Introduction

As global climate change becomes increasingly severe, achieving carbon neutrality has become a critical environmental goal for countries worldwide. The Paris Agreement established a long-term temperature goal to keep the global average temperature rise within 2 °C, aiming to limit warming to 1.5 °C by the end of the 21 st century1. Compared to 1850–1900, the global surface temperature for the ten-year average from 2011 to 2020 has risen by approximately 1.09 °C, and future global warming is projected to reach or exceed 1.5°C2. For a rapidly developing country like China, effectively controlling and reducing carbon emissions while ensuring sustained economic growth is an urgent challenge. The Chinese government has committed to reaching peak carbon emissions by 2030 and aims to achieve carbon neutrality by 2060. Against the backdrop of these ambitious targets, scientifically accurate assessments and predictions of provincial carbon emission trajectories and their potential for achieving carbon neutrality are crucial for formulating effective policies at both the local and national levels3[,4.

In recent years, the international community, along with various countries and regions, has been actively formulating policies and incentives aimed at reducing carbon emissions and promoting low-carbon development to accelerate the achievement of carbon neutrality goals5,6,7. However, these policies and measures often focus on macro-level economic development models, overlooking regional differences in economic development levels, industrial structures, and energy use8. This research approach limits our comprehensive understanding of carbon emission characteristics and reduction potential across different regions. Significant differences in economic development levels, industrial structures, energy consumption patterns, and policy environments among China’s provinces directly impact their carbon emission levels and reduction potential9. Therefore, a carbon forecasting model that integrates multiple driving factors and scenarios can provide policymakers with more precise and tailored data support, aiding in the development of more effective carbon reduction strategies. Therefore, a carbon forecasting model that integrates multiple driving factors and development scenarios can provide more precise and region-specific data support for policymakers, helping them formulate more effective carbon reduction strategies. Due to its rapid industrialization and urbanization, the Yangtze River Middle Reaches Urban Agglomeration has emerged as a focal point in national low-carbon transition policies10. National policy documents such as the Yangtze River Economic Belt Development Plan (2016) and the 14th Five-Year Plan (2021) explicitly emphasize ecological conservation and green transformation in this region, underscoring its strategic importance in achieving China’s carbon peaking and neutrality goals.

As understanding of carbon emission issues deepens, an increasing number of studies have begun to consider more complex socio-economic factors11. For example12, introduced the impact of international trade on carbon emissions, constructing a carbon emission forecasting model that incorporates trade factors to reveal new characteristics of carbon emissions in the context of globalization. Furthermore, the role of environmental policies and technological innovation in the emission reduction process has also garnered extensive attention13. Empirical analysis indicates that the development of clean energy technologies and policy support are key drivers in achieving carbon emission reductions. By evaluating renewable energy policies across Chinese provinces, the study identifies the synergistic effect of technological advancement and policy innovation in promoting a low-carbon transition. Additionally, social and cultural factors, along with changes in residents’ lifestyles, also have a significant impact on carbon emissions. The study by 14explores the impact of changing consumption patterns on the carbon footprint, highlighting the potential of raising public environmental awareness and promoting green lifestyles in achieving carbon neutrality goals. However, these studies often focus on single factors or specific national contexts, and there is a lack of integrated, regional-level forecasting frameworks that systematically incorporate diverse socio-economic and environmental drivers, particularly for inland urban agglomerations like the Yangtze River Middle Reaches.

Against the backdrop of extensive research, carbon emission forecasting serves as a critical foundation for studying carbon reduction and achieving carbon neutrality. The academic community has conducted numerous empirical analyses, proposing various carbon emission forecasting methods for different regions and industries based on diverse theories. To incorporate a wide range of socio-economic factors into carbon emission forecasting15, first introduced the IPAT model, representing the impact on carbon emissions as the product of affluence, technology level, and population, thus establishing a logical model linking human activities with carbon emissions; Building on the IPAT model16, proposed the STIRPAT model, which allows for the decomposition of population, affluence, and technology factors. This model enables the flexible inclusion of various factors in analyzing environmental impacts, making it highly adaptable for incorporating different variables based on actual conditions. However, when new characteristics are added, it may lead to overlapping explanations within the model17,18 extended the STIRPAT model by incorporating factors such as industrial structure, urbanization, and climate variability to examine the drivers of carbon dioxide emissions in prefecture-level cities across China. This approach integrates natural factors like climate into the assessment framework but does not address the region-specific selection of key factors. In recent years, various other methods have also been used to model and predict carbon emissions, such as employing Markov models to forecast the energy structure within different sectors of a region; Using the LEAP model to forecast energy demand across various sectors within a region19; and combining the extended STIRPAT model with an optimized Extreme Learning Machine (ELM) network, entropy method, and Zero-Sum Gain Data Envelopment Analysis (ZSG-DEA) model to explore carbon emission drivers, long-term reduction pathways, and carbon quota allocation; An ACE inventory was constructed, using kernel density estimation and conditional probability density estimation to explore the dynamic evolution patterns of ACE. Long Short-Term Memory (LSTM) networks were employed to train and forecast ACE under various scenarios2021 utilized the Tapio decoupling model to study the carbon emission peak states of 30 provinces in China. These methods bring innovation to carbon emission forecasting models from various perspectives, with each focusing on different aspects and based on assumptions of different development scenarios. Although significant progress has been made in carbon emission forecasting research, there remain gaps and limitations in studies on carbon peaking and carbon neutrality. Most current research focuses on single factors or specific regions, lacking systematic analysis involving multiple driving factors and complex scenarios. Moreover, while the STIRPAT model and other forecasting methods offer flexibility and applicability in carbon emission analysis, optimizing these models by integrating more real-world contexts to improve forecasting accuracy remains a challenge. Therefore, there is an urgent need to explore more comprehensive carbon emission forecasting models based on existing research to fill these knowledge gaps.

