The Estimation of a Remote Sensing Model of Three-Dimensional Green Space Quantity and Research into Its Cooling Effect in Hohhot, China

Abstract

With the continuous advancement of urbanization, the urban heat island effect has become increasingly prominent. Studying the cooling effect of urban green space can provide direction for improving the comfort of urban residents and reducing the harm caused by the urban heat island. In this paper, the band information was extracted from remote sensing images, and eight vegetation indices were calculated for Hohhot, such as normalized differential vegetation index (NDVI) and vegetation coverage (VC). The land surface temperature (LST) was calculated using the split-window algorithm, and the total 3D-GSQ, the three-dimensional green space quantity per unit area (3D-GSQP), and the single plant 3D-GSQ of different tree species were measured on the spot. The relationships between 3D-GSQP and eight vegetation indexes were processed by correlation analysis and regression analysis, and the remote sensing estimation model of 3D-GSQP was established. The correlation between 3D-GSQP and LST was established, and the cooling effect was analyzed in depth. The results indicate that: (1) The contributions to 3D-GSQ types of common tree in the built-up area of Hohhot are (from large to small): Salix babylonica, Populus hopeiensis, Gleditsia sinensis Lam, Salix matsudana, and Saphora japonica; and the 3D-GSQ of single types of shrubs are (from large to small): Amygdalus triloba, syringa oblata, Sorbaria sorbifolia, ligustrum lucidum, and Euonymus alatus. (2) The linear equation with the chlorophyll vegetation index (CVI) as the independent variable is suitable for the parameters of an inversion model for the 3D-GSQP (R² = 0.72), and the regression model of the two is: 3D-GSQP = −30.412 + 35.842 × CVI. (3) The 3D-GSQP in the study area is mainly in the range of 0–15.78 m³/m², and in such range it has an obvious negative relationship (R² = −0.73) with LST to the level of 0.01 (unilateral). When the 3D-GSQP increases by 1 m³/m², the LST decreases by 0.55 °C, and this result sufficiently indicates that the urban green has an obvious cooling effect. In the future, in order to improve the living environment of urban residents, in addition to considering the green space area, we should pay more attention to the requirements of the 3D-GSQP index.

1. Introduction

Sustainable Development Goal 11 of the United Nations’ (UN) 2030 Agenda emphasizes making cities and human settlements inclusive, safe, resilient, and sustainable [1]. According to the United Nations Population Organization statistics, the global urban population has increased rapidly since 1950. In 2018, 55% of the world’s total population lived in urban areas; and it is estimated that 68% will by 2050 [2]. Rapid urbanization has become the core element to achieving the Sustainable Development Goals. To achieve this goal, it is necessary to alleviate a series of environmental problems caused by rapid urbanization, such as local-scale biodiversity and dramatic changes in hydrological systems, and even urban ecological environment deterioration and local temperature rise.

Microclimate changes caused by local temperature increases impact the evapotranspiration of vegetation and the water and carbon cycles in urban ecosystems [3]. In addition, the rise in temperature plays an important role in the process of disturbing thermal power, which affects the location of precipitation [4]. There is also an interaction between urban heat island circulation and sea breeze circulation, accelerating the speed of sea breeze moving to the city center [5]. Elevated local temperatures can affect urban climate and cause severe drought in the region [6]. An elevated local temperature reduces indoor and outdoor thermal comfort, and even affects human health [7,8], causing epidemics such as dengue fever. Additionally, higher temperatures increase the risk of heat-related mortality [9].

Local temperature rise, as a local climate feature of a city, is closely related to local people’s heat source emissions and underlying surface pattern, in addition to local geographical and climatic characteristics [10]. Scholars have found that the decrease in vegetation and the increase in human heat release are the main reasons for the increase in urban heat island intensity. The urban expansion caused by urbanization also leads to changes in landscape configuration. The natural landscape is replaced by buildings, and the surface temperature rises, resulting in the heat island effect [11,12,13]. Reducing the amount of heat release is becoming more and more difficult. Under this premise, how to use urban green space to reduce the local temperature of the city has become a hot topic for scholars [14,15,16,17,18,19].

A large number of studies have shown that urban green space has the effect of alleviating the urban heat island effect. Part of the light absorbed by the leaves is released by long-wave radiation [20,21]. Part of it is released by sensible heat of heat conduction [22]. Most of the energy is released by transpiration and reduces leaf temperature, and only a small part of the energy is fixed by organic matter. Vegetation realizes heat energy exchange with air through photosynthesis and transpiration. The shading effect of plants can also reduce local temperature. The intensities of shading and transpiration are related to the scale of urban green space. Only when the green space reaches a certain scale can it regulate local microclimate. In addition, the cooling effect is also related to plant species and configuration. The cooling effects of various plants are as follows: broadleaf > shrub > conifer [23]. Trees with larger canopy volumes have better effects on reducing surface temperature than shrubs and lawns, and their shading effect can reduce solar radiation by 50–90% [24]; at the same time, trees have a large number of leaves and strong transpiration. In fact, the sum of the stem and leaf parts of green plants, that is, the green space quantity, as the dominant factor, determines the shading and transpiration of green space, and is the key to regulating the urban climate and improving the quality of the urban thermal environment [25]. Therefore, a scientific increase in urban the green space quantity is an effective way and important standard to improve the level of the modern urban landscape [26].

