Urban green spaces provide neighborhoods with a wide range of health, social, and economic benefits; they improve quality of life, promote physical and mental health, decrease crime rates, and increase property values [1
]. However, rapid urbanization has accelerated land use and land cover changes, and modified the structure and function of urban ecosystems [8
]. The major concern is that urban land conversions with large amounts of impervious structures increase the heat storage capacity of cities and change their microclimates, thus creating the urban heat island effect (UHI). UHI is defined as an isolated urban area where the temperature is relatively higher than the encompassing suburban and rural areas [10
]. UHI is of critical importance for substantial climate and health studies to understand the magnitude of its effects [11
]. Previous research indicates that the increased heat stress in urban areas has exposed residents to many health problems, including a high concentration level of volatile organic compounds (VOCs) and nitrogen oxides (NOx
), elevating the incidence of cardiovascular diseases, and heat-related mortality from episodes such as heat stroke [15
]. In addition, increasing heat in urban areas discourages residents from participating in physical activity due to uncomfortable thermal conditions experienced when walking, bicycling, or undertaking other outdoor activities [23
]. Many studies have reported on the importance of built environment conditions that promote physical activity which can prevent health risk factors for chronic diseases, such as cardiovascular disease, diabetes, and obesity [24
UHI can be quantitatively assessed by two factors: atmosphere temperature and surface temperature [27
]. Atmosphere temperature, as a more direct indicator of UHI, has been used to measure temporal changes through the data obtained from single weather stations or mobile measurements. Conversely, to quantitatively assess surface-based UHI, land surface temperature (LST) has been retrieved from remote-sensed thermal infrared data [28
]. LST distributions are easily mapped for specific spatial units with high coverage, while interpolation is required to map the distribution of air temperature [31
]. LST captures the energy radiated from diverse structures in built and natural environments in urban areas, including rooftops of buildings, impervious surfaces, and vegetation and water bodies [10
]. For this reason, it has been used widely to assess the relationships between land cover patterns and the spatial distributions of UHI [8
A substantial body of literature has investigated the relationship between green space and UHI, suggesting that urban green spaces have a cooling effect and thus mitigate UHI [8
]. Studies have reported that the closer and bigger green spaces are, the greater temperature differences exist between green areas and the surrounding built environment [36
]. In addition, some literature indicates that green spaces in an urban area with high building height to street width ratios intensified the cooling effect at night as heat trapped inside building canopies was released during night hours [37
]. Although early studies allowed for the empirical understanding of cooling effects by increased greenness [36
], they relied on limited observations and simply monitored temperature changes by varying locations inside and outside targeted green areas. Thus, some researchers advanced the methodological approaches to quantify the abundance of greenness and expanded their scope of statistical analyses to seek the spatial relationship between green space and LST. The normalized difference vegetation index (NDVI), a method of measuring the composition of urban green spaces, has been integrated with LST data at a pixel level [47
]. Since built environments in urban areas limit greenness by including human-built structures, such as roads and buildings, understanding the effect of neighborhood landscape spatial patterns helps to strategically develop mitigation plans for UHI effects at the neighborhood level.
To measure the influences of landscape spatial patterns on distribution of LST, landscape indices have been applied to quantify their density, shapes, and connectivity [9
]. However, more empirical studies are required to explore the role of the spatial patterns of urban landscape in UHI mitigation using neighborhood-level data. The relevant body of literature has focused primarily on simple correlation analysis between landscape indices and LST [8
], and the results have been inconsistent, depending upon spatial and temporal conditions [9
]. To illustrate, Li and his colleagues [8
] found that the percent cover of green space, mean patch area, and shape index had a negative correlation with LST, while patch density had a positive relationship in bivariate analysis. However, Zhou et al. [9
] found that patch density of vegetation showed no significant relationship, while edge density had a negative relationship with LST in multi-linear regression models. Another study suggested that the area-weighted parameter area ratio had a stronger relationship with temperature changes than patch density and shape index [34
]. Despite these results, previous researchers emphasize the significance of landscape indices on measuring landscape spatial patterns as well as their impact on increases in urban temperature. Moreover, most of the previous studies examining the relationship between landscape spatial patterns and LST were conducted in major cities in Japan and China [34
]; a few empirical studies have investigated the influences of landscape spatial patterns on LST by using cases in the United States [9
]. Though urban dwellers are directly affected by UHI in their living environments, previous studies have typically been conducted on a macroscopic scale (e.g., census track), not a neighborhood scale [8
]. Utilizing a macroscopic scale has several limits with regard to controlling size and capturing characteristics of particular neighborhoods, and thus comparisons are limited by unit differences when assessing the spatial configurations of neighborhood landscapes.
