Trend Analysis of the Impact of Ecological Governance on Industrial Structural Upgrading under the Dual Carbon Target

Abstract

The Chinese government faces significant challenges in achieving the goals of carbon peaking and carbon neutrality (dual carbon targets), particularly in the realms of implementing a low-carbon economy and achieving ecological balance. In order to assist the Chinese government in formulating more effective ecological governance policies, this paper focuses on 288 cities in China and proposes a predictive model combining gray forecasting, Backpropagation Neural Network, and threshold effect testing to forecast yearly ecological governance intensity. Under the premise of examining the predictive effect, fixed effects testing and threshold regression analysis were conducted to assess the future intensity of ecological governance. The empirical research results reveal that the increasing intensity of future ecological governance has a promoting effect on China’s upgrading of industrial structure, but this effect gradually diminishes. On the contrary, there is significant potential for optimizing industry’s internal structure. Efforts should be directed towards intensified governance, emphasizing energy-saving and emission reduction in high-carbon industries, and promoting environmentally and economically beneficial models. Our research provides a widely applicable method for studying the trend of research as it pertains to government decision-making effectiveness and valuable insights for governments to make more informed decisions in the pursuit of sustainable development.

1. Introduction

With the intensification of global climate change and environmental issues, sustainability has become a critical agenda worldwide. In the context of China’s dual carbon targets, achieving a low-carbon economy and ecological balance has become an urgent necessity for Chinese governments and businesses [1]. Ecological governance aims to establish a virtuous cycle of economic development and environmental protection by protecting natural resources, reducing environmental pollution, and promoting the use of renewable energy. Industrial structure upgrading refers to the process of adjusting and optimizing the proportion and structure of industries in the national economy, aiming for more efficient, higher value-added, and sustainable development. Exploring the relationship between ecological governance and industrial structural upgrading reveals that the implementation of ecological governance measures can promote the upgrading and transformation of industrial structure. Two modes of the role of ecological governance in promoting industrial structural upgrading, namely overall upgrading and internal optimization, have been validated [2]. However, there is currently a severe lack of assessment regarding the effectiveness of implementing policies after the government adopts the principles and recommendations of ecological governance for industrial structural upgrading.

The relevant research on the impact of government decisions on industrial structure in ecological governance includes the continuous refinement of indicator system design and policy evaluation. For example, Zhang et al. [3] quantified the environmental policies in the Beijing-Tianjin-Hebei region using a series of environmental regulatory indicators and studied their spatial and temporal effects on industrial structure. Li et al. [4] proposed an evaluation framework to study the comprehensive process of urban ecological governance. Hou et al. [5] assessed the industrial ecological efficiency of resource-based cities by using the Super-SBM model and investigated the influence of environmental regulations and financial pressure on industrial ecological efficiency. Zhou et al. [6] conducted an eco-efficiency study of 48 cities in the Bohai Sea region using data envelopment analysis, with a specific focus on the impact of industrial structural upgrading, which takes spatial heterogeneity. Li et al. [7] established an environmental evaluation model using a comprehensive evaluation approach and validated the suitability of the evaluation method by combining it with BP neural networks. Although certain studies incorporate trends, they may fall short of meeting the necessary criteria to effectively validate government decision recommendations.

Scholars have also employed various methods to explore the mechanisms behind the impact of ecological governance on industrial structure. Wei et al. [8] empirically analyzed the effects of environmental regulations, synergistic effects of environmental regulations, and roundabout production on the transformation and upgrading of industrial structure using a panel threshold model. Chen et al. [9] and Liu et al. [10] studied the impact of informal environmental regulations and public environmental regulations, respectively, on industrial structure upgrading. Yu et al. [11] and Shi et al. [12] used dynamic spatial panel models and geographic and time-weighted regression (GTWR) models, respectively, to study the impact mechanism of environmental regulations on industrial capacity utilization, as well as the relationship between industrial structure, urban governance, and haze pollution. Du et al. [13] and Cai et al. [14] estimated the impact of environmental regulations on green technology innovation by using the partially linear functional coefficient panel model and the panel Poisson fixed effect model, and considered the economic level heterogeneity and industry heterogeneity, respectively. These studies explore the mechanisms behind the impact of ecological governance on industrial structure using historical data and provide development recommendations for the government based on their findings.

