Environmental Influence on NbS (Nature-Based Solution) Mitigation of Diurnal Surface Urban Heat Islands (SUHI)

Abstract

Utilizing 58 Landsat-7 images taken over 10 years, the current study investigated the relationship between the mitigation of surface urban heat islands (SUHIs) by NbSs (Nature-based Solutions) and influential variables such as physical variables of NbSs, environmental variables of the streets, and meteorological variables. Parks and permeable pavements are the two types of NbS devices under examination. Reference (i.e., unaffected by any NbS) and experimental (i.e., affected by only one NbS) areas were selected to perform the analysis. Areas affected by large water bodies or more than one NbS device were excluded. The cooling effect caused by NbS was linked to the influential variables by multiple regression models. Key findings included the following: Firstly, the distance to an NbS is more important than the area of an individual NbS, implying that small and evenly distributed NbS devices might have better overall cooling effects than large but sparsely placed NbS devices. Secondly, NbSs do not significantly contribute to cooling in districts with grid-type streets, while exhibiting significant cooling for districts with complex street patterns. Older districts with complex street patterns should be the focus of NbS implementation, not newer, modern districts. However, NbS cooling is sensitive to several variables in districts with complex patterns. NbS installation in those districts requires careful planning to maximize engineering investment. Lastly, maintenance can be essential to sustain the cooling capacity of NbSs over time.

1. Introduction

In the built environment, urban heat islands alter thermal balance and generate abnormal increases in temperature in the urban space, increasing the risk of grid failure, increasing urban dwellers’ mortality and morbidity, and decreasing air quality [1]. Numerous strategies can curb its development. Focusing on engineering measures, reflective roofs, green roofs, shading structures, reflective paving, vegetation/plantation, and water bodies are the most common. Based on the analysis of Khare et al. [2], vegetation, reflective surfaces, and water bodies are considered the most effective and practical engineering measures.

Most studies of high-albedo surfaces are still limited in scale [3,4]. Therefore, vegetation and water bodies are the primary practical engineering means to fight urban heat islands. Creating large water bodies in existing cities is often not feasible; therefore, the current study focuses on the effect of vegetation or green space. Vegetation and green space can be considered forms of NbS (Nature-based Solution). In the urban setting, NbSs also include nature-inspired stormwater control devices such as BMPs (Best Management Practices) or LIDs (Low-Impact Developments) [5,6], which also reduce heat islands [6]. Through shading and evapotranspiration, NbSs could significantly reduce temperature in urban spaces [7,8]. Fernández [9] determined that the horizontal temperature influence of green space could reach as far as 200 m.

Remote sensing provides a cost-effective means to study NbSs’ spatial influence on urban heat islands by observing surface heat islands (SUHIs) derived from thermal infrared data. Utilizing spaceborne remote sensors, Addas et al. [10] found that vegetation greenness (in the NDVI: Normalized Differential Vegetation Index) and built-up land use affected heat islands, with the latter having a greater influence. Another study in India [11] further highlighted such an influence in a city with a hot and humid climate. With a decrease in vegetation, farms, and grassland and an increase in built-up area from 1990 to 2020, the highest and the lowest land surface temperature increased by 11.55 °C and 8.35 °C, respectively.

Among existing remote sensing studies, only a few have adopted a holistic approach to evaluate variables influencing the cooling capacities of NbSs. Many studies have focused only on the types of NbS [7], without examining other influential variables. Using airborne remote sensing data from Sydney, Australia, Bartesaghi-Koc et al. [12] identified that the spatial arrangement of plants in NbSs played an essential role in cooling capacity, and the pavement type could significantly affect NbSs’ cooling effects. In terms of variables influencing NBSs’ cooling capacity, the number of attributes considered by this study was very limited. In another study in Rome, Italy, spaceborne, remotely sensed data assisted Marando et al. [13] in studying NbSs’ environmental constraints on cooling capacity. Nevertheless, their conclusions were still limited to variables related to plant conditions, such as irrigation schedule, soil structure, NDVI, etc. A broader, holistic view of variables affecting the cooling capacities of NbSs was still elusive. Indeed, Yu et al. [14] considered the influence of the attributes of artificial landscapes on urban heat islands and found that floor area ratio, average building height, and space congestion degree were the most influential factors. However, variables directly related to NbSs were missing from the picture. A few recent researchers, such as Huang and Wang [15], started to consider a more holistic approach by focusing on variables of land use/land cover (a.k.a., landscape quality), which cannot be used as practical guidance on NbS placement based on field-level street attributes. Indeed, a recent review [8] clearly pointed out that the effect of urban morphology is still one of the gaps that require future research.

