Volume 10, Issue 5 e2021EF002401
Research Article
Open Access

Urbanization Contributes Little to Global Warming but Substantially Intensifies Local and Regional Land Surface Warming

Decheng Zhou

Decheng Zhou

Jiangsu Key Laboratory of Agricultural Meteorology, College of Applied Meteorology, Nanjing University of Information Science & Technology, Nanjing, China

Contribution: Conceptualization, Methodology, ​Investigation, Writing - original draft

Search for more papers by this author
Jingfeng Xiao

Corresponding Author

Jingfeng Xiao

Earth Systems Research Center, Institute for the Study of Earth, Oceans, and Space, University of New Hampshire, Durham, NH, USA

Correspondence to:

J. Xiao and G. Zhou,

[email protected];

[email protected]

Contribution: Methodology, Writing - review & editing

Search for more papers by this author
Steve Frolking

Steve Frolking

Earth Systems Research Center, Institute for the Study of Earth, Oceans, and Space, University of New Hampshire, Durham, NH, USA

Contribution: Writing - review & editing

Search for more papers by this author
Liangxia Zhang

Liangxia Zhang

Jiangsu Key Laboratory of Agricultural Meteorology, College of Applied Meteorology, Nanjing University of Information Science & Technology, Nanjing, China

Contribution: ​Investigation

Search for more papers by this author
Guoyi Zhou

Corresponding Author

Guoyi Zhou

Jiangsu Key Laboratory of Agricultural Meteorology, College of Applied Meteorology, Nanjing University of Information Science & Technology, Nanjing, China

Correspondence to:

J. Xiao and G. Zhou,

[email protected];

[email protected]

Contribution: Writing - review & editing, Supervision

Search for more papers by this author
First published: 04 May 2022
Citations: 15

Abstract

Increasing urbanization causes an urban heat island (UHI) effect and exacerbates health risks of heat waves due to global warming. The surface UHI (SUHI) in large cities has been extensively studied, yet a systematic evaluation on the impacts of urbanization on local-to global-scale land surface warming is lacking. We propose a new procedure to quantify the warming effects of urbanization at local, regional, and global scales using high-resolution satellite observations. We find strong local warming effects for 88% of the urban-dominated pixels across the globe and cooling effects for the rest of the urban lands on a diurnal mean timescale, with a global urban mean intensity of 1.1°C in 2015. The SUHI effects differ substantially by time of day, season, and climate zone, and are closely related to surface evapotranspiration. By extending local effects to the entire land surface, we estimate a diurnal mean warming of only 0.008°C globally. However, urbanization can have large warming effects regionally, especially in eastern China, the eastern United States, and Europe. In addition, we show that global urban expansion results in over three-quarters of SUHI effects in 1985–2015, and its effect will likely increase by 50%–200% by the end of this century. The SUHI-added warming could be up to 0.12°C in summer in Europe by 2100 under a fossil-fueled development pathway. Our results reveal that urbanization substantially intensifies local and regional land surface warming and that prioritized attention should be given to the SUHI-added warming in highly or rapidly urbanized regions.

Key Points

  • Impacts of urbanization on local-to global-scale land surface warming are quantified according to high-resolution satellite observations

  • Urbanization contributes little to global warming but greatly intensifies local and regional warming with large spatiotemporal variability

  • Urban expansion leads to over 3/4 of the current urban heating effects in 1985–2015 and likely a 50%–200% further increase globally by 2100

Plain Language Summary

Rapid urban expansion can increase temperature by the resultant urban heat island effect, therefore exacerbating negative impacts of global warming on the environment, human health, and energy consumption. However, the magnitude by which global urbanization intensifies land surface warming is not well understood. We proposed a new procedure to quantify the effects of urban land use on local-, regional-, and global-scale land surface warming using high-resolution satellite observations. It is shown that urban expansion substantially warms local and regional climates, with high spatial-temporal variability, but contributes little to global-scale warming, and that the impacts are projected to nearly triple under a fossil-fueled development pathway by the end of this century. This study offers a new perspective of methods to quantify the surface urban heat island effect and important insights for climate change assessments.

1 Introduction

We are living on a rapidly urbanizing planet, and urbanization is one of the most important megatrends in this century. The global urban population reached 4.2 billion in 2018 and is projected to be over 6.6 billion by 2050 (UN, 2019). To accommodate urban dwellers, more and more land surface has been (Gong et al., 2020) and will be (G. Chen et al., 2020) converted into artificial impervious surfaces, dramatically altering local and regional climate by modulating the land surface energy balance (Oke, 1982). The urban heat island (UHI) phenomenon represents a major alteration, leading to a suite of consequences such as precipitation changes (Shepherd, 2005), lower air quality (Grimm et al., 2008), higher morbidity and mortality (Patz et al., 2005), and higher energy consumption (X. Li et al., 2019). These effects can be aggravated by a warming climate (Krayenhoff et al., 2018; L. Zhao et al., 2021), particularly during heat waves (D. Li & Bou-Zeid, 2013; Jiang et al., 2019). Urban heating therefore has been regarded as one of the eight key risks caused by global change in the fifth IPCC assessment report (IPCC, 2014). Understanding UHI-added warming is crucial for not only formulating climate mitigation and adaptation strategies but also for gaining a more holistic view of human land use effects on the Earth's climate.