In sum, a lack of understanding regarding the cooling effects of landscape spatial patterns on neighborhoods limits urban design and planning geared toward strategically mitigating the increasing temperatures faced by neighborhood dwellers. The main objectives of this study, then, were to: (1) assess the relationship between LST and landscape spatial patterns by using multi-regression models after controlling for spatial autocorrelation; and (2) identify the most significant predictor among landscape indices to cooling effects of green spaces at a neighborhood scale.
represents the descriptive statistics for LST, the characteristics of the selected single-family houses, and landscape spatial patterns in each neighborhood. The mean LST, a dependent variable, was approximately 32.60 °C. This indicates the low level of fluctuation in average temperature in August in Austin, Texas. The average housing sale price was about $302,000 ranging from $10,900 to $7,750,000. The living area ranged from 26.66 m2
to 1270 m2
. The homes were approximately 33 years old and had approximately three bedrooms and 2.34 bathrooms, on average.
For the selected landscape indices, approximately 38% of the neighborhood areas within the 800 m buffer were covered by trees (PLAND), and more than 4000 tree and urban forest patches existed (NP), on average. The mean patch size (MPS) was approximately 240 m2, and the mean nearest distance (MNN) between two neighboring patches was 2.60 m. Most tree and urban forest patches were highly connected, according to the patch cohesion index (COHESION). Based on bivariate tests, all of the selected landscape indices were significantly correlated with LST. Some landscape indices are highly correlated with each other (the highest value of the correlation coefficient was 0.79 between PLAND and MPS); however, based on the VIF analysis, none of landscape indices showed the risk of multicollinearity. The range of VIF between the selected landscape indices was from 1.51 to 5.45, which was lower than the commonly accepted threshold of 10.
shows the final results of this research, displaying both Model 1 (OLS model) and Model 2 (spatial lag regression model). The overall results indicated a strong relationship between LST and the landscape indices. Model 1 explained about 54% of the variance in the relationship between LST and neighborhood landscape spatial patterns (R2
= 0.5350), while Model 2 explained 82% (R2
= 0.8201). Based on our final models, the OLS model overstated the relationship between landscape indices and LST. In controlling for spatial autocorrelation issues, while each variable in both the OLS and spatial lag models showed the same signs for dependent variable coefficients, absolute values of the coefficients in the spatial lag model were smaller than absolute values of the coefficients in the OLS model.
The levels of significance for all landscape indices in both of the final models were significant at the 0.01 level. From Model 1, according to the standardized coefficients (Beta) value to compare the explanatory power between independent variables, MPS and MNN were the most significant predictors among landscape indices to cooling effects of trees. The standardized coefficient values of MPS and MNN were −0.396 and 0.353, respectively.
Based on results from Model 2, the percent of tree cover (PLAND) showed a significant negative relationship to LST, which indicates that larger amounts of urban trees and forests in neighborhoods could be positively related to decreased surface temperatures. From measuring fragmentation, it was concluded that the number of patches (NP) was positively associated with LST, while the mean patch size (MPS) showed a significantly negative relationship to LST. These results indicate that more fragmented neighborhood landscape patterns could contribute to an increase in LST. The mean shape index (MSI) was positively related to LST at the 0.01 level, which means that more irregularly shaped neighborhood landscape patterns contributed less to the decrease in LST. The final spatial lag model indicated that there was a statistically negative correlation between the mean nearest neighbor distance (MNN) and LST. This finding suggests that less isolated landscape spatial patterns within an 800 m neighborhood buffer will be likely to reduce LST. Finally, the patch cohesion index (COHESION) measuring the physical connectedness of the tree patches showed a negative relationship; this indicates that well-connected landscape spatial patterns could be more beneficial to decreasing LST.
This study focused on the influences of neighborhood landscape spatial patterns on LST, using landscape indices to quantitatively measure the configurations of landscape structures of neighborhood trees and forests. This study, with a large sample size (approximately 16,000 neighborhood buffers), considered the effects of spatial autocorrelation in order to develop Model 2 (spatial lag regression model) due to the spatial overlap of each unit of analysis. The results from our spatial regression model suggest that larger green spaces (PLAND) and well-connected (COHESION) landscape spatial patterns are positively correlated with lower LST values in neighborhoods, while more fragmented (NP and MPS), irregularly shaped (MSI), and isolated (MNN) conditions are negatively related to the reduction of LST. These results are similar to the findings of previous studies. Li and his colleagues [8
] found that the percentage of landscape cover and mean patch size were negatively correlated with LST, while patch density showed a positive relationship in their bivariate analysis. Understanding the role of landscape spatial patterns on mitigating LST is important, as the results can be directly translated into urban planning and design guidelines. While previous studies found the proximity, location, and size of green spaces are significantly important factors to reduce temperatures [36
], those findings are limited to interpolating how the configuration of land cover features can contribute to mitigating LST. Instead of measuring only size and proximity of green spaces, using landscape indices to measure landscape spatial patterns in urban areas allows policy-makers, urban planners, and designers to understand what kind of spatial configurations of neighborhood landscapes should be considered for improving residents’ quality of life and health by reducing LST.
