Decoupling Analysis and Scenario Prediction of Port Carbon Emissions: A Case Study of Shanghai Port, China

Abstract

This study presents a comprehensive analysis of carbon emission trends and their driving factors at Shanghai Port, with a particular focus on the decoupling relationship between port economic development and carbon emissions, as well as forecasting the timeline for achieving the port’s carbon peak. The findings reveal distinct temporal patterns in emission growth: from 2009 to 2012, Shanghai Port experienced steady increases in carbon emissions, while from 2020 to 2023, it witnessed accelerated growth, primarily driven by fuel oil consumption. Using the Logarithmic Mean Divisia Index (LMDI) decomposition model, the study identifies operational revenue as the most significant contributor to carbon emission growth, while economic intensity emerges as the strongest inhibiting factor. Notably, the carbon-promoting effects of energy structure and efficiency improvements substantially outweigh the emission reductions achieved through enhanced economic intensity. The Tapio decoupling analysis indicates that during 2010–2023, neither operational revenue nor port cargo throughput capacity achieved stable decoupling from carbon emissions at Shanghai Port. Operational revenue exhibited alternating patterns of strong and weak decoupling, while cargo throughput showed more pronounced fluctuations, cycling through phases of decoupling and negative decoupling. Scenario-based predictions using the GRU-LSTM hybrid model provide critical insights: under the baseline scenario, Shanghai Port is projected to fail to achieve a carbon peak by 2035. However, both the low-carbon and enhanced mitigation scenarios project a carbon peak around 2026, with the enhanced scenario enabling earlier attainment of the target. These findings offer valuable theoretical foundations for formulating Shanghai Port’s carbon peak strategy and provide practical guidance for emission management and policy development at ports. The methodological framework and empirical results presented in this study may serve as a reference for other major ports pursuing similar decarbonization goals.

1. Introduction

Global climate change has emerged as one of the most pressing challenges of our era, characterized by escalating environmental crises, including increasing pollutant emissions, accelerated global warming, and rising sea levels, that threaten ecosystems and human societies worldwide. As critical energy-intensive nodes in global supply chains, ports contribute approximately 3% of total anthropogenic greenhouse gas emissions annually [1]. Shanghai Port exemplifies both the economic significance and environmental challenges confronting modern maritime infrastructure. As the world’s busiest container port, Shanghai Port capitalizes on its strategic geographic advantages and substantial economic capacity to function as a vital global logistics hub. It maintains commercial connections with 221 countries and over 500 ports worldwide, while achieving a record throughput exceeding 50 million twenty-foot equivalent units (TEUs) in 2024. However, this exceptional operational performance incurs significant environmental costs, with rapidly increasing cargo volumes driving substantial growth in carbon emissions. Consequently, research into Shanghai Port’s carbon peak timeline and decarbonization pathway is crucial not only for the port’s sustainable development but also for establishing a replicable framework to guide the global shipping industry’s transition to low-carbon operations.

Accurate carbon emission quantification forms the fundamental basis for assessing the environmental footprint of industrial activities and production processes, facilitating data-driven environmental impact assessments. Current carbon accounting methodologies can be classified into three primary categories: (1) the top-down approach, which estimates emissions using aggregate energy consumption data [2,3,4]; (2) the bottom-up methodology, which calculates emissions by aggregating energy consumption from individual operational activities [5,6,7]; and (3) lifecycle assessment (LCA), which comprehensively evaluates emissions throughout a product or service’s entire value chain—from raw material extraction and production to transportation, utilization, and final disposal [8,9]. Within the port sector specifically, carbon emissions predominantly originate from terminal operations and visiting vessel activities. Seminal studies in this field include [10]’s comprehensive emission inventory for Barcelona Port, which distinguished between maritime and terrestrial port emissions, and [11]’s validated vessel emission model for Felixstowe Port. Additionally, the applicability of the IPCC’s standardized emission factors for calculating port-related emissions in multimodal transportation networks was demonstrated by [12], establishing a robust framework for intermodal carbon accounting.

To systematically identify and quantify carbon emission sources across operational stages and establish a scientific foundation for targeted mitigation strategies, researchers have predominantly employed advanced decomposition methodologies including the LMDI and STIRPAT (Stochastic Impacts by Regression on Population, Affluence, and Technology) models. The STIRPAT framework has been extensively applied in regional emission analyses, including examinations of driving factors in China’s Yangtze River Delta [13], investigations of Brazilian energy-related emissions [14], and assessments of Algerian carbon dynamics [15]. The LMDI approach, originally developed by [16], offers distinct advantages through its assumption-free logarithmic decomposition that precisely isolates individual factor contributions while circumventing multicollinearity problems, thereby providing robust insights into complex emission drivers. This methodological strength is evidenced by [17]’s transportation sector analysis (2001–2018), which identified capital investment and technological advancement as primary emission catalysts and constraints, respectively. Logistics industry emissions were systematically assessed [18], alongside a comprehensive evaluation of China’s waterway transport emissions during 2002–2020 [19]. While sectoral and regional applications abound, as demonstrated by [17,20], port-specific emission factor decomposition remains understudied despite its critical value in pinpointing operational leverage points for maritime decarbonization. This research gap motivates our dual application of LMDI decomposition to Shanghai Port’s emission profile, which will not only elucidate the relative impacts of key driving factors but also generate actionable insights for developing port-specific carbon mitigation roadmaps, thereby contributing both methodological innovation and practical policy guidance for sustainable port operations.

The integration of factor decomposition with decoupling theory has emerged as a robust methodological framework for evaluating the efficacy of emission reduction policies, offering distinct advantages over conventional economic analyses through its ability to quantitatively assess the dynamic relationship between economic growth and carbon emissions. Building upon foundational work by [21], who pioneered the development of elasticity coefficients to precisely characterize decoupling states, this approach has become instrumental in contemporary research examining the nexus between economic development and environmental sustainability. Empirical applications demonstrate the framework’s analytical power through multiple dimensions: the identification of three distinct decoupling patterns in China’s transportation sector, revealing how energy efficiency gains increasingly constrain emissions while economic growth, private vehicle ownership, and freight turnover exhibit complex nonlinear impacts [22]; an assessment of decarbonization pathways aligned with Paris Agreement targets [20]; and comparative analyses of China–US decoupling trajectories [23]. Recent evidence from the Yangtze River Basin further highlights persistent challenges, with most provinces yet to achieve meaningful decoupling between maritime emissions and GDP growth [24]. Further research outcomes can be found in [25,26,27], as well as other studies. Despite the existing sectoral studies, a significant research gap exists in decoupling analyses specific to ports. This study fills that gap by utilizing Tapio’s decoupling framework to explore the relationship between emissions and economic growth at Shanghai Port. This novel application will yield crucial insights into whether and how the port can maintain economic competitiveness while achieving decarbonization, thereby generating both theoretical advances in maritime environmental economics and actionable policy recommendations for sustainable port development strategies that balance environmental and economic objectives.

Carbon emission forecasting serves as a critical foundation for ports to formulate forward-looking decarbonization strategies and optimize their industrial structures, with current predictive methodologies encompassing both traditional statistical models and emerging machine learning approaches. Traditional techniques include autoregressive integrated moving average (ARIMA) models [28,29], generalized autoregressive conditional heteroskedasticity (GARCH) models [30], grey correlation analysis [31], and the STIRPAT model [32], while advanced machine learning methods feature support vector regression (SVR) [22,33], long short-term memory (LSTM) networks [32,34], and gated recurrent units (GRU) [35,36]. These conventional approaches typically assume static conditions regarding policies, technologies, and economic development when extrapolating from historical data. In contrast, scenario analysis has emerged as a powerful complementary methodology that systematically explores emission trajectories under varying policy, technological, and socioeconomic conditions through carefully constructed alternative futures, thereby enhancing prediction robustness and policy relevance. This technique’s value is demonstrated by [37]’s projection of peak transportation sector emissions during 2040–2045 for China, residential building sector forecasts [38], provincial-scale analysis for Guangdong [39], agricultural emissions modeling [40], and innovative evaluation framework for ship emission reduction measures across intensity scenarios [41]. Despite the progress made in the field, a significant gap remains in the application of scenario forecasting to port carbon emissions through deep learning methods. This research fills this important gap by creating a comprehensive analytical framework that integrates scenario analysis with advanced machine learning models, allowing for more precise and actionable emissions projections for ports. Ultimately, this approach enhances data-driven decision-making in efforts to achieve maritime decarbonization.

Despite significant progress in carbon emission research, several critical methodological and analytical gaps persist in current studies. First, while factor decomposition techniques have been widely adopted, most analyses remain constrained by conventional STIRPAT model frameworks, failing to adequately examine the complex interdependencies and synergistic effects among multiple driving factors. Second, although decoupling theory has proven valuable for analyzing economy–emission relationships, its integration with factor decomposition methods remains notably underdeveloped in port-specific research, leading to incomplete assessments of emission reduction efficacy. Most critically, existing studies often artificially separate historical analysis from future projections. Current approaches either focus narrowly on retrospective factor decomposition or conduct isolated emission forecasts without incorporating decomposition insights into scenario parameter selection. This disconnect between diagnostic and predictive methodologies undermines both the accuracy of emission projections and the evidentiary foundation for policy formulation. These limitations collectively highlight the need for an integrated analytical framework that (1) incorporates an advanced analysis of factor interactions, (2) couples decomposition and decoupling approaches, and (3) systematically links historical emission drivers with forward-looking scenario modeling—particularly in the crucial yet understudied context of port operations.