To address these research gaps, this study applies an extended STIRPAT model coupled with Partial Least Squares (PLS) regression, enabling the integration of a comprehensive set of socioeconomic, industrial, and environmental drivers for scenario-based carbon emission analysis. Unlike the conventional IPAT model, this approach not only considers the classic drivers of population, affluence, and technology, but also incorporates variables such as industrial structure, urbanization rate, energy structure, and carbon intensity. By leveraging the strengths of PLS regression, the model effectively handles multicollinearity among predictors and quantifies the relative importance of each driver. Focusing on the three provinces of the Yangtze River Middle Reaches in China, the study systematically investigates the driving forces of carbon emission changes from 2001 to 2021, and projects future emission trajectories from 2022 to 2080 under multiple development scenarios.The research framework is shown in Figure 1.

Fig. 1
figure 1

Research framework.

Full size image

Data and methods

Study area

The Yangtze River Middle Reaches region, comprising Hubei, Hunan, and Jiangxi provinces, occupies a strategic location in central and southern China (108°21′–118°28′ E, 20°09′–33°20′ N), covering approximately 564,700 km² and encompassing 36 prefecture-level cities, 2 autonomous prefectures, 1 forest district, and 325 county-level administrative units (Fig. 2). The area features complex topography dominated by hills and mountains, interspersed with fertile plains, and is endowed with abundant water and mineral resources. Compared with the upper Yangtze plateaus and the lower Yangtze plains, this region’s diverse terrain and resource base present unique opportunities and challenges for development.

With a population density roughly twice the national average, the region is highly urbanized and economically dynamic, serving as a key component of the Yangtze River Economic Belt and the Central Region Rise strategy. The urban system is anchored by Wuhan and radiates outward, supporting balanced and coordinated economic growth. The regional economy remains heavily reliant on heavy industry, especially in cities such as Nanchang, Jiujiang, Wuhan, and Changsha. Despite national efforts to promote high-quality development, structural legacies have made industrial transformation and short-term carbon emission reduction challenging, highlighting the significant influence of urbanization, industrial structure, and energy intensity on regional carbon emissions.

Selecting the Yangtze River Middle Reaches as the study area reflects both its critical role in China’s economic development and its representativeness for carbon peaking and neutrality pathways. The three provinces, while geographically adjacent and sharing similar development trajectories, exhibit distinct differences in urbanization, energy structure, industrial composition, and emission intensity. Comparative analysis of these provinces enables a nuanced assessment of key drivers of carbon emissions and supports the formulation of targeted low-carbon development strategies.

Fig. 2
figure 2

Study area.

Full size image

The map was created by the authors using ArcGIS 10.8 software (Esri Inc., https://www.esri.com/en-us/arcgis/about-arcgis/overview). The administrative boundaries are based on public data from the National Geomatics Center of China.

Data source

The datasets employed in this study cover prefecture-level cities within the Yangtze River Economic Belt from 2001 to 2021. Carbon emission data (C) were calculated using provincial energy balance sheets from the China Energy Statistical Yearbook (National Bureau of Statistics, 2024 edition) and carbon content coefficients specified in the IPCC Guidelines for National Greenhouse Gas Inventories. Land use data, including the 1-km resolution National Land Use Type Distribution Grid (2001–2023), were obtained from the Geographic Remote Sensing Ecology Network (http://www.gisrs.cn) and cross-validated with the Land Use Change Survey (Ministry of Natural Resources, 2024) and RESDC (http://www.resdc.cn). Socioeconomic variables—population (TP), urbanization rate (UR), GDP per capita (RG), energy structure (ES), industrial structure (IS), energy intensity (EI), carbon intensity ((CI)), and tertiary industry proportion (PT)—were compiled from provincial statistical yearbooks, the China Statistical Yearbook (2025), and the CEADs Database. Carbon sinks were estimated by integrating land use data with IPCC default sequestration coefficients (Table 1). All datasets underwent temporal interpolation and spatial consistency checks to address missing values.

Table 1 Data type, name, and source.
Full size table

Ecological carbon sequestration and carbon emission accounting for the three provinces in the Yangtze river middle reaches

The carbon sequestration calculation method proposed by the Intergovernmental Panel on Climate Change (IPCC) is currently the most widely applied. 22Following the IPCC methodology, we first obtain the area of each land-use type within each province. Then, using specific carbon sequestration coefficients for different vegetation types, we calculate the ecological carbon sequestration of each type and sum them to obtain the total ecological carbon sequestration for each province.

First, the area (A) of each land-use type is obtained from the shapefile (shp) data.

$$\:{\text{A}}_{\text{j}}\:=\:{\sum\:}_{\text{i}=\:1}^{\text{n}}{\text{A}}_{\text{i}\text{j}}$$

Aj: the area of land type j;

Aij: the area of the ith polygon belonging to land type j;

n: the total number of polygons.

Using the carbon sequestration coefficients provided by the IPCC, the total ecological carbon sequestration for each province is then calculated.

$$\:{E}_{C}\:=\:\sum\:_{j\:=\:1}^{\:3}{S}_{j}\:\times\:{C}_{j}$$

Ec: ecological carbon sequestration;

j: land-use type, such as wetlands, forests, or grasslands;

Sj: land-use area of type j;

Cj: carbon absorption coefficient for each land-use type j.

The above formula is used to calculate the area of each land-use type multiplied by its corresponding carbon absorption coefficient (2.36, 0.07, and 1.9, respectively), and the results are then summed to obtain the total ecological carbon sequestration.

