4.1. Changing Climate in Boise
Like many other cities across the globe, Boise has experienced a dramatic temperature increase over the last few decades. According to a national study of temperature change from 1970 to 2018, Boise was ranked as the 13th fastest-warming city in the US [41
]. The heating trend has significant impacts on the social, ecological, and environmental conditions in the region. Based on the climate risk assessment conducted by the Langdon Group and the University of Idaho, the frequency of heat stress days with moderate risk will increase from 16 days (historical baseline) to 66 days each summer by 2050 [42
]. This is going to have an adverse impact on the population susceptible to heat-related diseases. A moderate drought is projected to occur every two years by the mid-21st century, compared with one in four years now. In addition, the frequency of exceptional drought will increase from one in every 12 years now to one in every three or four years. Further, the odds of very large wildfires in the region will increase by 400% by 2050, suggesting great potential and concerns for chronic air-quality issues within the Treasure Valley.
Our results indicate a major rise in the surface temperatures from 2001 to 2013. The mean LST in the region had increased by 2.9 °C from 2001 to 2008, and by 7.3 °C from 2008 to 2013, resulting in a total of a 10.2 °C increase over the 13 years. There is a greater rate of temperature increase for more recent years than a decade ago, and the trend is consistent for all land use types in the region. The warming trend is evident across the whole study area, but more prominent warming is seen in developed medium and high intensity areas, hay/pasture, and desert lands in East Boise.
4.2. The Interacting Effect of Land Cover Abundance and Spatial Pattern on the LST
While the type and composition of land cover features are key determinants of the LST, the landscape pattern offers additional insight on useful strategies for heat island mitigation. To further explore how the landscape pattern changes with different levels of land cover abundance, we created scatterplots of the local Moran’s I with G of NDVI and G of NDBI respectively (Figure 6
). Interestingly, although the local Moran’s I was not a very strong predictor for the variations in LST, there was a concave quadratic relationship of the local Moran’s I with both the Getis variables for all three years. A possible interpretation is that at a lower level of spatial concentration, an increase in the quantity of the land cover (increased Getis) results in a higher degree of landscape fragmentation (decreased local Moran’s I). When the quantity of land cover exceeds a critical point, a continued increase in the land cover abundance will cause the land cover features to cluster (increased local Moran’s I). For both vegetation and built-up areas, the critical point is met when the Getis score approximates zero. Since the Getis was standardized with respect to the mean value, a critical point is reached when the degree of land cover concentration approximates the region average.
Given the findings above, we speculated that the land cover abundance and spatial pattern should be linked to the LST in an interactive manner. Our hypothesis is that the local Moran’s I is statistically associated with the LST with a moderating effect from the Getis variable. In other words, the relationship of the LST with the local Moran’s I varies for different Getis scores. To test this hypothesis, we performed a hierarchical multiple regression and conducted an F test to evaluate the existence of an interaction effect of the local Moran’s I and G of NDVI on the LST. To alleviate multicollinearity, all the predictors were centered by subtracting the mean from the original value. We performed the analysis for three years and the results were reported for 2013 only.
Test results show that adding the interaction term results in a 0.009 change of the R square and a 20.449 change of the F value. The change was statistically significant at the 0.05 level, indicating that the interaction effect from the local Moran’s I and G of NDVI accounted for a significant amount of variance in the LST above and beyond the first-order effect.
To better understand the interacting effect, we created an interaction plot of the predicted LST against the local Moran’s I, using the G of NDVI as the moderating variable (Figure 7
). The samples were divided into two subgroups based on the G of NDVI scores: group 1 (blue) has a G of NDVI above 0 and group 2 (orange) has a G of NDVI below 0. The G of NDVI serves as a proxy of a specific land composition scenario. For example, when the G of NDVI is less than 0, the level of vegetation concentration is below the region average, and vice versa. A regression line was fit for each subgroup.
According to Figure 7
, the samples were divided into two distinct groups based on the G of the NDVI score. There was a negative relationship of the predicted LST with the local Moran’s I for the subgroup with the G of NDVI above zero. The relationship turned positive for the subgroup with the G of NDVI below zero. Both relationships were statistically significant and the subgroup with the G of NDVI above zero achieved a better model fit (R2
> 0.8) than the other subgroup (R2
The plot revealed interesting findings regarding optimal landscape designs to cool the city. When the green vegetation abundance is above the region average (G of NDVI > 0), a clustered spatial arrangement is more desirable than a dispersed pattern. This applies to areas such as low intensity developed lands, developed open space, cultivated crops, and hay/pasture. In these areas, planning clumped greenspace is more effective for lowering the surface temperatures. This finding is in line with several previous studies focusing on the spatial arrangement of green vegetation for cooling [8
]. A few of these studies, however, have considered the moderating effect of land cover abundance on the relationship between landscape patterns and LST.
When the abundance of green vegetation is below the region average (G of NDVI < 0), the local Moran’s I was positively related to the LST, suggesting that a dispersed landscape pattern is desirable for cooling. Specific land use types in this category include medium to high intensity developed areas, herbaceous, and shrub/scrub. Due to the low coverage of green vegetation, these areas tend to be warmer than the rest of the region. A dispersed spatial arrangement of buildings, for example, could help cool down some of the hottest areas such as those surrounding the Boise airport and downtown Boise. A similar result was reported in a case study in Phoenix, AZ that when controlling for the fraction, clustered paved surfaces result in an aggregate warming effect [14
]. This finding was corroborated in a more recent study for Bangkok, Jakarta, and Manila, in which a positive correlation was found between the LST and an aggregation index of impervious surfaces, suggesting a warming (cooling) effect from clustered (dispersed) impervious surfaces [44
4.3. Urbanization Impacts on the Spatiotemporal Pattern of LST
The Boise-Meridian metropolitan area originally developed along the railroads running across the cities. Railroads attracted commerce and industry to the region in the late 19th and early 20th centuries, which fundamentally affected the growth patterns of the urban areas. Figure 8
shows a satellite image of the region in 2013 together with the railroads and a three-kilometer (two-block distance) buffer along the railroads. Of all the developed areas in the study area, the railroad buffer contains 80% of the high intensity developed areas and 70% of the medium intensity developed areas. From 2001 to 2013, 71.1% of the area increase in the high intensity developed areas occurred in the buffer. This indicates that the urban development occurred mostly along the railroads within a two-block buffer distance.
To analyze the urbanization pattern by LULC type, we created the bar charts showing the area and rate of change for major LULC types in the region (Figure 9
). There was a continued growth in the urban areas and a continued decline in the agricultural lands. About 13% of cultivated crops and 5% of hay/pasture were converted into urban. Although the high intensity developed areas took the least land proportion in 2001, its area increased by 66.2% from 2001 to 2013, followed by medium intensity (30.8%) and low intensity developed areas (7.1%).
To better understand the impact of the LULC change on the LST, we created a bar chart showing the mean LST by the LULC type (Figure 10
). Consistent across all the LULC types, there was an ever-increasing trend in the LST from 2001 to 2013, with a greater LST increase from 2008 to 2013. The LULC type with the highest LST was the high intensity developed areas, followed by medium and low intensity developed areas, and developed open space. By 2013, the mean LST in all developed areas had reached 44.8 °C and the mean LST in the high intensity developed areas were as high as 48.4 °C.