To systematically address these research gaps, this study develops an integrated analytical framework focusing on Shanghai Port, encompassing four critical dimensions: carbon emission accounting, factor decomposition, decoupling analysis, and scenario forecasting. The research methodology progresses through several innovative stages: First, we employ a bottom-up approach to accurately quantify the port’s carbon emissions from both fossil fuel and electricity consumption perspectives, establishing a robust empirical foundation. Second, we advance beyond conventional STIRPAT limitations by developing a multi-dimensional LMDI decomposition model that captures complex interaction effects among emission drivers, while synergistically integrating decoupling theory to reveal the underlying mechanisms of economy–emission decoupling. Third, we pioneer the application of an ensemble modeling approach combining multiple single and hybrid models to overcome traditional ridge regression’s shortcomings in handling nonlinear emission patterns and abrupt changes. Finally, we innovatively utilize factor decomposition results to parameterize scenario indicators, implementing a novel GRU-LSTM hybrid model to project emission trajectories under three distinct development scenarios. This comprehensive approach not only provides Shanghai Port with multiple evidence-based pathways to achieve its carbon peak targets but also establishes a transferable methodological framework for port emission management globally. The study makes two primary contributions.

This study introduces an innovative analytical approach by combining LMDI decomposition with decoupling theory to examine Shanghai Port’s carbon emission drivers. The LMDI model quantitatively assesses key influencing factors, including energy intensity, operational efficiency, and fuel mix, providing more nuanced insights than conventional methods. The decoupling analysis complements this by evaluating the dynamic relationship between port economic growth and emissions across different development phases. Together, these methods offer a robust evidence base for formulating targeted emission reduction strategies at Shanghai Port, with methodological implications for global port decarbonization research.

This study overcomes traditional prediction limitations by developing a GRU-LSTM hybrid model that effectively captures nonlinear emission patterns and long-term trends. The model projects Shanghai Port’s carbon peak under three strategic scenarios, with integrated LMDI decomposition results enhancing forecast reliability. This advanced approach provides port administrators with data-driven insights for low-carbon planning while enabling policymakers to design targeted emission control measures. The methodology establishes a new benchmark for port-specific carbon prediction with global applicability.

The remainder of this paper is structured as follows: Section 2 details the methodological framework and model specifications employed in this study. Section 3 measures the carbon emissions of Shanghai Port, decomposes the influencing factors of carbon emissions, and examines the decoupling relationship between carbon emissions and economic growth. Section 4 forecasts Shanghai Port’s carbon emissions under three development scenarios. Finally, Section 5 summarizes key findings and discusses policy implications, while Section 6 provides actionable recommendations for port authorities and policymakers.

2. Materials and Methods

2.1. STIRPAT Model

The STIRPAT model represents an advanced statistical framework for analyzing anthropogenic environmental impacts. As an extension of the classic IPAT identity, this regression-based model quantitatively assesses how key socioeconomic drivers, including population dynamics (P), economic affluence (A), and technological level (T), collectively influence environmental indicators such as carbon emissions. The general STIRPAT model specification is expressed in Equation (1), as follows:

$$\begin{array}{c}\hfill I=a{P}^b{A}^c{T}^d\epsilon ,\end{array}$$

(1)

where I denotes the environmental impact indicator and a serves as the model scaling constant. The coefficients b, c, d are elasticity parameters that quantify the percentage change in environmental impact resulting from a 1% change in each corresponding factor. $\epsilon$ constitutes the stochastic error term, accounting for all unexplained variation, including measurement errors and omitted variables.

To facilitate model interpretation and enable conventional regression analysis, we apply logarithmic transformation to Equation (1), yielding the linearized specification shown in Equation (2):

$$\begin{array}{c}\hfill lnI=lna+blnP+clnA+dlnT+ln\epsilon .\end{array}$$

(2)

From the perspective of elasticity, every 1% change in P, A, and T leads to $b\%$, $c\%$, and $d\%$ change in $lnI$, respectively.

2.2. LMDI Decomposition Model

The LMDI decomposition model provides a robust analytical framework for factor decomposition in environmental and energy systems. As an enhanced version of traditional Divisia index decomposition, this approach offers perfect decomposition and handles zero values effectively through its logarithmic mean weighting scheme $L(X_t,X_{t-1})={\displaystyle \frac{X_t-X_{t-1}}{ln{X}_t-ln{X}_{t-1}}}$. The model quantifies the individual contributions of key driving factors, including population dynamics, economic activity, energy intensity, and fuel mix, to observed changes in target indicators such as carbon emissions or energy consumption. The general LMDI decomposition formulation can be expressed as Equation (3):

$$\begin{array}{c}\hfill \mathrm{\Delta }Y=\sum _{i}L(X_{i,t},X_{i,t-1})·(ln{Y}_t-ln{Y}_{t-1}),\end{array}$$

(3)

where $\mathrm{\Delta }Y$ represents the change in the indicator (such as carbon emissions, energy consumption, etc.). $X_{i,t}$ denotes the i-th influencing factor (such as population, economic level, energy structure, etc.) in period t. $Y_t$ refers to the value of the target variable in period t. $L(X_{i,t},X_{i,t-1})$ represents the logarithmic mean weighting.

2.3. Tapio Decoupling Model

The Tapio decoupling model is a tool used to analyze the relationship between economic development and environmental pressure. Its core idea is to describe, through the concept of “decoupling”, whether environmental pressures (such as carbon emissions, energy consumption, etc.) can be relatively or completely detached from the impact of economic growth. The Tapio Decoupling Model typically quantifies the relationship between economic activity (such as GDP) and environmental pressure (such as CO₂ emissions), using a “decoupling indicator” to measure the degree of decoupling. The decoupling indicator defined in Equation (4) is expressed as the ratio of the economic growth rate to the environmental pressure change rate.

$$\begin{array}{c}\hfill {ϵ}_t={\displaystyle \frac{\mathrm{\Delta }{C}_t/C_0}{\mathrm{\Delta }{G}_t/G_0}},\end{array}$$

(4)

where $\mathrm{\Delta }{C}_t=C_t-C_0$ is the difference in environmental indicators between period t and the base period. $\mathrm{\Delta }{G}_t=G_t-G_0$ is the difference in economic indicators between period t and the base period. If the decoupling indicator is greater than 1, this indicates relative decoupling; if it equals 1, this indicates that economic growth and environmental pressure are growing synchronously; if it is less than 1, this means that the growth rate of environmental pressure exceeds that of economic growth, suggesting that economic growth has not effectively reduced environmental pressure.

2.4. LSTM Model

Recurrent Neural Networks (RNNs) are a specialized class of deep learning models designed for sequential data processing, whose fundamental principle involves capturing dynamic temporal patterns through recurrent hidden states. The core mathematical formulation is expressed as (5):

$$\begin{array}{cc}\hfill h_t=& \sigma (W_h{x}_t+U_h{h}_{t-1}+b_h),\hfill \end{array}$$

(5)

where $h_t$ is the hidden state at time step t. $\sigma$ is the activation function. $W_h$ is the recurrent weight matrix. $U_h$ is the input weight matrix. $x_t$ is the input vector at time t. $b_h$ is the hidden layer bias term. The RNN suffers from the vanishing gradient problem due to its training via the backpropagation through time algorithm. To address the issue of gradient explosion, the LSTM neural network was first proposed by [42]. RNNs pass sequential information through a chain-like structure, but this process can lead to unstable gradients. LSTM resolves this by using a gating mechanism and a cell state, enabling long-term memory retention and the selective forgetting of information. The key innovation of LSTM lies in decoupling memory storage from the hidden state, allowing gradients to propagate efficiently through the cell state. The LSTM network structure consists of three gating modules and a cell state. The forget gate determines which useless information should be discarded from the cell state, while the input gate selects the information to be added. The main structure and basic steps of the LSTM algorithm are as follows.