Understanding how the internal (e.g., NbS type, NDVI, NbS size, etc.) and external variables (e.g., the height of buildings, the geometry of streets, meteorological conditions, etc.) at the field level affect NbSs’ cooling capacities is crucial in optimizing NbS deployment. This also guides future urban planning and design if urban heat islands are a concern. With Landsat imagery from 2010 to 2020, the current study provided a holistic understanding of the primary internal and external attributes affecting the diurnal cooling capacity of NbSs for SUHIs in humid, subtropical Taipei City.

2. Material and Methods

2.1. Site Description

Taipei City is the testing ground to study the relationship between various variables and the cooling capacity of NbSs. Figure 1 shows the geographical location of Taipei City in northern Taiwan. The city sits in a basin surrounded by mountain ranges and is not adjacent to the ocean. Tamsui River cuts through the mountain at the northwestern corner of the basin, providing the city with ocean access. Except for a few rivers, the city has no large water bodies. In the low-altitude part of the city, the annual mean temperature is around 21 °C. The annual accumulated precipitation is around 2400 mm [16]. The climate type is Cfa (temperate, hot summer with no dry season) in the Koppen climate classification system [17].

Taipei City is the largest city in Taiwan. However, urban sprawling is not an option due to its geographical limitations. Instead of expanding outward, Taipei City has become one of the most densely populated cities in the world [20]. There is no clear distinction between most zones in Taipei City, as zonings are mixed-use based on the practice in Taiwan [21].

27,000 m² of permeable surface had been implemented before 2005 [22]. Considering different types of NbS, including parks, green roofs, and urban farms, NbS coverage in Taipei City is far from negligible [23]. However, with one of the highest urban population densities in the world, heat islands are still a serious issue [24]. The tug-of-war between NbSs and heat islands made Taipei City an ideal candidate for the current study.

2.2. Data Sources

The current study utilized public data shown in

Table 1. Data sources utilized by the analysis.

Category	Description
Satellite data	Landsat 7 Collection 2, Level 2 data [25], providing thermal, NDVI, and solar azimuth data
Meteorological data	Temperature, precipitation, relative humidity, wind speed, wind direction, and solar irradiation of Taipei City regarding satellite data from 2010 to 2020 [16]
Area geographic data	GIS layer of large water bodies in Taiwan [26]
	The road network of Taipei City [27]
	Height of buildings [28]
Additional NbS information	Age, type, location, and area of green infrastructure devices in Taipei City [29]

. Collection 2, Level 2 Landsat 7 images [25], spanning from 2010 to 2020, with cloud cover lower than 30% (no cloud above Taipei City), were selected for the analysis. The thermal band is the focus of the current study. The raw resolution of the thermal band is 60 m, but the released data have been resampled to 30 m. Besides a simple conversion to temperature [25], no additional manipulation to satellite images was applied. It should be stressed that the temperature measured by Landsat imagery is the land surface temperature (LST), not the actual near-surface microclimate air temperature. The current study used satellite-derived LSTs to study SUHIs.

2.3. Methodology

2.3.1. Research Steps

The current study can be divided into four major phases. Figure 2 provides a summary of the steps, in which Step 1 (also Phase 1) involved identifying available data, Steps 2–4 (Phase 2) involved identifying experimental and reference sites, Step 5 (also Phase 3) included deriving the cooling levels at each experimental site, and Steps 6–8 (Phase 4) involved performing statistical analyses.

Phase 1 (Step 1: Select appropriate NbS types): The Taipei Green Map [29] provided information and locations of the following types of NbS: permeable asphalt, sidewalk permeable surfaces, parks, green campuses, green roofs, and urban farms. After evaluating the available data, permeable asphalt, sidewalk permeable surfaces, and parks were selected as research subjects, as the other three types of NbS (green campuses, green roofs, and urban farms) lack actual area information in the dataset. Nevertheless, the influences of green campuses, green roofs, and urban farms were still considered in Step 3 below.