Due to the lack of officially published provincial CO₂ emission data in China, we estimated carbon emissions for the study area based on fossil energy consumption and carbon content following the IPCC Guidelines for National Greenhouse Gas Inventories23,24,25,This method is widely recognized internationally and has scientific validity. Specifically, carbon emissions from energy consumption are calculated by multiplying energy consumption by the conversion coefficient to standard coal and the carbon emission factor, and then by the carbon content rate.

$$\:{E}_{energy}=\frac{44}{12}\sum\:_{i=\:1}^{\:8}{E}_{i}\:\times\:{F}_{i}$$

Eenergy: CO₂ emissions from different fuel types in each province;

i: type of energy, including the main fuels used in production and daily life, such as raw coal, coke, crude oil, diesel, fuel oil, gasoline, kerosene, and natural gas (eight types in total);

Ei: consumption of the ith energy type;

Fi: CO₂ emission factor for each energy type i.

PLS regression analysis

Before formally solving the model, we first determine the influencing factors to ensure that the model accurately reflects the annual total carbon emissions of each province. Appropriate variables are selected to improve model performance26. Based on previous studies, we identified eight variables from Table 2 that affect carbon emissions through social development, economic levels, and direct or indirect factors: total population, GDP per capita, energy intensity, energy structure, industrial structure, urbanization rate, carbon emission intensity, and the proportion of the tertiary industry.

Table 2 Driving factors in the STIRPAT Model.
Full size table

Before constructing the model, we need to determine the appropriateness of variable selection. To do this, we conducted a Pearson correlation analysis, as shown in Fig. 3, to assess the linear relationships between variables. The triangular section displays the Pearson correlations among various indicators. Land-use change carbon emissions (SC) are mainly influenced by the two indicators CI and TP, while industrial carbon emissions (IC) are significantly affected by UR, TP, ES, PG, CI, and PT. Traffic carbon emissions (TC) are primarily influenced by TP and PG.

Fig. 3
figure 3

Correlation Analysis Chart.

Full size image

To avoid the negative impact of multicollinearity among these independent variables on the model and to ensure accurate capture of the relationship between provincial carbon emissions and socio-economic factors, we applied Partial Least Squares (PLS) regression to obtain elasticity coefficients for each province’s driving factors. This approach enabled the construction of an extended STIRPAT model specific to each province and the calculation of each lnXi value. First, data for independent variables—including industrial structure, energy intensity, urbanization rate, total population, energy structure, GDP per capita, and the proportion of the tertiary industry—as well as the dependent variable of total carbon emissions, were extracted from the statistical yearbooks of each province. A logarithmic transformation was applied to all variables to linearize relationships and enhance model stability. The PLS regression model was then used to estimate the regression weight coefficients of each driving factor, determine the number of principal components, and calculate the intercept and R² to assess the model’s fit.

Subsequently, we employed the variable importance in projection (VIP) scores to interpret the model’s predictive mechanism and clarify the interactions among variables. As illustrated in Fig. 4, the VIP values for each variable exhibited corresponding trends with the increase or decrease in the number of principal components. When the number of principal components increased to a certain threshold, the VIP values tended to stabilize, at which point the corresponding number of components could be considered the optimal model selection. This process ensures that the model captures the key driving factors while avoiding redundant information, thereby enhancing both the predictive accuracy and interpretability of the model.

Fig. 4
figure 4

Variable importance in projection (VIP) values for key drivers under different numbers of principal components.

Full size image

STIRPAT model

Fig. 5
figure 5

STIRPAT Model Operation Flowchart.

Full size image

In Fig. 5, the Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model is widely used in carbon emission forecasting14, Based on the IPAT equation, scholars have identified the primary factors influencing carbon emissions as population size (P), economic level (A), and technology level (T), constructing an equation to represent environmental impact (I) from these three aspects:

$$\:I=a\times\:{P}^{b}\times\:{A}^{c}\times\:{T}^{d}\times\:e$$

Compared to the IPAT equation, the STIRPAT model not only allows for the inclusion of additional factors that influence energy consumption and carbon emissions but also enables analysis of the disproportional effects of these factors on the dependent variable 2427. We selected the STIRPAT model to forecast regional carbon emissions. Due to its analytical and explanatory capabilities, which can be enhanced by adding, removing, or decomposing factors, we further expanded the STIRPAT model by integrating additional driving factors. Taking the logarithm of both sides transforms it into a linear form, facilitating regression analysis. The regression coefficients in the equation reflect the elasticity relationship between the independent and dependent variables.

$$\:ln{C}_{k}=ln{a}_{1}+{a}_{2}ln{TP}_{k}+{a}_{3}ln{UR}_{k}+{a}_{4}ln{RG}_{k}+{a}_{5}ln{ES}_{k}+{a}_{6}ln{IS}_{k}+{a}_{7}ln{EI}_{k}+{a}_{8}ln{CI}_{k}+{a}_{9}ln{PT}_{k}+lne$$

In this equation, k represents the province, a1 is a constant, a2, a3, … a9 are regression coefficients, and e is the error term. Finally, exp(Ck) is used to back-calculate CO₂ emissions.

In the model validation stage, we performed five repetitions of five-fold cross-validation to assess the sensitivity of the regression model, using the average R² as the evaluation metric. Multiple rounds of cross-validation enabled a comprehensive evaluation of the model’s performance across different data subsets, thereby verifying its stability and generalization ability. As shown in Fig. 6, the average R² values remained consistently high across different folds and numbers of principal components. This indicates that the extended STIRPAT model constructed in this study demonstrates strong goodness-of-fit and predictive capability, effectively capturing the complex relationships between provincial carbon emissions and socioeconomic factors.

Fig. 6
figure 6

Mean R² values from five-fold cross-validation with five repetitions of Hubei, Hunan, and Jiangxi provinces.

Full size image

Scenario setting

Considering the impact of global economic downturn pressures, we set up three scenarios (BAU, CIS, and AIS) to simulate carbon emission forecasts under different economic development conditions, aiming for broader applicability across various development states. To ensure alignment with planning objectives, the parameters for different development scenarios from 2021 to 2080 for the three provinces in the Yangtze River Middle Reaches were set based on documents such as the National Medium- and Long-Term Development Plan28. and the Energy Production and Consumption Revolution Strategy (2016–2030)29. Key parameter values for each scenario are provided in Table 2.

