The Impact of Land Use Changes on Carbon Flux in the World’s 100 Largest Cities

Abstract

Urbanization has become an important player in the global carbon cycle, and land use change is the second largest source of carbon emissions. However, despite great advances in remote sensing and satellite imagery, there is no reliable estimate of the impact of land use change on changes in land carbon stock in global cities. This paper quantified the impact of land use change on land carbon flux in the world’s 100 largest cities by using annual land cover data based on LandSat 8 images and land carbon stock parameters provided by the IPCC (Intergovernmental Panel on Climate Change). It was found that significant urban expansion could be observed in 83 cities, while 29 cities showed a deforestation trend, and croplands in 42 cities have shrunk. Carbon stock reduced by more than 112 million tons in the 100 selected cities from 2013 to 2022 due to land cover change. A total of 39 cities showed significant negative trends in land carbon stock that were mainly caused by urban sprawl and shrinkage in forest or cropland, among which Kolkata, Chongqing, Seoul, Guangzhou, and Hefei showed the greatest decline. Because of the growth of forest and cropland, or reduction in barren land and grassland, 28 cities showed clear positive trends in land carbon stock. In order to increase urban land carbon stock, the urban planning of most cities should focus on the protection of forests or afforestation that replace barren land or grassland and should avoid mindless urban expansion.

1. Introduction

One of the most influential factors in human-induced global environmental change over the past two hundred years has been the shift in land and resource use associated with urbanization and the expansion of low density development [1,2,3]. Currently, over 50% of the world’s population lives in urban areas, and this is going to reach 70% by 2050 [4]. This proportion has already reached 81% in the US and 95% in California [5]. With urban population growth, global urban land cover may increase by 1.2 million km² in 2030, nearly tripling the world’s urban area in 2000 [6]. The creation and expansion of cities and urban areas are typically associated with significant carbon emissions through the removal of vegetation, the addition of impervious surfaces, and increases in local fossil fuel usage. Around the globe, rates of land cover change and losses in terrestrial carbon storages are expected to increase significantly in the coming decades as the population continues to grow [7,8,9,10]. Accurate quantification and mapping of these stocks are critical for integrating urban forestry in local climate action plans, carbon offset markets, and urban planning for sustainable development.

Currently, several studies have evaluated carbon flux caused by land cover change in several cities worldwide. For instance, Ghaffar et al. [11] assessed the change of carbon sources and sinks through land cover change in Bangkok, Thailand. This article discovered rapid urbanization and shrinkage of cropland in Bangkok Metropolitan Area; however, it did not form a quantitative analysis between the growth of carbon release and significant changes of urban land cover. Studies of Xu et al. [12], Wang et al. [13], and Gao et al. [14] only considered the loss of carbon sequestration capacity caused by the decline of greenspace when analyzing regional carbon balance in Guangzhou, Beijing-Tianjin-Hebei region, and Hubei province in China, respectively. However, excluding the amount of carbon dioxide greenspace absorbed by photo-synthesis, land carbon flux mainly refers to the change of carbon stored in biomass and soil [15]. At present, land use change is the second largest source of carbon emissions, after energy consumption [16]. Globally, about 1.1 Pg of carbon is released into the atmosphere each year from net land use change [17]. Although much research has analyzed land carbon flux (land carbon emission or sink) by calculating the difference of total land carbon stock caused by land cover change in respective regions [15,18,19,20], these current studies only focused on a single city/region to analyze land carbon flux using different data source and methods, and their conclusions are only associated with their case study area. Therefore, the results these studies produced may prove difficult to compare with other results. While the study of Churkina [21] summarized carbon flux in major urban areas globally, actual land cover datasets were not applied. This happened because, despite great advances in remote sensing technology, it was still difficult to obtain global land cover datasets with high spatial and temporal resolution. Without a global estimate, we cannot explain the variations in urban land cover amongst countries, nor can we predict the amount of land use change that will be needed to mitigate carbon emission in many of these countries.

This study aimed to understand the impact of land use changes on carbon stocks in global cities based on uniformed datasets and methods. Using the world’s 100 largest cities as case studies from 2013 to 2022, the Landsat 8 (30 m) database was used to identify the land use patterns of these cities in time. The carbon stocks of these cities were estimated with land use type coefficients under different climate zones provided by IPCC. Multivariate linear regression was used to estimate the extent to which land use changes have impacted the carbon stocks in these cities. The results from this study provide new estimates and insights into the impact of urban expansion and the effect of associated land cover changes on terrestrial carbon stocks.