Many studies simultaneously explore and evaluate the mechanisms involved. Yu et al. [15] examined the regional variations in how different types of environmental regulations impact China’s industrial structural transformation. Wu et al. [16] and Song et al. [17] used the Differences-in-Differences method (DID) to explore the impact of a low-carbon city pilot on industrial structure upgrading and urban ecological efficiency and its mechanism. Zheng et al. [18] and Wan et al. [19] constructed a multi-stage dynamic difference model and a doubly robust DID model, respectively, to study the impact mechanism of ecological compensation on industrial structure upgrading. Zhang et al. [20] used a spatial panel model to discover the regional heterogeneity in reducing haze effects through enhanced environmental regulations on industrial structural upgrading in China. Zhu [21] and Xiang et al. [22] explored the relationship between industrial structure optimization and energy structure as well as technological innovation using the applied panel threshold model. Xie et al. [23] developed a dynamic panel model to study the impact and mechanisms of environmental regulations on low-carbon transition. Zhang et al. [24] used the multi-period difference method to analyze the synergistic governance effects and mechanisms of China’s carbon trading system. Most relevant studies are limited by the availability of data, resulting in a lack of timeliness. This paper aims to address the issue by predicting future trends.

Forecasting indicator data can not only fill in missing data but also explore future trends. Trend analysis, based on the implementation of government policies and changes in the economy, technology, and market, can predict the intensity and development path of future ecological governance and help governments formulate more actionable policies. In this paper, based on the construction of an indicator system for the impact of ecological governance on industrial structure, a combination of gray forecasting models (GM), Backpropagation Neural Network (BP Neural Network), and threshold models is utilized to achieve yearly predictions of the intensity of ecological governance. The BP Neural Network is employed for predicting the level of industrial structural upgrading, while the gray forecasting model is used for predicting control variables. By examining the predictive effectiveness and conducting fixed effects testing and threshold regression analysis on the future intensity of ecological governance, the study aims to shed light on the development path of ecological governance in the future.

The contributions of this paper are as follows:

• The paper proposes a method that combines gray forecasting models and neural networks to predict the level of industrial structural upgrading and introduces a threshold regression model for yearly prediction of the intensity of ecological governance. It exhibits strong applicability in studying the trend of government decision-making effectiveness.
• Based on the predicted data, fixed effects and threshold regression analyses were conducted on the ecological governance intensity, verifying the mechanisms through which future ecological governance affects the industrial structure.
• By examining the trends of ecological governance intensity and industrial structural transformation, the study reveals the development path of future ecological environmental management.

The structure of the article is as follows. Section 2 introduces the data sources and methods used. Section 3 presents an empirical analysis of the trends of the impact of ecological governance intensity on industrial structure for prefecture-level cities nationwide. Finally, Section 4 concludes the study and provides recommendations based on the research findings.

2. Data Sources and Methods

2.1. Data Source and Processing

The indicator system of the threshold regression model for the impact of ecosystem governance on industrial structure constructed in this paper is shown in

Table 1. Indicator system.

Indicator Category	Indicator Name
Interpreted variables	Industrial Structure Upgrading Index (TS)
	Industrial Low-Carbonization Index (CS)
Core explanatory variable	Ecological Governance Efforts (EG)
Control variables	Population
	Economy
	Finance
	Tec-Edu
	Business
	Salary

[25,26]. The core explanatory variable is the intensity of ecosystem governance. The dependent variables are the industrial structure upgrading index and the industrial decarbonization index. Control variables include population, economy, finance, science and technology, commerce, and wages, which are used to reflect the comprehensive level of each city.

The data for this study were sourced from the statistical yearbooks of various cities, the National Bureau of Statistics, China Economic and Social Big Data Platform, and Carbon Emission Accounts and Datasets (https://www.ceads.net/, accessed on 21 June 2023). Among them, the first three statistical platforms provide access to data on population, economy, fiscal revenue, and other variables for each city. The Carbon Emissions Account and Dataset is jointly developed by researchers from multiple domestic and international research institutions, offering access to data related to carbon emissions. The time span covered is from 2007 to 2021. There are two methods used for handling missing values. Interpolation fitting techniques are applied to variables with a large number of missing values, while mean imputation is used for the remaining variables. Nine cities with a significant amount of missing data were excluded. As a result, the final number of studied prefecture-level cities is 288.