Business as Usual (BAU), the global financial environment and China’s macroeconomic policy direction remain largely unchanged. Economic development and ecological protection strategies follow the current trend, with a gradual population decline, an increase in urbanization rate, steady economic growth, and optimization of the energy industry structure.

Continued Improvement Scenario (CIS), China’s population declines more rapidly than expected, and the economic growth rate is faster than in the BAU scenario, returning to the pre-COVID-19 growth rate. In this scenario, carbon emissions decrease at an accelerated pace, and the timeline for carbon peaking and carbon neutrality is advanced.

Accelerated Improvement Scenario (AIS), China will further reduce energy intensity at a rate of 3.00%–7.50%, optimize the energy structure at a rate of 3.00%–5.50%, and enhance the industrial structure at a rate of 2.00%–3.25%. Additionally, China’s terrestrial ecosystems will be maximally protected and restored, leading to a significant reduction in carbon emissions.

Table 3 Driving factors and their assumptions under different scenarios from 2022 to 2080 (Unit: %).
Full size table

Based on the established model, solutions were derived under the BAU, CIS, and AIS scenarios. We took the median value of each data point in Table 3 as the rate of linear change for model influencing factors over time. All influencing factors in the model were represented as intermediate variables with time as the independent variable, resulting in three models that predict carbon emissions over time.

To determine the carbon peaking time, we calculated the maximum value of each model function expression after 2024, with the extreme point representing the time of carbon peaking in that scenario. For the carbon neutrality time, we first calculated the linear change in total ecological carbon sequestration over time. We then set up equations equating carbon sequestration and carbon emissions within the same scenario and solved for the time when carbon neutrality would be achieved.

Results

PLS regression results and VIP value analysis

The PLS regression results for the three provinces yield the partial least squares regression coefficients, intercepts, and correlation coefficients as shown in Table 4. The correlation coefficient (R²) is used to evaluate the contribution of variables to the model; values closer to 1 indicate stronger explanatory power of the model. With R² values greater than 0.9 in the table, this indicates a good fit for the model.

From the PLS regression coefficients, it can be observed that in Hubei province, the energy structure (ES) has a negative correlation with carbon emissions, while increases in the other factors contribute positively to carbon emissions. Therefore, Hubei can reduce carbon emissions by enhancing ES factor and decreasing the intensity of other factors. In Hunan province, the urbanization rate (UR), ES, industrial structure (IS), and energy intensity (EI) factors are negatively correlated with carbon emissions, while the others are positively correlated. Thus, Hunan can reduce carbon emissions by increasing ES, IS, and EI while reducing the intensity of other factors. In Jiangxi province, ES and EI factors are negatively correlated with carbon emissions, while the others are positively correlated. Hence, efforts should be made to enhance ES and EI factors while reducing the intensity of other factors.

Table 4 Weight coefficients of driving Factors, model Intercepts, and R² for different Provinces.
Full size table

Based on the different partial least squares regression coefficients for each indicator, the Variable Importance in Projection (VIP) values for the total carbon emissions from 2001 to 2021 were calculated and normalized (as shown in Table 5). Analyzing the VIP data for the driving factors of Hubei, Hunan, and Jiangxi provinces reveals that the impact of different indicators on total carbon emissions varies across provinces, indicating different structural influences on carbon emissions. Longitudinally, Hunan province has the highest standard deviation among its driving factor VIPs, followed by Hubei, and Jiangxi has the lowest. This suggests that the contribution of carbon emissions from various indicators in Jiangxi is the most balanced, making it the most challenging province to reduce carbon emissions compared to the other two.

Table 5 VIP values of driving factors for different Provinces.
Full size table

In Hubei province, the factor that contributes the most to carbon emissions is GDP per capita, followed by energy intensity, indicating that Hubei’s economic structure is still predominantly supported by the secondary industry. Accordingly, efforts should focus on promoting energy-saving and emission-reduction measures in industry while enhancing productivity and efficiency to reduce carbon emissions simultaneously. The factor contributing the least to carbon emissions is the industrial structure, suggesting that Hubei’s industrial structure optimization is relatively advanced, making its contribution to carbon emissions nearly negligible.

In contrast to Hubei, the increase in urbanization rate in Hunan province is the main driver of rising carbon emissions. While urbanization progresses, it is essential to avoid excessive urbanization, ensuring that the urban population levels are compatible with the level of urban development.

Similarly to Hubei, the VIP values of the factors in Jiangxi province are comparable, with both provinces exhibiting similar economic development models. The highest contribution rate is also attributed to GDP per capita, followed by energy intensity, indicating that as GDP per capita steadily increases, Jiangxi should focus on improving energy intensity to reduce carbon emissions.

Figures 6, 7, 8, 9, 10 and 11 illustrate the actual carbon emissions, fitted values, and prediction errors for Hubei, Hunan, and Jiangxi provinces from 2001 to 2021. The carbon emission trends across the three provinces exhibit general consistency, without a sustained year-on-year increase. In certain years, emissions declined; however, the magnitude of these reductions was smaller than the increases, indicating an overall upward trajectory. The close alignment between actual and fitted values in terms of monotonicity suggests that the model demonstrates strong robustness, generalizability, and resistance to external disturbances. From the perspective of prediction errors, the models effectively capture future carbon emission trends and total emissions, maintaining errors within 4%, thereby exhibiting high predictive accuracy and reliability.