UHI effects are usually quantified by climate models, meteorological observations, or remotely sensed land surface temperature (LST) (Chapman et al., 2017; D. Zhou et al., 2019; Mirzaei & Haghighat, 2010; Stewart, 2011), and each method has strengths and limitations. Climate models can be used to predict the warming effects of urbanization from local to global scales (Argüeso et al., 2014; D. Li et al., 2019; Georgescu et al., 20132014; Huang et al., 2021; Krayenhoff et al., 2018; L. Zhao et al., 20142021; Q. Cao et al., 2018). Modeling results are determined by the predefined scenarios and physical processes, and may differ substantially between models and parameterization schemes (L. Zhao et al., 2021). Meteorological observations provide the most direct evidence of the urbanization effects on climate warming at local and regional scales, though their results are sometimes controversial due to the restricted geographic coverage, a limited number of urban-free meteorological stations, and the difficulty in separating the contribution of urbanization from that of other external forcing factors (Chrysanthou et al., 2014; Kalnay and Cai, 2003; L. Zhou et al., 2004; Sun et al., 2016). With rapid advancements in remote sensing along with geospatial science, satellite-sensed LST has been extensively used in recent decades to detect the surface UHIs (SUHI) from local to global scales under a space-for-time substitution assumption because of the unparalleled advantages of low cost, consistency, repeatability, and global coverage of satellite remote sensing (Voogt & Oke, 2003). The SUHI now forms the main body of the current UHI literature (D. Zhou et al., 2019). However, a systematic evaluation of the contributions of urbanization to local-, regional-, and global-scale land surface warming is still lacking.

Previous studies focused primarily on local SUHI effects in large cities and/or urban core areas, covering only a small fraction of global artificial impervious surfaces (D. Zhou et al., 2019). These studies are mostly further undermined by two inherent limitations. First, urban and rural areas are required to be defined in order to quantify the SUHI intensity, but there are no ideal and uniform ways to delineate their boundaries even today (Schwarz et al., 2011). This in turn has led to ongoing debates on the magnitudes, patterns, and drivers of SUHI effects (C. Cao et al., 2016; D. Li et al., 2019; L. Zhao et al., 2014; Manoli et al., 2019; Martilli et al., 2020). Second, the studies ignored possible daytime warming effects of crop land use activities relative to natural vegetation covers (Alkama & Cescatti, 2016; D. Zhou et al., 2021; Yan et al., 2022; Y. Li et al., 2015) and the SUHI of urbanization in villages, towns, and small satellite cities (Heinl et al., 2015) in the surrounding reference areas. This may not only substantially underestimate the SUHI intensity of large cities (D. Zhou et al., 2018) but also make it impossible to gain a more holistic view of urban-induced land surface warming. Some studies assessed the warming effects of urban expansion considering total artificial impervious surfaces at a city or urban agglomeration scale (Du et al., 2020; Hu et al., 2015; M. Zhao et al., 2016). Huang et al. (2019) provided a global-scale satellite assessment of heat island intensification due to urban expansion, but they concentrated only on local warming intensification from the projected increases in urban cluster sizes between 2015 and 2050.

In this study, we aim to provide a comprehensive evaluation of the SUHI-added land surface warming effects from local to global scales using high-resolution satellite observations. We consider all artificial impervious surfaces as urban lands. The temperature effects of urbanization (δT) are defined as the LST differences between urban lands and natural vegetation covers based on the space-for-time substitution assumption. First, we quantify the local SUHI effect (referred as δT50 hereafter) of the urban-dominated lands, which are defined as the LST pixels with the urban area percentage (UAP) ≥50% and the sum of water bodies and croplands <5%, to reduce the effects of mixed land use on the SUHI signal. Biophysical and climatic controls on δT50 are then analyzed by examining the variability of δT50 with evapotranspiration (ET), surface albedo, precipitation, and air temperature. Third, the contributions of all urban lands to regional- and global-scale land surface warming (δTall) are estimated by extending the δT50 to the entire land surface. Last, changes in δTall due to past (1985–2015) and future urban expansion (2015–2100) are explored under the framework of the shared socioeconomic pathways (SSPs). These efforts can improve our understanding of the SUHI-added warming at local, regional, and global scales and inform general mitigation and adaptation plans for the heat risks of global warming.

2 Materials and Methods

2.1 Satellite LST Observations and Urban Extent Data

LST data were obtained from the 1-km Aqua/MODIS 8-day LST composites (MYD11A2, Version 6) from 2014 to 2016. Aqua has two overpass times: 1:30 (nighttime) and 13:30 (daytime) local solar hours. We ensured that only data with good quality were used by retaining data with emissivity error ≤0.02 and LST error ≤1 K. The resultant missing values were filled based on the linear relationship between LST and skin temperature (Tskin) from the ERA5-Land meteorological reanalysis dataset (Muñoz-Sabater, 2019) on a per-pixel basis (D. Zhou et al., 2021). The gap-filled LST was temporally averaged and thereby aggregated from the 8-day time step to a seasonal scale for the period 2014–2016. Spring, summer, autumn, and winter in the northern hemisphere refer to the periods of March–May, June–August, September–November, and December–February, respectively. The seasons in the southern hemisphere were defined oppositely as compared to the northern counterparts.

The urban extent data in 2015 were based on the 30-m Global Artificial Impervious Area (GAIA) dataset (Gong et al., 2020). The total urban area reached 692,404 km2 in 2015, accounting for approximately 0.52% of the global land surface. Urban areas are mainly distributed in the northern mid-latitudes, especially in eastern China, the eastern United States, and Europe (Figure 1g). The United States and China together contribute nearly half of global urban lands. The temperate and cold zones account for about 75% of total urban lands according to the Köppen-Geiger definitions (Beck et al., 2018). We calculated the UAP for each 1-km pixel from the GAIA data to ensure that the resolution is consistent with that of the LST data (Figure 1). Crop and water area percentages were also derived from the GFSAD30 Cropland Extent-Product and ESRI's World Water Bodies in earlier work (D. Zhou et al., 2021) in order to identify urban-dominated pixels. The resultant urban-dominated lands account for one-fourth of the global urban area (Figure 1f).