The Industry Structure Upgrading Index (TS) is represented by the ratio of the tertiary industry and the secondary industry. Among them, the secondary industry, also referred to as the manufacturing industry or industrial sector, is primarily associated with the production of tangible goods and physical products. The tertiary industry, commonly known as the service sector, encompasses diverse economic activities focused on providing services and non-material products. Furthermore, considering that industrial activities are the primary source of carbon emissions, this study utilizes the industrial low-carbon index (CS) to reflect the extent of structural optimization within industries. It begins by categorizing industries into high-efficiency and low-efficiency sectors based on their carbon emission return coefficients and determining industry weights. Subsequently, CS is calculated by employing a weighted ratio of the number of enterprises in high-efficiency and low-efficiency industries.

The ecological governance intensity (EG) is calculated by synthesizing nine tertiary indicators, including governance investment, pollutant treatment rate, and other cause indicators, as well as carbon emissions, industrial pollutant emissions, and other result indicators, comprehensively reflecting the investment intensity and governance effect of regional ecological governance [2]. Based on the economic level of cities, the cities were classified into three categories: developed cities, moderately developed cities, and underdeveloped cities. In this paper, Class I represents developed cities, and a higher category number indicates a lower level of economic development. The intensity of ecosystem governance is calculated using the coefficient of variation method on multiple indicator data.

2.2. Methods

2.2.1. Prediction Methods

Based on the threshold regression model that calculates the optimal governance intensity range on a yearly basis, a method for calculating future ecological governance intensity has been defined. The prediction of control variables utilizes the gray prediction model, while the prediction of the dependent variable applies the BP neural network.

It is important to note that the prediction of both the core explanatory variable and the dependent variable is conducted on an annual basis. Once the ecological governance intensity for a specific year is predicted, the neural network is then employed to predict the level of industrial structure upgrading. The prediction for the ecological governance intensity of the following year is based on the predicted results of the industrial structure upgrading level, which is then subjected to a threshold effect test to determine the optimal intensity range. The overall flowchart of the prediction model in this study is shown in Figure 1.

1.

Core Explanatory Variable—EG

The ecological governance intensity and optimal range for the previous period were calculated using the coefficient of variation method [27]. However, the future intensity needs to be calculated using a new prediction method defined in this paper. The impact of ecological governance on industrial structure follows a pattern of initial suppression, subsequent promotion, and further suppression [28,29]. The core of predicting the EG lies in simulating the government’s future efforts in ecological governance, where the government’s ecological governance intensity should remain within a reasonable range. Therefore, it is assumed that the future variation in ecological governance intensity will gradually approach the optimal intensity range.

The growth coefficient of ecological governance intensity is defined as consisting of two components: the natural growth coefficient and the target growth coefficient. Cities that currently have governance intensity within the optimal range will maintain the natural growth rate. If the governance intensity is outside the optimal range, both the natural growth rate and the target growth rate to reach the optimal range need to be considered simultaneously.

Let the ecological governance intensity of the i-th city in the j-th year be denoted as ${\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}$, and the optimal governance intensity range be represented as $[{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{i}},{\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{i}}]$. The growth rate of ecological governance intensity for the i-th city in the j-th year is denoted as ${\mathrm{g}}_{\mathrm{i}\mathrm{j}}$, and ${\mathrm{g}}_{\mathrm{i}}^{\prime }$ is the average natural growth rate.

$${\mathrm{g}}_{\mathrm{i}}^{\prime }=\frac{{\sum }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{g}}_{\mathrm{i}\mathrm{j}}}{\mathrm{n}}$$

(1)

${\mathrm{g}}_{\mathrm{i}}^{-}$ is the target growth rate for the increase in governance intensity of the city to the lowest point of the range, and ${\mathrm{g}}_{\mathrm{i}}^{+}$ is the target growth rate for the decrease in governance intensity of the city to the lowest point of the range.

$${\mathrm{g}}_{\mathrm{i}\mathrm{j}}^{-}=\frac{{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{i}}}{{\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}}-1$$

(2)

$${\mathrm{g}}_{\mathrm{i}\mathrm{j}}^{+}=\frac{{\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{i}}}{{\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}}-1$$

(3)

${\mathrm{G}}_{\mathrm{i}\mathrm{j}}$ is the growth coefficient of the i-th city in the j-th year, obtained by weighting the natural growth rate and the target growth rate.

$${\mathrm{G}}_{\mathrm{i}\mathrm{j}}=\mathsf{\alpha }{\mathrm{g}}_{\mathrm{i}}^{\prime }+{\mathsf{\beta }}_1{\mathrm{g}}_{\mathrm{i}\mathrm{j}}^{-}+{\mathsf{\beta }}_2{\mathrm{g}}_{\mathrm{i}\mathrm{j}}^{+}$$