Input gate ($i_t$): The input gate in LSTM networks serves as a critical regulatory mechanism that determines what new information should be incorporated into the cell state ${\tilde{c}}_t$. This gating function employs a sigmoid activation $\sigma$ to produce values between 0 and 1, representing the degree of information retention. As shown in Equations (6) and (7), the gate mechanism involves two complementary transformations:

$$i_t=\sigma (W_i{x}_t+U_i{h}_{t-1}+b_i),$$

(6)

$${\tilde{c}}_t=tanh(W_c{x}_t+U_c{h}_{t-1}+b_c).$$

(7)

Forget gate ($f_t$): The forget gate serves as the LSTM’s selective memory mechanism, determining which historical information should be retained or discarded from the cell state. As formulated in Equation (8), this gating operation employs a sigmoid activation to process both current inputs and previous hidden states:

$$\begin{array}{c}\hfill f_t=\sigma (W_f{x}_t+U_f{h}_{t-1}+b_f).\end{array}$$

(8)

By combining the results of the input gate and the forget gate, the cell state $c_t$ is updated by Equation (9):

$$\begin{array}{c}\hfill c_t=i_t\times {\tilde{c}}_t+f_t\times c_{t-1}.\end{array}$$

(9)

Output gate ($o_t$): The output gate serves as the final information filter in the LSTM architecture, selectively determining which aspects of the processed cell state should be propagated as the hidden state to subsequent time steps. This gating mechanism employs a sigmoid activation followed by a tanh transformation, as formalized in Equation (10):

$$\begin{array}{c}\hfill o_t=\sigma (W_0{x}_t+U_0{h}_{t-1}+b_0).\end{array}$$

(10)

2.5. GRU Model

The GRU represents a streamlined variant of LSTM networks that maintains comparable performance through reduced complexity. While LSTM employs three gating mechanisms (input, forget, and output gates) and separate cell/hidden states (Equations (6)–(9)), GRU utilizes just two gates, update ($z_t$, Equation (13)) and reset ($r_t$, Equation (11)) with a unified hidden state (${\tilde{h}}_t$ Equation (12)). This optimized architecture yields 30% fewer parameters than LSTM, enhancing computational efficiency while preserving temporal modeling capabilities, making it particularly advantageous for smaller datasets where parameter efficiency is crucial.

$$r_t=\sigma (W_r\times [h_{t-1},x_t]),$$

(11)

$${\tilde{h}}_t=tanh(W_h\times [r_t\times h_{t-1},x_t]),$$

(12)

$$z_t=\sigma (W_z\times [h_{t-1},x_t]).$$

(13)

3. Decomposition Effect of Carbon Emission

3.1. Carbon Emission Calculation of Shanghai Port

The “Carbon Emission Accounting Guidelines for Port Operating Enterprises” identify two primary carbon accounting methodologies: the fuel consumption method and the material balance method. While the material balance approach theoretically calculates emissions by comparing carbon content in input versus output materials, its practical application is limited by data availability challenges and computational complexity, often resulting in reduced accuracy. In contrast, the fuel consumption method has become the predominant approach in port emission studies because of its operational simplicity and the stable, well-defined emission sources typical of port facilities. This method directly calculates emissions by applying standardized emission factors to the measured consumption of energy fuels and electricity. Port carbon emissions originate from two distinct pathways: (1) direct emissions from fossil fuel combustion in operational equipment and (2) indirect emissions from purchased electricity used in auxiliary production processes, with the fuel consumption method effectively capturing both emission streams through its comprehensive factor-based calculation framework.

The calculation formula for carbon emissions from energy consumption is presented in Equation (14):

$$\begin{array}{c}\hfill E_1=\sum _i{D}_i\times C_i,\end{array}$$

(14)

where $E_1$ represents the direct carbon emissions generated by the port’s consumption of fossil fuels. $D_i$ is the consumption of the i-th type of fossil fuel. $C_i$ is the carbon emission factor of the i-th type of fossil fuel. According to the “IPCC Carbon Emission Calculation Guidelines (2006)” the method for calculating the carbon emission factor is presented as Equation (15):

$$\begin{array}{c}\hfill C_i=\sum _i{V}_i\times L_i\times F_i\times {\displaystyle \frac{44}{12}},\end{array}$$

(15)

where $V_i$ represents the lower heating value of the i-th energy fuel. $L_i$ is the carbon content per unit of heating value for the i-th energy fuel. $F_i$ indicates the carbon oxidation rate. The molecular weight ratio of carbon dioxide to carbon is 44/12. The carbon emission coefficients of different fossil fuels are shown in

Table 1. The carbon emission coefficients of different fossil fuels.

Fossil Fuel	${\mathit{V}}_{\mathit{i}}$ (MJ/kg)	${\mathit{L}}_{\mathit{i}}$ (tC/GJ)	${\mathit{F}}_{\mathit{i}}$	SC	MC
Diesel	42.652	0.0202	98%	1.4571	3.0959
Fuel oil	41.816	0.0211	98%	1.4286	3.1705
Liquefied natural gas	51.498	0.0172	99%	1.8620	3.1829
Natural gas	38.931	0.0153	99%	1.3300	2.1622

SC: converted standard coal coefficient; MC: carbon emission coefficient.

.

Port operations consume substantial amounts of electricity. Although the port itself does not directly emit CO₂ during electricity use, the combustion of fossil fuels in power generation produces carbon emissions. Consequently, the port’s electricity consumption indirectly contributes to carbon emissions and must be included in emission inventories to comprehensively assess the port’s carbon footprint. The calculation method for carbon emissions from electricity consumption is as shown in Equation (16).

$$\begin{array}{c}\hfill E_2=\sum _i{D}_{ie}\times F_{ie},\end{array}$$

(16)

where $E_2$ represents the indirect carbon emissions generated by the electricity consumed at the port. $D_e$ represents the electricity consumption of type-i equipment. $F_e$ represents the carbon emission coefficient of electricity for type-i equipment. The power emission factor is selected from the “2023 Emission Reduction Project China Regional Power Grid Baseline Emission Factor”. The power emission factor of Shanghai Port is 7.921t CO₂/10,000 kWh.

The total carbon emissions of a port are equal to the sum of the direct and indirect carbon emissions generated within the scope of its production and operation and by its affiliated equipment, defined as Equation (17).

$$\begin{array}{c}\hfill E_3=E_1+E_2,\end{array}$$

(17)

where $E_3$ represents the total carbon emissions of Shanghai Port. $E_1$ represents the direct carbon emissions of Shanghai Port. $E_2$ represents the indirect carbon emissions of Shanghai Port.

Analysis of

Table 2. Carbon emissions from energy consumption in Shanghai Port (unit: ton).

Year	Diesel		Fuel Oil		Power		E
	${\mathit{E}}_{\mathbf{11}}$	Proportion	${\mathit{E}}_{\mathbf{12}}$	Proportion	${\mathit{E}}_{\mathbf{2}}$	Proportion
2009	338,174.44	46.65%	169,197.91	23.34%	217,574.03	30.01%	724,946.38
2011	383,433.40	48.21%	169,958.80	21.37%	241,923.18	30.42%	795,315.39
2012	402,897.33	47.07%	179,736.33	21.00%	273,234.90	31.92%	855,868.54
2013	393,383.63	45.98%	176,363.01	20.61%	285,829.29	33.41%	855,575.93
2014	378,504.73	44.72%	183,122.30	21.63%	284,847.08	33.65%	846,474.12
2015	372,511.07	42.33%	224,955.73	25.56%	282,637.12	32.11%	880,103.93
2016	342,911.17	39.93%	209,135.44	24.35%	306,796.17	35.72%	858,842.78
2017	317,481.45	38.21%	201,913.26	24.30%	311,453.72	37.49%	830,848.43
2018	275,141.92	35.41%	212,033.18	25.66%	321,758.94	38.94%	826,376.34
2019	267,968.72	32.95%	234,926.64	28.13%	324,935.26	38.91%	835,003.82
2020	292,584.22	32.73%	175,310.44	21.41%	375,423.72	45.86%	818,702.87
2021	233,396.80	22.83%	434,516.00	42.50%	354,464.75	34.67%	1,022,377.56
2022	230,520.71	20.77%	472,754.19	42.59%	406,616.61	36.64%	1,109,891.52
2023	232,629.02	20.21%	51,089.56	46.14%	387,313.14	33.65%	1,151,031.71
2024	241,953.87	19.74%	606,101.22	49.45%	377,522.78	30.80%	1,225,577.87

$E_{11}$ represents the carbon emissions from the consumption of diesel; $E_{12}$ represents the carbon emissions from the consumption of fuel oil.

and Figure 1 reveals three distinct phases in Shanghai Port’s carbon emission trajectory from 2009 to 2023. The initial growth phase (2009–2012) saw steady emission increases driven by port expansion, infrastructure development, and rising throughput. A transitional phase (2013–2019) followed the Shanghai International Port Group’s implementation of green port initiatives, yielding measurable reductions from 2014 through the deployment of energy-efficient equipment and operational optimizations. Most notably, the post-pandemic surge (2020–2023) exhibited rapid emission growth, primarily attributable to the following: (1) global trade recovery driving throughput to record levels; (2) increased vessel calls at this global hub port; and (3) the IMO 2020 sulfur regulations shifting bunker demand to low-sulfur fuels. This triphasic pattern demonstrates how operational scale, regulatory changes, and mitigation measures interact to shape port emission profiles.

Figure 2 presents the evolving composition of carbon emissions at Shanghai Port from 2009 to 2023, revealing three distinct trends: the proportion of diesel-related carbon emissions exhibited an overall decline from 46.3% to 19.7%, following an initial fluctuation period (2009–2014) where emissions peaked at 47.9% in 2010; the consistent post-2014 reduction demonstrates the effectiveness of Shanghai Port’s clean energy initiatives, particularly the replacement of diesel equipment with LNG and electric alternatives.

Fuel oil emissions displayed a V-shaped trajectory, reaching a low of 20.47% in 2012 before rebounding to 49.3% by 2023. This resurgence, particularly pronounced during the post-pandemic trade recovery (2020–2023), reflects the increased demand for large vessels, expanded transportation operations, and growth in bonded fuel services, establishing fuel oil as the port’s primary emission source.