2. Materials and Methods

The method flowchart applied in this study is demonstrated in Figure 1 below.

2.1. Selection of Cities

Both the population density and size of urban areas were used to choose global cities in this study. In previous studies, spatial datasets of population density were widely applied as the base to extract urban areas in a global or national range [22,23,24]. Among these, studies of Weiss et al. [23] and Mori et al. [24] extracted urban area worldwide based on the Land Scan 2020 global population density database [25], where areas with a population density greater than 1000/km² and a total population of at least 10,000 within a continuous area were identified as urban areas. According to the same database, we added the size of urban area as another metric because if only population density was used, most of the selected top cities would be only in Asia. As there is a high population density in rural regions of India, Bangladesh, Pakistan, and Indonesia, these areas were excluded. In addition, the urban areas with limited availability of remote sensing images because of high cloud cover were excluded to ensure the quality of the analysis. Consequently, the top 100 largest cities selected covered seven continents, although Asia and North America included more cities than other continents (Figure 2a).

To better capture the pattern of land use change in all cities, 15 km buffers were created around the selected urban areas to include the land use changes in the adjacent areas rather than the administrative divisions. In addition, spatially interconnected cities were treated as a single study area. For instance, the largest study site among the 100 cities is a delta city cluster of the Pearl River in China that is headed by Guangzhou but also contains four nearby cities: Dongguan, Shenzhen, Foshan, and Zhuhai (Figure 2b). Three other selected study sites that contain more than one city are as follows: Wuxi and Changzhou, Hangzhou and Shaoxing in China; Essen and Dortmund in Germany.

2.2. Land Use Classification

The study period of this research was set from 2013 to 2022 to reflect the land use change in recent decades. Most of the existing global land cover datasets were generated based on SPOT or MODIS satellite images, which have low spatial resolution (300 m^–1 km) [26,27]. Land use datasets based on higher resolution images, such as Landsat (30 m resolution), have insufficient temporal span and are only available in specific years. For example, the only available global land use datasets based on Landsat images are FROM-GLC produced by Tsinghua University [28], which only include data from 2010, 2015, and 2017. This study classified land cover into 16 classes (

Table 1. Land use classification.

Land Use Classification	Subclass
Open Water
Perennial Snow/Ice
Developed Area	Open Space
	Low Intensity
	Medium Intensity
	High Intensity
Barren Land
Forest	Deciduous Forest
	Evergreen Forest
	Mixed Forest
Grassland	Shrub/Scrub
	Herbaceous
Cropland	Hay/Pasture
	Cultivated Crops
Wetland	Woody Wetlands
	Emergent Herbaceous Wetland

) based on Landsat 8 (30 m) using a pre-trained deep learning package in the ArcGIS Pro 3.0 environment [29]. Landsat 8 images were selected within the same period (within two months each year) to reduce the impact of seasonal variation. This method can not only produce land use data with a high spatial resolution for continuous years, but it also has an acceptable classification accuracy. The classification accuracy was tested for land cover data in Brisbane in 2017, which featured an overall accuracy of 86.08% and a Kappa coefficient of 0.804. These two indicators were 61.18% and 0.487 from FROM-GLC. (Refer to the distribution of accuracy test points and confusion matrices in Appendix A).

To generalize land use change in all study sites, 16 land cover types were grouped into six land cover groups. Specifically, developed area includes four subclasses, forest includes three subclasses, grassland includes shrub and herbaceous, cropland includes pasture and cultivated crops, wetland includes two subclasses, and barren land is its own group. Perennial snow/ice and waterbody were omitted since they occupied small areas and varied slightly.

After generating land use datasets for each city, Pearson correlation was applied between area of each land use group and time to detect trends of land use changes. Correlations with a coefficient above 0.3 were considered evidence of a strong trend [30].