Details are in the caption following the image

Spatial distribution of urban area percentage (UAP) at a 1 × 1 km2 spatial resolution in 2015. (b)–(e) Enlarged view of the UAP distributions (b1, c1, d1, e1) and the urban-dominated lands (b2, c2, d2, e2) in the eastern United States, Europe, North India, and eastern China, respectively. (f) Area of total urban and urban-dominated lands in each three-degree latitudinal bin. (g) Urban areas in different Köppen-Geiger climate zones.

2.2 Estimating Local SUHI Effect Due To Urbanization

The δT50 (i.e., local SUHI effect) was measured by the LST differences between urban-dominated lands (LSTurban) and surrounding natural vegetation covers (LSTref). Positive and negative δT50 values indicate a warming and cooling effect, respectively. Pixels dominated by natural vegetation and with the sum of urban, crop, and water coverage less than 5% were regarded as the reference. The reference pixels were identified by the 250-m MODIS Land Cover Type Product (MCD12Q1; Collection 6).

LSTurban was directly extracted from the gap-filled LST data. LSTref was usually represented by the elevation-corrected mean LST of reference lands in a small grid scale (e.g., 50 × 50 km2) according to the space-for-time assumption. The method is effective when the reference pixels are relatively even distributed in a grid. However, it cannot quantify the δT50 of all urban-dominated lands due to the lack of adequate reference pixels and the complex terrain surface in some areas. For example, about 45% of the urban-dominated pixels have fewer than 125 reference pixels (5% of the total land pixels) in a 50 × 50 km2 grid scale (Figure S1 in Supporting Information S1). Therefore, a cokriging interpolation method (Davis, 1986) was used to predict LSTref from the LST observations of the surrounding reference land pixels by taking elevation as the secondary attribute. The cokriging method can consider the spatial dependence between known and unknown pixels and the co-variation with more densely sampled elevation. It also allows the estimation of δT50 on a per-pixel basis at a high-spatial resolution of 1 km. Global 1-km elevation data were obtained from a recently released global elevation product (Amatulli et al., 2018).

We used the universal cokriging algorithm embedded in ESRI's ArcGIS 10.5 software with searching neighbors between 20 and 50 in each of the eight sectors. Theoretically, the interpolation tool can predict LSTref at once globally, but it exceeded our computing capacity due to the large number of reference pixels (∼108). Therefore, we divided the global land surface into 2.5° × 2.5° regions and simulated the LSTref in each region separately. To avoid possible low interpolation accuracy in the fringe areas of some regions caused by low density and uneven distribution of reference pixels, the reference land pixels in eight neighboring regions were also utilized during the interpolation process for each target region. Cross-validation was used to evaluate the interpolation accuracy. Overall, our method can reasonably predict LSTref with annual mean root-mean-square error of 0.9°C during the day and 0.5°C at night globally. Further details can be found in our previous work (D. Zhou et al., 2021).

In addition, we tested the sensitivity of the estimated δT50 to the size of regions by predicting LSTref in much smaller regions of 50 × 50 km. We roughly screened the regions with a total of 500 or more reference pixels in it together with its eight neighboring regions in order to reduce the possible interpolation errors caused by the lack and/or uneven distribution of the reference pixels. A total of 4,622 grids were finally chosen, covering only 6.2% of total global urban-dominated pixels. The results show that the annual mean δT50 estimated at the scale of 50 × 50 km relates closely to that of 2.5° × 2.5°, with mean absolute errors of 0.87 and 0.36°C during the daytime (Pearson's r = 0.72, P < 0.001) and nighttime (Pearson's r = 0.73, P < 0.001), respectively (Figure S2 in Supporting Information S1). However, the larger regions are preferred since they can ensure the full coverage of urban-dominated lands and the availability of the minimum number of neighbors in the eight sectors for interpolation.

2.3 Analyzing the Spatial Variations of δT50 and Their Drivers

We calculated seasonal and annual mean δT50 for both day and night. The diurnal mean effect was represented by the mean of day and night δT50. A δT50 larger than the 99th percentile or lower than the 0.1st percentile was considered an outlier and excluded from the following analyses. The mean δT50 for different climate zones and latitudes was calculated. The impacts of biophysical (surface ET and albedo) and climatic factors on the δT50 values were then examined. The ET data were derived from the 500-m MODIS gap-filled ET product (MYD16A2GF, 8-day composites, Collection 6) between 2014 and 2016 (Running et al., 2019). The gap-filled ET product has replaced the poor-quality data with good-quality values based on the Quality Control (QC) label for every pixel. The albedo was based on the shortwave white sky albedo from the 1-km, daily MODIS gap-filled albedo product (MCD43GF, Collection 6) (Schaaf, 2019). The background precipitation and air temperature data were obtained from the 1-km WorldClim 2.0 product (Fick & Hijmans, 2017). All the data were aggregated to seasonal and annual timescales at a 1-km resolution. ET and albedo changes (δET and δAlbedo) due to urbanization were quantified by the same method applied to the δT50. The changing trends of δT50 with above drivers were examined by a binning method. We calculated the binned mean δET, δAlbedo, precipitation, and temperature by 5 mm, 0.001, 5 mm, and 0.1°C intervals, respectively. The form of the δT50 tendency with a driving variable was explored by a linear and/or piecewise linear regression method.