As the main index of urban green land research, the conception of the green space quantity was firstly raised by Watson, the British ecologist, in 1947; after that, there has been more and more research on the green space quantity performed by domestic and foreign scholars. Currently, the definitions of the green space quantity made by Chinese scholars are mainly categorized into two categories. The first category focuses on the research on leaf area, i.e., the total amount of green plants in a unit area, and it is calculated using half of the total area of all leaves of the plant [27], and it is called two-dimensional the green space quantity; the other category focuses on the research of plant volume, i.e., the space occupied by the stems and leaves of plant [6], and it is called three-dimensional green space quantity (3D-GSQ) [28]. Affected by human and material resources, the scale of urban green space research based on field measurement is mostly small- and medium-sized.

Based on this, this paper attempts to establish a large-scale method for studying the 3D environment of green space. The remote sensing images were used to invert the NDVI; FVC and eight other vegetation indexes; and LST in the urban area of Hohhot, China. Combined with the field measurement data of the total 3D green volume of the sample plot and the green volume of different tree species in the sample plot, the 3D green volume remote sensing estimation model was established after statistical analysis and processing, which laid the foundation for studying the spatial distribution of the 3D environment of large-scale green space. The relationship between the three-dimensional green space quantity per unit area (3D-GSQP) and LST was further analyzed to explore the cooling effect of 3D-GSQ.

The research on the urban thermal environment only relying on FVC, NDVI, greening rate, green space proportion [17,29,30], patch situation, and other green space indicators cannot fully reveal the mode and mechanism of how green space improves the urban thermal environment. Studying the spatial distribution of 3D-GSQ and its cooling effect aims to provide an important guarantee for the scientific and rational layout of urban green space and the construction of more targeted planning indicators, but also to alleviate urban heat islands, construct urban livable environments, and realize urban green development [31].

2. Materials and Methods

2.1. Study Area

The city of Hohhot is located in the middle part of Inner Mongolia (110°46′–112°10′ E, 39°35′–40°51′ N). It is located in the Tumochuan Plateau and is located north, next to Yinshan Mountain (as Figure 1). There are altogether 350 kinds of wild animals in the governing area, and they belong to more than 40 families. In terms of natural vegetation, the forest vegetation transforms to meadow vegetation from southeast to northwest gradually, and the halophytic vegetation, mire vegetation, and desert vegetation also grow here. As a typical city located in arid–semi-arid regions, Hohhot is not only an important node in the ecological security barrier in the northern border of China, but also has an important role in Hohhot–Baotou–Erdos–Yulin urban agglomeration, which will be vigorously cultivated and developed in The Syllabus of the Fourteenth Five-year Planning of China. The ecological situation of Hohhot is not only related to the ecology and development of Inner Mongolia Autonomous Region, but also to the ecological security of North China, Northeast China, Northwest China, and even the whole of China.

In recent years, urbanization has become faster, and the tropical island effect of Hohhot is becoming more and more obvious [32,33,34,35]. Studies have shown that the urban heat island of Hohhot has doubled in recent years. From 1997 to 2007, the area of the high temperature zone of surface temperature increased by nearly 10 times. The diffusion trend of the urban heat island kept spatial consistency with urban expansion.

2.2. Remote-Sensing Image Data

Considering the features of seasonal variation for vegetation growth in the region, the optimal time period for researching the cooling effect of 3D-GSQ of the urban green should be from July to October. Meanwhile, according to the requirement for using the remote sensing images that the cloud coverage rate should be less than 2%, the Landsat OLI data on 20 September 2018 obtained from the website of the USA Geological Surveying Bureau were selected for this research (USGS, https://earthexplorer.usgs.gov/ accessed on 3 July 2019). This research involved the images in two scenes and the data verification was performed with the existing land coverage data on the research area and the historical Google Earth images of different periods. Precision can reach 89.3%. The kappa coefficient is 0.83. The data such as the land surface temperature (LST), index of vegetation coverage (VC), and chlorophyll vegetation index (CVI) were interpreted and obtained from Landsat OLI data.

2.3. Research Method

2.3.1. The Calculation of Land Surface Temperature (LST) and Vegetation Index Based on Remote Sensing Images

Vegetation indexes quantitatively show the vitality of vegetation. They are more sensitive than using a single band to detect green vegetation. Each vegetation index has its specific expression of green vegetation. Among them, NDVI is sensitive to green vegetation and is used to study crops and semi-arid areas [36]. VC mainly expresses vegetation cover. The soil-adjusted vegetation index (SAVI) and modified soil-adjusted vegetation index (MSAVI), based on physical knowledge, combine the interactions of electromagnetic wave radiation, atmosphere, vegetation cover, and soil background. The enhanced vegetation index (EVI) and atmosphere resistance vegetation index (ARVI) are mainly measured using thermal infrared remote sensing, and the influence of atmospheric scattering is corrected by increasing the blue band. The difference vegetation index (DVI) is sensitive to the change in soil background. As a chlorophyll content index, the chlorophyll vegetation index (CVI) can maintain the sensitivity to chlorophyll content. The above vegetation indices reflect soil, atmospheric scattering, and vegetation. We selected the above vegetation indices. LST uses Qin Zhihao’s single window algorithm [37]. The specific formulae for the indexes and LST are shown in

Table 1. The formulae for different indexes.