2.3. Carbon Stock Calculation

The unified carbon stock calculation technique recommended by IPCC was adopted in this study to calculate and analyze carbon stock changes of different land use types in cities, which has been widely applied and justified in previous studies [11,22,31]. According to IPCC 2019 [32], carbon stock changes for each land cover type can contain one or more carbon stock/emission component, such as above-ground biomass, mineral soil, waste carbon stock, dead wood carbon stocks, and CO₂ emission in wetland. Specifically:

$$C_{soil\left(Wetland\right)}=({SOC}_{Ms}-{SOC}_E)\times A_{Wetland}$$

(1)

$$C_{soil\left(Settlement\right)}={SOC}_{AB}\times A_{Settlement}$$

(2)

$$C_{soil\left(Forest\right)}=({SOC}_{AB}+{SOC}_{MS}+{SOC}_{LC}+{SOC}_{DW})\times A_{Forest}$$

(3)

$$C_{soil\left(Grassland\right)}={SOC}_{MS}\times A_{Grassland}$$

(4)

$$C_{soil\left(Cropland\right)}=({SOC}_{AB}+{SOC}_{MS})\times A_{Cropland}$$

(5)

In Equations (1) to (5) above, C_soil refers to the annual carbon stocks of different land-use soils; A represents the area of corresponding land use type; SOC_MS represents the carbon stock of mineral soil at a 0–30 cm depth; SOC_E means carbon emission component from wetland; SOC_AB represents carbon stock from above-ground biomass; SOC_LC represents the carbon stock component of waste; SOC_DW represents the carbon stock of dead wood.

Datasets from the Reports of the Intergovernmental Panel on Climate Change [32], the 2013 Supplement (wetlands) [33], and the research of Godwin et al. [34] were used to calculate carbon stock/emission components mentioned above for each land use type in different continents and in different climate zones. As a result,

Table 2. Carbon stock/emission components for different land use types.

Land Use Classification	Subclass	Carbon Stock and Emission Components (Tons Ha-1)
Developed area		Above Ground Biomass (SOCAB)
	Open Space	61.86
	Low Intensity	58.22
	Medium Intensity	52.3
	High Intensity	42.29
Forest		Above Ground Biomass (SOCAB)
		Tropical		Subtropical				Temperate
		Africa	America	Asia	Africa	America	Asia	Asia	Europe	America and Oceania
	Deciduous	236.6	187.3	67.7	70	100	180	200
	Evergreen	190	195	433.5	35.1	74.6	250.2	170.4	214.7	185.9
	Mixed	69.6	127.5	184.6	65.2	115.9	70.9	116	162	128.9
		Mineral Soils (SOCMS)
		Tropical		Subtropical			Temperate
		dry	moist	dry		moist	cool and dry	cool and moist	warm and dry	warm and moist
	All types	21	40	24		64	43	81	24	64
		Litter (SOCLC)
		Tropical		Subtropical				Temperate
	Deciduous	4.3		5.6				23.9
	Evergreen	14.8		6.8				66.3
	Mixed	5.9		8.7				47.8
		Dead Wood (SOCDW)
		Tropical		Subtropical				Temperate
	Deciduous	8.4		4.1				23.6
	Evergreen	3.4		10.9				22.1
	Mixed	8		13.2				23
Grassland		Mineral Soils (SOCMS)
		Tropical		Subtropical			Temperate
		dry	moist	dry		moist	cool and dry	cool and moist	warm and dry	warm and moist
	All types	21	40	24		64	43	81	24	64
Cropland		Above Ground Biomass (SOCAB)
		Tropical and Subtropical					Temperate
	Hay/pasture	22.1					26.1
	Cultivated Crops	58.2					69.9
		Mineral Soils (SOCMS)
		Tropical		Subtropical			Temperate
		dry	moist	dry		moist	cool and dry	cool and moist	warm and dry	warm and moist
	All types	21	40	24		64	43	81	24	64
Wetland		Mineral Soils (SOCMS)
		Tropical		Subtropical			Temperate
		dry	moist	dry		moist	cool and dry	cool and moist	warm and dry	warm and moist
	All types	22	68	74		135	87	128	74	135
		Wetland Carbon Emission Component (SOCE)
		Tropical and Subtropical				Temperate
		dry		moist		cool	warm and dry		warm and moist
	All types	2.95		2.77		1.02	1.7		1.46

was used to calculate the carbon stock/emission components for different land use types. It should be noted that barren land was not included in the calculation above since it does not contain any carbon stock component and its carbon stock is considered to be zero [32].

For each city, the carbon stock was calculated each year during the study period based on annual land cover datasets. Pearson correlation was applied again between land carbon stock and time to detect the trend of changes. Correlations with a coefficient above 0.3 were considered evidence of a strong trend [30].