2.4 Calculating Regional- and Global-Scale LST Changes Due To Urbanization

To quantify LST changes due to the urban land uses at regional and global scales, we extended the δT50 to the global land surface based on a linear relationship between δT and UAPs in a region with the same background climate. We grouped the global land surface into small grids of 50 × 50 km2 and assumed that the climate background was uniform within each grid. A linear relationship between the δT and UAP has been shown in previous studies (D. Zhou et al., 2019; H. Li et al., 2018). We further tested this linear assumption using the grids containing three or more pixels for each 1% UAP interval and 50 or more UAP levels. The δT pixels with the sum of cropland and water body fraction larger than 5% were excluded. A total of 603 grids were used for the test. The δT changed linearly and significantly with UAP for more than 82% of the grids (Figure 2). The linear trends are weaker at night than in the day, likely due to much lower SUHI effects and more complicated drivers at night (as discussed later). The linear relationship is also evidenced by relating δT50 to the UAPs of global total urban-dominated pixels (Figure S3 in Supporting Information S1). These together confirm the rationality of the linear assumption in this study.

Details are in the caption following the image

Linear regression analysis of land surface temperature (LST) changes (δT) with urban area percentage (UAP) for the 50 × 50 km2 grids containing ≥50 UAP levels. The UAP is evenly divided into 100 levels.

Grid cells can be grouped into two types based on their UAP: urban-dominated pixels (Surban) and others (Sother). The δTall in each grid can be expressed as the area-weighted mean of the temperature effects of the urban-dominated pixels and others as follows:
urn:x-wiley:23284277:media:eft21029:eft21029-math-0001(1)
where δTurban and δTother represent the urban-induced LST changes in Surban and Sother, respectively. The δTall was assumed to be zero for grids without urban lands. The δTurban can be directly calculated from the δT50 estimates. The δTother was estimated as:
urn:x-wiley:23284277:media:eft21029:eft21029-math-0002(2)
where UAPurban and UAPother represent the mean UAP for Surban and Sother, respectively, and δTmax indicates the potential maximum LST change due to urbanization (UAP equals to 100%). For each grid with fewer than five urban-dominated pixels, δTmax was interpolated from that of the adjacent grids with an inverse distance weighted method. We used the small threshold for the number of urban-dominated pixels (5) to minimize the number of grids requiring interpolation of δTmax. Two larger thresholds (10 and 20) were also used to examine the sensitivity of the results to the threshold. We find that the resultant latitudinal patterns of δTall are highly consistent among the different thresholds (Figure 3). The sensitivity of the results to the grid size was also explored using one smaller grid size (40 × 40 km2) and one larger grid size (60 × 60 km2). We find highly similar results when bigger and smaller grid sizes were used (Figure 3). We then estimated the contributions of urbanization to national-, continental-, and global-scale land surface warming according to the δTall based on the 50 × 50 km2 grids.
Details are in the caption following the image

Sensitivity of the estimated regional mean land surface temperature (LST) changes (δTall) to grid sizes and minimum number of urban-dominated pixels in each 50 × 50 km2 grid being employed to extend the LST changes of the urban-dominated pixels (δT50) to δTall.

2.5 Quantifying LST Changes Due To Past and Future Urban Expansion

Impacts of past (1985–2015) and future (2015–2100) urban expansion on regional and global mean LST changes were also investigated with a linear assumption between the changes of δTall and UAP based on 50 × 50 km2 grids. The LST change due to urban expansion in a grid was computed by multiplying δTmax in a grid by the UAP increase. The UAP changes in 1985–2015 were calculated from the 30-m GAIA data (Gong et al., 2020). The future UAP changes under the SSPs were obtained from Chen et al. (2020), who predicted future urban expansion according to historical Global Human Settlement Layer (GHSL) (Pesaresi et al., 2016) at a 1-km resolution. We only show LST changes under the SSP3 (a regional rivalry pathway contrary to global cooperation) and SSP5 (a fossil-fueled development pathway) scenarios because urban expansion under the other SSPs in general falls between that of SSP3 and SSP5 (G. Chen et al., 2020). Note that the total urban area estimated by the GHSL is 23% higher than that by the GAIA in 2015, though they are closely related across the grids (Figure 4). The per-grid UAP derived from the GAIA is generally larger than that from the GHSL in eastern China and the eastern United States, while the opposite occurs in Europe and most other regions. Thus, the past and future urban expansion and the resultant LST changes are not exactly comparable in terms of magnitudes. The purpose of this analysis is to reveal the general trends and patterns of regional LST changes due to urban expansion.

Details are in the caption following the image

Comparison of the mean urban area percentage (UAP) calculated from the GAIA product (Gong et al., 2020) with that from the Global Human Settlement Layer (GHSL) dataset (Pesaresi et al., 2016) at a 50 × 50 km2 grid level for the year 2015.

3 Results

3.1 Global Distributions of the Local Warming Effects Due To Urbanization

We find a warming effect during the daytime in 2015 for 84% of the urban-dominated pixels across the globe (Figure 5a). The warming is particularly evident in eastern China, the eastern United States, and Europe. Concurrently, urbanization cools the land surface in some arid regions like central Asia, northern India, the Middle East, and the central-west United States. On average, the urban-dominated lands increase LST by 1.7°C annually. The δT50 in general decreases with rising latitude in the northern hemisphere and fluctuates greatly by latitude in the southern hemisphere (Figure 5b). More specifically, the tropical region shows the largest warming (2.9°C), which is about 60% higher than that in the temperate and cold regions and nearly five times of that in the cold and arid areas (Figure 5g). In addition, the δT50 differs greatly by seasons and is characterized by much more intensive warming in summer than in winter, particularly in the middle-high latitudes (Figures 5b and 5g).