Electricity’s emission share showed complex dynamics, peaking at 33.4% (2013) and 45.7% (2019) before declining post-2020. These fluctuations mirror the port’s transition from diesel to electricity through initiatives like LED lighting retrofits and equipment upgrades, while recent reductions highlight progress in electricity system optimization and clean energy adoption, despite the concurrent growth in fuel consumption due to increased trade volume.

3.2. Decomposition of Factors Influencing Carbon Emissions

3.2.1. Comparison of Decomposition Models

Port carbon emissions are determined by an integrated system of five key factors, each representing distinct dimensions of port operations and energy use: (1) carbon emission intensity (EF) measures CO₂ output per unit of economic or energy metric, serving as the core efficiency indicator; (2) fossil energy structure (EI) quantifies the carbon potential of the energy mix, where higher fossil fuel ratios directly elevate emission baselines; (3) energy intensity (ET) reflects the energy–GDP conversion efficiency, with lower values signaling better energy utilization; (4) economic intensity (TG) captures the energy dependence of value creation processes; and (5) operating income (G) represents the absolute scale of economic activity. These variables were systematically selected based on their theoretical significance, empirical measurability, and established relationships in the port decarbonization literature, forming a comprehensive analytical framework.

The STIRPAT model is typically expressed as shown in Equation (18):

$$\begin{array}{c}\hfill C=a\times E{F}^b\times E{I}^c\times E{T}^d\times T{G}^e\times G^f\times \epsilon ,\end{array}$$

(18)

where C represents the total carbon emissions of Shanghai Port. $EF$ represents the carbon emission intensity. $EI$ denotes energy structure. ET stands for the energy intensity. $TG$ represents the economic intensity, and G is the operating income. a is a scaling constant. b, c, d, e, and f are the regression coefficients, reflecting the intensity of the impact of each factor. $\epsilon$ is the random error term.

To enable coefficient estimation, we apply logarithmic transformation to Equation (18) and obtain the linearized specification shown in Equation (19), with the regression results presented in

Table 3. Regression coefficients of the STIRPAT model.

Variable	Coefficient	Standard Error	p Value	VIF
$lna$	9.864	0.266	0.001	—
$lnEF$	0.996	0.033	0.002	2.16
$lnEI$	−0.807	0.104	0.001	3.13
$lnET$	0.976	0.039	0.002	4.59
$lnTG$	−0.008	0.013	0.575	3.76
$lnG$	−1.087	0.187	0.001	4.44

:

$$\begin{array}{c}\hfill lnC=9.864+0.996lnEF-0.807lnEI+0.976lnET-0.008lnTG-1.087lnG.\end{array}$$

(19)

The model demonstrates excellent explanatory power, with an $R^2$ of 0.997, confirming that 99.7% of carbon emission variability is captured by the selected drivers. The overall model passed the significance test, and the individual variables $lnEF$, $lnEI$, $lnET$, and $lnG$ are all significant, suggesting that these are key factors influencing carbon emissions. $lnTG$ shows statistical insignificance, suggesting limited explanatory contribution. Diagnostic tests confirm model robustness: all variance inflation factors (VIFs) remain below 5.0, effectively ruling out multicollinearity concerns. The elasticity coefficients indicate that carbon emissions are the most responsive to operating income and carbon intensity, with 1% increases leading to 1.087% decreases and 0.996% increases in emissions, respectively, highlighting these as pivotal leverage points for emission mitigation strategies.

We decomposed the influencing factors of carbon emissions in Shanghai Port into the above five factors, and the formula for the influencing factors of carbon emissions of Shanghai Port using the LMDI decomposition model is presented in Equation (20):

$$\begin{array}{c}\hfill C_t=\sum _i{\displaystyle \frac{C_{t,i}}{E_{t,i}}}\times {\displaystyle \frac{E_{t,i}}{E_t}}\times {\displaystyle \frac{E_t}{T_t}}\times {\displaystyle \frac{T_t}{G_t}}\times G_t,\end{array}$$

(20)

where $C_t$ represents the total carbon emissions of Shanghai Port in the t-th period. $C_{t,i}$ represents the carbon emissions of the i-th type of energy consumption in the t-th period. $E_{t,i}$ represents the consumption of the i-th type of energy in the t-th period. $E_t$ is the total consumption of energy in the t-th period. $T_t$ represents the throughput of Shanghai Port in the t-th period. $G_t$ is the operating income of Shanghai Port in the t-th period.

Let $E{F}_{t,i}=C_{t,i}/E_{t,i}$ be the carbon emission intensity of the i-th type of energy in the t-th period and $E{I}_{t,i}=E_{t,i}/E_t$ represent the consumption proportion of the i-th energy source in the t-th period. $E{T}_t=E_t/T_t$ is the energy consumption per unit throughput. $T{G}_t=T_t/G_t$ represents economic intensity. Hence, Equation (20) can be rewritten as Equation (21):

$$\begin{array}{c}\hfill C_t=\sum _{i}E{F}_{t,i}\times E{I}_{t,i}\times E{T}_t\times T{G}_t\times G_t.\end{array}$$

(21)

A comparative analysis of model performance metrics in

Table 4. Comparison of the decomposition effects between LMDI model and STIRPAT model.

Moldes	SSE	${\mathit{R}}^2$	MAPE	p Value
				$\mathit{EF}$	$\mathit{EI}$	$\mathit{ET}$	$\mathit{TG}$	$\mathit{G}$
LMDI	10.08	0.9922	0.91%	0.000	0.000	0.000	0.000	0.000
STIRPAT	12.47	0.9863	1.71%	0.002	0.001	0.002	0.575	0.001

demonstrates the superior abilities of the LMDI approach in Shanghai Port’s carbon emission analysis: the LMDI model achieves a higher goodness-of-fit $R^2$, indicating stronger explanatory power for emission variations. This superiority is further confirmed by (1) SSE decreases of 20%, reflecting better variable fitting; and (2) a reduced mean absolute percentage error, demonstrating enhanced prediction accuracy. Crucially, while the STIRPAT model shows insignificant results for carbon intensity, all LMDI variables maintain statistical significance, providing complete factor decomposition. These comparative results establish LMDI as the preferred methodology for Shanghai Port’s emission analysis, offering more reliable factor decomposition, better-fitting predictions, and comprehensive explanatory abilities for policy-relevant emission drivers.

3.2.2. Decomposition Effect Based on LMDI Model

The LMDI model can be divided into additive and multiplicative decomposition methods based on different contexts; these methods are used to analyze the contributions of various effects. Most scholars use the additive decomposition method when studying the factors influencing carbon emission intensity, as this is intuitive and computationally simple. Therefore, this paper adopts the additive decomposition method to analyze the factors affecting port carbon emissions from 2010 to 2023. The total effect formula for Shanghai Port’s carbon emissions based on the LMDI additive decomposition method is shown in Equation (22):

$$\begin{array}{c}\hfill \mathrm{\Delta }{C}_t=\mathrm{\Delta }{C}_{t,EF}+\mathrm{\Delta }{C}_{t,EI}+\mathrm{\Delta }{C}_{t,ET}+\mathrm{\Delta }{C}_{t,TG}+\mathrm{\Delta }{C}_{t,G},\end{array}$$

(22)

where $\mathrm{\Delta }{C}_{t,EF}$ is the carbon emission intensity effect. $\mathrm{\Delta }{C}_{t,EI}$ is the energy structure effect. $\mathrm{\Delta }{C}_{t,ET}$ is the energy efficiency effect. $\mathrm{\Delta }{C}_{t,TG}$ is the economic intensity effect. $\mathrm{\Delta }{C}_{t,G}$ is the operating income effect. The specific expression is shown in Equations (23)–(27):

$$\begin{array}{cc}\hfill \mathrm{\Delta }{C}_{t,EF}=& \sum _i\omega (C_{t,i},C_{0,i})ln{\displaystyle \frac{E{F}_{t,i}}{E{F}_{0,i}}},\hfill \end{array}$$

(23)

$$\begin{array}{cc}\hfill \mathrm{\Delta }{C}_{t,EI}=& \sum _i\omega (C_{t,i},C_{0,i})ln{\displaystyle \frac{E{I}_{t,i}}{E{I}_{0,i}}},\hfill \end{array}$$

(24)

$$\begin{array}{cc}\hfill \mathrm{\Delta }{C}_{t,ET}=& \sum _i\omega (C_{t,i},C_{0,i})ln{\displaystyle \frac{E{T}_{t,i}}{E{T}_{0,i}}},\hfill \end{array}$$

(25)

$$\begin{array}{cc}\hfill \mathrm{\Delta }{C}_{t,TG}=& \sum _i\omega (C_{t,i},C_{0,i})ln{\displaystyle \frac{T{G}_{t,i}}{T{G}_{0,i}}},\hfill \end{array}$$

(26)

$$\begin{array}{cc}\hfill \mathrm{\Delta }{C}_{t,G}=& \sum _i\omega (C_{t,i},C_{0,i})ln{\displaystyle \frac{G_{t,i}}{G_{0,i}}},\hfill \end{array}$$

(27)

where $\omega (C_{t,i},C_{0,i})={\displaystyle \frac{C_{t,i}-C_{0,i}}{ln\left(C_{t,i}\right)-ln\left(C_{0,i}\right)}}$.

