Introduction

Global warming has resulted in severe eco-environmental issues, making urgent measures to reduce greenhouse gas emissions necessary. As of February 2021, 124 countries had committed to achieving carbon neutrality by 2050 or 2060 (Chen 2021). China, being the world’s largest carbon emitter, has witnessed a continuous increase in emission levels (Xia et al. 2023). Despite this, China has actively taken responsibility for reducing carbon emissions and has committed to achieving carbon neutrality by 2060. Given that the main population and human activities are concentrated in urban areas, a majority of carbon emissions originate from these regions (Li et al. 2021). These emissions predominantly arise from industries, residential consumption, urban traffic and commerce, among other sources. Furthermore, the urban fringe serves as a hotspot for land-use changes due to the ongoing expansion of Chinese cities (Xia et al. 2020). The expansion often results in the occupation of a large area of green land. Land-use changes significantly impact the carbon balance, affecting both terrestrial ecosystem carbon (Lai et al. 2016; Simmonds et al. 2021) and anthropogenic carbon emissions (Wu et al. 2022). Therefore, optimising land use appears to be an effective approach to enhancing carbon sinks and mitigating carbon emissions. Cities exhibit diverse land use, encompassing various functional areas, landscapes and spatial patterns, all of which can influence carbon sink/source dynamics. Consequently, two key issues are pertinent to the establishment of low-carbon cities: accurate spatial monitoring of the city’s carbon sink/source capacity and determining effective strategies to reduce carbon emissions and enhance carbon sinks through land-use control.

Many attempts have been made to spatially monitor terrestrial carbon sinks and anthropogenic carbon emissions. However, there are still many shortcomings. First, evaluations of terrestrial ecosystem carbon are usually conducted at large scales, such as continents (Winkler et al. 2023), countries (Uchale et al. 2023) and provinces (Ran et al. 2023), and the resolution is often not high. Additionally, due to the use of different data sources and models, such as the Ecosystem process modelling method (He et al. 2019), the Atmospheric inversion method (Wang et al. 2021), the Eddy covariance method (Yu et al. 2014) and the Inventory method (Piao et al. 2009), the amount of carbon sink/source in China still faces significant uncertainties; carbon sink values varied between 0.1 Pg C/yr and higher than 1.0 Pg C/yr (Chuai et al. 2018). For calculating anthropogenic carbon emissions, both the ‘top-down’ and ‘bottom-up’ methods are used, each with advantages and disadvantages. The former method can provide more detailed data, improving calculation accuracy, but is usually not suitable for use on a large scale. The accuracy of the latter method is relatively coarse, but it allows for studies on a large scale (Wu et al. 2023; Lian et al. 2024).

As for spatial carbon emissions monitoring, it is typically conducted at the administrative unit level. Some researchers have attempted to simulate carbon emissions spatially using night-light data (Zhao et al. 2018; Jung et al. 2022). However, this approach has limitations, particularly in high-emission areas like industrial factories, which often exhibit weak night-light values. Other studies use carbon satellite data to estimate carbon emissions spatially based on carbon concentration data. Commonly used data sources include GOSAT, OCO-2, OCO-3 and China’s TanSat, launched in 2016. Although this method allows monitoring at large scales and over multiple periods, the resolution remains coarse (Hakkarainen et al. 2019; Kiel et al. 2021; Shim et al. 2019). Researchers have developed the China high-resolution emission database using power plants, the statistical data and the national industry investigation data. This database achieves a resolution of 1 km (Cai et al. 2018, 2019). Spatial data, such as population distribution, road network density and urban big data like cell phone signal data and points of interest, are also crucial indicators for the current spatially refined simulation of carbon emissions at the local scale (Yao et al. 2023; Luo et al. 2023). While previous studies have made significant efforts to enhance the accuracy of spatial carbon emission monitoring, existing research remains insufficient for urban areas, requiring more auxiliary data for improvement. Moreover, virtually no studies have been conducted at scales smaller than the city level. To effectively manage low carbon land use, there is a need for a comprehensive carbon monitoring map encompassing both anthropogenic carbon emissions and terrestrial carbon sink/source.

The modelling of land use and land cover change (LUCC) has consistently been a prominent research topic, driven by various factors including physical, environmental and socioeconomic considerations (Aburas et al. 2019). Future land-use change simulations are typically conducted under different scenarios, considering the demand for various land-use types and the spatial driving forces influencing land distribution (Liu et al. 2017). Numerous models have been developed to simulate the land-use change process, including cellular automata (CA) models, the CLUE-S model (Mei et al. 2018), the slope, land use, exclusion, urban extent, transportation and hillshade model (KantaKumar et al. 2011), the land transformation model (Tayyebi et al. 2013), the artificial neural network (ANN) (Qiang and Lam 2015) and the support vector machine (Mustafa et al. 2018). Each model has its advantages and disadvantages, and some models can be integrated to produce a more effective simulation. Many studies have adjusted future land-use structures for low-carbon land use optimisation according to the carbon sink/source capacity of each land-use type, with the linear program model being a commonly employed method (Wang et al. 2021). Additionally, some scholars, using spatial land-use change models, have simulated land-use changes toward low-carbon objectives (Yao et al. 2023). While these approaches consider carbon sequestration and carbon emission reduction targets, the simulations often focus on urban-scale control or specific land-use type control or protection, making them relatively simple. Given the significant spatial variations in the carbon sink/source capacity of even the same land-use type, a more detailed and accurate low-carbon land-use optimisation requires adjustments at the land-patch scale. Therefore, a new approach must be developed to integrate high-resolution carbon sink/source monitoring and LUCC simulation models.