2.4. Evaluation of the Impacts of Land Use Change on Carbon Stocks

According to the correlation analyses in Section 2.2 and Section 2.3, cities that showed a significant trend in carbon stock change were further analyzed by stepwise multiple regression. For these cities, area changes of land use groups with a clear trend were selected as independent variables, and they were introduced into the model one by one. The annual land carbon flux was set as the dependent variable. After eliminating land use groups with insignificant influence, the multivariable regression models with a p value lower than 0.05 [35] were generated and classified into several groups to generalize how land use changes impact carbon flux in the selected cities.

3. Results

3.1. Average Annual Change of Six Land Use Types of 100 Cities from 2013 to 2022

The area of 100 study sites around the world ranges from 19,548 km² (the Pearl River Delta city cluster) to 1687 km² (Cape Town in South Africa). There are seven cities with areas exceeding 7500 km². Five of them are located in Asia: Guangzhou (Pearl River Delta city group), Tokyo, Kolkata, Dhaka, and Jakarta. The areas of 20 cities range from 5000 to 7500 km² and are mainly in North America, China, and Southeast Asia. The other 73 cities with areas less than 5000 km² are distributed throughout all continents in the world.

Most cities that were selected (83 out of 100 cities) had a significant trend of urban expansion during the 10 years (Figure 3). A total of 25 cities showed a relatively high urban expansion speed, with an average annual growth of 20 to 50 km², mainly located in China, India, and the US. Rapid urban expansion could be found in 19 cities that are also mainly distributed in China and India. Among them, Guangzhou (Pearl River Delta city group) had the highest rate of urban expansion, which was 193 km² annually. The annual urban expansion area of 39 cities was less than 20 km². These are mainly located in Europe, the Middle East, and America. Three cities in South Africa and two cities in Australia showed slower growth speeds (10.4 km² and 6.9 km² annually in Sydney and Brisbane, respectively). Compared with the study of Novotný et al. [36] that predicted urban expansion in 43 megacities from 2015 to 2030, the annual urban expansion rates produced by our study are similar in corresponding cities.

It seems that only Tokyo, Japan, where the developed areas decreased by around 31 km² annually during the study period, showed that the city was well developed, and the urban green space showed a slight expansion trend. The urban areas in 16 cities had no obvious trends of change, and they are mainly distributed in Europe and in North/South America.

Significant change trends in the forest were observed for half of the cities studied. A total of 16 cities showed slight forest expansion, ranging from 0.2 to 17.7 km². Five cities experienced significant forest expansion, with Beijing, China, experiencing the highest forest growth rate of 41.3 km² annually. Eleven cities had high forest decreasing rates, whose forests reduced below 20 km². Kolkata in India had the highest reduction rate, with a −193.7 km² decrease of forest annually. Moderate forest reduction (−20 to −1.5 km² annually) during the study period was shown in 18 cities that are scattered in all continents. Moreover, the other 50 cities had no significant trend of change in forest in the given period.

A relatively significant increase trend of grassland was observed in 15 cities that are mainly located in Europe, among which the grassland expanded by 140.7 km² annually in Cairo, the most rapid growth. A decrease trend of grassland was shown in 39 cities, with most of them distributed in India and China. For 17 cities, the reduction area was above 30 km ² each year. New Delhi in India had the highest annual reduction, with −217.3 km² of grassland shrinkage annually. The other 46 cities had no obvious trend in grassland change.

There was a relatively significant increase/decrease trend of cropland in 57 cities. Among these, cropland expanded by 105.4 km² each year in New Delhi, much higher than all other cities. The area of cropland in 25 cities showed a moderate decrease trend, with an average annual decrease of less than 20 km². The other 17 cities had a higher decreasing rate, with the fastest shrinkage in cropland being in Beijing, with a 109.1 km² decline each year.

Wetlands tend to be distributed in coastal cities or cities with waterbodies of a certain scale, where clear trends of change are more likely to be observed. An increase trend was shown in 25 cities, of which Kolkata expanded the most, with an annual increase of 111.2 km². There was an obvious shrinking trend in 22 cities, among which wetland in Dhaka in Bangladesh featured the fastest reduction, with an annual decrease of 178 km², while 53 cities had no obvious trend of wetland change.

Among the 49 cities showing relatively obvious trends in barren land, slow barren land expansion was identified in 24 cities that are scattered around the world. Seoul in South Korea had the fastest expansion, reaching 9.8 km² each year. Only five cities had an annual decrease of more than 10 km², including Cairo, Amman, Riyadh in the Middle East, Karachi in Pakistan, and Chongqing in China. In contrast, more than half (51) of the 100 cities had no clear trend of change in the barren land.