Reducing fragmentation, improving connectivity, and bonding isolated landscape patterns are the most important objectives of ecological planning. Larger patch sizes, well-connected conditions, and less fragmented and isolated landscape patterns are the main indicators of better ecological quality in landscapes [79
]. Following previous theories and guidelines from landscape ecology research, the findings of this study show that ecologically healthy neighborhood landscapes positively contribute to mitigating LST in urban neighborhoods. Our results support that ecological planning should be considered to contribute to environment-public health planning and policy. The current planning framework may miss this connection, which is potentially important to reducing UHI by creating ecologically healthy neighborhood green spaces. The full benefits of neighborhood landscapes shaped by larger and well-connected urban green spaces appear to be significant. Yet, increasing size and enhancing connectivity of green space can often be limited in urban areas due to the existing land use. For compensating the deficiency of urban greenery to mitigate LST, small green spaces as ecological stepping stones can play an important role to increase the size of green space and build green networks in urban areas. Previous studies have also supported the potential ecological contributions of small green spaces to urban planning by assessing school green areas [95
], roof gardens [96
], and roadside green space [97
Our findings suggest that planning policy and guideline development should consider landscape spatial patterns to improve public health in neighborhoods. Uncomfortable walking and cycling conditions with higher LST tend to keep residents from being involved in outdoor physical activity, which can increase the risk of obesity, cardiovascular diseases, and heat-related mortality [16
]. The results of this study imply that if urban forest comprehensive/management plans for neighborhoods would guide larger and well-connected tree areas with less fragmented and isolated patterns, it will likely contribute to reducing LST significantly. This can help to enhance physical and mental health conditions of residents. Improving the quality of life and health of community members is one of the most important goals of urban and landscape planning. As one of the essential elements in neighborhood environments, a better spatial pattern of neighborhood landscapes can enable reduction in LST and should be considered in existing or future planning policies to maximize benefits of allocating and designing neighborhood green spaces.
This study developed two regression models—an OLS model (Model 1) and a spatial lag regression model (Model 2). In Model 2, the effects of spatial autocorrelation were controlled for, which occurred due to the geographical overlap of neighborhood buffers. Both models reported similar estimates for all independent variables. However, Model 2 reduced statistical bias in estimations; the effects of landscape spatial patterns on LST were less exaggerated in Model 2. For example, the influence of the mean nearest neighborhood distance (MNN) and the patch cohesion index (COHESION) on LST was overestimated by about two times in the OLS model. This indicates that if spatial autocorrelation had not been controlled for, it would likely have led to biased outcomes. This result implies that better modeling performances could be achieved when spatial autocorrelation effects are controlled for. Further studies would enhance these findings and suggest solid solutions for healthy neighborhood environments with lower LST.
There are several limitations to this study. First, although we collected a large sample size to analyze the relationships among landscape spatial patterns and LST, the study area was limited to a single city: Austin, Texas. Thus, the findings of this study may not be generalizable to other cities, especially those in geographic areas in the Northern U.S. that have cooler summer temperatures than the study city. In addition, this research used only one land use type, single-family residential areas. Future research should be expanded by estimating the relationship between LST and landscape spatial patterns in diverse land uses including multi-family residential areas. Second, this study employed DOQQ aerial photographs to measure neighborhood landscape spatial patterns. The DOQQ imagery could not detect the full layers of landscape structures with three-dimensional information; it captured only two-dimensional features. Thus, alternative media capable of fully detecting landscape structures under the tree canopy should be considered in future research. Third, this study tested only one size of buffer to measure neighborhood landscape spatial patterns. Future research should perform sensitivity analyses considering different spatial resolution of the data and various scenarios for determining the spatial extension which will improve robustness of estimating results from landscape spatial pattern analysis. In addition, this is a cross-sectional study analyzing the LST data from a selected day. Future research should consider seasonal changes in order to develop more accurate models. Finally, although we selected the most appropriate landscape indices for representing the characteristics of neighborhood landscape spatial patterns, the final model may not fully cover potential control variables such as pavement materials, level of perviousness, and emissivity of neighborhood structures. These factors indicating built environment conditions should be considered to predict more robust estimations when measuring the relationship between urban green spaces and LST.