For Hubei Province, as shown in Fig. 6, actual carbon emissions fluctuate due to factors such as climate variability, international market volatility, and natural disasters, rather than following a steady growth pattern. These fluctuations introduce uncertainty regarding the province’s ability to achieve its carbon peaking and neutrality targets ahead of schedule. Between 2001 and 2021, Hubei’s total carbon emissions increased by nearly 1.5 times—the highest growth among the three provinces—reflecting significant industrial expansion and rapid economic development.

An analysis of the model’s fitting performance for Hubei Province indicates that prediction errors remain generally low, within ± 3%, confirming its strong predictive capability (Fig. 7). However, certain years exhibit relatively larger errors, particularly in 2005, which recorded the highest deviation. This discrepancy may be attributed to a surge in industrial investment aimed at enhancing development potential, leading to industrial production growth that significantly exceeded projections. Additionally, in 2017, extreme weather events, such as flooding, disrupted industrial activities, causing actual carbon emissions to fall below the predicted values, thereby contributing to a notable prediction error in that year.

Fig. 7
figure 7

Actual, Fitted Carbon Emissions, and Prediction Error in Hubei Province (2001–2021).

Full size image

The growth of carbon emissions in Hubei Province is driven by multiple factors. As illustrated in Fig. 8, economic expansion, industrial structure, and energy intensity are the primary contributors. From 2000 to 2021, the rapid increase in per capita GDP (represented by the deep red section) has been the dominant driver of rising carbon emissions. Additionally, Hubei’s industrial structure is heavily reliant on heavy industries with high energy consumption, leading to significant contributions from industrial structure (blue) and energy intensity (orange). In contrast, urbanization rate (purple) and total population (green) have had relatively stable impacts on carbon emissions. Although ecological factors (yellow) have shown some growth, their overall contribution remains minor.

Moving forward, while maintaining economic growth, Hubei Province should focus on optimizing its industrial structure, reducing the proportion of high-carbon sectors, and accelerating the adoption of green and low-carbon technologies. Enhancing energy efficiency will be crucial in effectively controlling carbon emissions and ensuring the steady progression toward the province’s carbon peaking and neutrality goals.

Fig. 8
figure 8

Contributions of Different Factors to Carbon Emissions in Hubei Province (2000–2021).

Full size image

As shown in Fig. 9, the total carbon emissions and growth rate in Hunan Province are lower than those in Hubei Province. Over the past two decades, Hunan’s carbon emissions increased by 180%, reaching approximately 450 million tons in 2021, which accounted for around 37% of the total carbon emissions in the middle and lower reaches of the Yangtze River. Compared to Hubei, Hunan’s per capita carbon emissions remain at a lower level and, as the province is currently in the late stage of industrialization, total emissions are expected to continue rising as industrialization progresses.

In terms of model performance, the prediction model for Hunan Province demonstrates superior overall accuracy compared to that of Hubei, with a lower relative error (Fig. 8). Although the error rate remains generally low, a noticeable deviation occurred in 2017. This anomaly may be reasonably attributed to the severe flooding that affected large parts of Hunan during that year. According to official reports from the Ministry of Emergency Management, the 2017 floods caused widespread disruptions to industrial operations and energy supply chains, which are key drivers of carbon emissions. Given the scale and duration of the event, it is plausible that such an external shock contributed to the divergence between predicted and actual emissions. It should be noted, however, that due to data and methodological limitations, a formal statistical outlier test was not conducted in this study, and this explanation is based primarily on documented evidence and contextual analysis. Future research could further validate such anomalies using quantitative outlier detection methods to strengthen the robustness of model evaluation. Furthermore, the non-monotonic trend observed in Hunan’s carbon emission data suggests that the province may be influenced by a broader set of complex and irregular factors compared to Hubei.

Fig. 9
figure 9

Actual, Fitted Carbon Emissions, and Prediction Error in Hunan Province (2001–2021).

Full size image

As illustrated in Fig. 10, carbon emissions in Hunan Province are influenced by multiple factors, with urbanization rate (purple) and energy intensity (orange) being the primary drivers. In contrast, the effects of industrial structure, ecological factors, and energy structure are relatively minor. Compared to Hubei Province, Hunan’s industrial structure is more diversified, with a higher proportion of agriculture and services, which has mitigated the growth rate of carbon emissions to some extent.

Looking ahead, Hunan Province can further reduce carbon emission intensity and enhance sustainable carbon management by promoting industrial upgrades and expanding the use of clean energy.

Fig. 10
figure 10

Contributions of Different Factors to Carbon Emissions in Hunan Province (2000–2021).

Full size image

As shown in Fig. 8, Jiangxi Province exhibits the lowest total carbon emissions and growth rate among the three provinces in the middle reaches of the Yangtze River. However, its carbon emissions have demonstrated significant volatility over the past two decades, rising from 150 million tons in 2001 to nearly 290 million tons in 2021, with six distinct phases of fluctuation. This variability presents challenges for carbon emission forecasting and indicates that Jiangxi’s carbon emissions are influenced by a diverse and complex set of factors.

From a model performance perspective, Jiangxi Province exhibits relatively larger prediction errors (Fig. 11), with substantial deviations in multiple years. These discrepancies may be attributed to shifts in the province’s economic structure and specific external events, such as policy adjustments or natural disasters. Compared to Hunan Province, Jiangxi’s prediction errors are more pronounced, highlighting the need for more precise model calibration and the inclusion of a broader range of variables in provincial carbon emission forecasts to improve prediction accuracy.

Fig. 11
figure 11

Actual, Fitted Carbon Emissions, and Prediction Error in Jiangxi Province (2001–2021).

Full size image

As illustrated in Fig. 12, the growth of carbon emissions in Jiangxi Province is driven by multiple factors. Compared to Hubei and Hunan, Jiangxi’s industrial structure is predominantly composed of agriculture and light industry, resulting in lower energy consumption intensity and, consequently, a relatively lower carbon emission per unit of GDP.