Phase 2 (Steps 2–4: identify experimental and reference sites): This phase of action aimed to identify experimental (influenced by NbS) and reference sites (not influenced by NbS) so that the SUHI mitigation effect at each experimental site can be calculated. This phase contains three steps.

(1)

Step 2 kept areas unaffected by large waterbodies. Large waterbodies can cool the surrounding areas up to 740 m [30]. Therefore, this study excluded areas within 780 m of rivers from subsequent analysis.

(2)

Step 3 kept areas affected by only one NbS device. Fernández [9] determined that the horizontal temperature influence of green space can reach as far as 200 m. This study assumes an influential radius of green infrastructure as 180 m. Based on this assumed influential radius, areas influenced by more than one NbS were eliminated from the analysis, so any cooling effect can be confidently tagged to a single NbS.

(3)

Step 4 selected experimental and reference areas. After the previous steps, two kinds of areas are left: affected by only one NbS, or unaffected by any NbS. “Experimental areas” were selected from the former areas, and “reference areas” were selected from the latter, with the following criteria being observed:

(a)

Only street areas were selected;

(b)

Each experimental area has an NbS adjacent to its center;

(c)

For each experimental area, a reference area must be available nearby.

A total of 11 experimental areas were identified, as shown in Figure 3. Each experimental area typically had a corresponding reference area with the same number, with the exceptions of experimental areas #1 and #10, which shared the same reference area (#1). The reference areas were selected to be as close to the corresponding experimental areas as possible and to have similar street characteristics based on field observations. All street environmental variables of the same pairs of experimental and reference areas are identical.

Phase 3 (Step 5—compute NbS cooling effects): A total of 1842 experimental area cells (distributed in 58 Landsat images) were available for analysis. Cooling was simply computed by the temperature difference between the reference and experimental areas. At each cell $i$ in the experimental area, the magnitude of temperature mitigation ${∆T}_{i,mitigated}$ was calculated by Equation (1). In Equation (1), $T_i$ was the temperature of the cell $i$ in the experimental area, the associated reference area has n cells, and the temperature of cell j in the reference area is presented by $T_{ref,j}$. Only the temperature of ground surface cells (excluding visible vegetation, tree canopies, and rooftops) was considered in Equation (1).

$${∆T}_{i,mitigated}={\displaystyle \frac{\sum _{j=1}^n{T}_{ref,j}}{n}}-T_i$$

(1)

Phase 3 (Steps 6–8—identify experimental and reference sites): This phase of action aimed to generate statistical models to evaluate relationships between variables and cooling effects. This phase contains three steps.

(1)

Step 6 (select potential urban space characteristics variables): In addition to the type of NbS, initial variables in

Table 2. Potential NbS physical variables selected in this study.

NbS Physical Variable	Variable Symbol	Unit
NbS type	$NbS$	Unitless
NbS area	$A_{NbS}$	${\mathrm{m}}^2$
Distance from NbS	${Dist}_{NbS}$	$\mathrm{m}$
NbS age	${Age}_{NbS}$	Years
NbS NDVI	${NDVI}_{NbS}$	Unitless

,

Table 3. Potential street environmental variables selected in this study.

Street Environmental Variable	Variable Symbol	Unit
NDVI of the experimental area	${NDVI}_{exp}$	Unitless
Height of surrounding buildings	$H$	$\mathrm{m}$
Street aspect ratio (height of building/street width)	$H/W$	Unitless
Angle between street and wind directions	${\theta }_{wind-street}$	Degrees
Angle between the direction to NbS and the wind direction	${\theta }_{wind-NbS}$	Degrees
Angle between the solar azimuth and street directions	${\theta }_{sun-street}$	Degrees

and

Table 4. Potential meteorological variables selected in this study.

Meteorological Variable	Variable Symbol	Unit
Ambient hourly air temperature	$T$	$℃$
Hourly relative humidity	$RH$	%
Hourly wind speed	$U$	$\mathrm{m}/\mathrm{s}$
Hourly solar irradiation	$SR$	MJ/m2
2-day (48-h cumulative) accumulated precipitation	$P_{2d}$	$\mathrm{m}\mathrm{m}$
3-day (72-h cumulative) accumulated precipitation	$P_{3d}$	$\mathrm{m}\mathrm{m}$
5-day (120-h cumulative) accumulated precipitation	$P_{5d}$	$\mathrm{m}\mathrm{m}$

were selected based on the literature. These initial variables were further refined in Step 7.