To address these research gaps, this study selects Shangyu District, located in Shaoxing City, Zhejiang Province, as an example. The objectives of this research include: (1) calculate regional anthropogenic carbon emissions and terrestrial carbon sink/source; (2) monitor the carbon sink/source in high resolution; (3) analyse land use change and the influence on total carbon; (4) optimise land use under the target of increasing carbon sink and reducing carbon emissions. Innovations include: (1) developing a high-resolution carbon monitoring model by incorporating various data sources and methods; (2) exploring a low-carbon land use model that further promotes optimisation at the land-patch scale.

Methods and data

Study area

Shangyu District is located at 120°38′26″E to 121°06′12″E and 29°43′59″N to 30°15′56″N, northeast of Shaoxing City, in Zhejiang Province, China, one of China’s typical developed regions (Fig. 1). The climate condition here is the East Asian monsoon climate, with good hydrothermal conditions. It has a population of more than 800,000. Although the tertiary industry is developed, the secondary industry is still a pillar industry, contributing more than 50% to the increase in Gross Domestic Product. Shangyu is rich in forest, with a woodland percentage of 41.12%; cropland and water have the second and third largest land areas, with percentages of 21.09% and 18.67%, respectively. Built-up land accounts for 17.33%.

Fig. 1
figure 1

Location of Shangyu county.

Full size image

Data sources

Research data in this study include energy-consumption data, land-use data, night-light remote sensing images, industry data, buildings data, net ecosystem productivity (NEP), carbon flux from inland water and agricultural input indices. The land-use data with detailed land use types is obtained from the third national land survey. Energy consumption data comes from China’s Energy Statistical Yearbook, and for carbon emissions allocated to Shangyu District, the percentage of industry output value or population that accounts for Zhejiang Province is used. The night-light images with a resolution of 130 m are from the Luojia1-01 satellite. Industry enterprise data, including the latitude and longitude, industry sectors, production and operation situations and a vector layer of buildings, are provided by the urban planning department. The NEP uses the method from previous study (Ye and Chuai 2022). Carbon flux capacities for inland water are obtained from the study of Ran et al. (2021). The agricultural input indices and industry products are from Shangyu Statistical Bulletin (Table 1).

Table 1 Data sources.
Full size table

Anthropogenic carbon emissions calculation

Carbon emissions from energy consumption

Carbon emissions generated by energy consumption are calculated by multiplying the carbon emissions coefficient and the quantity of energy consumed. Additionally, the carbon emissions of various sectors were considered in the calculation:

$${C}_{e}=\mathop{\sum }\limits_{i=1}^{n}{E}_{i}\times {\theta }_{i}$$
(1)

where \({C}_{e}\) is the carbon emissions amount for a certain sector, \({E}_{i}\) and \({\theta }_{i}\) are the energy consumption amount and the carbon emissions coefficient for energy type i. The detailed coefficients are shown in Table S1 (SI-1).

Carbon emissions from industrial production processes

This study mainly considered the production of glass, aluminium, cement, steel, lime and the synthesis of ammonia, the calculation is shown in Eq. (2):

$${C}_{{\rm{industry}}-i}={{\rm{Q}}}_{{\rm{industry}}-i}\times {V}_{{\rm{industry}}-i}$$
(2)

where \({C}_{{\rm{industry}}-i}\) represents carbon emissions from industrial product with the type of i, \({Q}_{{\rm{industry}}-i}\) is the amount of industrial product with the type of i and \({V}_{{\rm{industry}}-i}\) indicates the carbon emission coefficient for industrial product with the type of i, which is quoted from the study of Zhao et al. (2015).

Carbon emissions from agricultural production

The main agricultural inputs that can generate carbon emissions have been considered. The carbon emission coefficient method has also been used here:

$${C}_{a}=\mathop{\sum }\limits_{j=1}^{n}{T}_{j}\times {\sigma }_{j}$$
(3)

where Ca is the amount of carbon emissions from agricultural production, Tj is the total amount of energy consumption or agricultural resources used with the carbon source j and \({\sigma }_{j}\) is the carbon emission coefficient for the carbon source j. The carbon emission items in the grain production stage are shown in Table S2 (SI-2).

Terrestrial ecosystem carbon sink/source

The NEP index is used to describe the terrestrial ecosystem’s carbon sink/source capacity. This is calculated by subtracting soil heterotrophic respiration from net primary productivity (NPP). The method developed in the former study (Ye and Chuai 2022) was employed here but with an NPP resolution of 500 m, which is more suitable for use at the county level.

Main explorations

This study comprises two main explorations: the high-resolution carbon monitoring model and a low-carbon land use model. The carbon monitoring model helps generate a net carbon sink/source model, forming the basis for low-carbon land use optimisation. The low-carbon land-use model involves optimising the land-use structure, integrating the net carbon sink/source and traditional spatial constraints as the spatial driving forces for land use change, and optimising land transfer rules according to the spatial driving forces. Different land-use types are then allocated spatially at the land-patch scale. The framework is illustrated in Fig. 2.

Fig. 2
figure 2

Main explorations.

