Estimating local-scale urban heat island intensity using nighttime light satellite imageries

Highlights

• Propose a new approach to local-scale UHI intensity estimates by using nighttime light satellite imageries.

• Spatial regression models was used to estimate UHI intensity due to the presence of residual spatial autocorrelation.

• Estimate results tend to be weaker from district level to neighbourhood level.

• District-level UHI intensity was successfully estimated (approximately R² = 0.7, MAE = 1.16 °C, and RMSE = 1.74 °C).

Abstract

Urban heat island (UHI) effect tends to harm health, increase anthropogenic energy consumption, and water consumption. Some policies targeting UHI mitigation have been implemented for a few years and thus needs to be evaluated for changes or modifications in the future. A low-cost approach to rapidly monitoring UHI intensity variations can assist in evaluating policy implementations. In this study, we proposed a new approach to local-scale UHI intensity estimates by using nighttime light satellite imageries. We explored to what extent UHI intensity could be estimated according to nighttime light intensity at two local scales. We attempted to estimate district-level and neighbourhood-level UHI intensity across London and Paris. As the geography level rises from district to neighbourhood, the capacity of the models explaining the variations of the UHI intensity decreases. Although the possible presence of residual spatial autocorrelation in the conventional regression models applied to geospatial data, most of the studies are likely to neglect this issue when fitting data to models. To remove negative effects of the residual spatial autocorrelation, this study used spatial regression models instead of non-spatial regression models (e.g., OLS models) to estimate UHI intensity. As a result, district-level UHI intensity was successfully estimated according to nighttime light intensity (approximately R² = 0.7, MAE =1.16 °C, and RMSE =1.74 °C).

Introduction

The urban heat island (UHI) effect refers to the difference in temperature between an urban area and the rural surroundings of a conurbation (Azevedo, Chapman, & Muller, 2016). The higher temperatures experienced in urban areas compared to the surrounding countryside has enormous consequences for the health and wellbeing of people living in cities (Mohajerani, B, akaric, & Jeffrey-Bailey, 2017). UHI tends to harm health (McMichael, Woodruff, & Hales, 2006; Patz, Campbell-Lendrum, Holloway, & Foley, 2005), increase anthropogenic energy consumption (Rosenfeld, Akbari, Romm, & Pomerantz, 1998), and water consumption (Guhathakurta & Gober, 2010). Infrared satellite imageries or airborne photos have been used to estimate UHI intensity since they can directly derive the temperature of the surface or the temperature of the atmosphere (Chui, Gittelson, Sebastian, Stamler, & Gaffin, 2018; Coutts et al., 2016; Fabbri & Costanzo, 2020; Wang et al., 2019). Apart from some direct measurements like infrared pictures, indirect measurements like visible light pictures are of high potential as well. UHI is contributed by built environment characteristics and heat emissions of economic activity in urban areas. Built environment characteristics, such as impervious surface area density, building area density, building height, and road density, have impacts on both surface temperature and air temperature since impervious surface materials of buildings and roads can influence surface albedo, emissivity, and evapotranspiration (Oke, 1987; Pigeon, Legain, Durand, & Masson, 2007). Constructed impervious surfaces alter sensible and latent heat fluxes, causing urban heat islands (Changnon, 1992). Constructed surfaces are more likely to increase temperature than natural surfaces (Changnon, 1992); whilst urban green spaces can reduce UHI effects (Debbage & Shepherd, 2015; Oke, 1987). In the urban areas crossed by the major roads, the traffic is the major source (Pigeon et al., 2007). Aside from vehicle heat emissions, asphalt concrete on road surface is a contributor to the UHI (Mohajerani et al., 2017). Heat emissions are generated by economic activities including domestic, commercial & industrial, and transportation activities. Urban areas have more heat emissions than rural areas due to more economic activities. Urban areas have more domestic activities as urban areas are more densely populated. Urban areas also have more commercial & industrial and transportation activities since urban areas are commercial & industrial centres and have larger traffic volumes. A better understanding of how UHI is contributed by built environment characteristics and economic activity at the local levels can help to inform policy and planning decisions for UHI reduction. Particularly, UHI is usually strongest at night (Azevedo et al., 2016); for example, a study revealed that in Paris the magnitude of the night-time UHI was up to 7 °C more than the daytime UHI (Lac et al., 2013). Compared to nighttime UHI, daytime UHI is less noticeable and far more complicated to characterise (Azevedo et al., 2016). Recent studies demonstrate that nighttime light satellite imagery data may be a good proxy for night-time population (Anderson, Tuttle, Powell, & Sutton, 2010; Bagan & Yamagata, 2015; Liu, Sutton, & Elvidge, 2011; Lo, 2001; Pozzi, Small, & Yetman, 2003; Sutton, 1997; Zhuo et al., 2009) or economic activity (Ghosh et al., 2010; Mellander, Lobo, Stolarick, & Matheson, 2015; Zhang & Seto, 2011). Nighttime light intensity (NTLI) is positively related to constructed impervious surface area (ISA) density and energy consumption due to economic activities at night (Amaral, Câmara, Monteiro, Quintanilha, & Elvidge, 2005; Shi et al., 2014; Townsend & Bruce, 2010). As both constructed impervious surface area (ISA) density and nighttime consumption make contributions to UHI, NTLI is likely to be positively related to UHI intensity at night.