3.2. Average Annual Change in Carbon Stock within 100 Cities

According to Equations (1) through (5) and Table 2, land carbon stocks per unit area in forest, grassland, cropland, and wetland vary due to different climate conditions. In general, forests had the highest carbon stocks on a global scale, followed by cropland and wetland. The stock volumes were the lowest in developed area and grassland, except for barren land, whose land carbon stock is zero. The land use changes that led to carbon fluxes varied in 100 cities. The expansion of land use types with higher carbon stocks was used to occupy land use types with lower carbon stocks, resulting in increased land carbon stocks; in contrast, negative land carbon fluxes tend to result from increases in developed area or grassland and decreases of forest, cropland, or wetland.

The total volume of carbon stock in all study sites decreased by 112,211,487.9 tons in the past ten years due to land cover change. Specifically, 39 cities had significant decrease trends in carbon stock from 2013 to 2022 that are distributed in all continents in the world and mainly in China (Figure 4). A total of 21 cities showed a relatively low rate of decline (within 300,000 tons of carbon stock per year). In these cities, Houston, Dallas, Milan, Budapest, and Phoenix, they lost cropland, which was mainly transformed into developed area. In New York, Monterrey, and Sydney, urban expansion mainly occupied the original area of forest. Both forest and cropland decreased in Guatemala, Santiago, and Denver. The land carbon stock decreased by 135 tons km² per year in San Francisco, which is close to an average of 1.2 Mg C ha⁻¹ yr⁻¹ in Seattle produced by the study of Hutyra et al. [18]. Among the remaining 18 cities whose decrease rates were higher than 300,000 tons, Kolkata in India had the sharpest downtrend in carbon stock, with 2,503,007 tons of net carbon release per year caused by shrinkage in forest and cropland. Forest reduction was one of the main reasons that caused negative carbon fluxes in these cities.

In contrast, a significant increase trend could be observed in 28 cities, with many of them located in India and the Middle East, while others are scattered across the US, China, South Africa, and Australia. There were 17 cities with growth rates lower than 300,000 tons annually. Among them, carbon stock increased since the land cover converted from grassland to developed area in San Diego, Tehran, Tashkent, Bengaluru, Hyderabad, Ankara, and Johannesburg. In cities with an arid climate, including Las Vegas, Riyadh, Karachi, and Amman, the replacement of barren land with developed area also resulted in carbon stock increase in a moderate rate. In the other 11 cities with a higher increase (rate more than 300,000 tons each year), Xi’an had the highest growth rate in carbon stock where developed area and forest expanded significantly and replaced grassland, reaching an average of 803,478 tons annually. Significant carbon stock growth in Beijing, Moscow, Tokyo, and Paris was also caused by forest expansion.

The volumes of carbon stock in 33 cities did not show a significant trend of change. They are mainly distributed in North America and Europe, and a few cities are in Northern China. Grassland shrinkage could be observed in nine cities, among which grassland in Qingdao, Istanbul, Asuncion, Jinan, Austin, Bogota, and Kabul were mainly converted to urban areas. Cropland in Quanzhou, Taiyuan, Guadalajara, St Petersburg, Toronto, and Fukuoka was reduced significantly and changed to other land cover types. (Refer to Appendix B for specific data of average annual carbon flux changes and land use changes in all cities).

3.3. Regression Models between Carbon Flux and Land Cover Change in the Selected Cities

Among the 67 cities demonstrating significant changes in land carbon stock, the carbon stock changes in 60 cities could be well explained by changes in one or two land cover groups. These 60 cities were divided into five groups and 15 subclasses based on land use types that significantly contributed to changes in carbon stock (Figure 5,

Table 3. The 60 cities corresponding to carbon stock changes caused by different land use changes.