Index Name	Formula
Land Surface Temperature (LST)	$LST=\left\{a_{\lambda }\left(1-C_{\lambda }-D_{\lambda }\right)+\left[b_{\lambda }\left(1-C_{\lambda }-D_{\lambda }\right)+C_{\lambda }+D_{\lambda }\right]T_{\lambda }-D_{\lambda }{T}_{\lambda }\right\}÷C_{\lambda }$
Normalized Differential Vegetation Index (NDVI)	$NDVI=\left({\rho }_{NIR}-{\rho }_{Red}\right)÷\left({\rho }_{NIR}+{\rho }_{Red}\right)$
Vegetation Coverage (VC)	$VC=\left({\rho }_{NDVI}-{\rho }_{NDV{I}_{\mathit{soil}}}\right)÷\left({\rho }_{NDVI\mathit{veg}}-{\rho }_{NDV{I}_{\mathit{soil}}}\right)$
Soil Adjustment Vegetation Index (SAVI)	$SAVI=\left({\rho }_{NIR}-{\rho }_{Red}\right)÷\left({\rho }_{NIR}+{\rho }_{Red}+L\right)\times \left(1+L\right)$
Modified Soil Adjustment Vegetation Index (MSAVI)	$MSAVI=\left[\left(2{\rho }_{NIR}+1\right)-\sqrt{{\left(2{\rho }_{NIR}+1\right)}^2-8\left({\rho }_{NIR}-{\rho }_{Red}\right)}\right]÷2$
Enhanced Vegetation Index (EVI)	$EVI=2.5\times \left({\rho }_{NIR}-{\rho }_{Red}\right)÷\left({\rho }_{NIR}+6{\rho }_{Red}-7.5{\rho }_{Blue}+1\right)$
Atmosphere Resistance Vegetation Index (ARVI)	$ARVI=\left[{\rho }_{NIR}-\left(2{\rho }_{NIR}-{\rho }_{Blue}\right)\right]÷\left[{\rho }_{NIR}+\left(2{\rho }_{NIR}-{\rho }_{Blue}\right)\right]$
Differential Vegetation Index (DVI)	$DVI={\rho }_{NIR}-{\rho }_{Red}$
Chlorophyll Vegetation Index (CVI)	$CVI=\left({\rho }_{NIR}÷{\rho }_{Green}\right)\times \left({\rho }_{Red}÷{\rho }_{Green}\right)$

.

In this table, a_λ and b_λ are constants, whose values are −67.35535 and 0.458606, respectively; C_λ and D_λ are intermediate variables; T_λ is brightness temperature (K); ρ_NDVIveg and ρ_NDVIsoil are maximum and minimum values of NDVI; ρ_NDVI is value of NDVI; ρ_NIR is the reflectivity of the near-infrared band of remote sensing images; ρ_Red is the reflectivity of the infrared band of remote sensing images; L is the soil adjustment coefficient, value 0.5; ρ_Blue is the reflectivity of the blue band of remote sensing images; ρ_Green is the reflectivity of the green band of remote sensing images.

2.3.2. The Method for Sample Plot Selection Based on the Vegetation Coverage Index (VC)

We researched the regional VC with the ENVI 5.2 software (New York, NY, USA) and performed grading of the calculation results with the principle of density slicing: 2 or 3 sample areas were selected in each VC grade, and each sample area was divided into different sample plots according to the borders of internal land parcels. If the plant density inside the sample area was very large and it was difficult to measure variables plant by plant in the green land, three sample plots that were 10 × 10 m were selected to measure variables plant by plant. If it was convenient to enter the sample area and easy to measure plant by plant, it was divided into many sample plots according to the borders of internal land parcels, and the total 3D-GSQ of each sample plot was measured. When selecting the sample area, it should be far away from a water body, plants, and the urban main roads, so as to eliminate the effects of water bodies, industrial combustion, wasted water, wasted gas, and automobile exhaust on LST.

There are mainly two methods for researching 3D-GSQ. The first is to use “the three-dimensional quantity reckoned with two-dimensional quantity,” on the basis of artificial decisions and measurement of the coverage area, number of plants, crown and stem size, and other characteristic data of different kinds of trees in the image. The computer will simulate the volumes of different kinds of trees and obtain the 3D-GSQ. With this method, the previous 2D data can be converted into 3D data [38]; however, this method requires remote sensing images with high resolution. The other method is to use “the three-dimensional quantity deduced by three-dimensional quantity,” by dividing the plants by their volumes and expected sizes after growth. Then, the 3D-GSQ of plants can be obtained using calculation with the statistical data of crown pattern, crown height, crown breadth, and other indexes of plants, taking the volume formulae of different morphs into consideration. After the weighted average value is obtained from the sample areas with the same level of plant coverage, the 3D-GSQ of a single plant can be obtained, and the 3D-GSQ of the current level of vegetation coverage can be obtained. Finally, the 3D-GSQ with different levels of vegetation coverage and the total 3D-GSQ of the area under study can be obtained. The method of “deducing three-dimensional quantity with three-dimensional quantity” is relatively simple, and its accuracy is also higher.

The method of “deducing three-dimensional quantity with three-dimensional quantity” is used in the measurement of 3D green quantity [39], and the 3D-GSQ of each sample area should be practically measured. At the same time, the areas corresponding to VC of different levels in various sample plots should be obtained with the calculations of ArcGIS 10.2 software (Redlands, CA, USA). Then, the 3D-GSQP corresponding to VC of various levels and the total 3D-GSQ of the sample plot can be obtained by weighing. Finally, the total 3D-GSQ of the research area can be obtained. The concrete investigation and measurement data include the twig height, crown height, and crown breadth of arbor; the crown height and crown breadth of a single plant of the arbor; the relevant volumetric indexes of an artificial bush; and the height of herbaceous plants.