In Table 2 (NbS physical variables), NbS area, distance from NbS, and NbS age were first to be considered. Because NDVI (Normalized Difference Vegetation Index) could indicate the health of vegetation in NbSs and evapotranspiration could be the primary cooling mechanism of NbSs [32], the NDVIs of NbSs were also considered. These factors influenced NbSs’ cooling effects [33]. NbS types were also found to influence cooling efficiency [34]. It is worth noting that the elimination processes delineated in Step 2 and Step 3, combined with the limitation found in Step 1, significantly reduced the NbS types available in subsequent analyses. Only two types of NbS existed: sidewalk permeable surfaces and parks.

Table 3 considered the street environment. The street morphologies are well known to affect heat island intensities [35,36]. Therefore, the height of surrounding buildings and street aspect ratio (building height/street width) were among the potential variables. The current study also considered the angle formed by the solar azimuth and the street direction as it could control solar irradiation reaching the street level. The NDVI of the experimental area was also among the potential variables to be investigated regarding the competition between NbS cooling and street-level cooling. Besides the factors above, wind direction was also considered important in the literature [37,38,39]. Two wind-related variables were considered: the angle between street orientation and wind direction and the angle between direction to NbS and wind direction (i.e., for a certain cell in the experimental area, the angle formed by the direction to its NbS and the direction of the wind). Table 2 does not contain land use variables because of the mixed-use zoning practice in Taipei City.

Meteorological variables are considered in Table 4. Standard meteorological variables such as relative humidity, wind direction, solar irradiation, and precipitation were selected [40]. The ambient (large-scale mean) air temperature was found to be important in controlling the urban microclimate [41]. Precipitation with different antecedent periods (2-day, 3-day, and 5-day precipitation) was included because Tu and Chen [41] found that antecedent accumulated rainfall can influence permeable surfaces’ cooling capacity.

(2)

Step 7 (refine variables to avoid overfitting): The current study adopted the hybrid forward-selection procedure proposed by Tu et al. [42], which considers both p-values and VIF (variance inflation factor) of variables added into a multiple regression model to prevent the model from overfitting. No transformations were applied to the variables, and the thresholds (VIF: 10, p-value: 0.25) used by Tu et al. [42] were adopted.

(3)

Step 8 (multiple regression analysis): This step derived the contribution of individual variables to the cooling capacity with multiple linear regression equations. The coefficient signs in the equations represent the trends (positive or negative correlations) of the variables’ influences on the cooling capacities of green infrastructure. The influence of each variable was computed by the method reported by Judd et al. [43] and Tu et al. [42], based on the proportional change of goodness-of-fit if the variable of interest was removed from the equation.

2.3.2. Scenarios Analyzed

The following sets of data were analyzed to investigate how NbS cooling is influenced under different scenarios:

(1)

Comprehensive: All available data.

(2)

Hot season: Because the summer SUHI was the primary concern in subtropical Taiwan, hot season (April to September) data were further scrutinized.

(3)

Different street patterns: Many cities incorporated both old and new districts. Old districts usually grew organically, with newer districts exhibiting gridded street patterns. The current study hypothesized that the cooling of NbSs was affected by nearby street patterns. The street patterns were determined by the coefficient of street complexness, whose calculation can be explained by Figure 4. Centering on each experimental area, an “area of interest” spanning a 300 m radius was drawn. Two main boulevards which are perpendicular (or nearly perpendicular) to each other can be selected in each area of interest. For Street segment (small streets within street blocks) k, two angles can be formed with the boulevards, and the smaller one was selected as $∆{\theta }_k$ to represent that street segment. For example, ${\theta }_1$ represents Street segment 1 and ${\theta }_3$ represents Street segment 2 in Figure 4. If the street segments and/or boulevards are not straight within the area of interest, $∆{\theta }_k$ is determined by their averaged directions within the area of interest. Within the area of interest, the length the Street segment k is $l_k$. The coefficient of street complexness is defined as $c$ in Equation (2):

$$\mathrm{c}={\displaystyle \frac{\sum _{k=1}^m(∆{\theta }_k·l_k)}{m}}$$

(2)

Because the length of street segments is restricted by the radii of areas of interest, the angle $∆{\theta }_k$ is the most decisive term in Equation (2). If most street segments are parallel to either one of the boulevards, c would be small; on the other hand, if street segments have random directions, the value of c would be high. Subsets of experimental areas with high c values and low c values were analyzed separately in this scenario.

