall type | − | + | − | − | + | − | +/− |

change of temperature | − | + | − | − | + | − | +/− |

The above mentioned hypothetical relationships are as follows:

(1) The impact of green space change on air pollution change and microclimate changes.

The number and area of green space changes negatively affect changes in air pollution emissions and the microclimate. Thus, the “change of landscape” latent variables of the “change of green space” perspective are assumed to be opposite to the “change of air pollution emission,” the “change of rainfall type” and “change of temperature” latent variables.

A large green space is synergistically helpful in reducing air pollution, the temperature and changes of rainfall type. Therefore, reducing the size of the green area and the percentage of the largest patch negatively impacts air quality, temperature and rainfall type. Keeping with the above statement, for this paper, we assumed the “change of air pollution emission,” “change of rainfall type” and “change of temperature” latent variables were affected by the “change of area” and “change of largest patch percentage” latent variables of the “change of green space” perspective.

The aggregate effect of green space is the same as the scale effect of large green space; it can reduce air pollution, the temperature, and the change of rainfall type. Thus, the “change of aggregation” latent variables of the “change of green space” perspective are assumed to be opposite to the “change of air pollution emission,” “change of rainfall type” and “change of temperature” latent variables.

The greater the nearest neighboring distance of green space, the more dispersive the patches and the less their effect in reducing air pollution, the temperature and the change of rainfall type. Moreover, the fragmentation of green space also has the same effect as the proximity of green space. Thus, the “change of fragmentation” and “change of proximity” latent variables of the “change of green space” perspective are assumed to be comparable to the “change of air pollution emission,” the “change of rainfall type” and the “change of temperature” latent variables.

(2) The impact of air pollution change on microclimate change

Airborne particulate and sulfate aerosol reduce the volume of solar radiation and temperature through solar short wave radiation scattering (Because solar radiation enters into the atmosphere in the form of the short wave radiation, the more airborne particulate there are, the more the short wave radiation is reflected directly back into space. Thus, the above situation reduces the solar radiation reaching the earth surface). Air pollutants form easily in clouds, and clouds can reflect sunlight. In addition, clouds can warm through absorbing thermal radiation, as well as cool by diverting thermal radiation. The effect depends on the height and type of clouds [

49,

50,

51,

52,

53]. Thus, the “change of air pollution emission" latent variables are assumed to affect the “change of temperature” latent variables.

Air pollutants are the source of cloud condensation nuclei (CCN). When air pollutants increase, the formation of rain is more difficult due to the number of cloud droplets increasing while the size of cloud droplets becomes smaller. Therefore, the formation of rain requires a substantially greater number of cloud droplets. Such a situation results in a change in the level of total rainfall and the number of rainy days, decreasing the number of light rainy days while increasing the frequency of torrential rainy days [

52,

54]. Thus, the “change of air pollution emission” latent variables are assumed to affect the “change of rainfall type” latent variables.