Arbor: The crown height was measured with a height indicator and the crown breadth was obtained by measuring the diameter of its projection at high noon with tape twice.

Bush: The crown diameter of individual round single plant, the cross section, the crown height of the individual flat bush, and the volume and other relevant indexes of artificial bushes were measured with tape.

Herbaceous plant: The average height of the herbaceous plant was measured with tape, and the area of herbaceous sample plot was measured with tape or a portable GPS device.

We imported the data of 3D-GSQ practically measured for the formula for volumetric quantity [39], and the total 3D-GSQ of each sample plot (V_Y) could be obtained. The areas corresponding to different levels of VC (S_VC) and the total area of sample plot (S_Y) can be obtained by interpretation of remote sensing images, and the quotient of the two is the area proportion of the pattern spots of VC at different levels in the sample plot. By multiplying the proportion of the pattern spots with the total 3D-GSQ of the sample plot, the 3D-GSQ corresponding to various levels of VC in the sample plot (V_ni) is obtained, i.e.,

$${\mathrm{V}}_{\mathrm{ni}}={\mathrm{V}}_{\mathrm{Y}}\times \frac{{\mathrm{S}}_{\mathrm{VC}}}{{\mathrm{S}}_{\mathrm{Y}}}$$

(1)

in which n represents VC of different levels, and its value range is 1–5: 0–0.1 is represented by 1, 0.1–0.3 is represented by 2, 0.3–0.5 is represented by 3, 0.5–0.7 is represented by 4, and 0.7–1.0 is represented by 5; i represents the number of sample plots with the same level of VC.

The sum of the 3D-GSQ corresponding to the same VC level in all the sample plots (V_n1 + V_n2 + V_n3 + ... + V_ni) is taken, and the average of the 3D-GSQ corresponding to the same level of VC is obtained (V_nz) when it is divided by the sum of the areas of various kinds of VC in all the sample areas by interpretation (S_nz). The 3D-GSQ per unit area corresponding to various levels of VC (V_n) is obtained, i.e.,

$${\mathrm{V}}_{\mathrm{nz}}={\mathrm{V}}_{\mathrm{n}1}+{\mathrm{V}}_{\mathrm{n}2}+{\mathrm{V}}_{\mathrm{n}3}+\dots +{\mathrm{V}}_{\mathrm{ni}}$$

(2)

$${\mathrm{V}}_{\mathrm{n}}=\frac{{\mathrm{V}}_{\mathrm{nz}}}{{\mathrm{S}}_{\mathrm{nz}}}$$

(3)

The areas corresponding to VC of different levels (S_n) are multiplied by interpretation with the 3D-GSQ per unit area (V_n) corresponding to VC of various levels, and the total 3D-GSQ of the urban area of Hohhot (V_z) is obtained; i.e.,

$${\mathrm{V}}_{\mathrm{z}}={\mathrm{V}}_1\times {\mathrm{S}}_1+{\mathrm{V}}_2\times {\mathrm{S}}_2+\dots +{\mathrm{V}}_{\mathrm{n}}\times {\mathrm{S}}_{\mathrm{n}}$$

(4)

2.3.3. The Method of Data Analysis

1.

Establishing the regression model

Extract various kinds of vegetation indexes corresponding to VC of various levels of the sample plot from the images after pre-treatment and establish the regression model of the practically measured 3D-GSQP in the sample plot and various vegetation indexes, including the nomadic linear regression model, multivariate linear regression model, and non-linear regression model; the accuracy of the model is evaluated using the absolute error, root-mean-square error, and relative error.

2.

The correlation analysis

Select sample points in the research area with the ArcGIS10.2 software randomly and calculate the 3D-GSQP and LST values corresponding to the sample points with the 3D-GSQP regression model and LST inversion formula, and then analyze the correlation between the two with the SPSS Statistics 19 software (Armonk, NY, USA) (as in Figure 2).

3. Results

3.1. Estimation of the 3D-GSQ Model

3.1.1. The Results of the Measurement and Calculation of the Sample Plot

The sample plot was divided into different levels based on the results of the VC calculations and the principle of density slicing, and five levels were obtained, 0.1, 0.3, 0.5, and 0.7, as demarcation points: 0–0.1, 0.1–0.3, 0.3–0.5, 0.5–0.7, and 0.7–1.0. These were the results of the spatial distribution of the VC level (as shown in Figure 3). Different sample areas according to different levels of VC were selected, and finally, 13 sample areas and 88 samples plots were selected (as in Figure 4).

1.

The spatial distribution of the vegetation coverage degree (VC) in the research area

The sequence of areas corresponding to the VC of various levels (from larger to smaller) is: (0.7–1.0) > (0.3–0.5) > (0.5–0.7) > (0.1–0.3) > (0–0.1). The built-up area is dominated by VC with the levels of 0.3–0.5 and 0.5–0.7, which is mainly collected in the moderate and advanced residential quarters and campus, and the land parcels corresponding to the VC level of 0.7–1.0 are mainly collective in tree gardens, botanical gardens, and other parts with a high afforestation rate and the forest and plowland at the northern and southern sides of Yinshan Mountains to the north of Hohhot. The land parcels corresponding to VC with the level of 0–0.1 in the research area are mainly collectively distributed within the Second Ring Road, and they are mainly the land used for man-made building and traffic facilities. Being affected by the resolution ratio of images, the green land with large areas exceeding 30 × 30 m in the built-up area mainly exists in the form of parks, and the parks also include buildings and roads, etc., and they affected the overall VC value of the pixel to a certain extent.