The LMDI decomposition analysis reveals distinct temporal patterns and varying magnitudes of influence among factors affecting Shanghai Port’s carbon emissions from 2009 to 2023 (

Table 5. The contributions of various influencing factors of carbon emissions in Shanghai Port.

Year	$\mathbf{\Delta }{\mathit{C}}_{\mathit{t},\mathit{EF}}$	$\mathbf{\Delta }{\mathit{C}}_{\mathit{t},\mathit{EI}}$	$\mathbf{\Delta }{\mathit{C}}_{\mathit{t},\mathit{ET}}$	$\mathbf{\Delta }{\mathit{C}}_{\mathit{t},\mathit{TG}}$	$\mathbf{\Delta }{\mathit{C}}_{\mathit{t},\mathit{G}}$
2010	0.1808	2.6846	−6.5153	1.2017	5.0236
2011	1.0489	−0.9203	−0.3906	5.0375	6.5586
2012	1.0833	−0.9323	−0.5087	4.9687	−1.6379
2013	0.2298	−2.1413	4.3406	0.847	1.9027
2014	−0.9664	−5.7072	13.4305	−5.0017	−22.4497
2015	2.7008	−2.3716	−1.5131	11.9579	−1.7695
2016	1.3401	−2.3696	1.1707	5.4124	−4.4131
2017	1.1645	−5.1553	5.6639	4.3089	−34.8694
2018	0.1048	−5.8637	7.7591	0.0581	9.7557
2019	5.1184	4.5643	−17.454	18.6903	6.7864
2020	−8.4411	−41.472	54.4045	−34.0751	3.0752
2021	1.8957	−8.1769	2.1313	7.77	−54.7919
2022	−2.8049	−5.8899	6.2331	−12.4166	−23.4176
2023	−2.7465	−5.5324	5.4932	−13.2425	42.6094
Cumulative effect	−0.0918	−79.2836	74.2452	−4.4834	−67.6375
Cumulative CR	−0.12%	−102.63%	96.11%	−5.80%	−87.56%

The unit of cumulative effect is ten thousand tons. CR: contribution rate.

, Figure 3). The factor contributions, ranked by absolute cumulative impact, demonstrate the following: (1) energy structure effect emerges as the dominant mitigation factor, with consistent negative values indicating successful clean energy adoption; (2) the energy intensity effect shows volatile positive contributions, reflecting periodic efficiency challenges; (3) the production income effect delivers stable annual reductions through industrial upgrading; (4) the economic intensity effect transitions from positive to negative, marking a structural shift to low-carbon growth; and (5) the carbon intensity effect exhibits high variability but net reduction, underscoring the importance of sustained low-carbon technology deployment. These findings collectively demonstrate that operational modernization has become the primary driver of emission reductions, offsetting the carbon-increasing effects of economic expansion. The decomposition results quantitatively validate Shanghai Port’s transition pathway from carbon-intensive growth to sustainable port operations.

3.3. Carbon Emission Decoupling Effect

Carbon emission decoupling analysis quantitatively evaluates the relationship between economic growth and environmental impacts at Shanghai Port. This approach measures the decoupling elasticity index $\epsilon$ to identify three primary states—decoupling, coupling, and negative decoupling—which are further classified into eight specific statuses; see

Table 6. Classification of Tapio decoupled status.

Classification	Decoupled Status	$\mathbf{\Delta }\mathit{C}$	$\mathbf{\Delta }\mathit{G}$	$\mathbf{\Delta }\mathit{\epsilon }$	$\mathit{\epsilon }$
Connection	Expansion connection (I)	>0	>0	>0	$0.8<\epsilon <1.2$
	Recessionary connection (II)	<0	<0	<0	$0.8<\epsilon <1.2$
Decoupling	Strong decoupling (III)	<0	>0	>0	$\epsilon <0$
	Weak decoupling (IV)	<0	<0	<0	$0<\epsilon <0.8$
	Recessionary decoupling (V)	<0	<0	<0	$\epsilon >1.2$
Negative decoupling	Expansive negative decoupling (VI)	>0	>0	>0	$\epsilon >1.2$
	Strong negative decoupling (VII)	>0	<0	<0	$\epsilon <0$
	Weak negative decoupling (VIII)	<0	<0	<0	$0<\epsilon <0.8$

. The analysis determines whether economic expansion occurs alongside proportional, reduced, or increased emissions, providing critical insights into the port’s sustainable development progress. By examining these decoupling states, policymakers can assess the effectiveness of low-carbon strategies and identify areas needing improvement in Shanghai Port’s green transition.

To quantify the economic drivers of decoupling at Shanghai Port, three key indicators are analyzed: revenue, total assets, and port cargo throughput capacity. The Tapio decoupling model calculates the elasticity indices for each variable, expressed as Equation (28):

$$\begin{array}{c}\hfill {\epsilon }_t={\displaystyle \frac{\mathrm{\Delta }C}{\mathrm{\Delta }T}}+{\displaystyle \frac{\mathrm{\Delta }C}{\mathrm{\Delta }G}}+{\displaystyle \frac{\mathrm{\Delta }C}{\mathrm{\Delta }A}}:={\epsilon }_T+{\epsilon }_G+{\epsilon }_A,\end{array}$$

(28)

where $\mathrm{\Delta }C=(C_t-C_0)/C_0$ is the rate of change in carbon emissions, $\mathrm{\Delta }G=(G_t-G_0)/G_0$ is the rate of change in revenue, and $\mathrm{\Delta }T=(T_t-T_0)/T_0$ is the rate of change in port cargo throughput capacity. $\mathrm{\Delta }A=(A_t-A_0)/A_0$ is the rate of change in total assets. ${\epsilon }_T$ is the Tapio decoupling index of port carbon emissions and port cargo throughput capacity. ${\epsilon }_G$ is the decoupling index of carbon emissions and operating income of the port. ${\epsilon }_A$ is the decoupling index of the port’s carbon emissions from its total assets.

The decoupling elasticity index between Shanghai Port’s carbon emissions and its economic indicators was calculated using the Tapio decoupling model; see

Table 7. The decoupling elasticity index between the carbon emissions and economic indicators.

Year	$\mathbf{\Delta }\mathit{C}/\mathbf{\Delta }\mathit{T}$		$\mathbf{\Delta }\mathit{C}/\mathbf{\Delta }\mathit{G}$		$\mathbf{\Delta }\mathit{C}/\mathbf{\Delta }\mathit{A}$
	${\mathit{\epsilon }}_{\mathit{T}}$	Status	${\mathit{\epsilon }}_{\mathit{G}}$	Status	${\mathit{\epsilon }}_{\mathit{A}}$	Status
2010–2011	0.5426	IV	0.2912	I	0.5804	IV
2011–2012	−0.0017	III	−0.0123	III	−0.0138	III
2012–2013	1.4042	V	−0.6211	III	−0.1327	III
2013–2014	1.7955	VI	0.6213	I	−5.3400	VII
2014–2015	−0.9643	III	−0.5506	III	0.5085	VIII
2015–2016	−0.5247	III	−0.1819	III	−16.8559	III
2016–2017	−0.0301	III	−0.0340	III	−0.0637	III
2017–2018	0.6075	IV	0.4651	I	−0.0503	III
2018–2019	0.3833	VIII	1.2926	V	0.4770	VIII
2019–2020	−0.8926	VII	2.5537	VI	−4.7425	VII
2020–2021	0.2725	IV	0.9031	I	0.8872	I
2021–2022	0.4239	IV	0.5767	I	−0.4593	VII
2022–2023	0.7657	IV	0.5489	I	0.6650	IV

. The results reveal the following trends.

The relationship between the operating income and carbon emissions of Shanghai Port generally exhibits a “strong-weak” decoupling pattern. From 2010 to 2012, carbon emissions grew at a slower rate than operating income, indicating decoupling. However, between 2013 and 2014, the rapid expansion of import and export trade led to a surge in carbon emissions, resulting in negative decoupling. Following the implementation of the “Green Port Three-Year Action Plan (2015–2017)”, decoupling was re-established. However, another phase of negative decoupling emerged from 2018 to 2020, where operating income saw modest growth while carbon emissions rose sharply, reflecting excessive energy consumption and a high-carbon economic model. Around 2020, the global pandemic caused a sharp decline in trade, reducing operating income, yet carbon emissions continued to steadily increase. After 2021, as port operations normalized, carbon emissions increased at a slower pace than operating income, signaling a shift toward low-carbon development and effective emissions control.