Full size image

Carbon monitoring

The terrestrial ecosystem modelling used grid satellite and field observation data, enabling the production of NEP at a grid scale (Ye and Chuai 2022). Anthropogenic carbon emissions allocation in spatial terms uses land use and auxiliary data. The details are as follows: First, the relationship between carbon emissions from different sectors and land-use types is established; then, each land-use type was assigned its total amount of carbon emissions. Second, multiple auxiliary data are employed to allocate carbon emissions spatially for different land-use types. For rural residential land and cropland, the average value of the entire region described the carbon emission levels for all land patches. For commercial land and urban residential land, we allocate the carbon emission level based on the area and the number of floors for each building. The carbon emissions level is spatially allocated for traffic land by using the night-light data. For industrial land, the geographical locations for main industries are obtained, and information such as industry types and production and operation are also collected. The emission density weights (per output value of carbon emissions) of different industry sectors are calculated using the Shangyu Statistical Yearbook; these weights are assigned to different enterprises based on their categories. By considering geographical locations and the distribution of industrial land, the spatial distribution of carbon emissions could be simulated. Finally, the carbon emission images for different land-use types are merged into one layer in ArcGIS (Table 2).

Table 2 Carbon monitoring in spatial terms.
Full size table

After the simulation of carbon emissions and the modelling of the terrestrial ecosystem sink/source in spatial terms, net carbon could be calculated by subtracting the carbon sink from carbon emissions, which could be illustrated at a grid scale.

Low-carbon land-use structure optimisation

The linear programming model was employed to optimise the land-use structure. This model has two targets: defining constraint conditions and establishing a target function. In this study, minimising the total amount of carbon emissions from different land uses is set as the target (Eq. (4)), and the constraint conditions are set as the area demand for different land-use types in 2030 (Eqs. (5)–(17)). The model was executed using Lingo software, version 20.0:

$${\rm{Min}}\,{\rm{C}}=\,{\rm{Min}}\left(\mathop{\sum }\limits_{i=1}^{n}{E}_{i}\times {X}_{i}\right)$$
(4)

Where Xi is the area of land-use category i, referring to cropland X1, woodland X2, grassland X3, water area X4, urban green land X5, rural residential land X6, urban residential land X7, commercial land X8, industrial land X9, traffic land X10, public facility land X11 and other built-up lands X12. Ei is the net carbon emissions intensity of land-use category i, indicating the per unit area of total carbon emissions minus the terrestrial ecosystem carbon sink.

Then, land-use policies, economic social development and the carbon neutrality requirements are all considered. Year 2020 is set as the initial year and year 2030 as the target year.

The first constraint condition is set according to Shangyu’s total land area:

$$\mathop{\sum }\limits_{i=1}^{n}{X}_{i}=1400$$
(5)

The area of cropland is determined by urban expansion and rural residential land consolidation. We assume that urban land expansion will mainly occupy cropland, as in the previous year. Idle rural residential land will be converted into cropland, and expanded woodland will occupy cropland due to geographic spatial proximity. Finally, the constraints are set as follows:

$$301.21\le {X}_{1}\le 304.81$$
(6)

Forests have the highest carbon sink capacity; therefore, the conservation of woodland must be strengthened, and the decreasing trend must be stopped. The area in 2020 was set as the lowest value, at 575.81 km2. According to the forest development plan, the afforestation task in Shangyu is about 7 km2 from 2020 to 2025, which is 14 km2 in 10 years to 2030 at the same increasing rate; we set it as the highest value, assuming all afforestation was made on newly increased woodland:

$$575.81\le {X}_{2}\le 589.81$$
(7)

Grassland has a relatively stable fluctuation. For ecological conservation, we assume the area will increase until 2030. The increasing rate in recent years is selected to predict its high value in 2030:

$$24.99\le {X}_{3}\le 27.86$$
(8)

Between 2000 and 2020, the water area decreased by 50 km2. We use the decreasing rate between 2000 and 2020 to predict the low-value water area in 2030. However, after 2020, the decreasing trend has slowed; therefore, we use the changing trend after 2020 to predict its high value in 2030:

$$236.81\le {X}_{4}\le 256.5$$
(9)

The urban green land area is very low and needs to be increased. The area in 2020 is set as the low value, and the high value is set according to the plan of the Urban Garden Bureau in Shangyu:

$$1.6\le {X}_{5}\le 2.5$$
(10)

Since 2017, the population of Shangyu has been decreasing. The population directly determines urban and rural residential land requirements. The decrease was noticeable before 2021 but slowed down after 2021. We use the population decreasing trend between 2017 and 2022 as the high decreasing rate and the decreasing rate between the latest years of 2021 and 2022 as the low decreasing rate. In this way, the population in 2030 will reach 63.8 × 104 and 69.17 × 104, respectively. The urbanisation rate has always been increasing, with a value of 67.78% in 2020. We assume that, in 2030, the urbanisation rate will be close to the low value of developed countries, at 75%, and the rural population will have an obvious decrease. Furthermore, we assume that all idle rural residential land will be consolidated. For urban residential land, we also considered the immigrant population. We use the recent land area increasing rate to predict the land in 2030 as the low value and use the changing trend between 2010 and 2020 to predict its high value in 2030:

$$37.56\le {X}_{6}\le 40.71$$
(11)
$$25.88\le {X}_{7}\le 27.88$$
(12)

Commercial land in recent years is still increasing slowly; we set its value according to the scale of the urban population:

$$6.87\le {X}_{8}\le 7.54$$
(13)