Details are in the caption following the image

Spatial distributions of local land surface temperature (LST) changes due to urbanization of the urban-dominated pixels (δT50). (a), (c), (e) Annual mean daytime, nighttime, and diurnal mean δT50. Histograms show the frequency of the main δT50 bins and the red lines represent the means. The urban-dominated pixels are shown in the form of ESRI Shapefile of points. (b), (d), (e) Mean δT50 for each three-degree latitudinal bin. The red shaded area shows the 95% confidence interval of the annual mean effect. (g) Mean δT50 in different climate zones, with the error bar representing the 95% confidence interval of the polar region. The ranges of the 95% confidence intervals are less than 0.033°C and not shown in the figure for the other climatic zones. The mean δT50 differs from zero (P < 0.0001, two-tailed t-test) in all the latitudinal and climate zones but the spring daytime and winter diurnal mean effects in the polar region. Antarctica is not shown in this and all the other figures.

A prevalent warming effect is also observed in most urban-dominated lands (75%) at night, but in a much lower magnitude and smaller seasonality than that in the day (Figures 5c, 5d, and 5g). Some regions such as the central-north United States and northeastern China present evident cooling effects. On average, daytime δT50 predominates the diurnal mean effect (Figure 5e) and approximately 88% of the urban-dominated pixels warm the terrestrial surface. The warming on average reaches 1.1°C globally, with the strongest in the tropical zones and the weakest in the polar regions. Note that nearly 40% of the urban-dominated pixels show a cooling effect in the day and/or at night (Figure 6). Also, the season with the maximum warming or cooling effect differs greatly over space, especially at night (Figure 7). These together suggest the complicated diurnal and seasonal cycles of the SUHI effects.

Details are in the caption following the image

Spatial distributions of the signs of the annual mean δT50 during the daytime and nighttime.

Details are in the caption following the image

The seasons with maximum warming and cooling for the annual-warming and annual-cooling urban-dominated pixels. Spr, Sum, Aut, and Win are spring, summer, autumn, and winter, respectively.

3.2 Changes of Surface Biophysical Characteristics and Their Controls on the δT50

Surface ET and albedo, two major biophysical factors, are greatly altered by urbanization (Figure 8). Most urban-dominated pixels (76%) present lower ET than the reference base conditions and the mean annual δET is −65.7 mm globally (Figure 8a). The resultant rannual ET decreases in urban areas are clearly larger in lower latitudinal regions than in higher latitude regions (Figure 8b). On average, ET decreases in the tropical regions are 2.7 times of those in the temperate and cold regions and 14.5 times of those in the arid zone (Figure 8e). The δET varies greatly by season in the cold and temperature regions but is negligible there in the winter (Figure 8e). By contrast, on average, urbanization leads to a slight increase in albedo, although the increase of albedo exhibits substantial spatial heterogeneity (Figures 8c and 8d). Albedo increases are larger in tropical and cold regions than in the temperate zone (Figure 8f), while arid and polar regions on average experience a slight decrease of albedo. The δAlbedo in general differs slightly by season globally except in the polar region, where the albedo decreases are larger in winter and spring, possibly due to snow presence on the reference lands in these seasons.

Details are in the caption following the image

Spatial distributions of evapotranspiration (δET) and surface albedo (δAlbedo) changes due to urbanization for urban-dominated pixels. (a), (c) Annual total δET and mean δAlbedo. Histograms show the frequency and the red lines indicate the global means. (b), (d) Mean δET and δAlbedo for each three-degree latitudinal bin. The red shaded area shows the 95% confidence interval of the annual mean effect. (e), (f) Mean δET and δAlbedo in each climate zone with the error bar representing the 95% confidence interval. The mean values are significantly (P < 0.0001, two-tailed t-test) different from zero in all climate (except the polar region) and latitudinal zones.

The changes of surface ET might contribute largely to the SUHI effects, and a higher δT50 corresponds to a larger ET decrease along with latitude bins or climate zones (Figure S4 in Supporting Information S1). This can be shown by the significant negative relationship between the binned δT50 and δET, particularly during the daytime (R2 = 0.89, N = 126, P < 0.01; Figure 9a). The δT50 decreases with δAlbedo at night (Figure 9b), suggesting that albedo plays an important role in regulating the nighttime SUHI effects. However, the generally positive δAlbedo in the tropical, cold, and temperature zones (Figure 8e) contradicts to the positive δT50 estimates (Figure 5g), indicating that other factors (e.g., anthropogenic heat emissions and heat storage capacity) likely make stronger contributions to nighttime warming effects than albedo. In addition, our results show the strong influence of background climate on δET (Figure 9c) and therefore on daytime δT50 (Figure 9e). This is shown by a larger ET decrease and LST increase with a higher annual precipitation and mean temperature. By contrast, the background climate contributes little to nighttime δT50 (Figure 9f), possibly owing to small temperature effects (Figure 5c) and strong controls by other factors.

Details are in the caption following the image

Variations of annual mean δT50 with biophysical and climatic factors. (a), (b) Variations of daytime and nighttime δT50 with δevapotranspiration (ET) and δAlbedo. (c), (d) Variations of δET and δAlbedo with annual precipitation and mean annual temperature. (e), (f) Variations of daytime and nighttime δT50 with annual precipitation and mean annual temperature. The light gray regions have no data. The line charts adjacent to the axes show the trend of δT50 (a), (b), (e), (f), δET (c), and δAlbedo (d) with the corresponding factor (N = number of bins). The light gray areas in the line charts indicate one standard deviation. Piecewise linear regression (green dashed line) is used to examine the possible non-linear climatic effects.