The decoupling elasticity between Shanghai Port’s total assets and carbon emissions exhibits significant fluctuations, with periodic rises and declines reflecting year-to-year variability in their relationship. While certain years demonstrate negative decoupling or coupling, the overarching trend reveals a gradual shift toward decoupling. Notably, from 2010 to 2011, carbon emissions grew at a much faster rate than total assets, signaling a failure to achieve green growth and underscoring substantial challenges in emission control. Between 2012 and 2017, however, carbon emissions declined despite asset growth, marking notable progress in decarbonization and the increasing efficacy of emission-reduction policies. Post-2020, as the economy recovered from the pandemic, carbon emissions increased sharply alongside rapid asset expansion. This surge in emissions occurred at an even higher rate, highlighting the tension between economic stimulus and environmental sustainability. This trend emphasizes the critical need for port managers to prioritize green development strategies during recovery efforts, ensuring that economic growth and environmental protection are pursued in tandem rather than at cross-purposes.

The decoupling elasticity between Shanghai Port’s cargo throughput capacity and carbon emissions demonstrates significant cyclical fluctuations, following an alternating pattern of “decoupling–negative decoupling” phases. During 2010–2013, cargo throughput expanded by over 10% while carbon emissions remained stable, establishing clear decoupling. However, 2013–2015 and 2018–2022 witnessed negative decoupling, largely attributable to booming import–export trade that overwhelmed the port’s energy capacity. This operational strain forced increased reliance on road transport, elevating both transportation frequency and environmental pressure. Notably, even during the 2019–2022 throughput decline, carbon emissions persisted in rising, revealing the port’s sustained operational intensity during pandemic conditions while advanced emission-reduction technologies were gradually implemented. By 2023, throughput recovery to pre-pandemic levels, which was coupled with emissions growth falling below throughput expansion, achieved weak decoupling, demonstrating Shanghai Port’s evolving capacity to balance economic growth with environmental objectives through technological and operational improvements.

4. Scenario Prediction of Carbon Emissions

4.1. Prediction Comparison

4.1.1. Evaluation Criteria

In the process of model evaluation, multiple evaluation metrics are typically used to ensure a comprehensive and objective analysis of the model’s predictive ability. These metrics can measure the deviation between the model’s predicted values and the actual observations from different perspectives, helping researchers assess the accuracy and reliability of the model. Let $y_t$ represent the actual value, ${\widehat{y}}_t$ represent the predicted value, and n be the number of samples.

Mean Square Error (MSE): The mean squared error (MSE) measures the discrepancy between the model’s predicted values and the actual observations. The calculation is defined as follows:

$$\mathrm{MSE}={\displaystyle \frac{1}{n}}\sum _{t=1}^n{(y_t-{\widehat{y}}_t)}^2.$$

Mean Absolute Percentage Error (MAPE): The MAPE calculates the percentage of the prediction error for each sample point relative to the actual value, then averages the percentage errors across all sample points. The formula for MAPE is as follows:

$$\mathrm{MAPE}={\displaystyle \frac{1}{n}}\sum _{t=1}^n{\displaystyle \frac{y_t-{\widehat{y}}_t}{y_t}}\times 100\%.$$

Mean Absolute Error (MAE): The MAE is the average of the absolute differences between the predicted values and the actual values. The calculation formula is as follows:

$$\mathrm{MAE}={\displaystyle \frac{1}{n}}\sum _{t=1}^n|y_t-{\widehat{y}}_t|.$$

Coefficient of Determination $R^2$: $R^2$ is a statistical measure that explains how well the independent predictors in a regression model can predict the dependent variable. It is calculated as follows:

$$R^2=1-{\displaystyle \frac{S{S}_{error}}{S{S}_{total}}},$$

where $S{S}_{error}$ is the sum of the squared residuals. $S{S}_{total}$ is the total sum of squares.

4.1.2. Prediction Results of Single Models

This study employs four modeling approaches—ARIMA, SVR, LSTM, and GRU—for predicting Shanghai Port’s carbon emissions, selected based on their respective strengths in time series modeling, nonlinear relationship capture, and prediction stability. Each model underwent rigorous hyperparameter optimization to enhance predictive accuracy.

ARIMA modeling: The Augmented Dickey–Fuller (ADF) test confirmed non-stationarity in the original carbon emission time series. First-order differencing successfully transformed the series into a stationary process. Through a comprehensive analysis of autocorrelation (ACF) and partial autocorrelation (PACF) functions, coupled with AIC and BIC evaluations, the optimal specification was determined as ARIMA(1,1,1).

SVR: Utilizing a radial basis function (RBF) kernel, our SVR implementation effectively handles nonlinear relationships through implicit high-dimensional space mapping. After grid search optimization, we established $C=100$ (regularization parameter controlling error tolerance) and $\gamma =0.1$ (kernel coefficient determining data point influence scope). The selected $\gamma$ value ensures appropriate model smoothness while maintaining pattern recognition capabilities.

LSTM network: The architecture incorporates 64 units per LSTM layer, providing sufficient complexity to learn the temporal dependencies in emission patterns. The model underwent 500 training epochs with adaptive moment estimation (Adam) optimization, achieving convergence through iterative weight updates that minimized the prediction error function.

GRU network: While maintaining identical parameterization to LSTM (64 units, 500 epochs), GRU’s simplified architecture, featuring only update and reset gates, demonstrates computational advantages. This streamlined design reduces training time by approximately 30% compared to LSTM while maintaining comparable predictive performance, as evidenced by our validation experiments.

As evidenced by the evaluation metrics presented in

Table 8. Prediction results of single models.

Indicators	ARIMA	LSTM	GRU	SVR
MSE	33.2406	0.6885	1.4384	10.384
MAE	5.0861	0.6369	1.1169	3.265
MAPE	0.08182	0.01176	0.03207	0.0429
$R^2$	0.8312	0.9965	0.9927	0.9015

and Figure 4, our comparative analysis reveals significant differences in predictive performance across the four models. The SVR model demonstrates superior performance to the ARIMA model, exhibiting both lower prediction errors and a higher coefficient of determination $R^2$, confirming that machine learning approaches outperform traditional time series methods in capturing the complex patterns of port carbon emissions.

More notably, the deep learning models LSTM and GRU achieve even better results, with prediction errors lower than SVR and $R^2$ values exceeding 0.95. This performance gap suggests that as dataset complexity increases, the limited capacity of kernel-based methods like SVR becomes apparent, while recurrent neural networks maintain their advantage in handling nonlinear temporal dependencies.

Interestingly, while both LSTM and GRU show excellent performance, the LSTM model marginally outperforms its GRU counterpart, with lower prediction errors values and a slightly higher $R^2$. This difference, though modest, indicates that LSTM’s more complex gating mechanism provides a better handling of long-term dependencies in our emission prediction task, despite GRU’s theoretical advantage in computational efficiency.

4.1.3. Prediction Results of Hybrid Models

Shanghai Port’s carbon emissions exhibit complex dynamics shaped by both persistent structural drivers and transient market forces. Long-term factors such as carbon taxation policies and technological innovation interact with short-term variables including trade policy shifts, economic volatility, and climate anomalies. To capture these multi-scale temporal patterns, we developed a hybrid GRU-LSTM neural network that synergistically combines both architectures’ strengths. The LSTM component excels at modeling gradual, long-term trends through its sophisticated memory gating system, while the GRU module efficiently processes rapid fluctuations via its simplified gate structure. This dual approach achieves comprehensive temporal representation without compromising computational performance.

The GRU-LSTM hybrid prediction model was configured with the following parameter settings: Following feature selection and target variable identification, all input data underwent normalization to ensure consistent scales. Temporal dependencies were captured using a sliding window approach with a five-year lookback period, where historical data from five consecutive years served as the input to predict emissions for the subsequent year. The model architecture incorporates a GRU layer (128 units) to efficiently process short-term temporal patterns while reducing computational complexity, complemented by an LSTM layer (128 units) to extract long-term dependencies and enhance memory retention. Regularization was implemented through a dropout layer (rate = 0.15), with model optimization performed using the Adam optimizer (learning rate = 0.001) and mean squared error as the loss function. The utilized data were trained from 2009 to 2023, incorporating early stopping (patience = 20) to prevent overfitting while ensuring convergence. To demonstrate the predictive superiority of the GRU-LSTM architecture, comparative analyses were conducted against LSTM-SVR and GRU-SVR hybrid models, with detailed performance metrics provided in

Table 9. Prediction results of hybrid models.

Indicators	GRU-LSTM	LSTM-SVR	GRU-SVR
MSE	0.5352	2.5663	1.0354
MAE	0.6688	1.1753	0.7231
MAPE	0.0122	0.0352	0.0145
$R^2$	0.9973	0.987	0.9947

.

The comparative results presented in Table 9 and Figure 5 reveal distinct performance differences among the hybrid models: LSTM-SVR demonstrates the poorest performance with both the highest prediction error and lowest goodness-of-fit, attributable to SVR’s inability to effectively leverage the temporal features extracted by LSTM. While GRU-SVR shows improved accuracy over LSTM-SVR, confirming GRU’s superior feature extraction capability, it still underperforms relative to our proposed GRU-LSTM model, which achieves optimal prediction accuracy with the smallest error margins and highest $R^2$ values, thereby establishing itself as the most reliable framework for Shanghai Port’s carbon emission forecasting.