Industrial land is the main contributor to carbon emissions; the area should be strictly restricted. According to the 14th Five-Year Plan for Industry, Shangyu will continue to increase the number of industrial enterprises. One way is to expand the industrial land-use scale; the other is to improve industrial land-use efficiency. During future development, intensive land use must be implemented. Here, we assume more intensive industrial land use will be strengthened, factories with low efficiency will be shut down and the low land area will be set as the same in 2020. For the high land-use area, we use half of the increasing rate in recent years to make a prediction for 2030, which is 63.05 km2:

$$58.06\le {X}_{9}\le 63.05$$
(14)

Traffic land is very important for people’s travel. According to the low and high populations in 2030, the traffic land needed in 2030 is set as follows:

$$65.1\le {X}_{10}\le 70.1$$
(15)

Public service land in recent years is still increasing slowly; we set its value according to the scale of the urban population:

$$12.67\le {X}_{11}\le 14.58$$
(16)

Other built-up land mainly includes irrigation facility land and urban square land. The low and high values are set according to the low and high cropland areas in 2030, respectively, which need to increase water facility land to meet the needs of the newly increased cropland.

$$15.36\le {X}_{12}\le 15.51$$
(17)

The area is limited for unused land, and we assume it will all have been used by 2030, with no unused land left.

Low-carbon land-use simulation in spatial

After the quantitative land-use structure is determined, spatial land-use optimisation will be made, including two steps of optimising land probability-of-occurrence by considering carbon sink/source capacity and allocating future land in spatial.

Optimising land probability-of-occurrence

We improve the land probability-of-occurrence by integrating the influence from the traditional spatial factors and the carbon sink/source influence. Traditional driving factors include natural elements (including elevation, slope and slope direction) and socio-economic elements (distance from highways, urban trunk roads and railways). Then, carbon as a new spatial constraint is adopted. The rule is that land transfer probability decreases with the carbon sink capacity increasing, while an increasing trend with the carbon emission capacity increasing. The ANN algorithm is used to estimate the land probability-of-occurrence. The ANN includes three layers: the input layer, the hidden layer and the output layer. In the input layer, each neuron corresponds to an input variable, such as traditional spatial constraints, carbon sink/source capacity in spatial and socio-economic variables. The equation can be shown as:

$${\rm{X}}={[{x}_{1},{x}_{2},\mathrm{.....}.{x}_{n}]}^{T}$$
(18)

Where Xi is the neuron i in the input layer and n is the number of the neurons. T represents transposition.

In the hidden layer, neuron j received the signal from all the input neurons on grid cell P at time t is checked by Eq. (19):

$${{\rm{net}}}_{j}(p,t)=\mathop{\sum}\limits _{i}{w}_{i,j}\times {x}_{i}(p,t)$$
(19)

Where netj(p,t) is neuron j l received signal. wi,j is an adaptive weight between the input layer and the hidden layer. xi(p,t) is the ith variable associated with the input neuron i on grid cell P at training time t.

The connection between the hidden layer and the output layer is determined by an activation function, which is finished by the sigmoid, as shown below:

$${\rm{sigmoid}}({{\rm{net}}}_{j}(p,t))=\frac{1}{1+{e}^{-{{\rm{net}}}_{j}(p,t)}}$$
(20)

The value of the jth neuron will generate a value that represents the probability-of-occurrence for the jth land use type at the grid cell. The probability-of-occurrence of land use type k on grid cell P at training time t is written as P (p, k, t), which is checked by Eq. (21):

$$p(p,k,t)=\mathop{\sum}\limits_{j}{w}_{j,k}\times {\rm{sigmoid}}({{\rm{net}}}_{j}(p,t))=\mathop{\sum}\limits_{j}{w}_{j,k}\times \frac{1}{1+{e}^{-{{\rm{net}}}_{j}(p,t)}}$$
(21)

where wj,k is an adaptive weight between and the output layer and the hidden layer; then, the ANN model is established and used to estimate the probability-of-occurrence for each land use type in a specific grid cell. In general, the spatial resolution of the data is 90 m*90 m in the description, involving a total of 172,839 grids. To ensure the accuracy of the model, 10% of the data were used for training and 90% for testing.

Allocating future land in spatial

Based on the optimised land probability-of-occurrence, land use changes are eventually predicted using FLUS with an adaptive inertia mechanism to simulate future land use patterns. The detailed land allocation algorithm process is shown below:

$$T{P}_{p,k}^{t}={P}_{p,k}\ast {\omega }_{p,k}^{t}\ast \,\mathrm{intertia}_{k}^{t}\ast (1-S{C}_{c\to k})$$
(22)

where \(T{P}_{p,k}^{t}\) is the combined probability of grid cell P to convert from the original land use type to the target type k at the time point of t. Pp,k is the probability-of-occurrence of land use type k on grid cell p; Intertiap,k is the inertial coefficient of land use type k at time t.\(S{C}_{c\to k}\) is the land transfer cost coefficient for land use type S and C. \({\omega }_{p,k}^{t}\) indicates the neighbourhood effect of land use type k on grid cell P at iteration time t:

$${\omega }_{p,k}^{t}=\frac{\sum N\ast N{\rm{con}}({C}_{p}^{t-1}=k)}{N\ast N-1}\,*\, {w}_{k}$$
(23)

where \(\sum N\ast N{\rm{con}}({C}_{p}^{t-1}=k)\) indicates the total number of grid pixels occupied by the land use type k at the last iteration time t − 1 within the N × N window. \({w}_{k}\) is the variable weight among the different land use types:

$$\mathrm{intertia}_{k}^{t}=\,\mathrm{intertia}_{k}^{t-1}\ast \left|\frac{\max ({D}_{k}^{t-1},{D}_{k}^{t-2})}{\min ({D}_{k}^{t-1},{D}_{k}^{t-2})}\right|$$
(24)

where \({D}_{k}^{t-1}\) is the difference between the macro demand and the allocated amount of land use type k until iteration time t − 1.The overall process is shown in Fig. 3.