3.3 Contribution of Urbanization to Regional- and Global-Scale Warming

Urbanization could largely contribute to regional warming, especially in eastern China during the daytime (Figure 10a). The greatest daytime warming reaches 3.2°C at a 50 × 50 km2 level. More intensive warming is observed in the northern middle-high latitudes, particularly Europe, because of their larger UAPs (Figures 10b, 10d, 10f, and 10h). Though urbanization contributes only slightly to national-scale LST changes (<0.1°C) for most countries (Figures S5a–S5c in Supporting Information S1) due to their small percentages of urban lands (Figure S5d in Supporting Information S1), some highly urbanized small countries in Europe and Asia experience extensive warming due to urbanization, with the largest being 1.7°C annually in Singapore during the daytime. Globally, the annual mean LST increases by 0.013°C in the daytime and 0.004°C at night, with a diurnal mean intensity of only 0.008°C (significantly different from zero by two-tailed t-test). The top 10 countries contribute about 80% of the urban-induced warming (Table 1). China alone accounts for one-third of the total urban-induced warming effects with about one-fourth of global urban lands.

Details are in the caption following the image

Regional and global mean LST changes due to urbanization (δTall). (a), (c), (e) Daytime, nighttime, and diurnal mean δTall at a 50 × 50 km2 regional level. (g) Urban area percentage (UAP). The light-gray grids represent regions without urban lands (δTall = 0). (b), (d), (f), (h) Variations of δTall (daytime, nighttime, and diurnal mean) and UAP by latitude and climate zone. All mean δTall estimates are significantly different from zero (P < 0.0001, two-tailed t-test). The δTall for Antarctica was not assessed due to no urbanization there. AS, NA, EU, AF, SA, OA, and AU represent Asia, North America, Europe, Africa, South America, Oceania, and Australia, respectively.

Table 1. Contributions of the Top 10 Countries to Global Urban Lands and Urban-Induced Annual Warming Effects
Rank Urban area Daytime Nighttime Diurnal mean
Country Contribution (%) Country δTall Contribution (%) Country δTall Contribution (%) Country δTall Contribution (%)
1 USA 24.4 China 0.063 35.7 China 0.014 31.1 China 0.036 34.9
2 China 24.3 USA 0.037 20.7 India 0.015 11.0 USA 0.019 18.8
3 India 4.5 Japan 0.230 4.8 USA 0.004 9.3 India 0.014 4.6
4 Russia 3.4 Brazil 0.007 3.4 Italy 0.055 3.8 Japan 0.113 4.0
5 Brazil 3.0 India 0.017 3.2 France 0.025 3.6 Brazil 0.004 3.5
6 France 2.3 Germany 0.135 2.9 Germany 0.042 3.5 Germany 0.085 3.1
7 Italy 2.2 France 0.069 2.5 Brazil 0.002 3.3 France 0.045 2.8
8 Germany 2.2 Indonesia 0.017 1.9 Russia 0.001 2.7 Russia 0.001 1.9
9 Canada 1.9 Russia 0.002 1.7 Egypt 0.008 2.0 Italy 0.062 1.8
10 Mexico 1.7 Malaysia 0.079 1.5 Pakistan 0.008 1.7 Indonesia 0.009 1.7
  Total 69.8 Total 78.3 Total 72.0 Total 77.1

3.4 Impacts of Urban Expansion on Regional- and Global-Scale Warming

Figure 11 summarizes the past (1985–2015) and future (2015–2100) UAP increases and their impacts on δTall. Global urban land area increased by 2.9 times from 1985 to 2015, with bigger increases in Europe, Asia, and North America (absolute increase of UAP >0.45%) than in the remaining regions of the world (<0.2%; Figures 11a and 11c). In particular, the absolute UAP increase exceeds 50% in some regions in eastern China. LST increases due to urban expansion are estimated to be 0.01°C in the day and 0.003°C at night globally, accounting for about three fourths of the current urban warming effects. Europe and Asia experience much stronger warming intensifications (>0.018°C) than the other continents (<0.009°C; Figures 11b and 11d).

Details are in the caption following the image

Impacts of past and future urban expansion on the regional and global land surface temperature (LST) changes. (a), (b) Spatial variations of UAP changes in the past (1985–2015) and future (2015–2100) and their resultant δTall changes at the 50 × 50 km2 grid level. The light-gray grids represent regions without urban lands. (c), (d) Continental and global means of UAP and δTall changes in the past and future. SSP3 and SSP5 represent the shared socioeconomic pathways with the maximum and minimum future urban expansion, respectively.

Projected changes of UAP and δTall in the future differ greatly by both socioeconomic scenario and region (Figure 11). Under SSP3, the absolute UAP increase is projected to be 0.26% globally by 2100 and the relatively large increases will occur in Asia (0.34%) and Africa (0.32%). This will lead to a 0.0053 and 0.0019°C increase of global mean LST during the day and night, respectively. Under SSP5, the UAP will double by 2100 (increase from 0.52% to 1.06%). Europe (1.8%), Oceania (1.7%), and North America (1.2%) have much larger increases than the other regions (<0.34%). As a result, global mean LST is expected to increase by 0.0116 and 0.0023°C in the day and night, respectively, and much larger increases are found in the eastern United States and Europe than in other countries/regions. In particular, daytime LST increases will exceed 0.04°C in Europe and Oceania. Considering the current warming and the seasonal variations, the urban-induced daytime warming in Europe in summer could be up to 0.12°C by 2100 under SSP5 (Figure S6 in Supporting Information S1).