(4)

It is given that $\mathsf{\alpha }+{\mathsf{\beta }}_1+{\mathsf{\beta }}_2=1$, with the initial assignment of $\mathsf{\alpha }={\mathsf{\beta }}_1+{\mathsf{\beta }}_2=0.5$. ${\mathsf{\beta }}_1$ is the weight of the target growth rate for cities below the optimal range, and ${\mathsf{\beta }}_2$ is the weight of the target growth rate for cities above the optimal range. Since the current governance intensity of each city cannot be simultaneously below and above the optimal range, at most one of ${\mathsf{\beta }}_1$ and ${\mathsf{\beta }}_2$ can be greater than 0. When the current ecological governance intensity is below the optimal range, ${\mathsf{\beta }}_1>0$ and ${\mathsf{\beta }}_2=0$. When it is above the optimal range, ${\mathsf{\beta }}_1=0$ and ${\mathsf{\beta }}_2>0$. When it falls within the optimal range, ${\mathsf{\beta }}_1={\mathsf{\beta }}_2=0$. In this case, the growth coefficient is entirely determined by the natural growth rate.

${\mathrm{D}}_{\mathrm{i}\mathrm{j}}$ represents the minimum difference between the current ecological governance intensity value and the boundary of the optimal range. If it falls within the optimal range, ${\mathrm{D}}_{\mathrm{i}\mathrm{j}}$ is equal to 0. Normalize the values of ${\mathrm{D}}_{\mathrm{i}\mathrm{j}}$, representing the values below the optimal range and above the optimal range to [0, 0.5], denoted as ${\mathsf{\beta }}_1$ and ${\mathsf{\beta }}_2$.

$${\mathrm{d}}_{\mathrm{i}\mathrm{j}}=\left\{\begin{array}{c}{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{i}}-{\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}, {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}<{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{i}}\hfill \\ \hfill 0, {{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{i}}\le \mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}\le {\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{i}}\hfill \\ {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}-{\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{i}}, {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}}>{\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{i}}\hfill \end{array}\right.$$

(5)

Therefore, the ecological governance intensity of the i-th city in the (j+1)-th year is determined.

$${\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}+1}={\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{j}} \ast {\mathrm{G}}_{\mathrm{i}\mathrm{j}}$$

(6)

2.

Control Variables

In the evaluation framework established in this paper, the control variables include population size, economic level, fiscal level, education level, commercial economy, and human capital level. These indicators are relatively stable and do not exhibit significant fluctuations between adjacent years. They generally show a steady upward trend. The prediction of the control variables is conducted for each city, using the patterns of indicator changes over the past 15 years to forecast their future values. Gray prediction models [30] are used to forecast the control variables.

The commonly used gray prediction models include the first-order and second-order gray prediction models. The GM(1,1) model is established based on a first-order ordinary differential equation, which generally describes a monotonic changing process. For non-monotonic oscillating sequences or sequences with a saturated S-shaped pattern, the GM(2,1) model can be used. In this study, a combination of the first-order and second-order models is adopted. The ratio-to-expected-value test is used to determine if the indicator sequence is suitable for the GM(1,1). If it is not suitable, the GM(2,1) is chosen.

3.

Interpreted Variables—TS, CS

The changes in the indices of industrial structure upgrading and industrial decarbonization are influenced by both the core explanatory variables and control variables [2]. Therefore, it is not suitable to use a single time series forecasting model to predict the dependent variables. The BP neural network can construct a mapping network and adjust the weights through backpropagation, continuously minimizing the errors and training a stable and reliable model.

Based on the data from 2006 to 2021, using the core explanatory variables and control variables as inputs and the dependent variables as outputs, a BP neural network model is trained. Then, using the core explanatory variables and control variables from 2022 to 2025 as inputs, these variables are fed into the neural network model to obtain the predicted data for the dependent variables.

2.2.2. Panel Threshold Regression Model

In this paper, multiple 15-year panel threshold models were examined by shifting the years based on the predicted data.