Fig. 3
figure 3

Flowchart of low-carbon land-use simulation.

Full size image

Results

Carbon monitoring

Carbon changes in time series

Between 2000 and 2020, the total amount of anthropogenic carbon emissions showed a rapidly increasing trend, with the amount rising from 51 × 104 to 265 × 104 t. Carbon emissions from industry contributed the most, accounting for around 80% of the total carbon emissions for all the years. Besides industry, carbon emissions from traffic were relatively high and showed a quick increase. The amounts were also increasing for carbon emissions from cropland, commercial land, urban residents and rural residents, and the increase was especially obvious for urban residents (Fig. 4a). For the terrestrial ecosystem, it acted as a carbon sink overall. As for the changing trend among different years, it showed obvious fluctuation but also exhibited an increasing trend; the highest carbon sink occurred in the year 2019, at 13 × 104 t (Fig. 4b). On the whole, the average annual carbon sink could offset 3% of regional anthropogenic carbon emissions.

Fig. 4: Carbon sources and sinks accounting and changes.
figure 4

Changes in anthropogenic carbon emissions (a) and terrestrial ecosystem carbon sink (b).

Full size image

Carbon monitoring in spatial terms

In 2019, the net carbon sink/source value (carbon emissions minus carbon sink) was checked at the grid scale. Figure 5a shows that carbon emissions with high intensity were well determined by the distribution of land-use types. The highest emissions occurred intensively in the northeast, in Shangyu’s industrial park, with some dispersed distribution around the urban region; the average emission intensity was 3.7 × 104 t/km2. The urban area also showed high carbon emission levels, but the intensities were much lower than in industrial land. For example, urban residential and commercial land had similar intensity values, around 0.4 × 104 t/km2. For rural residential land, the emission intensity was about 0.12 × 104 t/km2, and the lowest values occurred in cropland, at only 0.02 × 104 t/km2. For other green land, there were nearly no anthropogenic carbon emissions. By integrating the terrestrial carbon sink/source, Fig. 5b shows that the net carbon sink/source values spatially had an obvious regional pattern. Areas with a net carbon sink are concentrated in the south, accounting for about 33% of the region. The net carbon emissions are mainly located in the north, with some patches in the south.

Fig. 5: The spatial distribution of carbon sources and sinks.
figure 5

Spatial distribution of carbon emission intensity (a) and the net carbon sink/source values (b) (t/grid).

Full size image

Land-use change and its influence on carbon

Land-use change

From 2000 to 2020, cropland, woodland, grassland and water areas all decreased, with decreasing areas of 33.29, 15.84, 2.13 and 49.78 km2, respectively. Meanwhile, the area of all built-up land increased. A total of 14.5% of the total land area saw its land-use type change, with the changing area amounting to 193 km2. The transfer-out of cropland shows the highest area, at 96.44 km2, mainly due to built-up land expansion; rural residential land occupied 32.21 km2, and other built-up land (mainly including industrial land and traffic land) occupied 33.7 km2. The transfers to water area and urban land are also noteworthy, at 16.96 and 13.44 km2, respectively. Spatially, the transfer to rural residential land has a wide distribution across Shangyu, at the periphery of the initially residential land-located region. The transfer to other built-up land is mainly located in the north. The transfer to urban land is mainly situated at the edge of the original urban region. As for the water area, the transfer has a relatively wide distribution but is more intensive in the southwest corner. The occupation of the water area also shows high values, with a total of 73.54 km2; of this, 60.98 km2 is the transfer to cropland, intensively located in the north. The occupation of woodland is much lower, at 16.06 km2, and the distribution is dispersed. For other land transfer types, the areas are usually low (Fig. 6).

Fig. 6: Land use transfers.
figure 6

Spatial distribution of mainland transfer types (a) and the land transfer matrix (b) from 2000 to 2020.

Full size image

Carbon changes caused by land-use change

Here, we conducted a comprehensive analysis by considering both terrestrial carbon sinks and anthropogenic carbon emissions to examine the carbon variations caused by land-use transfer (Table 3). Generally, land-use transfer causes an annual increase of 77.72 × 104 t in carbon emissions, of which 76.28% is from the transfer-out of cropland. Particularly, the transfer from cropland to other built-up land has increased carbon emissions by 50.08 × 104 t/year. The transfer-out of water area and woodland contributed the second and third highest increases in carbon emissions, at 10.34 × 104 and 7.39 × 104 t/year, respectively, and the main contributors are also the transfers to other built-up land. The transfer-out of urban land and other built-up land contributed little to the decrease in carbon emissions, whereas the transfer-out of rural built-up land increased carbon emissions by 0.7 × 104 t/year.