3. Results

Table 5. Seasonal summary of mean meteorological and surface temperature mitigation data.

Season	Spring	Summer	Fall	Winter
$T$ ($\mathrm{℃}$)	24.4	32.2	28.1	19.5
$RH$ (%)	62	62	63	64
$U$ (m/s)	5.5	2.5	1.9	2.1
$SR$ (MJ/m2)	1.66	1.94	1.61	1.18
$P_{2d}$ (mm)	6.6	2.2	8.0	1.5
$P_{3d}$ (mm)	13.6	5.6	12.9	6.8
$P_{5d}$ (mm)	25.3	12.1	40.6	7.9
Surface temp. reduced by NbS ($\mathrm{℃}$)	0.98	1.56	0.64	0.92

summarizes the meteorological data and LST reduction by NbS in different seasons from 2010 to 2020. Summer is the hottest season due to the high solar irradiation intensity, so LST reduction by NbS is the strongest, which will be explained in the analysis section.

Table 6. The comprehensive model (r = 0.41) for LST reduction by NbS using all available data, with essential variables (p-value < 0.05 and influence ≥ 0.1) in boldface.

Variable	Coefficient	p-Value	VIF	Influence
Intercept	3.88	<10−4	-	-
$NbS$ *	−0.75	<10−4	2.57	0.056
$A_{NbS}$	5 × 10−5	0.29	2.49	0.0029
${\mathit{D}\mathit{i}\mathit{s}\mathit{t}}_{\mathit{N}\mathit{b}\mathit{S}}$	−0.013	<10−4	1.99	0.10
${Age}_{NbS}$	−0.043	<10−4	3.43	0.059
${NDVI}_{exp}$	1.20	0.15	1.05	0.0053
$H$	0.026	0.17	5.03	0.0048
$H/W$	−0.45	0.32	5.05	0.0026
${\theta }_{wind-street}$	0.0090	<10−4	1.07	0.052
${\theta }_{wind-NbS}$	−6 × 10−4	0.65	1.13	6 × 10−4
${\theta }_{sun-street}$	−0.016	<10−4	1.23	0.071

* Coefficient is −0.75 for parks and 0.75 for sidewalk permeable surface.

provides the “comprehensive” (all data, all seasons) multiple regression equation for LST reduction by NbS based on all available data. The variables originally selected in Table 2, Table 3 and Table 4 went through Step 6 and Step 7 of the methodology before multiple regression was performed. The parameter “influence” is the influence of individual variables [42,43] calculated by dividing the reduction of goodness-of-fit (when the variable of interest is removed) by the original goodness-of-fit. Essential variables with statistical significance and higher influence (defined as influence larger than or equal to 0.1 in this study) are in boldface in Table 6.

As mentioned, the diurnal SUHI in summer is the primary concern in subtropical Taipei City.

Table 7. The hot season model (r = 0.5) for LST reduction by NbS with essential variables (p-value < 0.05 and influence ≥ 0.1) in boldface.

Variable	Coefficient	p-Value	VIF	Influence
Intercept	0.27	0.79	-	-
$NbS$ *	−0.99	<10−4	1.60	0.061
${\mathit{D}\mathit{i}\mathit{s}\mathit{t}}_{\mathit{N}\mathit{b}\mathit{S}}$	−0.017	<10−4	1.86	0.11
${\mathit{A}\mathit{g}\mathit{e}}_{\mathit{N}\mathit{b}\mathit{S}}$	−0.071	<10−4	2.55	0.10
${NDVI}_{exp}$	2.55	0.081	1.04	0.011
$H$	0.12	10−4	4.51	0.053
$H/W$	−2.85	<10−4	4.54	0.060
${\theta }_{wind-street}$	0.016	<10−4	1.10	0.073
${\theta }_{wind-NbS}$	−0.0034	0.078	1.12	0.011
SR	2.30	<10−4	1.09	0.12
$P_{3d}$	0.024	0.0017	1.05	0.036

* Coefficient is −0.99 for parks and 0.99 for sidewalk permeable surface.

presents the result based on the hot season (April–September) data. Variables with statistical significance and higher influence are in boldface in Table 7. The variables originally selected in Table 2, Table 3 and Table 4 went through Step 6 and Step 7 of the methodology before multiple regression was performed.