Among the contributing factors, per capita GDP (deep red) and urbanization rate (purple) are the primary drivers of carbon emission growth, whereas energy intensity (orange) and industrial structure (blue) play relatively minor roles. Additionally, the impact of ecological factors (yellow) has gradually increased in recent years, likely reflecting the strengthening of ecological protection policies.

However, as Jiangxi’s economy continues to develop, particularly with the advancement of industrialization, carbon emissions are expected to rise further. Therefore, while promoting economic growth, Jiangxi Province must prioritize the adoption of green and low-carbon technologies, regulate industrial carbon emission intensity, and optimize energy consumption structures to prevent a trajectory of high carbon emissions.

Fig. 12
figure 12

Contributions of Different Factors to Carbon Emissions in Jiangxi Province (2000–2021).

Full size image

Carbon emission forecasting for the three provinces in the Yangtze river middle reaches

As illustrated in Fig. 13, we combined the extended STIRPAT model with scenario analysis to project the future carbon emissions of Hubei, Hunan, and Jiangxi provinces from 2022 to 2080. The projections are based on three distinct scenarios: the Baseline Development Scenario (BAU), the Low Development Scenario (CIS), and the Rapid Development Scenario (AIS). This approach allows for an assessment of how varying socioeconomic development levels and policy interventions influence future carbon emission trajectories in the three provinces. While all three provinces exhibit an initial increase followed by a decline in carbon emissions, notable differences emerge in the timing of peak emissions and the subsequent rate of decline, reflecting variations in their socioeconomic and industrial contexts.

To ensure the scientific validity and comparability of the emission reduction assessment under different scenarios, the average annual emission reduction rate (AERR) for each province was calculated using the following formula:

$$\:AERR=\frac{{E}_{peak}-{E}_{neutral}}{{E}_{peak}\times\:({Y}_{neutral}-{Y}_{peak)}}\times\:100\text{\%}$$

Where \(\:{E}_{peak}\)represents the carbon emissions at the peak year (\(\:{Y}_{peak}\)), and \(\:{E}_{neutral}\) is the carbon emissions in the year when carbon neutrality (\(\:{Y}_{neutral}\)) is achieved or targeted. The denominator \(\:({Y}_{neutral}-{Y}_{peak)}\) is the number of years between the emission peak and the carbon neutrality year. This approach standardizes the rate of reduction, allowing for direct comparison of decarbonization pace across different scenarios and regions.

Under the BAU scenario, which assumes a continuation of current development patterns with minimal policy changes, carbon emissions in all three provinces are projected to reach their highest levels. The CIS scenario envisions a moderately sustainable development path, with improved energy efficiency and emission controls, resulting in lower emissions than the BAU scenario but still showing significant growth before reaching peak levels. In contrast, the AIS scenario represents an ambitious strategy centered on rapid decarbonization and the adoption of green technologies, leading to the lowest projected emissions among the three scenarios and a clear shift toward a low-carbon economy.

The differences in these carbon emission trajectories highlight the varying commitments of the three provinces to environmental sustainability. The projections indicate that although each province is expected to reach peak emissions at different times, the AIS scenario stands out for its potential to achieve carbon neutrality through proactive policy measures and technological innovation.

The timeframes for carbon peaking and neutrality vary significantly across the three provinces, with a time span of over 25 years between peak emissions and full decarbonization. These disparities reflect differences in development pathways, industrial structures, and policy implementation strategies. Hubei Province, characterized by rapid industrialization and a transition toward stricter environmental regulations, exhibits the fastest decline in carbon emissions among the three provinces. Under the BAU scenario, Hubei’s average annual reduction rate is 4.68%, indicating a moderate decarbonization pace. The CIS scenario shows an improved rate of 5.35%, driven by enhanced sustainability measures. The AIS scenario, which assumes the most ambitious decarbonization efforts, achieves the highest average reduction rate of 6.05%, facilitated by substantial investments in clean energy, energy efficiency, and stricter industrial and transportation emissions standards.

Hunan Province, with a slightly later stage of industrialization than Hubei, follows a similar but less aggressive decarbonization trajectory. Under the BAU scenario, its average reduction rate is 4.76%, while the CIS scenario sees an improvement to 5.34%, attributed to stricter environmental policies and technological advancements. Under the AIS scenario, Hunan’s reduction rate reaches 6.02%, primarily due to the rapid expansion of renewable energy and the transition to low-carbon technologies. However, slower industrial restructuring and continued reliance on coal limit the extent of its emission reductions compared to Hubei.

Jiangxi Province, with a more traditional industrial base and a slower pace of industrial transformation, exhibits the lowest decarbonization rates among the three provinces. Under the BAU scenario, its average annual reduction rate is 4.86%, slightly higher than that of Hunan. The CIS scenario sees a moderate increase to 5.20%, as the province adopts more energy-efficient practices and cleaner technologies. Under the AIS scenario, Jiangxi achieves an average reduction rate of 6.36%, driven by growing investments in green infrastructure and a transition toward sustainable energy sources. Nevertheless, its slower shift toward green technologies results in a less pronounced decline in emissions compared to Hubei and Hunan.

The differences in carbon emission reduction rates and timeframes underscore the challenges posed by varying economic structures, resource availability, and levels of policy intervention in each province. While Hubei is expected to achieve carbon neutrality sooner, Hunan and Jiangxi face greater obstacles due to their continued dependence on conventional energy sources and slower industrial transitions. These findings suggest that achieving carbon peaking and neutrality requires region-specific strategies, emphasizing the need for tailored climate policies that account for local conditions and developmental contexts.

Fig. 13
figure 13

a) Carbon Emission Forecast for Hubei Province, b) Carbon Emission Forecast for Hunan Province, c) Carbon Emission Forecast for Jiangxi Province.