Table 3 Annual carbon variations caused by land-use transfer between 2000 and 2020 (104 t/year).
Full size table

Low-carbon land-use simulation and the influence on carbon

Land-use structure optimisation and carbon reduction

Through land-use structure optimisation, carbon emissions will increase only by 7154 t/year, compared with 2020 under the historical carbon source/sink intensity. This is much lower than the land-use change induced carbon emissions between 2000 and 2020, as illustrated in Table 3. The optimised land-use structure has significantly increased the area of ecological land, including cropland, woodland, grassland and urban green land, all of which will reduce carbon emissions except for cropland. This is because agricultural activities will bring about anthropogenic carbon emissions. The area increase of woodland will contribute the most to the carbon sink, at 2058 t/year. Rural residential land consolidation will lead to a decrease in its total land area, contributing to a carbon emissions reduction of 18,515 t/year, the highest compared with other land-use area changes. For increasing carbon emissions, high amounts come from built-up land, especially urban residential land and traffic land. Although their expanded land area is low, the increased carbon emissions are noticeable due to their high carbon emissions intensity (Table 4).

Table 4 Optimised land-use structure in 2030 and the resulting carbon changes compared with the land-use structure in 2020.
Full size table

Land-use simulation in spatial terms and further carbon sink increasing

Under the land-use structure in Table 4, the spatial distribution of land use is simulated for 2030 under two scenarios. One scenario only considers traditional constraints in spatial terms, and the other scenario integrates traditional spatial constraints and the carbon sink/source capacity at the land patches scale. Figure 7a shows the simulated low-carbon land-use pattern under the latter scenario. The consolidation of rural residential land into woodland and cropland shows the highest transferred land area. Moreover, the transfer between cropland and woodland and from water area to cropland also shows a high transfer area. Figure 7b shows the mainland transfer adjustments compared with the land-use pattern under the first scenario. Because the land-use structure is the same for the two scenarios, the adjustment is made only at the patches scale; the total area of individual land-use types will not change. This shows that the adjustment was primarily made between cropland and woodland, cropland and water, cropland and rural residential land, rural residential land and traffic land. Generally, further low-carbon land optimisation at the patches scale increased the carbon sink by 129.59 t C compared with step two, which only considered traditional spatial constraints without carbon sink/source capacity. For the transfer out optimization, woodland can decrease the most carbon sink loss, at 166.78 t C. Conversely, for the transfer in optimization, cropland will bring the most significant carbon sink increase, at 100.59 t C.

Fig. 7: Low-carbon land use simulation.
figure 7

Spatial distribution of low-carbon land use (a) and the land transfer compared with the simulation without optimisation at the land patches scale (b) in 2030. The influence on the carbon sink is shown in the histogram.

Full size image

Discussion

Research meaning

Land-use change can significantly affect carbon cycle (Chuai et al. 2015, 2018). However, due to the coarse resolution of carbon source/sink assessments, low-carbon land use adjustments typically remain at the level of land-use type adjustments. Although spatial simulations are conducted, the carbon capacity at the land-patch scale has not yet been thoroughly explored (Yao et al. 2023). These limitations result in low-carbon land spatial planning and urban planning lacking practical guidance for land-patch adjustments, keeping low-carbon land use more theoretical than practically applicable. There is still a significant gap to bridge for practical implementation. Addressing these issues, our study develops a high-resolution carbon monitoring model and a new low-carbon land-use model, providing support to advance low-carbon land use into practical applications. This has significant implications for guiding the implementation of land spatial planning and urban planning.

Carbon calculation

The simulation of carbon emissions in spatial terms is usually conducted at the country, province or city scale (Cai et al. 2018, 2019), but there is little research at the county scale, largely due to the fact that basic data accessibility is less convenient compared with studies at the country and province scales. In this study, both the ‘top-down’ and ‘bottom-up’ methods are used to comprehensively calculate anthropogenic carbon emissions. The integrating use of these two methods is widely adopted in other studies (Wang et al. 2013; Yang et al. 2021). In using the ‘top-down’ method, we allocate the total carbon emissions amount according to sector output values or the population of Shangyu, accounting for its parent city. Sectors include agriculture, urban residents and rural residents. Although this method may introduce some bias, it is limited because the same sector in nearby regions, especially within the same city, usually has similar carbon emission levels. For the ‘bottom-up’ method, focusing on sectors such as industry and traffic, its advantages lie in the direct data accessibility for Shangyu, which can result in higher accuracy. Generally, the carbon sink in Shangyu can offset only a small percentage of carbon emissions, a finding consistent with most regions in China (Wang et al. 2021; Piao et al. 2009), indicating that the pressure to reduce carbon emissions and increase carbon sinks in Shangyu is high. Although the amount of carbon sinks is low compared to carbon emissions, it shows an overall increasing trend. The carbon sink capacity is well determined by climate change (Chuai et al. 2018; He et al. 2017) and also benefits from strengthened ecological conservation policies, particularly in recent years.