4.2. Scenario Prediction

4.2.1. Scenario Settings

Through scenario simulation analysis, this study identifies potential carbon peaking timelines for Shanghai Port, offering scientific support for achieving emission reduction targets. The results demonstrate that China’s port sector will undergo a phased decarbonization process: medium-term projections indicate a gradual decline in carbon emissions driven by the widespread adoption of clean energy technologies and comprehensive modernization of port infrastructure; long-term forecasts suggest substantial emission reductions through the rigorous implementation of green policies and optimization of whole-process carbon management measures, ultimately progressing toward carbon neutrality objectives.

Based on these projections, we establish three distinct development scenarios. (1) Baseline (BL) Scenario: Maintaining current technological levels and emission intensities, Shanghai Port’s carbon footprint exhibits inertial growth alongside economic expansion and throughput increases. (2) Low-Carbon (LC) Scenario: Incorporating energy structure optimization, shore power promotion, and intelligent technology adoption, this pathway achieves emission stabilization at reduced levels through coordinated government–enterprise efforts. (3) Enhanced Emission Reduction (ER) Scenario: Stringent carbon policies are implemented, clean energy deployment is accelerated, and energy systems are fundamentally restructured to enable early peaking and large emission cuts. The scenario design reflects progressively ambitious climate actions, from business-as-usual development to transformative decarbonization strategies, providing policymakers with a graduated framework for port emission management.

Table 10. The predefined values of economic indicators under the three scenarios.

Year	GDP			Emission Intensity			Energy Structure
	LC	ER	BL	LC	ER	BL	LC	ER	BL
2024	5.00	5.00	5.00	6.25	6.23	6.27	334,482.41	323,692.62	366,851.63
2025	5.40	5.49	5.44	6.01	5.98	6.05	349,181.52	309,757.76	382,973.23
2026	5.69	5.68	5.68	5.78	5.73	5.84	364,891.24	323,693.78	400,203.21
2027	5.98	5.97	5.98	5.56	5.49	5.64	356,991.35	326,216.22	387,766.45
2028	6.28	6.25	6.30	5.35	5.27	5.43	373,578.26	341,373.23	405,783.278
2029	6.60	6.55	6.64	5.15	5.07	5.25	391,110.57	357,394.11	424,826.96
2030	6.90	6.85	6.99	4.95	4.88	5.07	409,588.23	353,093.28	459,021.26
2031	7.22	7.16	7.34	4.78	4.69	4.90	429,011.31	369,837.28	429,011.25
2032	7.54	7.46	7.71	4.61	4.52	4.75	410,640.02	387,396.27	449,379.67
2033	7.88	7.78	8.09	4.45	4.34	4.59	430,116.57	389,539.42	470,693.47
2034	8.23	8.12	8.49	4.29	4.18	4.45	450,456.78	407,960.81	492,952.65
2035	8.76	8.52	8.92	4.14	4.03	4.31	471,660.92	427,164.58	516,157.21

presents the predefined economic indicators for the three scenarios.

1.

GDP

As China’s most dynamic economic hub, Shanghai continues to lead national growth with a 5.0% year-on-year GDP increase in 2024 (Shanghai Statistical Bureau, 2025), prompting our scenario-specific projections: the Business-as-Usual scenario adopts progressively moderated growth rates of 5.3% (2025–2030) and 5.0% (2030–2035), reflecting sustained economic momentum; the Policy-Controlled scenario incorporates climate-conscious adjustments with 5.0% (2025–2030) and 4.7% (2030–2035) growth rates; while the ambitious Emission-Constrained scenario implements 4.7% (2025–2030) and 4.3% (2030–2035) growth rates, systematically balancing economic development with decarbonization requirements through this tiered projection framework.

2.

Container cargo throughput capacity

As a crucial metric for assessing port-related carbon emissions, container cargo throughput capacity demonstrates a strong positive correlation with emission levels. Official planning documents establish clear benchmarks for Shanghai Port’s development trajectory: “the 14th Five-Year Plan for Shanghai’s Comprehensive Transportation Development” mandates a minimum throughput of 47 million TEUs by 2025, while the Shanghai Port Master Plan projects significant growth to 65 million TEUs by 2035 and 75 million TEUs by 2050. These established container cargo throughput capacity targets serve as fundamental economic indicators for evaluating the port’s future development and its corresponding carbon emission pathways.

3.

Carbon emission intensity

China’s port decarbonization policies have established progressively stringent targets, with the “13th Five-Year Plan” mandating 2% reductions in both energy consumption and CO₂ emissions per unit throughput from 2020 levels, followed by the 14th Five-Year Plan’s additional 3.5% reduction requirement for ship-related emissions. Shanghai Port has demonstrated significant compliance, achieving a 120,000-ton absolute emission reduction and a 34.7% intensity decline by 2023. Building on these policy benchmarks and empirical results, this study establishes scenario-specific decarbonization pathways: the baseline scenario adopts conservative reduction rates of 3.5% (2025–2030) and 3.2% (2030–2035); the low-carbon scenario implements moderately accelerated targets of 3.8% and 3.5%; and the intensified mitigation scenario pursues aggressive 4.1% and 3.8% reductions, systematically reflecting the trade-offs between operational continuity and climate ambition through this graduated framework.

4.

Energy structure

According to data from the Ministry of Transport, in the past year, the green and smart transformation of China’s ports has further accelerated. The energy structure of ports, which was mainly based on fossil fuels such as diesel, fuel oil, and coal, is gradually transitioning towards electrification and renewable energy. During the “12th Five-Year Plan” period, the proportion of coal and other fossil fuels in the energy consumption structure decreased by 5.2%. According to the “Energy Conservation and Carbon Reduction Action Plan for 2024–2025”, it is expected that the proportion of non-fossil energy will reach about 20% by 2025, and the target is to achieve 25% by 2030 and further increase this figure. In Shanghai Port, over the past five years, diesel and fuel oil accounted for 65% of the total energy consumption, electricity accounted for 16%, and new energy (such as photovoltaic, wind energy, and LNG) accounted for 14%. In the future energy consumption scenarios, under the baseline scenario, the proportion of fossil energy consumption will be 63% from 2025 to 2030, 58% from 2030 to 2035, and 53% from 2035 to 2040; under the low-carbon scenario, it will be 58% from 2025 to 2030 and 53% from 2030 to 2035; under the enhanced low-carbon scenario, it will be 53% from 2025 to 2030 and 48% from 2030 to 2035.

4.2.2. Results of Scenario Prediction

This study employs a comprehensive set of input variables, including GDP growth, container cargo throughput capacity, energy consumption patterns, and carbon emission intensity, across three distinct development scenarios (baseline, low-carbon, and enhanced emission reduction) to forecast Shanghai Port’s carbon peaking trajectory through our optimized GRU-LSTM hybrid model. The multi-scenario analysis, spanning 2025–2035, systematically evaluates how varying levels of policy intervention and technological adoption may influence the port’s decarbonization pathway. As illustrated in Figure 6, the model’s predictions reveal critical inflection points and scenario-dependent peaking timelines, providing valuable insights for port authorities to align operational growth with climate targets. The integration of economic, operational, and environmental indicators through this advanced neural network architecture offers a robust analytical framework for strategic decision-making in port sustainability management.

An analysis of Figure 6 and

Table 11. Prediction results of carbon emission for Shanghai Port under three scenarios.

Scenarios	2024	2025	2026	2027	2028	2029	2030	2031	2032	2033	2034	2035
LC	130.78	134.45	136.03	133.68	130.05	129.12	122.00	121.65	122.03	120.92	119.05	118.76
ER	124.21	126.76	125.46	122.32	120.87	117.98	114.32	111.89	109.59	110.00	108.89	106.78
BL	135.26	138.32	140.43	144.54	143.27	141.65	140.58	145.16	145.98	140.32	138.75	137.64

reveals that Shanghai Port’s carbon emissions follow stable growth trajectories across all three scenarios during the initial phase. However, under the combined influence of policy interventions, regulatory frameworks, and technological advancements, these emission pathways begin exhibiting marked divergence over time, ultimately developing distinct directional trends that reflect each scenario’s underlying assumptions.

Baseline Scenario: Sustained high-emission growth trajectory. The baseline scenario projects continuous rapid growth in carbon emissions from 2024 onward, with only a marginal reduction anticipated by 2035. This pathway assumes robust economic expansion largely decoupled from stringent low-carbon policies, where increasing cargo throughput and container demand drive persistent growth in fossil fuel consumption. The energy mix remains predominantly conventional due to limited policy intervention, with clean energy adoption failing to achieve critical mass. While incremental technological improvements yield modest efficiency gains, the resulting emission intensity reductions prove insufficient to offset absolute emission growth. However, the synergistic effects of strengthened policies and declining costs of renewable energy collectively constrain the growth of carbon emissions. The baseline scenario, predicated on the continuation of current trends, results in a progressive deceleration of emission growth rates, ultimately reaching equilibrium by or after 2035.

Low-Carbon Scenario: Stabilized emission trajectory. The low-carbon scenario projects a gradual emission decline post-2023, with peaking anticipated by 2026 and stabilization at approximately 1.2 million tons through 2035. This pathway reflects balanced policy intervention, where moderated GDP growth and optimized container cargo throughput capacity combine with energy structure improvements, shore power adoption, and clean energy deployment to achieve emission stabilization. While effectively preventing further emission increases and maintaining operational growth, this scenario’s moderate measures limit the short-term reduction potential. The projected 2028 peak and subsequent plateau demonstrate compliance with medium-term “Five-Year Plan” targets, showcasing achievable decarbonization through current policy frameworks without radical economic disruption.