Establish a single-threshold model and a double-threshold model for TS.

$$\begin{gathered} {\mathrm{T}\mathrm{S}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}<\mathsf{\gamma }\right)+{\mathsf{\beta }}_2\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\ge \mathsf{\gamma }\right)+{\mathsf{\beta }}_3\times \\ \ln \left({\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_4\times \ln \left({\mathrm{E}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_5\times \ln \left({\mathrm{F}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e}}_{\mathrm{i}\mathrm{t}}\right)+ \\ {\mathsf{\beta }}_6\times \mathrm{l}\mathrm{n}\left({\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{E}\mathrm{d}\mathrm{u}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_7\times \mathrm{l}\mathrm{n}\left({\mathrm{B}\mathrm{u}\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{e}\mathrm{s}\mathrm{s}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_8\times \mathrm{l}\mathrm{n}\left({\mathrm{S}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}} \end{gathered}$$

(7)

$$\begin{gathered} {\mathrm{T}\mathrm{S}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}<{\mathsf{\gamma }}_1\right)+{\mathsf{\beta }}_2\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathsf{\gamma }}_1<{\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}<{\mathsf{\gamma }}_2\right)+ \\ {\mathsf{\beta }}_3\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\ge {\mathsf{\gamma }}_2\right)+{\mathsf{\beta }}_4\times \ln \left({\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_5\times \ln \left({\mathrm{E}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+ \\ {\mathsf{\beta }}_6\times \ln \left({\mathrm{F}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_7\times \ln \left({\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{E}\mathrm{d}\mathrm{u}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_8\times \ln \left({\mathrm{B}\mathrm{u}\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{e}\mathrm{s}\mathrm{s}}_{\mathrm{i}\mathrm{t}}\right) \\ +{\mathsf{\beta }}_9\times \mathrm{l}\mathrm{n}\left({\mathrm{S}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}} \end{gathered}$$

(8)

Establish a single-threshold model and a double-threshold model for CS.

$$\begin{gathered} {\mathrm{C}\mathrm{S}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}<\mathsf{\gamma }\right)+{\mathsf{\beta }}_2\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\ge \mathsf{\gamma }\right)+ \\ {\mathsf{\beta }}_3\times \ln \left({\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_4\times \ln \left({\mathrm{E}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_5\times \ln \left({\mathrm{F}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e}}_{\mathrm{i}\mathrm{t}}\right)+ \\ {\mathsf{\beta }}_6\times \mathrm{l}\mathrm{n}\left({\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{E}\mathrm{d}\mathrm{u}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_7\times \mathrm{l}\mathrm{n}\left({\mathrm{B}\mathrm{u}\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{e}\mathrm{s}\mathrm{s}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_8\times \mathrm{l}\mathrm{n}\left({\mathrm{S}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}} \end{gathered}$$

(9)

$$\begin{gathered} {\mathrm{C}\mathrm{S}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}<{\mathsf{\gamma }}_1\right)+{\mathsf{\beta }}_2\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathsf{\gamma }}_1<{\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}<{\mathsf{\gamma }}_2\right)+ \\ {\mathsf{\beta }}_3\times {\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\left({\mathrm{E}\mathrm{G}}_{\mathrm{i}\mathrm{t}}\ge {\mathsf{\gamma }}_2\right)+{\mathsf{\beta }}_4\times \ln \left({\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_5\times \ln \left({\mathrm{E}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+ \\ {\mathsf{\beta }}_6\times \ln \left({\mathrm{F}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_7\times \ln \left({\mathrm{T}\mathrm{e}\mathrm{c}\mathrm{E}\mathrm{d}\mathrm{u}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\beta }}_8\times \ln \left({\mathrm{B}\mathrm{u}\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{e}\mathrm{s}\mathrm{s}}_{\mathrm{i}\mathrm{t}}\right)+ \\ {\mathsf{\beta }}_9\times \mathrm{l}\mathrm{n}\left({\mathrm{S}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{y}}_{\mathrm{i}\mathrm{t}}\right)+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}} \end{gathered}$$

(10)

On the basis of selecting the appropriate threshold model, the optimal ranges of ecological governance intensity were calculated for each city and each year. Finally, the future ecological governance paths for different types of cities were analyzed.

3. Empirical Analysis

3.1. Forecasting Results

3.1.1. Forecasting Results of Core Explanatory Variable

Considering the accuracy of the forecasting model, this study used data from 2006 to 2021 as a basis to predict the indicator data for the next four years, specifically from 2022 to 2025. The predicted values of ecological governance intensity for each city were calculated using the growth coefficients defined in Section 2.2.1. The average intensity values for the years 2020 to 2025 are shown in Figure 2. The data for 2020 and 2021 represent the average values of the actual governance intensity, while the data for 2022 to 2025 represent the average values of the predicted results. The optimal range of ecological governance intensity, which needs to be noted, dynamically changes over time and is determined by the threshold values tested in subsequent contexts. The prediction of ecological governance intensity and the testing of threshold effects are conducted alternately. The information presented here represents the final predicted results.