4 Discussion and Conclusions

This study provides a comprehensive view of urban induced-warming from local to global scales. Our results demonstrate a much higher warming intensity in the day than at night, which is contradictory to the similar magnitudes or a slightly lower magnitude at night as found by previous studies (Chakraborty & Lee, 2019; Clinton & Gong, 2013; D. Zhou et al., 2014; Peng et al., 2012). Also, we find the largest warming in the low-latitudinal regions, while previous findings showed that the middle-high (D. Zhou et al., 2014; Peng et al., 2012) or temperate regions (Imhoff et al., 2010) had the largest warming. The disparities can be attributed to the fact that the climatic effects of croplands were ignored in previous studies. One recent study showed that croplands had strong daytime warming effects, especially in the tropical regions, which in turn could lead to lower SUHI estimates if croplands were used as the rural reference (D. Zhou et al., 2021). Meanwhile, the prevalent nighttime cooling effects of croplands may result in an overestimation of the SUHI effect at night. For example, we estimate the δT50 relative to adjacent croplands at a 50 × 50 km2 grid level and indicate that the δT50 might be largely overestimated (e.g., eastern China) or underestimated (e.g., arid zone) in the daytime and mostly overestimated at night (particularly in eastern China; Figure 12).These together highlight the importance of considering crop land use effects when assessing SUHI effects (Yan et al., 2022), and that prioritized attention should be given to the urban-induced warming in tropical regions where heat risks are already the highest (Liao et al., 2018).

Details are in the caption following the image

The δT50 difference between this study and that computed using cropland as a reference. The latter was estimated by the LST difference of urban-dominated lands relative to the average LST of adjacent croplands at a 50 × 50 km2 grid scale. (a), (b) Daytime and nighttime δT50 differences.

The SUHI effect has been widely attributed to a decrease of latent heat flux in the day and an increase of heat storage at night, which is mainly determined by surface biophysical characteristics (D. Li et al., 2019; L. Zhao et al., 2014; Manoli et al., 2019; Peng et al., 2012). Anthropogenic heat emissions can aggravate the SUHI effect regardless of the time-of-day by adding an energy input to the urban system (Oke, 1982; Rizwan et al., 2008; Shepherd, 2005). In contrast to previous studies (D. Zhou et al., 2014; Peng et al., 2012), a significant positive relationship is found between daytime and nighttime δT50 globally in our study (Pearson’s r = 0.22, P < 0.0001, Figure S7 in Supporting Information S1), suggesting that there are possibly some common drivers for the day and night. For example, ET and hence the background climate played a dominant role in controlling both the daytime and nighttime SUHI effects (Figure 9). Urbanization can reduce heat dissipation by lowering ET and therefore decrease latent heat flux and increase heat storage (Oke, 1982; Peng et al., 2012; Shepherd, 2005). Precipitation influences the SUHI effects by modulating soil moisture and vapor pressure deficit (Manoli et al., 2019; L. Zhao et al., 2014), which are key factors controlling ET. Precipitation changes can also alter plant growth and leaf area index, which will in turn influence ET and albedo. However, it remains controversial whether a threshold exists for the background climatic effect (D. Zhou et al., 2016; Manoli et al., 2019) or not (D. Li et al., 2019; L. Zhao et al., 2014). Our results suggest that the climate effects have thresholds (Figure 3e), but that these threshold values are much larger than previous estimates (D. Zhou et al., 2016; Manoli et al., 2019). The disagreement might be caused by the use of the different definitions of SUHI effects and/or the use of different cities. Further studies are required to elucidate the role of the background climate on SUHI effects.

As proposed elsewhere (D. Zhou et al., 2014; Peng et al., 2012), we find that albedo effects are statistically significant only at night. Albedo likely affects SUHI indirectly through reducing the daytime heat storage, which is supported by the negative relationship between δAlbedo and δT50 at night. Albedo differences are especially large in the North China Plain and central-north United States, and at least partially contribute to the nighttime cooling effects in these regions. Nevertheless, albedo should not be considered as the dominant factor for widespread nighttime warming effects globally since most urban lands have higher albedo than the reference areas. The key role of albedo is to counteract the warming effects induced by other factors such as ET, convection efficiency, and anthropogenic heat emissions (L. Zhao et al., 2014). Similar to a previous study (Manoli et al., 2019), our study finds that albedo effects are more evident in dry regions and that their contributions are negligible in wet regions compared to those of ET. This can be partially seen in the non-linear correlation between δAlbedo and precipitation (Figure 9d).

By extending local SUHI effects of crop-dominated pixels to the entire land surface, our study shows that urbanization makes a small contribution to overall global warming. The satellite-sensed diurnal mean warming that we find is less than 1% of the observed global warming of air temperatures (1.09°C) from 1951 to 2010 (Stocker et al., 2013). Given the higher sensitivity of clear-sky LST to urbanization than the all-sky air temperature (Voogt & Oke, 2003), SUHI-added warming at the height of a standard weather station should be lower than the estimate in this study. Therefore, it is unlikely that increasing urbanization can strongly influence land surface warming at the global scale (Alcoforado & Andrade, 2008; Hansen et al., 2010; Parker, 2006). However, urbanization effects should not be ignored in local and regional climate change assessments and mitigations because of high population exposures to SUHI-added warming. For example, we estimated the population exposures to the SUHI-added warming and found that the proportion of the people exposed to the SUHI effects is much higher than that of the land surface exposed (Figure 13). More than one-fifth of the global population is exposed to warming of >1.0°C during the daytime at a local scale and more than one-third of the population is exposed to warming of >0.1°C at a regional scale of 50 × 50 km2. The exposures could be even more serious than the present estimate given the possible underestimation of population density for urban pixels based on current population products (CIESIN, 2018; Gao, 2020). More details can be found in the Supporting Infomation Text S1. We caution that more attention should be paid to the SUHI-added warming in eastern China, Europe and the eastern United States, which are projected to experience rapid increases of SUHI effects in future under a fossil-fueled development pathway (SSP5). It should be noted that a larger SUHI-added warming intensity does not necessarily mean a higher urban heating risk. Indeed, heat stress and mitigation strategies depend mainly on background climates and thermal characteristics of urban areas (Martilli et al., 2020).