5 Main Groups	Subgroups	Cities	Average R2 of Models
Forest driven	Forest and developed area	Chongqing, Tokyo, Nanchang, Xi’an, Sydney, Guatemala, Denver, Kochi	0.9305
	Forest and cropland	Nagoya, Santiago, Adelaide, Moscow, Seoul, Beijing	0.9747
	Forest and wetland	Washington, Maputo	0.9975
	Forest	Guangzhou, Kolkata, New York, London, Paris, Hangzhou, Surabaya, Monterrey, Birmingham, Shenyang, Durban	0.7612
Grassland driven	Grassland and forest	San Diego, Hefei	0.92
	Grassland and barren land	Tehran	0.865
	Grassland	Los Angeles, Johannesburg, San Francisco, Rio de Janeiro, Hyderabad, Lahore, Bengaluru, Goiania, Ankara, Tashkent, Rome	0.8056
Developed area driven	Developed area and barren land	Las Vegas, Dubai	0.97
	Developed area and cropland	Milan, Brisbane, Chennai	0.819
	Developed area and grassland	Nanjing, Shijiazhuang, Changsha	0.837
	Developed area	Wenzhou, Wuhan	0.5135
Cropland driven	Cropland and developed area	Shantou, Phoenix	0.771
	Cropland and forest	Zhengzhou	0.759
	Cropland	New Delhi, Jakarta, Bangkok	0.753
Barren land driven	Barren land and developed area	Karachi, Riyadh, Amman	0.9903

). The land carbon fluxes in such cities were mainly driven by the change in forest, grassland, developed area, cropland, and barren land, respectively.

According to the standardized regression coefficients (Beta), there were 27 cities whose carbon stock changes were mainly driven by the change of forest within their respective extents (Table 3), and the coefficients of forest were all positive in the respective regression models. In eight cities where developed area change was the second independent variable that impacted land carbon fluxes, the land carbon fluxes were negatively correlated with developed area in Chongqing, Nanchang, Xi’an, Tokyo, and Sydney. Among these, only Xi’an and Tokyo had increased carbon stock due to reforestation. In six cities where land carbon fluxes were also driven by, and positively correlated with, the change of cropland in addition to forests, the land carbon stock reduced in Nagoya, Seoul, and Santiago, where both forest and cropland shrunk in such cities. In Adelaide, Moscow, and Beijing, forest expansion and cropland reduction were the main causes of positive land carbon fluxes. In Washington and Maputo, land carbon stock reduction was driven by the change in forest and wetland. These two cities both showed forest shrinkage within the study period. The wetland reduced in Maputo and expanded slightly in Washington, which could not offset the negative effects of forest loss on land carbon flux. In the other 11 cities where forest change was the sole independent variable, land carbon stock decreased in Guangzhou, Hangzhou, Shenyang, Surabaya, Kolkata, London, Birmingham, New York, and Monterrey with forest shrinkage. A significant forest expansion trend could be observed in Paris and Durban, which increased carbon stock in these cities.

Land carbon stock changes in 14 cities were mainly caused by, and negatively correlated with, changes in grassland. Among them, forest changes in San Diego, US, and Hefei, China, was the second influencing factor, which was positively correlated with carbon fluxes. Among them, slight forest expansion and grassland shrinkage increased land carbon stock moderately in San Diego; the rapid decline in carbon stock in Hefei was caused by deforestation and grassland expansion. Change in barren land was the second independent variable in Tehran. The conversion of grassland and barren land to other land use types resulted in a positive land carbon flux in this city. In the remaining 11 cities, significant grassland decreased, leading to the growth of land carbon stock in Los Angeles, Johannesburg, Rio de Janeiro, Ankara, Tashkent, Hyderabad, Bengaluru, and Lahore. In contrast, negative land carbon fluxes observed in San Francisco, Goiania, and Rome were caused by grassland increase.

Changes in developed areas was the main variable leading to carbon stock changes in 10 cities. It negatively correlated with land carbon fluxes in these cities, except for Las Vegas and Dubai, where carbon stock increased due to developed area expansion and the reduction of barren land. Among three cities where carbon stock changes were also driven by cropland changes, urban expansion occupying the original cropland led to the reduction of carbon stock in Milan. In Chennai, India, land carbon stock increased due to rapid growth in cropland that offset the negative effect of expansion in developed area. Grassland changes was the second influencing factor that negatively correlated with carbon stock in Shijiazhuang, Nanjing, and Changsha; only Shijiazhuang featured positive land carbon flux in the past ten years. Carbon stock decreased in the remaining two cities, Wenzhou and Wuhan, which could only be explained by the rapid expansion of developed area.

Land carbon stock changes in five cities were mainly caused by changes in cropland. Among them, in Shantou and Phoenix, carbon stock reduction was also affected by the change in developed areas. However, changes in cropland and developed area were both negatively correlated with carbon flux in Shantou, in contrast to Phoenix, where the coefficients of these two variables were positive. In addition to the shrinkage of cropland, the decrease in land carbon stock in Zhengzhou was also caused by forest degradation. For Jakarta and Bangkok, which showed negative carbon fluxes, their land carbon fluxes were only driven by cropland shrinkage.