2.

The situation of vegetation in the sample plot

The site surveying indicated that, besides the young arbor and bush in the newly built parks, the other vegetation in the sample plot was mainly mature vegetation. The common trees include: Sophora japonica, Populus hopeinesis, Salix babylonica, Ulmus pumila, Salix matsudana, Saina chinensis, Pinus tabulaeformis, Gleditsia sinensis Lam, Picea asperata, Amygdalus davidiana, Sabina chinensis, Fraxinus chinensis, Lariz gmelini, Pinus sylvestris var. Mongolica, Euonymus maackii, Platycladus orientalis, Armeniaca vulgaris, etc. The common bushes include Syringa oblata, Sorbaria sorbilfolia, Buxus sinica var. Parvilfolia, Rosa xanthina, Ulmus pumila cv. Jinye, Euonymus alatus, Amygdalus triloba, Ligustrum lucidum, etc.

The average 3D-GSQ of different vegetation is different, the 3D-GSQ of herbaceous plants is dependent on its area and height, and the 3D-GSQs of arbors and bushes are also different because of their features. The contributions of the 3D-GSQ per plant of the common tree in the sample plot (from bigger to smaller) are: Salix babylonica (255.51 m³/plant), Populus hopeiensis (93.73 m³/plant), Gleditsia sinensis Lam (67.55 m³/plant), Salix matsudana (62.55 m³/plant), and Saphora japonica (56.51 m³/plant). The contributions of the 3D-GSQ per plant of the common shrub in the sample plot (from bigger to smaller) are: Amygdalus triloba (65.29 m³/plant), syringa oblata (22.49 m³/plant), Sorbaria sorbifolia (13.99 m³/plant), ligustrum lucidum (11.03 m³/plant), and Euonymus alatus (4.05 m³/plant).

The 3D-GSQP corresponding to the different VC in the research area is as follows (

Table 2. The statistics of three-dimensional green space quantity (3D-GSQ) in the urban area of Hohhot.

Category	VC Level
	0–0.1	0.1–0.3	0.3–0.5	0.5–0.7	0.7–1.0
Total 3D-GSQ of sample plot (m3)	0.00	24,705.18	35,386.86	13,129.67	58,289.79
3D-GSQP of sample plot (m3/m2)	0.00	0.17	0.58	0.48	17.14
Corresponding total area in research area (km2)	5220.90	25,625.97	57,994.83	56,640.78	62,729.55
Corresponding 3D-GSQ in research area (km3)	0.00	43.73	333.62	273.63	10,754.39
Total 3D-GSQ in research area (km3)	11,405.36

):

The statistical results indicate that the 3D-GSQP corresponding to the level of 0.5–0.7 is lower than the 3D-GSQP corresponding to the level of 0.3–0.5, and the VC of the former is larger than the VC of the latter. For the sample plot with a VC of 0.7–1.0, its 3D-GSQP is obviously larger than that of the other sample plot. The vegetation in this sample plot is mainly arbor, bush, and herbaceous plants with large 3D-GSQP values.

From the discussion above, we can see that VC cannot present the status of vegetation totally, and a high value of VC does not completely equal a large 3D-GSQ. The total 3D-GSQ and 3D-GSQP for the VC level of 0.5–0.7 are obviously lower than those for the VC level of 0.3–0.5. The sample plot corresponding to this level is dominated by newly built parks, and the green land configuration in the park is arbor, bush, and herbaceous plants. However, the crown height and crown breadth of arbor and bushes are normally small as their cultivation times are relatively short.

3.1.2. Model Estimation

We extracted the 3D-GSQP of 88 sample plots and their corresponding values of the vegetation index and established the regression model of GDP and the vegetation index. Fifty-eight portions of data were selected randomly for the regression simulation and the error analysis was performed for the remaining 30 portions of data.

1.

The situation of vegetation in the sample plot

We established the nomadic regression model between eight vegetation indexes and the 3D-GSQP of 58 sample plots and performed the significance testing of the regression results with the F test. The results are shown in

Table 3. The statistics of 3D-GSQ in the urban area of Hohhot.

Vegetation Index	Equation	a	b	R2	F	Sig.
SAVI	$y=bx+a$	−4.43	22.10	0.42	39.83	0.00
VC		−23.44	43.35	0.42	41.03	0.00
NDVI		−4.46	31.96	0.42	41.03	0.00
MSAVI		−4.77	22.68	0.35	30.18	0.00
EVI		−0.49	−9.19	0.10	6.29	0.00
DVI		−2.85	2.11	0.23	16.84	0.00
ARVI		−6.85	17.49	0.57	73.76	0.00
CVI		−30.41	35.84	0.72	140.61	0.00

.

The main parameters of the simulation equation, the test statistics of R² and F, and the test significance of F are listed in

Table 4. The statistics of 3D-GSQ in the urban area of Hohhot.