Full size image

As shown in Fig. 13a, the projected carbon emissions for Hubei Province from 2021 to 2080 vary under different scenarios. In the BAU scenario, which extends the existing emission trajectory based on 2021 levels, carbon emissions are expected to peak in 2037 at approximately 636 million tons. The CIS scenario, assuming slightly more favorable development conditions than BAU, predicts an earlier peak in 2032 at around 579 million tons. Under the AIS scenario, emissions are projected to peak before 2030 at approximately 545 million tons, with the province anticipated to achieve carbon neutrality by 2060.

Figure 13b presents the carbon emission projections for Hunan Province from 2021 to 2080. Under the BAU scenario, emissions are expected to peak in 2033 at approximately 485 million tons. The CIS scenario, representing a moderate development pathway, forecasts a peak in 2030 at around 432 million tons. In the AIS scenario, which assumes the most rapid development pace, carbon emissions are expected to peak before 2030 at approximately 330 million tons. However, unlike Hubei, Hunan Province is not expected to achieve carbon neutrality by 2060 under this scenario.

According to Fig. 13c, the projected carbon emissions for Jiangxi Province from 2021 to 2080 also exhibit distinct trajectories across scenarios. In the BAU scenario, Jiangxi’s emissions are expected to peak in 2036 at approximately 347 million tons. The CIS scenario projects an earlier peak in 2030 at around 308 million tons. Under the AIS scenario, emissions are expected to peak before 2030 at approximately 288 million tons, with carbon neutrality potentially achievable by 2060.

To systematically evaluate the reliability of long-term forecasts, this study employs a combined Bootstrap-Monte Carlo simulation framework to quantify uncertainty. First, 1,000 bootstrap samples of parameter perturbations are generated based on the residual distribution of historical data. Second, for each scenario, the driving factors in Table 2 are subjected to normally distributed random disturbances of ± 15%. Finally, Monte Carlo simulations are conducted to calculate the 95% confidence intervals for carbon emission projections from 2022 to 2080 (represented by the shaded areas in Fig. 13). The key results show that the mean confidence interval width for predicted carbon neutrality years is ± 4.7%, and the maximum deviation in peak emission years ranges from ± 2 years under the BAU scenario to ± 1 year under the AIS scenario. These analyses demonstrate that the model not only exhibits excellent historical fitting performance, but also maintains the long-term predictive uncertainty within a reasonable range, fully meeting the precision requirements for policy-making.

Discussion

Analysis of differences in driving factors among three provinces

The empirical results of the study show that the dominant factors and change mechanisms leading to carbon emissions in different provinces are significantly different. Carbon emissions in Hubei Province are mainly driven by RG (0.828), CI (0.846) and TP (0.708), and economic growth is the main driver of carbon emissions growth. As a major economic province in central China, Hubei has a strong industrial base, with a large proportion of manufacturing and heavy chemical industries. In recent years, the urban circle economy with Wuhan as the core has also developed rapidly30, driving high economic growth while significantly increasing carbon emissions. At the same time, ES progress has a suppressive effect on carbon emissions, indicating that the proportion of renewable energy and clean energy in Hubei Province has gradually optimized. Hunan is strongly driven by TP (1.117). The accelerated urbanization process in Hunan has led to population concentration and urban expansion31. A large number of rural people have moved to cities, which has promoted energy demand and carbon emissions in infrastructure, transportation, construction and other fields. In contrast, the correlations of ES (−0.240), IS (−0.155), and EI (−0.118) are negative and small, indicating that some progress has been made in these areas, but their inhibitory effect on carbon emissions is not significant and their overall influence is limited. This also reflects that Hunan’s current low-carbon development still faces structural constraints. Jiangxi’s IS (1.601), PT (1.385), and CI (0.999) have the highest correlation with carbon emissions. This shows that high-energy-consuming industries and traditional manufacturing industries still occupy an important position in Jiangxi’s economy, and the upgrading and transformation of traditional industries is still in the critical stage. In recent years, the expansion of Jiangxi’s service industry has been accompanied by the rapid growth of urbanization and modern consumption. Energy consumption and emissions in areas such as transportation, logistics, information and communications, and tourism are also increasing, and energy consumption in electricity and transportation has increased significantly32,33. The empirical results obtained in this article are helpful in formulating differentiated paths that are tailored to local conditions and implemented by province. This will help to give full play to each country’s respective advantages, reduce resistance to emission reduction, and provide realistic and feasible samples and experiences for achieving carbon peak and carbon neutrality goals in the region and even the country.

Policy measures of the three provinces in the middle reaches of the Yangtze river

The empirical results demonstrate significant differences in the timing of carbon peaking and neutrality among the three provinces, underscoring the need for region-specific strategies and climate policies tailored to local conditions and developmental contexts. Under the BAU scenario, none of the provinces are able to achieve carbon peaking or carbon neutrality targets. In Hubei, GDP per capita (RG, 0.828) is the most influential driver; only under the AIS scenario can Hubei achieve both carbon peaking (at approximately 545 million tons) and carbon neutrality. This suggests that Hubei should accelerate the green transformation of traditional high-energy industries34, foster the development of strategic emerging sectors and high-tech manufacturing, and intensify the deployment of renewable energy to enhance the cleanliness and intelligence of its energy system35.

In Hunan, total population (TP, 1.117) is the primary driver. Although the AIS scenario advances the carbon peaking timeline, achieving carbon neutrality remains highly challenging. Hunan should prioritize the development of green cities and low-carbon infrastructure, improve the coordination between population agglomeration and environmental carrying capacity, and vigorously promote green transportation, green buildings, and smart city initiatives to enhance per capita energy efficiency. At the same time, efforts to protect and enhance ecological carbon sinks must be strengthened through ecosystem conservation and restoration36,37. Only a multidimensional approach that addresses both key drivers and ecological systems can effectively close the gap to carbon neutrality38.