In three cities from the last group (Karachi, Riyadh, and Amman) located in arid regions, the growth in land carbon stock was mainly driven by the shrinkage of barren land, similar to Las Vegas and Dubai mentioned above. (Refer to Appendix C for all parameters of regression models corresponding to 60 cities).

4. Discussion

According to land use change analysis, Significant urban expansion could be observed in 83 cities, implying rapid urbanization is a global phenomenon. While 29 cities showed a deforestation trend, forest expansion only occurred in 21 cities. Cropland in 42 cities shrunk, while only 15 cities expanded their areas for agriculture. As the result of these urban land use changes, 112,211,487.9 tons of net carbon was released into the atmosphere from these 100 cities in the past 10 years.

A significant reduction trend of land carbon stock could be observed in 39 cities that are distributed on all continents and are mainly concentrated in Asia. They include, for example, Seoul, Guangzhou, Chongqing, Bangkok, Jakarta, and Kolkata. Most of them featured rapid urbanization in the past 10 years, where a decrease in forest and/or croplands that were occupied by developed areas, which was the main reason for a negative land carbon flux. Land use conversions that led to carbon stock decrease in these cities could be generalized into two main patterns. The first pattern featured urban expansion that occupied forest, mainly including Chongqing, Shenyang, Nanchang, Surabaya, Kolkata, and Kochi in Asia and Washington, New York, Denver, Monterrey, Santiago, London, and Birmingham in America and Europe. Although the urban expansion rates and forest decreasing rates of cities in developed countries are generally lower than that in developing countries, clear reduction trends of carbon stock in all the cities mentioned above require forest reservation and the limitation of urban sprawl. The second land use change pattern refers to cropland reduction that was converted to developed area, mainly including Guangzhou, Shantou, Wuhan, Bangkok, Seoul, Phoenix, and Milan. With the development of such cities, the primary industry (agriculture) might be occupied by secondary or tertiary industries [37]. To maintain total land carbon stock, blind urban sprawl should also be avoided through greenspace preservation or planning.

Positive carbon fluxes (increased land carbon stock) in 28 cities were mainly caused by an increase in forest or cropland and by shrinkage in grassland or barren land. They included, for example, Paris, Tokyo, New Delhi, Chennai, Las Vegas, and Dubai. Some of them are well-developed, such as Paris and Tokyo, where developed area remained unchanged, and green space was preserved; the others still featured significant urban expansion, accompanied by barren land shrinkage or agricultural development that increased total land carbon stock. Such land use conversions in these cities could also be generalized into three patterns.

First, forest growth in cities such as Paris, Tokyo, Moscow, Adelaide, Durban, Beijing, and Xi’an increased their carbon stock. The first five cities are well-developed, where developed area changes were insignificant. Beijing and Xi’an still featured rapid urban expansion; however, the increase in forest area in these two cities offset the negative impact of urbanization on carbon stock. These cities should keep current land management policies that might be beneficial for a positive land carbon flux. The second pattern showed that cropland increased with grassland shrinkage, including several India cities, such as New Delhi and Chennai. Due to population explosion [38], these cities planned more land for agriculture, increasing their overall carbon stock. However, with urban development in the future, developed areas may occupy these agriculture lands and cause negative land carbon flux. The final pattern refers to urban sprawl that occupied grassland or barren land in cities with relatively arid climate. These cities mainly include Tashkent, Tehran, Karachi, Las Vegas, Dubai, Riyadh, and Amman. In contrast to other cities where urban expansion had a negative impact on carbon stock or shrinkage in grassland or barren land, the land use types with the least carbon stock caused positive carbon fluxes in corresponding cities. For the cities of the last two land use change patterns, more green space should also be planned with the expansion of agricultural land or urban area, which can not only continuously increase the local land carbon stock but also reduce the impact of global climate change [39,40].

Land carbon stock had no obvious trend in 33 cities that are mainly located in North America, Europe, and East Asia, including Osaka, Shanghai, Dhaka, Cairo, and Mexico City. In some of these cities, changes in all land use types were insignificant. In other cities, conversion between grassland and developed areas had no significant effect on total carbon stock.