Therefore, this study attempts to estimate the UHI intensity according to nighttime light intensity (NTLI) offered by open nighttime light satellite imageries. A less costly and time-consuming approach to estimating UHI intensity over space and time is urgently needed for not only researchers but also policy-makers. How to mitigate UHI is vital for improving health and wellbeing of people living in cities. On the one hand, policy-makers are investing more and more on building material research to lower impervious surface area density; and one the other hand, they encourage urban planners to build more green spaces to reduce UHI. Some policies targeting UHI mitigation have been implemented for a few years and thus needs to be evaluated for changes or modifications in the future. A low-cost approach to rapidly monitoring UHI variations can assist in evaluating policy implementations. In this study, we propose a new approach to local-scale UHI intensity estimates by using nighttime light satellite imageries. We choose London as the case study since London is a representative city in the existing UHI studies (e.g., Giridharan & Kolokotroni, 2009; Doick, Peace, & Hutchings, 2014; dos Santos, 2020).

To the best of our knowledge, this paper is the first study devoted to local-scale UHI intensity estimates using nighttime light intensity directly as the explanatory variable. As satellite overpass time of nighttime light imageries is at night and nighttime UHI is more noticeable than daytime UHI, we selected the annual average nighttime surface UHI intensity to represent the annual average UHI level across London. The year 2015 is selected due to the data availability. Besides, we further choose Paris as a comparable city with London in this study. Owning to similar climatic and meteorological characteristics, economic development levels, energy-related policies, the associations of nighttime light intensity and UHI intensity might be fairly similar between London and Paris. Besides, since Paris is likely more polluted than London (Font, Guiseppin, Blangiardo, Ghersi, & Fuller, 2019), the potential influence of air pollution or sky turbidity conditions on UHI intensity estimates can be discussed in this study.

We attempted to establish models to estimate annual average UHI intensity (urban-rural surface temperature difference) at both the district and neighbourhood levels. We explored to what extent UHI intensity could be estimated according to nighttime light intensity at two geography levels. Moreover, we attempt to replace conventional regression models (non-spatial regression models) with spatial regression models in this study. Although the possible presence of residual spatial autocorrelation in the conventional regression models applied to geospatial data, most of the studies are likely to neglect this issue when fitting data to models. Therefore, in this study, we attempted to use spatial regression models to estimate UHI intensity according to NTLI if spatial autocorrelation existed in the residuals of non-spatial models estimated. The matrix exponential spatial specification (MESS) models which has analytical, computational, and interpretive advantages over conventional spatial autoregressive models were used in this study. Besides, this study can also pave a new way for estimating UHI intensity in some cities where accurate UHI data is missing.