The high volume of anthropogenic carbon emissions primarily comes from the industrial sector, and it is still on a rapidly increasing trend. This is because the main industries in Shangyu are medicine, textiles, chemicals and printing and dyeing factories, which belong to energy-intensive industries that generate a considerable amount of carbon emissions. Thus, optimising industrial structures and developing low-carbon buildings are critical for carbon emissions reduction; these phenomena are quite common in China (Chang et al. 2023). The local government should encourage enterprises with high efficiency but low emission levels; industrial land use should be more intensive, and enterprises with low efficiency but high-emission levels should be shut down, with their land-use rights revoked. In addition, carbon emissions from other sectors are low but are all increasing, especially in the transport sector. This is determined by both population growth and increasing energy and resource intensity. Key measures for carbon emission reduction include energy structure adjustment and the promotion of clean energy use (Zhou et al. 2023). Electric vehicles, public transit and photovoltaics can be further encouraged. For cropland, the coordination between the increasing agricultural economy and carbon emissions needs further study in terms of planting structure optimisation and improvements in agricultural management measures (Han et al. 2021, 2024).

Carbon monitoring in spatial

By integrating various data, this study developed a method to check carbon emissions in high resolutions. The basic framework includes two steps: the first is to assign carbon emissions to different land-use types. For a county-scale simulation, this study has detailed land-use classifications, which can help to finish the first total carbon emissions amount allocation for different land-use types. This can eliminate the bias that comes without the distinction of carbon emissions on different land-use types, such as the use of night light for the whole region’s carbon emissions simulation in spatial terms (Gao et al. 2023; Lu and Liu 2014; Shi et al. 2016). This method has less bias than the dominant carbon emissions from industry, which usually do not have high night-light values. It is also better than the method of inverse carbon emissions by carbon satellites, which use the original carbon concentration data and usually have low resolution; accuracy is also affected by weather conditions (Jin et al. 2023; Wang et al. 2022; Yang et al. 2017).

The second step in the simulation is to furtherly allocate carbon emissions for the same land-use types. For commercial land and urban residential, vectorised building height and land area data can improve the carbon emissions resolution from the grid to the vector, which is the biggest improvement compared with previous studies (Cai et al. 2018, 2019; Luo et al. 2023). For traffic land, although we lack direct traffic flow data on each road, the 130-m night-light data is used. This resolution is much better than commonly used data, such as the DMSP and the VIIRS, which have resolutions of 1 km and 500 m, respectively (Andries et al. 2023; Zhang et al. 2021). Because most vehicles turn their lights on during night travel, this method is feasible, as confirmed by other studies (Andreano et al. 2021; Kabanda 2020). The industry is the dominant carbon emitter. We obtained the Point of Interest information for each enterprise, which also includes information on industry type and production and operations. This can help assign carbon emissions intensity and the final amount. This method can provide guidance for industry carbon emissions checks without survey data, which would require a high cost in terms of labour and economics. For agricultural and rural residential carbon emissions, the study area is not large, and the difference in agricultural land is not obvious; therefore, we did not make further distinctions. For more detailed checking, field investigations are needed in future studies. For terrestrial ecosystem carbon, not only land with vegetation coverage is considered but also the carbon flux from inland water. The NEP simulation has also improved, both from the aspect of improving models based on field observations and by enhancing the NPP resolution compared with the former study (Ye and Chuai 2022). For carbon flux from water, the original data are from Ran et al. (2021), which is the most authoritative data covering China’s main streams, rivers, lakes and reservoirs.

Land use change

The main land-use transfer type is from cropland to built-up land, which is the dominant transfer type in China (Huang et al. 2021; Li et al. 2021), driven by urbanisation and economic development (Cui et al. 2022; Guo et al. 2022). Some special phenomena have occurred in Shangyu: although the urbanisation rate is increasing, rural residential land has expanded the most. This land expansion took place between 2000 and 2020 and represents a waste of land use as well as a potential direction for land-use optimisation, since the rural population is decreasing, and there is vast idle homestead land. In addition, although the registered population in urban areas began to decrease in 2017, urban land is still increasing. Despite a portion of the area having a low population, there is space for future urban land control. One reason is that the urbanisation rate in Shangyu is already high; the second reason is that fertility rates in China have been low in recent years (Li et al. 2020; Lu et al. 2023). Another obvious type of land transfer is from water areas to cropland. The north of Shangyu borders Hangzhou Bay, which belongs to the East Sea; through the reclamation of tidal flats, cropland can be increased. Due to differing carbon sink/source capacities among land-use types, even small changes in land area may result in high amounts of carbon emissions, especially in industrial land. For ecological land, especially woodland, a high carbon sink capacity is usually due to high biomass (Guo et al. 2013; Zhang et al. 2013). However, human disturbance, combined with climate change, may convert carbon sinks to carbon sources, not only in forests but also in grasslands and croplands (Zhang et al. 2015). Therefore, appropriate land management needs to be further explored.

Low carbon land use optimisation

The core objective of low land-use optimisation is to increase the area of land with carbon sink capacity, especially for land and areas with high carbon sink capacity, and to control areas with high carbon emissions intensity. Compared with most of the former studies, which only considered the land-use type scale (Liu et al. 2017; Cao et al. 2024), the low-carbon land use in our study has several advantages. The first is that more detailed land-use types are considered, especially for urban regions. Since city areas have diverse land-use types, each with significantly different carbon emission intensities, a detailed urban land-use distinction is much more meaningful for guiding low-carbon city development. Second, this study incorporates detailed carbon sink/source capacity in spatial terms, as constraints for land-use change probability. By integrating these with other traditional constraints, such as the economy, population, terrain and infrastructure (Qian et al. 2020; Vinayak et al. 2021), it can truly facilitate low-carbon land use at the patch scale. Third, machine learning is adopted in the land simulation model, which can help optimise the land transfer rules (Aburas et al. 2019; Mehmood et al. 2023). According to the application in our study, the rapid increase in land use change-caused carbon emissions trend has been well controlled, and the further land patches scale optimisation further increased carbon sink by 129.59 t C/year, indicating our model is more effective compared with the methods used in previous studies (Zhang and Zhang 2023; Yao et al. 2023). What is more, this study did not consider the anthropogenic carbon emissions reduction caused by land patch scales optimisation, since we think not only the carbon emissions level but also the efficiency needs to be scientifically considered to determine land transfer probability. If incorporated into the effect of carbon emissions, the carbon reduction effect from our low carbon land use model will be more significant. For the application, the model developed in our study is useful to serve for low carbon land use planning or city planning. It is much helpful since the mentioned planning all need to adjust land use at a patches scale, while, due to the resolution limitation, former studies and technology can only help at land-type scale, which more makes low carbon land use at theoretical level, while, our method will promote it to practice.