Enhanced Emission Reduction Scenario: Peak and declining trend. This scenario assumes that the immediate implementation of stringent emission-reduction policies and accelerated deployment of low-carbon technologies would rapidly suppress the emission growth trend in the short term. Consequently, port-related carbon emissions are projected to decline significantly after 2025, peak around 2026, and subsequently stabilize, ultimately being contained at approximately 1.1 million tons by 2035. This ambitious trajectory results from comprehensive multi-stakeholder efforts combining stringent government policies and societal engagement, implemented through a robust policy framework featuring elevated carbon taxation, rigorous emission standards, and binding energy consumption caps. These measures catalyze the rapid transformation of the energy infrastructure through large-scale clean energy deployment and the accelerated phase-out of conventional fuels, while parallel smart technology integration optimizes port logistics by minimizing energy-intensive handling and transportation processes. The synergistic effect of these interventions enables Shanghai Port to achieve early carbon peaking while sustaining operational efficiency, establishing this as the most effective model for reconciling port development with climate objectives through systemic energy transition and technological innovation.

5. Conclusions and Discussion

The carbon emissions of the Shanghai Port, as the largest port in China, serve as a critical indicator of both the energy consumption and environmental impact of the port industry, while also reflecting its profound influence on the green development of global supply chains. Against the backdrop of China’s “peak carbon and carbon neutrality” goals, the Shanghai Port occupies a pivotal position in the implementation of emission reduction policies. A comprehensive examination of the port’s current carbon emission status, influencing factors, and potential reduction pathways offers valuable insights for ports worldwide, facilitating the global shipping industry’s transition toward a low-carbon and sustainable future. This study leverages historical energy consumption data from the Shanghai Port to calculate its total carbon emissions, analyzing key characteristics and trends. By applying the LMDI model, the research decomposes the driving factors behind these emissions, focusing on five critical dimensions: carbon emission intensity, energy structure, energy efficiency, economic intensity, and operational revenue. Furthermore, the Tapio decoupling model is employed to assess the relationship between economic growth and carbon emissions, evaluating whether decoupling has been achieved and the extent of the dependence between these variables. Finally, a scenario analysis is conducted across three frameworks with the GRU-LSTM model used to forecast carbon peak timelines under each scenario, thereby informing future policy and mitigation strategies for the port.

An analysis of the Shanghai Port’s carbon emissions reveals distinct temporal patterns. Between 2009 and 2012, emissions grew steadily at an average rate of approximately 66,000 tons annually, whereas the period from 2020 to 2023 witnessed accelerated growth, driven largely by increased fuel oil consumption. The port’s energy-related emissions primarily originate from three sources: fuel oil, petroleum products, and electricity. Particularly, while diesel’s contribution to total emissions has consistently declined, emissions linked to fuel oil and electricity have exhibited sustained growth, with fuel oil remaining the dominant emission source throughout the study period. This evolving emission profile underscores the persistent challenges in transitioning the port’s energy systems toward cleaner alternatives.

The LMDI decomposition analysis identifies four key factors with divergent impacts on emissions. Energy structure, energy efficiency, and operating revenue exert positive driving effects on emission growth, whereas economic intensity demonstrates a significant inhibitory effect. A quantitative assessment of contribution rates reveals that operating revenue is the most substantial promoter of emission increases, while economic intensity plays the dominant role in emission suppression. Crucially, the combined growth effect from energy structure, efficiency, and revenue factors outweighs the reduction impact attributable to economic intensity, resulting in net positive carbon emission growth during the study period. These findings highlight the complex interplay between economic development and environmental pressures in port operations, suggesting that mitigating emissions requires addressing multiple interconnected drivers.

The Tapio decoupling analysis of the Shanghai Port’s economic and environmental relationship from 2010 to 2023 indicates an incomplete dissociation between growth indicators and carbon emissions. Operational revenue exhibits intermittent decoupling, alternating between strong and weak states, while cargo throughput displays more pronounced cyclical volatility, oscillating between decoupling and negative decoupling phases. These patterns suggest that neither operational revenue nor cargo throughput capacity has achieved sustained decoupling from carbon emissions, pointing to persistent challenges in reconciling port expansion with emission reduction objectives. The differential behavior of these economic indicators underscores the complexity of the port’s decarbonization process, where short-term improvements do not necessarily translate into long-term sustainability.

Scenario-based projections using the GRU-LSTM model reveal significant variations in the port’s carbon peaking timeline depending on policy interventions. Under the baseline scenario, emissions continue growing beyond 2035 without reaching a peak, whereas both the low-carbon and enhanced emission reduction scenarios facilitate peaking around 2026, with the latter achieving marginally earlier results through more aggressive decarbonization measures. These findings emphasize the critical role of comprehensive emission reduction strategies in aligning port development with China’s climate targets. Notably, intensified policy action could accelerate peak attainment by approximately nine years compared to unmitigated growth scenarios, demonstrating the transformative potential of proactive mitigation efforts.

Future research will focus on developing an advanced predictive framework that integrates artificial intelligence algorithms, big data analytics, and system dynamics modeling across multiple temporal and spatial scales to enhance the accuracy of carbon emission decoupling projections. This multidisciplinary approach will incorporate granular scenario simulations to quantify the dynamic interactions between policy interventions, technological innovations, and market fluctuations, complemented by a robust uncertainty quantification to strengthen model reliability. Additionally, the study will establish an integrated assessment methodology combining real-time monitoring data with longitudinal sustainability evaluations, ultimately constructing a sophisticated green transition model. This model will not only provide theoretical foundations and operational strategies for the Shanghai Port’s decarbonization but also contribute actionable insights for global port sustainability transformations, supporting the broader transition toward a low-carbon future in maritime logistics.

6. Future Mitigation Strategies

This study provides a theoretical foundation for emission control strategies and carbon peak targets at Shanghai Port. Considering the changes in the policy environment, technological advancements, and market dynamics, Shanghai Port must continuously drive improvements and innovations across multiple dimensions. By implementing these measures, Shanghai Port will not only establish a solid framework for its sustainable development but also contribute valuable knowledge to the global decarbonization efforts of ports.

In terms of carbon emission governance, regulatory agencies should implement a comprehensive carbon management framework that includes market mechanisms, such as carbon taxes and emission trading systems, to incentivize the adoption of green technologies. Industry-specific emission caps for terminal operations, logistics, and transportation should be established to control overall emissions, along with mandatory energy efficiency standards to optimize the energy use of port equipment and vehicles. The policy framework should combine strict control with proactive incentives: progressive carbon tax penalties for exceeding emission limits, alongside the introduction of a “green port” certification program that offers tax reductions and subsidies to compliant enterprises. This dual approach of regulatory pressure and economic incentives will accelerate the industry’s low-carbon transition while maintaining operational competitiveness.

To promote the use of clean and renewable energy, Shanghai Port should implement a multidimensional clean energy strategy. First, it should require all vessels docking for more than two hours to use shore power, incentivizing compliance through tiered electricity subsidies. Second, priority should be given to the installation of high-capacity shore power systems at major container and bulk cargo terminals, aiming for widespread coverage by 2030. Third, an integrated renewable energy infrastructure should be developed, including the installation of rooftop photovoltaic systems on all warehouse facilities, the construction of offshore wind farms in nearby seas, and the establishment of hydrogen refueling stations for port equipment. These measures aim to significantly increase the share of renewable energy while building a distributed energy network to reduce dependence on the grid. This transformation requires coordinated investment in smart microgrid technologies and energy storage systems to ensure reliability, with a phased elimination of fossil fuel equipment in line with the progress of renewable energy capacity expansion.

In promoting the construction of a smart port, Shanghai Port should leverage advanced technologies to implement an integrated smart port ecosystem. First, an AI-based scheduling system should be deployed, utilizing real-time AIS data and machine learning algorithms to optimize vessel berthing, cargo loading and unloading sequences, and truck routes, reducing equipment idle times. Second, a cognitive energy management platform should be established, with IoT sensors deployed across all electrical infrastructure to enable dynamic load-balancing and predictive maintenance, saving energy. Third, a neural network-based energy demand forecasting model should be developed, with a 48-h forecasting window to optimize grid scheduling accuracy. Fourth, a pilot project for automated electric truck fleets, combined with route optimization algorithms, should be launched to reduce empty container transport. This digital transformation, supported by 5G connectivity and edge computing infrastructure, aims to significantly reduce energy intensity by 2030, creating a replicable model for smart, low-carbon port operations.

Establishing a multidimensional probabilistic assessment framework that evaluates implementation likelihood across technological, policy, economic, and social acceptance dimensions. This integrated approach combines historical data analysis, expert elicitation, and computational modeling to conduct comprehensive assessments of low-carbon, baseline, and enhanced emission reduction scenarios. By quantifying uncertainties in each dimension through Monte Carlo simulations and decision tree analysis, we systematically evaluate scenario success probabilities while identifying critical drivers and potential risks, thereby providing robust scientific support for climate policy decision-making.