		SAVI	VC	NDVI	MSAVI	EVI	DVI	ARVI	CVI	3D-GSQP
SAVI	Pearson correlation	1	1.00 **	1.00 **	0.99 **	−0.84 **	0.96 **	0.92 **	0.73 **	0.65 **
	Significance (unilateral)		0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
	N	58	58	58	58	58	58	58	58	58
VC	Pearson correlation	1.00 **	1	1.00 **	0.99 **	−0.83 **	0.95 **	0.92 **	0.73 **	0.65 **
	Significance (unilateral)	0.00		0.00	0.00	0.00	0.00	0.00	0.00	0.00
	N	58	58	58	58	58	58	58	58	58
NDVI	Pearson correlation	1.00 **	1.00 **	1	0.99 **	−0.83 **	0.95 **	0.92 **	0.73 **	0.65 **
	Significance (unilateral)	0.00	0.00		0.00	0.00	0.00	0.00	0.00	0.00
	N	58	58	58	58	58	58	58	58	58
MSAVI	Pearson correlation	0.99 **	0.99 **	0.99 **	1	−0.86 **	0.97 **	0.90 **	0.69 **	0.59 **
	Significance (unilateral)	0.00	0.00	0.00		0.00	0.00	0.00	0.00	0.00
	N	58	58	58	58	58	58	58	58	58
EVI	Pearson correlation	−0.84 **	−0.83 **	−0.83 **	−0.86 **	1	−0.96 **	−0.58 **	−0.37 **	−0.32 **
	Significance (unilateral)	0.00	0.00	0.00	0.00		0.00	0.00	0.00	0.01
	N	58	58	58	58	58	58	58	58	58
DVI	Pearson correlation	0.96 **	0.95 **	0.95 **	0.97 **	−0.96 **	1	0.77 **	0.55 **	0.48 **
	Significance (unilateral)	0.00	0.00	0.00	0.00	0.00		0.00	0.00	0.00
	N	58	58	58	58	58	58	58	58	58
ARVI	Pearson correlation	0.92 **	0.92 **	0.92 **	0.90 **	−0.58 **	0.77 **	1	0.84 **	0.75 **
	Significance (unilateral)	0.00	0.000	0.00	0.000	0.00	0.00		0.00	0.00
	N	58	58	58	58	58	58	58	58	58
CVI	Pearson correlation	0.73 **	0.73 **	0.734 **	0.69 **	−0.37 **	0.55 **	0.84 **	1	0.85 **
	Significance (unilateral)	0.00	0.00	0.00	0.00	0.00	0.00	0.00		0.00
	N	58	58	58	58	58	58	58	58	58
3D-GSQP	Pearson correlation	0.65 **	0.65 **	0.65 **	0.59 **	−0.32 **	0.48 **	0.75 **	0.85 **	1
	Significance (unilateral)	0.00	0.00	0.00	0.00	0.01	0.00	0.00	0.00
	N	58	58	58	58	58	58	58	58	58

** Significantly correlated at the level of 0.01 (unilateral).

. The table shows that the correlations between EVI, MASVI, and DVI and 3D-GSQP are relatively low, and the inversion result is not obvious, and this means that the turbulence of these three parameters related to the inversion of 3D-GSQP is small in this time period. The inversion accuracies are SAVI, NVDI, VC, and 3D-GSQP are consistent, and the difference is not very obvious.

The ARVI and CVI are more suitable for inversion parameters. The nomadic linear regression model established with the CVI and 3D-GSQP is better, and this means that there is a closer relation among the ARVI, CVI, and 3D-GSQP.

2.

The multivariate linear regression between the vegetation index and 3D-GSQP

We analyzed the correlations between eight vegetation indexes and the 3D-GSQP of 58 sample plots with the SPSS software (as Table 4) and eliminated the disturbance of the parameter collinearity in the simulations with the method of collinearity diagnosis. Finally, the optimal regression model of 3D-GSQP was obtained with the method of gradual regression (see Table 4; the analysis procedure is omitted here).

The correlation coefficient between the EVI and 3D-GSQP is small, at −0.22, and there is a negative correlation with a low grade. The results are consistent with the results of nomadic linear regression. The correlation coefficients between CVI and ARVI and 3D-GSQP are very high: 0.85 and 0.75, respectively. The correlations between VC and NDVI and 3D-GSQP are consistent, and this means that the two parameters have consistent degrees of effect on 3D-GSQP. From the correlation coefficient table, it can be seen that the parameters participating in the regression simulation are significantly correlative when the (unilateral) confidence coefficient is 0.01.

In order to avoid the disturbance of collinearity of parameters as independent variables, collinearity diagnosis was performed for the parameters continually. The diagnosis results indicate that the R² value of the multivariate linear regression model equation of eight vegetation indexes is high (0.82); however, the eigenvalues of all the parameters in the regression simulation were all 0.00, and the conditional index was 130.14; this means that there is collinearity among the independent variable parameters in the regression equation, and such a regression equation cannot be used for a 3D-GSQP inversion simulation.

Therefore, gradual regression was further used to screen all the vegetation indexes. The results show that only the CVI, among all the vegetation indexes participating in the model regression, was reserved as the sole independent variable, and all the other variables were eliminated. Then, the multivariate regression equation became the nomadic linear regression equation with the CVI as the independent variable, and its results are consistent with those of nomadic linear regression, for the correlation coefficients are both 0.85. This means that the regression model with 3D-GSQP can be established with the CVI as the independent variable instead of seven other indexes, and it also verified the accuracy of the nomadic linear regression model generated with 3D-GSQP with the CVI as the independent variable. Therefore, the nomadic linear regression model with the CVI and 3D-GSQP was obtained (refer to Figure 5), and the regression equation is:

$$(3D\text{-}GSQP)=-30.412+35.842\times CVI$$

(5)

3.