The multiple regression models for gridded and complex street patterns are provided in

Table 8. The multiple regression models for LST reduction by NbS with different street patterns, with essential variables (p-value < 0.05 and influence ≥ 0.1) in boldface.

Model (r = 0.4) for Grid-Type Streets (c ≤ ~500)
Variable	Coefficient	p-Value	VIF	Influence
Intercept	3.27	<10−4	-	-
$A_{NbS}$	8 × 10−5	0.55	3.37	0.0026
${Age}_{NbS}$	0.044	0.0036	1.12	0.085
${NDVI}_{exp}$	1.32	0.29	1.07	0.0084
$H$	0.042	0.044	2.46	0.032
$\mathit{H}/\mathit{W}$	−4.85	<10−4	5.11	0.16
${\theta }_{wind-street}$	0.0081	0.0031	1.12	0.069
${\theta }_{wind-NbS}$	−0.0050	0.0050	1.15	0.062
$P_{5d}$	0.0053	0.0010	1.01	0.038
Model (r = 0.62) for Complex Streets (c > ~500)
Variable	Coefficient	p-Value	VIF	Influence
Intercept	−4.65	<10−4	-	-
${\mathit{A}}_{\mathit{N}\mathit{b}\mathit{S}}$	7.1 × 10−4	<10−4	5.17	0.14
${\mathit{D}\mathit{i}\mathit{s}\mathit{t}}_{\mathit{N}\mathit{b}\mathit{S}}$	−0.035	<10−4	3.00	0.27
${\mathit{A}\mathit{g}\mathit{e}}_{\mathit{N}\mathit{b}\mathit{S}}$	−0.12	<10−4	1.33	0.55
${\mathit{N}\mathit{D}\mathit{V}\mathit{I}}_{\mathit{N}\mathit{b}\mathit{S}}$	6.27	<10−4	1.45	0.12
$\mathit{H}$	0.27	<10−4	8.50	0.13

. The variables originally selected in Table 2, Table 3 and Table 4 went through Step 6 and Step 7 of the methodology before multiple regression was performed. Among the eleven experimental areas, five areas have lower coefficients (c ranges from 176 to 482) and six areas have higher (c ranges from 640 to 1173) coefficients of street complexity. Variables with statistical significance and higher influence are in boldface in Table 8.

Even though the statistical models in Table 6, Table 7 and Table 8 are in the form of multiple linear equations, one should bear in mind that these equations should not be used to predict actual temperature reductions by NbS. Instead, the models were used to evaluate the relationships (i.e., positive or negative correlation) between variables and cooling, and the relative importance of each variable for the overall cooling benefit.

4. Discussion

4.1. Summary of Seasonal Data

Table 5 shows a higher magnitude of cooling in summer, which conforms to the fieldwork done by Tu and Chen [41], as shading and evapotranspiration serve as the primary cooling mechanisms provided by NbSs [44]. Vegetation shading blocks sunlight in nearly a fixed portion [45]. As solar irradiation increases in summer, the canopy blocks more solar irradiation energy, meaning a higher magnitude of cooling. On the other hand, evapotranspiration often increases with higher solar irradiation [46]; thus, the magnitude of cooling provided by evapotranspiration also increases in summer.

However, the exhibited LST in summer is still higher, as the net solar irradiance received by the ground is higher even after NbS cooling. Solar irradiation dominates LST particularly in the hot season [47].

4.2. Comparing the Comprehensive and Hot Season Models

For the analyses below, variables bearing statistical significance and strong influence (defined as influence ≥ 0.1) are the focus of discussion. The comprehensive model (utilizing all data) has only one variable satisfying these criteria: the distance between the experimental area and NbS (${Dist}_{NbS}$), which is also one of the influential variables in the hot-season model. It matches the literature findings [48]. Higher ${Dist}_{GI}$ reduces the cooling capacity of NbSs in general, while the cooling capacity diminishes faster with increasing ${Dist}_{NbS}$ in the hot season.