Details are in the caption following the image

Cumulative percentages of the land area (short dash) and people (solid) exposed to the urbanization-induced warming effects in 2015 and in 2100 under SSP3 and SSP5 at a 50 × 50 km2 regional scale. The impacted land and people were also estimated at a local (or pixel) scale for the year 2015.

The contribution of urbanization to warming that we estimated with satellite derived LST data is generally lower than that estimated with air temperature based on meteorological observations. For example, Sun et al. (2016) estimated a 0.49°C urban-induced diurnal mean warming of air temperature in China over the period 1961–2013, which is much higher than the urban-induced land surface warming (0.036°C) estimated in our study. Park et al. (2017) found that urbanization contributed 3%–45% to air temperature warming in Korea. The disparity might be mainly due to the limited availability of UHI-free air temperature observations. Taking China’s 2,413 national meteorological stations as an example, we find that the absolute UAP in the 3 × 3 km2 neighborhood region of these stations on average increased by 33.4% between 1985 and 2015 (Figure S8 in Supporting Information S1). This undoubtedly can overestimate the contribution of urbanization to warming if proper corrections are not made.

The effects of urbanization on land surface warming deserve further investigation. First, satellite observations do not consider land-atmosphere feedbacks and non-local effects (e.g., impacts by remote regions through advection). These effects should be explored in future studies because they can alter the temperature effects of land use changes (L. Chen & Dirmeyer, 2020). Second, the possible SUHI changes due to climate change and urban land management might lead to certain biases in the estimated SUHI changes due to urban expansion (e.g., by assuming the same warming potential in the past and future for each grid). The effects of climate change and urban land management are beyond the scope of this study. Third, the MODIS LST data that we used have coarse resolution (1 km) and are not applicable for block-scale urban planning and continuous city-specific studies (e.g., climate variability between different urban typologies) (Martilli et al., 2020). Future studies could benefit from the use of finer-resolution LST data including new observations from ECOSTRESS (Chang et al., 2021a; Hulley et al., 2021). Fourth, though the cokriging interpolation method used in this study can estimate the LSTref of all the urban-dominated pixels, the lack of reference land covers in a large surrounding area of some urban lands may lead to certain biases of the estimated LSTref, and therefore the warming effects of urbanization due to the large variations of background climates. Fifth, unlike Aqua and Terra, new generation geostationary satellites (e.g., GOES-R, Himawari-8) offer high-frequency LST observations throughout the day and night. LST observations from GOES-R have been recently used to study how SUHI intensity varies over the diurnal (i.e., diel) cycle (Chang et al., 2022; Chang et al., 2021b). Future studies could use these high-frequency LST observations to reveal how the contribution of urbanization to land surface warming varies throughout the day and night. Finally, we anticipate that the complementary and/or synergistic use of satellite remote sensing, meteorological observations, and high-resolution climate model simulations in future efforts will likely better elucidate the effects of urbanization on climate warming and the underlying biophysical mechanisms.

Our study provides the first multi-scale (local, regional, continental, global) satellite evidence of the warming effects induced by global urban land use. The methods proposed in this study offer new perspectives for future SUHI studies, including avoidance of delineating urban/rural boundaries, elimination of cropland land use effects, reduction of mixed land use effects, and consideration of full urban land uses. Our results can provide insights for climate change assessments and mitigation options. We find that SUHI makes a small contribution to global-scale land surface warming. Meanwhile, we stress the importance of considering the SUHI-added warming in regional climate change assessments and mitigation for highly urbanized regions, but the warming effects of urbanization are lower than previously estimated warming based on meteorological observations. We affirm that increasing ET (e.g., green roofs, irrigation, and tree planting) is possibly the most effective mitigation strategy (D. Li et al., 2019; Manoli et al., 2019), but caution should be taken to avoid or minimize the possible risk of water scarcity (Bastin et al., 2019) and the reduction of thermal comfort caused by increasing air humidity (Jendritzky et al., 2012).

Acknowledgments

This work was supported by the National Natural Science Foundation of China (42061144004 and 42130506) and the National Key R & D Program of China (2021YFB2600102). J. X. was supported by the University of New Hampshire.

    Data Availability Statement

    MODIS data are provided by the Land Processes Distributed Active Archive Center (LP DAAC, https://lpdaac.usgs.gov/) managed by the NASA Earth Science Data and Information System (ESDIS) project. GAIA data are provided by Tsinghua University (http://data.ess.tsinghua.edu.cn/). DEM data are provided by the EarthEnv project (https://www.earthenv.org/topography). Climate data are provided by WorldClim, University of California (http://www.worldclim.org/). Population density in 2015 is provided by CIESIN at Columbia University and population densities under SSP3 and SSP5 in 2100 are provided by NASA’s Socioeconomic Data and Applications Center (https://sedac.ciesin.columbia.edu/). We thank an anonymous reviewer for constructive comments on our manuscript.