The non-linear regression model between the vegetation index and 3D-GSQP

For the correlation analysis between the vegetation index and 3D-GSQP, it can be seen that all eight vegetation indexes are significantly correlated with 3D-GSQP, and considering the results of the regression analysis above, the CVI was selected for the simulation of the non-linear inversion model, to create a better regression simulation model that has a more accurate inversion of 3D-GSQP and is more sensitive to its dynamic variation (see Figure 6 and

Table 5. Simulation results of the vegetation index and 3D-GSQP regression per unit area.

Vegetation Index	Regression Equation	R2	F Value	Sig.	Constant Number	b1	b2	b3
CVI	Logarithmic function equation	0.67	113.83	0.00	5.89	35.11
CVI	Reciprocal equation	0.62	90.53	0.00	39.53	−33.26
CVI	Quadratic equation with one unknown	0.79	106.09	0.00	41.98	−108.37	69.67
CVI	Cubic equation with one unknown	0.79	104.07	0.00	17.01	−35.43		21.77
CVI	Composite function equation	0.86	342.17	0.00	0	14,142.04
CVI	Power function equation	0.87	364.21	0.00	1.68	9.71
CVI	S-curve equation	0.86	343.38	0.00	10.19	−9.54
CVI	Growth function equation	0.86	342.17	0.00	−9.18	9.56
CVI	Exponential function equation	0.86	342.17	0.00	0	9.56
CVI	Logistic function equation	0.86	342.17		9705.54	0.00

).

Table 5 indicates the results of regression simulation of the CVI and 3D-GSQP and the main forms of non-linear model equations. Among them, the accuracy of the power function equation, S-curve equation, growth function equation, and exponential function equation is high, and their corresponding R² values are 0.87, 0.86, 0.87, and 0.86, respectively. The composite function equation and logistic function equation also have high accuracy; however, their constant numbers and b1 values are prominent, and they are not beneficial for the overall accuracy. The accuracies of the logarithmic equation and reciprocal function are not very good, and the corresponding R² values are 0.67 and 0.62, respectively.

In Figure 4, it can be seen that the points in the scatter diagram are mainly distributed near those of the power function curve equation, S-curve equation, growth function equation, and quadratic equation with one unknown, and considering the R² value, the CVI in the 3D-GSQP inversion model can be used as an output; therefore, the non-linear regression model of 3D-GSQP can be expressed as:

$$\ln (3D\text{-}GQP)=1.682+9.713\times \ln (CVI),$$

(6)

$$\ln (3D\text{-}GSQP)=\left(10.193-9.542\right)÷CVI,$$

(7)

$$\ln (3D\text{-}GSQP)=-9.18+9.557\times CVI,$$

(8)

$$(3D\text{-}GSQP)=41.975-108.367\times CVI+69.671\times CV{I}^2$$

(9)

4.

The test of the inversion model.

The data of the remaining 30 sample plots were regarded as the test sample to calculate the absolute error, root-mean-square error, and relative error and test the model accuracy (see

Table 6. The statistics of 3D-GSQ in the urban area of Hohhot.

	Absolute Error	Root-Mean-Square Error	Relative Error
Regression model of linear equation with one unknown	−0.25	1.13	85.70%
Regression model of power function	0.99	1.56	86.70%
Regression model of S-curve	−0.16	0.42	86.00%
Regression model of growth function	−0.16	0.39	85.90%
Regression model of quadratic equation with one unknown	−0.50	1.35	83.40%

).

The calculations show that there are differences between the predicted values obtained by the linear regression model and non-linear regression model and the actual measured value. The regression model of the power function produced the highest accuracy of 86.7%; however, its average error and root-mean-square of error are larger than those of the other models. Compared with the other three non-linear models, the average error and root-mean-square of error of the regression model of the growth function are the smallest, and they are −0.16 and 0.39, respectively, and its accuracy is the highest (85.9%). During this experiment, the regression model of the growth function was the best model out of the four non-linear models.

The corresponding CVI values of this sample plot did not exceed 1.36, though some CVI values of the research area obtained by the remote sensing inversion exceed this value. If the regression model of the growth function is selected, it will have a larger error than other models when the CVI value obtained by remote sensing inversion exceeds this value; therefore, this model was not suitable for the simulation. Finally, the linear equation with one unknown was selected as the regression model.

$$(3D\text{-}GSQP)=-30.412+35.842\times CVI$$

(10)

3.2. The Research on the Relation between 3D-GSQP and LST

3.2.1. The Spatial Distribution of Sample Points

We generated 3000 sampling points in the research area randomly with ArcGIS10.2 software. The sampling points in the Daqing Mountains, water area, main urban roads, and industrial area were discarded, and the information on the remaining 1764 sampling points was collected (Figure 7).

3.2.2. Establishing the Relation Model between 3D-GSQP and LST

We calculated the 3D-GSQP of 1764 random sampling points with the 3D-GSQP inversion model and established the regression model of the two with the LST obtained with inversion after the correlation analysis.

Table 7. Correlation analysis between 3D-GSQP and LST.

		3D-GSQP	LST
3D-GSQP	Pearson correlation	1	−0.73 **
	Significance (unilateral)		0.00
	N	1764	1764
LST	Pearson correlation	−0.73 **	1
	Significance (unilateral)	0.00
	N	1764	1764

** Significantly correlated at the level of 0.01 (unilateral).

indicates that 3D-GSQP and LST are significantly correlated at the level of 0.01 (unilateral), and the correlation coefficient is 0.73; therefore, there is an obvious negative correlation.