Besides ${Dist}_{GI}$, the strength of solar irradiation ($SR$) is essential in the hot season model. Higher $SR$ renders higher NbS cooling, as explained by Section 4.1.

On the other hand, the influence of ${Age}_{GI}$ shows an intriguing trend, with older NbSs generating lower LST cooling in the hot season model (Table 7). One would expect the opposite as older NbSs should have established vegetation [49]. To explain this result, the authors provided the following three hypotheses:

(1)

The timeframe is too short: The timeframe of the current study is less than ten years, so the growth of trees might not be noticeable. Ziter et al. [50] found that tree canopy provides nonlinear cooling effects, so that a significant cooling effect shows only when the ratio of canopy cover is high. However, if this is the only reason, the cooling effects should be kept steady, not degrade.

(2)

Intensive utilization: Due to the ultra-high population density, parks serve as residents’ only outdoor recreation space. The authors’ observation is that the understory vegetation of many parks in Taipei City had worn down quickly over time. Figure 5 below shows the understory condition of NbS #5.

(3)

Lack of maintenance of permeable surfaces: Moreover, Taipei City had not developed a schedule for permeable surface maintenance (personal communication with the Public Works Department of Taipei City Government), which means that the permeable surface might become clogged over time, reducing its water infiltration, and thus its LST-reduction capabilities [52].

It should be stressed again that the three explanations above are just hypotheses that need to be validated in future research. Whatever the actual causes are, NbSs in Taipei City exhibits decreasing cooling capacity in hot seasons, which is a trend that requires attention.

4.3. Comparing Models with Grid-Type and Complex Street Patterns

When the street pattern is gridded, the only variable satisfying the criteria (statistical significance and high influence) is the aspect ratio ($H/W$), with higher aspect ratios reducing the cooling efficiency of NbSs. It is worth noting that no NbS-related variables were deemed important under this scenario. This implies that the geometry of the urban canyon controls cooling.

Districts with complex street patterns are quite different. NbS cooling capacity is sensitive to many variables. Larger (higher $A_{NbS}$) and newer (lower ${Age}_{NbS}$) NbSs with better vegetation conditions (higher ${NDVI}_{NbS}$) and closer to the observer (lower ${Dist}_{NbS}$) have higher cooling capacities. It is intriguing that the age of NbSs (${Age}_{NbS}$) is the most influential variable in this scenario and also exhibits a trend of decreasing cooling efficiency over time.

For streets with complex patterns, the influence of building height ($H$) also stands out: higher building height increases NbS cooling efficiency. Higher building height ($H$) usually means higher aspect ratios ($H/Ws$) in general. This is opposite to the result from grid-type street patterns, where higher aspect ratios ($H/Ws$) reduce NbS efficiency.

The authors hypothesize that such a discrepancy is caused by wind patterns in different urban canyon shapes. In grid-type streets, ground-level wind can be strengthened by flow jetting between buildings [53]. If the wind direction is parallel or close to parallel to the direction of straight urban canyons, ground-level wind speed is accelerated. Higher buildings (with higher aspect ratios) increase such an acceleration [54]. Therefore, ground-level heat is often taken away by such strong wind, reducing the “baseline” temperature at grid-type streets. On the other hand, wind can only access the ground from the top in districts with complex street patterns. Higher buildings hinder such access, thus increasing ground-level “baseline” temperature for streets with complex patterns. Utilizing the explanation in Section 4.1, higher baseline temperature provides higher NbS cooling in districts with complex patterns, and lower baseline temperature provides lower NbS cooling in grid-type streets. The reasoning can be seen as a flow chart in Figure 6. One should realize that the magnitude of cooling is not equivalent to the exhibited temperature. Again, the explanation provided here is only a hypothesis that requires validation by future research.

4.4. The Effect of Distance Across Scenarios

The distance from NbS (${Dist}_{NbS}$) significantly influenced several scenarios (except for the grid-like street pattern model), as predicted by the literature [48]. To evaluate only the influence of ${Dist}_{NbS}$ in the scenarios, the variable $NbS$ was set to zero and other variables were set to mean values, as

Table 9. Urban space characteristics variables assumed in the derivation of the relationship between distance and UHI reduction.