Cities support livable, sustainable, and resilient dwelling conditions by providing various ecological services, such as livelihood, reduced pollution, water regulation, carbon sequestration, biodiversity, recreation, shading, and heat stress relief [41,42]. Therefore, cities are complex adaptive systems, and the impact of land use change on land carbon stock should be considered from the whole-of-system perspective. For example, deforestation and de-agriculture imply less land carbon stock, and the fragmented land structure of urban green space also threatens the native plants and animals that mainly live in the “urban green-islands” in the metropolitan area [20]. These land cover changes degrade ecological services and provide less support for livelihood in developing countries. In addition, urban forests play an important role in the carbon cycle of the whole ecosystem. According to the study of McPherson et al. [43], urban forests are estimated to account for 2 percent of the total C stored and sequestered annually by trees, and they are responsible for 20 percent of the total reduction of carbon emissions in California. For the cities that contributed most to increasing atmospheric carbon concentration by land use change, including Kolkata, Chongqing, Guangzhou, and Seoul, urban management should focus on the limitation of rapid urban expansion and the protection of forests. In the group of 33 cities that did not show an obvious trend, not much should be conducted without a significant change of land use, but for other cities, a close eye should be kept on the changes of land use. For cities with positive land carbon fluxes due to barren land shrinkage or cropland expansion, more green space should also be planned since it can store a substantial amount of carbon and increase the adaptive environment for urban residents [13,43]. It is hoped that those well-developed cities continue to increase land carbon stock in the future by land planning and management. Furthermore, more than half of the population will live in urban areas by 2050 [44], and the global temperature is projected to warm by more than 1.5 degree on average by 2100 [45]. These challenges should also be considered in future urban use planning.

Compared to current studies that often focus on several cities at discontinuous time points [11,12,13,14,15,16,17,18,19,20,46,47,48], this study expanded study areas to the 100 largest cities that are distributed around the world and generated annual land cover datasets with high spatial resolution from 2013 to 2022 to analyze the changing trends for six land use groups in these cities. This study reflected on the recent comprehensive trend of the impact of urban land use change on land carbon stock in the world. It will assist the urban land use planning and management in these cities to increase the total volume of land carbon storage and will enable comparison and learning from each other based on carbon mitigation efforts between cities.

There are several limitations related to data sources and methods applied in this research. First, this research only considered carbon flux caused by the change in land carbon stock. Carbon dioxide emitted by other anthropogenic activities, such as energy consumption and industrial production processes, was not included. Since carbon sequestration by the photosynthesis of greenspace or land carbon flux can only offset a small proportion of carbon emission by human activities [12,13], our study cannot tell whether cities are carbon neutral. The second issue refers to the reliability of datasets. In a few cities located in tropical or coastal regions, high-quality satellite images were unavailable within the same month each year due to high cloud cover. When calculating carbon stock parameters for each land use type, data from IPCC database were uniform within the same climate type in a continent, which varied among different cities [49,50,51]. Additionally, in contrast to taking spatial visualization of changes in land carbon stocks within limited study extents [12,13,18,19,20], we were not able to take a spatial analysis of land carbon flux within each city due to the large amount of study sites. Thus, further studies require more reliable and accurate datasets, and carbon emission from anthropogenic activities should be calculated to assess carbon neutrality.

5. Conclusions

This study applied uniformed land use data and methods of calculating land carbon stock and analyzed the impact of land use changes on carbon stocks in 100 cities worldwide. According to the results of this research, the volume of carbon stock reduced by 112,211,487.9 tons in total among 100 selected cities from 2013 to 2022 due to land cover change. A total of 33 cities did not show a significant trend of annual carbon flux change within the study period. Among the remaining cities, 28 cities had an obvious increasing trend in carbon stock, among which Xi’an in China had the highest growth rate, reaching an average increase of 803,478 tons annually. In contrast, Kolkata in India had the sharpest downtrend of carbon flux among 39 cities that showed a significant decreasing trend, with a reduction of 2,503,007 tons of carbon stock annually. There was no clear relationship between changes in carbon fluxes and the geographical distribution and population changes of the corresponding cities. Regarding the results of stepwise multiple regression between annual carbon flux and land use change in 60 cities, land use changes were generalized into five patterns. The carbon stock changes were mainly positively correlated with forest and cropland areas and negatively correlated with urban areas and grassland areas. Therefore, in order to increase carbon stock in the future, the urban planning of corresponding cities should focus on the protection of forests or afforestation that replace barren land or grassland and should avoid rapid urban expansion. Although cropland expansion is also conducive to the growth of carbon stock, it is not appropriate to plan for more cropland in the future as it may lead to other environmental issues.