Samples were obtained from the feature database of 1764 points randomly sampled with ArcGIS10.2 software and from the numerical distribution of 3D-GSQP in the database (refer to Figure 8). The range of the total data is 0.01–32.9 m³/m², and they were mainly gathered within 15.78 m³/m². The proportion of such data in the total data is 85%. Further regression model analysis can draw the regression equation between 3D-GSQP and LST:

$$LST=-0.554\times (3D\text{-}GSQP)+43.601$$

(11)

3.2.3. The Research on the Cooling Effect of 3D-GSQP

The correlation analysis and the regression model further verified the negative correlation between 3D-GSQP and LST, i.e., when 3D-GSQP increases, LST will decrease gradually.

According to the growth rules of vegetation in Hohhot, the data of 3D-GSQP were mainly collected in the range of 0.01–15.78m³/m², and the calculation with the regression equation of LST in this range shows that when 3D-GSQP increases by 1 m³/m², the corresponding LST will decrease by 0.554 °C.

4. Discussion

Our study found that green space has a cooling effect, which is consistent with previous studies [40,41,42,43,44,45,46,47,48,49,50,51]. In the process of urban construction, in order to improve the comfort of urban residents and reduce the harm of urban heat islands, the government issued a number of policies to increase the green space rate and increase the area of green space. The fundamental reasons why green space can reduce local temperature are evaporation and shading [20,21,22,23,24,25,26]. These mainly depend on the volume of green space, as confirmed in our research results. An increase in green space volume will reduce the local temperature.

However, we also found that in the process of urban construction, emphasizing the increase in green area cannot directly increase the 3D-GSQP, and the final cooling effect will be different (refer to Figure 9). To increase the 3D-GSQP, there are three aspects to consider: (1) In the case of a limited number of trees, try to select trees with large GQ per plant. (2) Maximize the number of trees of the same species. (3) If the numbers of trees and tree species must remain unchanged, we should optimize the tree species collocation plan. When selecting tree species and quantity, economic conditions should be considered, and economic tree species with large GQ per plant should be selected as often as possible for urban greening.

In this study, it was found that, when the VC in the study area is 0.5–0.7, the corresponding 3D-GSQP is smaller than when the VC is 0.3–0.5 (refer to Figure 10). It may be because most green spaces corresponding to VC of 0.5–0.7 grade are new parks, and the ages of these trees and shrubs are low. Although the greening area can meet the national requirements, the actual volume of forest and shrubs is small, and the comfort is poor. Therefore, 3D-GSQP can be used as an indicator to further evaluate the greening level and the ecological effect of green space.

Our research has several limitations. Firstly, the effect of plant hollow on the 3D-GSQ was not considered in the establishment of the regression model of 3D-GSQP. Existing studies have found that hollow trees are important structural components of forest ecosystems. With the increase in tree diameter at breast height, the probability of hollow trees will increase [52]. Secondly, when testing the accuracy of the regression model, the selected sample points were different spatially, but for the same period of time; there was no accuracy test for different time scales, which will have directly affected the accuracy of the inversion accuracy test.

On the basis of this study, we can continue to study the cooling effect of 3D-GSQP at different times in different seasons, and for days and evenings, in the future. This study is based on an instantaneous LST acquisition time on 20 September 2018; however, previous studies have shown that the transpiration of tree species is different in different seasons and different times of day and night [53,54]. LST also varies in different seasons and periods [16,17,18], so it is necessary to further study the differences in cooling effects of 3D-GSQP.

5. Conclusions

(1)

The 3D-GSQ per plant of main trees in the built-up area of Hohhot

The sequence of the 3D-GSQ per plant of the common tree in the built-up area of Hohhot (from bigger to smaller) is: salix babylonica (255.51 m³/plant), populus hopeiensis (93.73 m³/plant), Gleditsia sinensis Lam (67.55 m³/plant), salix matsudana (62.55 m³/plant), and saphora japonica (56.51 m³/plant); the sequence of the 3D-GSQ per plant of the common shrubs (from bigger to smaller) is: amygdalus triloba (65.29 m³/plant), syringa oblata (22.49 m³/plant), Sorbaria sorbifolia (13.99 m³/plant), ligustrum lucidum (11.03 m³/plant), and Euonymus alatus (4.05 m³/plant).

(2)

The estimation of the remote-sensing model of 3D-GSQP

The linear equation with the CVI as the independent variable is suitable as a parameter of the 3D-GSQP inversion model (R² = 0.72), and the regression model is 3D-GSQP = −30.412 + 35.842 × CVI.

(3)

The cooling effect of 3D-GSQ

The 3D-GSQP and LST have larger ranges in the built-up area. The maximum and minimum values of 3D-GSQP are 32.9 and 0.01 m³/m², respectively, and the maximum and minimum values of LST are 50.8 and 23.85 °C, respectively. The 3D-GSQP in the research area is mainly in the range of 0–15.78 m³/m², and the proportion of it in the sampling points is 85%. The regression model of the two is LST = −0.554 × (3D-GSQP) + 43.601, and they are obviously correlated at the level of 0.01 (unilateral). When 3D-GSQP increases by 1 m³/m², the LST will decrease by 0.55 °C.

The requirements in China for the construction of urban forests in the current stage are restricted in 2D space indexes, including area, proportion, etc. The research revealed that 3D-GSQP can reflect the utility of urban forest more accurately, and it can not only be beneficial for the management and control of an urban city, but also restrain the city’s problems in a more accurate and effective way, so as to improve the residential environment.