Variables	Value Assumed
NbS type ($NbS$)	0
NbS area ($A_{NbS}$)	2783
NbS age (${Age}_{NbS}$)	25.39
NbS NDVI (${NDVI}_{NbS}$)	0.41
NDVI of the experimental area (${NDVI}_{exp}$)	0.08
Height of surrounding buildings ($H$)	27.62
Street aspect ratio ($H/W$)	1.04
The angle between street and wind directions (${\theta }_{wind-street}$)	42.6
The angle between the direction of NbS and the wind direction (${\theta }_{wind-NbS}$)	93.46
The angle between the solar azimuth and street directions (${\theta }_{sun-street}$)	58.58
Solar irradiation ($SR$)	1.55
3-day accumulated precipitation ($P_{3d}$)	11.71

shows. Figure 7 provides the one-to-one relationship between ${Dist}_{GI}$ and cooling for the comprehensive model (all data), the hot season model, and the complex street pattern model.

Figure 7 shows similar decreasing trends for all three models, with the UHI reduction tapering off at approximately 100–200 m from NbS. Such a finding matches the observation by Fernández [9] that the horizontal temperature influence of green space can reach as far as 200 m.

Among the three lines in Figure 7, NbS efficiency drops quickly with distance in districts with complex patterns. The cooling effect is almost non-existent at merely 120 m, which is only approximately 60% of the maximum distance in the other two scenarios. Complex street patterns might interfere with how air circulates, thus hindering the distance that cooled air can reach.

5. Conclusions

The general benefits of NbSs for the urban microclimate are well known, but variables affecting NbS cooling have been less investigated. Using Landsat 7-derived LST to evaluate SUHIs, the current study analyzed the relation between NbS-induced temperature reduction and the surrounding urban space characteristics variables. The main findings are summarized below:

• The distance to NbS (${Dist}_{NbS}$) is important to the cooling effect in most situations. On the other hand, the area of NbS ($A_{NbS}$) is not as crucial. Balany et al. [7] indicated that the effect of an increased NbS area is often insignificant and non-linear. This finding implies that having small NbS devices distributed evenly can generate better overall cooling effects compared to having larger NbS devices that are sparsely placed, if the same total NbS area is the same. This is even more crucial to older districts where the NbS cooling effect cannot reach far because of the complex street patterns.

The smallest NbS considered by the current study is 419 m². Therefore, caution should be exercised when applying the conclusions to NbSs smaller than that criterion.

2.

When the street pattern is highly regulated (grid-type), NbSs might not be the primary cooling mechanism. It was hypothesized in this study that wind concentrated by the urban canyon provided the most cooling effects. On the other hand, NbSs are important for cooling districts with complex street patterns. When resources are limited, NbS planning should be focused on older districts with complex patterns, not newer districts with grid-like patterns.

3.

Probably due to localized NbS cooling in districts with complex patterns, NbS cooling is sensitive to several variables. NbSs that are larger (higher $A_{NbS}$), newer (lower ${Age}_{NbS}$), with better vegetation conditions (higher ${NDVI}_{NbS}$), and closer to the observer (lower ${Dist}_{NbS}$) have higher cooling capacity. NbS installation in those districts needs careful planning to maximize the efficiency of the engineering investment.

4.

The results imply that NbS maintenance is crucial for sustaining the cooling function of NbSs. The importance of maintenance has also been highlighted by past studies [55].

The current study contains several limitations. The data were limited to a span of 10 years, which might have made observing long-term effects of tree canopy growth difficult. Taipei City is subtropical so the results may not apply to cities with other climates. Only two types of NbS were included in the analyses, limiting the applicability of the results. Lastly, Taipei City is in a basin surrounded by mountains, and such a special geographic setting might have skewed the results into downplaying wind-related variables. Therefore, wind-related variables might play an important role in SUHI mitigation [37,38], but such an effect does not show in Taipei City [41].

The current study can also serve as a springboard for future research. For example, how significant is a lack of maintenance to the degradation of the SUHI mitigation capability (or vice versa) of NbSs? How do we quantify the influence of maintenance? What is the mechanism to trigger enhanced ground-level ventilation? How is it related to aspect ratios and street patterns? Is there a way to predict it based on basic urban morphological attributes such as aspect ratio or street width?