For Jiangxi, the proportion of the tertiary industry (PT, 1.385) is the most significant driver. Under the AIS scenario, carbon neutrality may be achievable by 2060, indicating that Jiangxi should focus on deep greening and high-efficiency management of the service sector. By advancing structural optimization, improving energy efficiency, promoting green consumption, and implementing targeted policy incentives, the province can facilitate the transformation of the service sector toward high-tech, high value-added, and low-carbon development, thereby injecting new momentum into high-quality and sustainable regional growth39,40. Developing customized policy pathways based on each province’s dominant drivers not only addresses regional emission reduction bottlenecks, but also provides scalable and transferable models for achieving carbon peaking and neutrality in central China and nationwide4.

Representativeness and influence of regional dimensions

As one of China’s core regions in terms of economy, population, and technological strength, the Yangtze River Middle Reaches—encompassing Hubei, Hunan, and Jiangxi—plays a pivotal strategic role in both national and global carbon governance and green transition41. This region not only bridges the developed east and emerging west, but also features a large and diverse economy, rapid urbanization, and robust innovation capacity. By systematically analyzing carbon emission dynamics and carbon neutrality prospects driven by multiple factors, this study addresses a key research gap in central China’s emission mechanisms and collaborative governance. The findings provide theoretical and methodological innovation to support differentiated policy formulation, regional emission reduction, and economic transformation. Furthermore, these insights offer valuable references for other transitioning regions in China and worldwide, enriching the scientific foundation for global carbon neutrality and green urban governance.

Limitations and future prospects

  1. 1.

    In this study, we adopted the carbon sink calculation method recommended by the IPCC, which estimates carbon sequestration by multiplying the area of each land use type by its corresponding fixed carbon absorption coefficient. This approach is currently the most mainstream and authoritative method, widely applied in carbon emission and carbon sink research worldwide. However, the limitations of the IPCC method should not be overlooked. Future research should explore more refined and dynamic carbon sink estimation techniques. For example, integrating remote sensing and GIS technologies could enable dynamic monitoring and assessment of carbon sink changes. The interaction between natural variables such as temperature, precipitation, and humidity affects the strength and stability of carbon sinks42,43, environmental changes and natural factors affecting carbon sink capacity could be explicitly incorporated into the analysis44,45. By introducing more regional and spatiotemporal variables, carbon sink calculations will be more accurate, thus providing stronger data support for the formulation of low-carbon development strategies and carbon neutrality policies.

  2. 2.

    We estimated the carbon emissions in the study area based on fossil energy consumption and carbon content, following the IPCC Guidelines for National Greenhouse Gas Inventories. While this method has been widely used in many studies, neglecting the factor of energy efficiency may lead to an overestimation of carbon emissions46. Future research should further improve carbon emission accounting methods, particularly with respect to energy efficiency. Indicators such as “energy intensity” (the ratio of energy consumption to economic output) and “energy utilization efficiency” can be incorporated as influential factors in the calculation47,48. By refining sector-level energy efficiency data and considering the actual impacts of technological advancement and policy interventions, more sophisticated energy system models—such as life cycle analysis or input-output models—can be employed to estimate the efficiency of different energy sources49. This would enable a more accurate assessment of how energy efficiency affects carbon emissions.

  3. 3.

    This study selected three scenarios based on current macroeconomic and policy trends. However, real-world development trajectories are subject to considerable uncertainty and variability, particularly due to factors such as regional development imbalances, differences in policy implementation effectiveness, and global economic fluctuations. Therefore, future research could further expand scenario design by incorporating a wider range of variables and scenario combinations. For example, more customized scenario models could be developed to account for differences in industrial transformation rates, technological innovation processes, and the stringency of environmental policy enforcement, thereby simulating the future carbon emission trajectories under various policy pathways. Such an approach would more comprehensively reflect the complex dynamics of socioeconomic development and provide policymakers with richer options for policy responses and future scenario forecasts.

  4. 4.

    Future research should strive to obtain more detailed county-level data, which would enable a more accurate capture of local carbon emission characteristics and trends. For instance, by acquiring finer-grained regional data on land use, energy consumption, and industrial activities, the model could better reflect local differences and achieve higher-precision regional forecasts. This, in turn, would support the targeted implementation of low-carbon development policies and facilitate progress toward carbon neutrality goals.

Conclusion

This study focuses on Hubei, Hunan, and Jiangxi provinces in the Yangtze River Middle Reaches region, utilizing the extended STIRPAT model and PLS regression analysis to explore the driving factors and future trends of carbon emissions under different scenarios. Based on the research findings, the following conclusions can be drawn:

(1) There are significant differences among the provinces in the Yangtze River Middle Reaches regarding the timing of carbon emission peaks and achieving carbon neutrality. Hubei is expected to reach its carbon emission peak in 2030 and achieve carbon neutrality before 2060, while Hunan and Jiangxi will have relatively later carbon neutrality timelines. This indicates that each province needs to develop differentiated emission reduction policies based on its economic structure and development stage to effectively advance the carbon neutrality process.

(2) From the analysis of driving factors, indicators such as GDP per capita, urbanization rate, energy intensity, and carbon emission intensity are closely related to carbon emissions, but the degree of influence of these driving factors varies significantly between provinces. Hubei’s carbon emissions are more influenced by economic growth and energy structure, while Hunan and Jiangxi are significantly affected by urbanization rate and industrial structure adjustments. Therefore, when formulating emission reduction strategies, provinces should fully consider their economic and social characteristics and propose targeted reduction measures.

(3) To achieve carbon peaking and carbon neutrality goals, the Yangtze River Middle Reaches should prioritize measures such as optimizing industrial structure, adjusting energy structure, enhancing technological innovation, and ecological protection, particularly in accelerating the reduction of energy intensity and increasing the proportion of the tertiary industry, to promote the green transformation of the regional economy and ensure gradual achievement of carbon reduction targets under various scenarios.