Conclusion and recommendations

Although attempts have been made to improve the accuracy of carbon emissions and carbon sinks in spatial terms, there is still a gap that needs to be bridged to effectively guide the implementation of low carbon land use. Our study mainly has the following contributions. First, it was conducted at the county scale, with carbon data accessibility not being as easy as in city or province scales. The method for both carbon sink and anthropogenic carbon emissions calculation can provide some guidance for other county-scale studies. Second, a high-resolution carbon monitoring model has been developed; the accuracy and resolution of carbon monitoring have been greatly improved, and it can effectively support low carbon land use at the land patch scale. Third, the traditional low carbon land use model can only make the carbon capacity of a certain land use type as a constraint, while our developed model promoted the optimisation to land patch scale, which can guide better low carbon land use and promote the practical application of low carbon land use planning or city planning. Key findings include:

Between 2000 and 2020,

  1. (1)

    Both carbon emissions and carbon sinks are on an increasing trend in Shangyu. There is a substantial gap in reaching the regional carbon neutrality goal, as the carbon sinks can only offset a small percentage of carbon emissions.

  2. (2)

    Spatially, the north shows net carbon emissions, with high-emission values more frequently occurring in the northeast. The southern region acts more as a carbon sink, especially in the southeast.

  3. (3)

    Land-use type transfer is obvious between 2000 and 2020, with the conversion of cropland to built-up land being the dominant transfer type and the main driver of increasing carbon emissions due to its high carbon emission intensity.

  4. (4)

    Through land-use structure optimisation, the increasing trend in carbon emissions will be well controlled between 2020 to 2030, and further land-scale optimisation will directly increase carbon sink significantly.

  5. (5)

    There is a high potential to reduce carbon emissions through land use control, with optimisation aspects, including but not limited to land use structure optimisation and adjustment of land patches. To better apply low-carbon land use, the accurate check of carbon sink/source in spatial terms is the fundamental basis.

Recommendations are proposed as the following. To keep improving the carbon monitoring accuracy and resolution, local governments should strengthen the monitoring and collection of basic data, such as traffic flow data, energy consumption, industry products and other enterprise data, as well as detailed agricultural inputs data for different crops in geographical locations, etc. In the interior urban region, an investigation for urban vegetation is well needed, including the spatial distribution, vegetation types and growth status; this is critical for a more detailed urban ecosystem carbon sink/source simulation. Secondly, during land use planning and city planning, the low carbon target should be a mandatory condition throughout the life cycle of the planning, requiring continuous evaluation of land use and its carbon impact. Thirdly, the government should implement land use policies where practices promoting carbon sequestration are encouraged, such as protecting ecological land. Conversely, land use with high carbon emissions but low efficiency should be subject to penalties or revocation.

Limitations and future directions

The innovations in both high-resolution carbon monitoring and low-carbon land optimisation are novel endeavours. However, certain limitations exist. Firstly, the ‘top-down’ method in anthropogenic carbon emissions may not be as accurate as the ‘bottom-up’ approach, and results for certain sectors may exhibit some bias (Hutchins et al. 2017). Nevertheless, the error is limited because the same sector in nearby regions, especially within the same city, usually has similar carbon emission levels. Secondly, carbon emissions simulation in spatial terms from transportation and cropland is still fundamental compared to monitoring, which can achieve vector resolution. Thirdly, low-carbon land use optimisation only considers the absolute carbon sink/source capacity as a constraint to influence land transfer rules, which may not align well with the actual land demand. Potential trade-offs between land use objectives such as carbon, economic benefit, biodiversity, etc., are not comprehensively considered.

Several improvements can be suggested for future studies. Firstly, the accuracy can be further enhanced; for instance, the consideration of the residential vacancy rate is crucial when conducting urban residential carbon emission simulations. Through field observations, a more detailed recognition of carbon emissions from croplands can be achieved (Zhong et al. 2023). Obtaining traffic flow data and enterprise carbon accounting data can significantly contribute to improving carbon monitoring accuracy and resolution. Secondly, in inter-urban regions, more detailed land-use types could be further subdivided with data support, such as differentiating land use within various industrial sectors or considering traffic flow on different types of roads. Thirdly, in the future, land use associated with carbon emissions may need to consider factors like land-use efficiency, suitability and modification of land transfer rules based on more complex constraints. Fourthly, multi-objective land use should be considered, and more intricate land use optimisation, not solely limited to carbon capacity, should be explored. The trade-offs between carbon, economic benefits, biodiversity, etc., need to be further investigated.