From the national average perspective, the ecological governance intensity shows a clear upward trend and gradually increases toward the optimal range calculated by the threshold model. When considering city classes, Class I cities have already reached the optimal range of EG values in 2020–2021. Therefore, their changes in the next four years are not significant and mainly involve fluctuations around the current intensity value. On the other hand, Class II and Class III cities have intensity values in the first two years that are noticeably lower than the optimal range. Hence, their intensity will gradually strengthen and move toward the optimal range in the coming years.

3.1.2. Forecasting Results of Control Variables

The control variables were predicted using the GM(1,1) + GM(2,1) approach. The GM(1,1) model was suitable for most cities’ indicator data, while a few cities did not pass the ratio test and thus, the GM(2,1) model was selected for them. The model selection is shown in

Table 2. Selection of gray prediction models for three types of cities.

Variable	Class I City		Class II City		Class III City
	GM(1,1)	GM(2,1)	GM(1,1)	GM(2,1)	GM(1,1)	GM(2,1)
Population	281	7	286	2	288	0
Economy	288	0	288	0	288	0
Finance	286	2	288	0	288	0
Tec-Edu	286	2	287	1	287	1
Business	275	13	280	8	281	7
Salary	281	7	283	5	287	1

. According to the results in the table, the majority of cities that used the GM(2,1) model belonged to Class I. This is because the indicator data of Class I cities have higher absolute values compared to that of other cities, and their growth rates have slowed down in recent years, making the second-order gray prediction model more suitable. The average errors for each variable were all below 0.05, indicating good predictive performance.

3.1.3. Forecasting Results of Interpreted Variables

Based on the indicator data from 2006 to 2021, with the core explanatory variables and control variables as the inputs and the dependent variable as the output, the neural network model has seven nodes in the input layer and two nodes in the output layer. The number of nodes in the hidden layer is determined using the formula $\mathsf{\mu }=\sqrt{\mathrm{n}+\mathrm{m}}+\mathrm{a}$, where $\mathsf{\mu }$ is the number of nodes in the hidden layer, n is the number of nodes in the input layer, and m is the number of nodes in the output layer. The value of a is a number between 1 and 10. The total sample size in the model is 4320. The data from 2006 to 2019 are selected as the training set, while the data from 2020 and 2021 are used as the testing set. The prediction model for 2021 is retrained based on the predicted indicator data for 2020. The specific parameters for the neural network structure are shown in

Table 3. Neural network structure parameters.

Name	Parameter Settings
Number of input layer nodes	7
Number of hidden layer nodes	10
Number of output layer nodes	2
Input layer to hidden layer transfer function	Logsig function
Implicit layer to output layer transfer function	Purelin function
Activation function	Sigmoid function
Learning function	Learngdm function
Training function	Trainlm function
Training frequency	10000
Target accuracy	0.0001
Learning rate	0.15

.

The prediction performance for the years 2020 and 2021 was evaluated through a series of tests. The higher the value of the coefficient of determination (R-squared), the closer it is to 1, indicating a higher level of predictive effectiveness. A smaller mean squared error (MSE) and mean absolute error (MAE) also suggest better prediction accuracy. The results of these tests are presented in

Table 4. Evaluation of neural network prediction performance.

Variable	R2	MSE	MAE
TS	0.9474	0.0517	0.0064
CS	0.9907	0.0233	0.0142

. The MSE and MAE values are both less than 0.01, and R² values are greater than 0.9 for both years. These test results indicate that the predicted values have a small range of errors, showing a high level of agreement with the actual values and demonstrating good predictive performance.

The training results for the TS and CS are illustrated in Figure 3. It is important to note that the numerical values on the x-axis represent city identifiers and do not hold any specific meaning. From the figures, it can be observed that the model’s training performance is reliable, thus making it suitable for predicting indicator data for the years 2022 to 2025.

3.2. Threshold Regression Analysis Based on Predicted Data

Threshold analysis was conducted by moving one year at a time over a span of 15 years. The results of the threshold analysis from 2007 to 2021 were used as the optimal governance intensity and regression coefficients for the year 2021. This process was repeated for the years from 2022 to 2025. The Hausman test was performed on all cities for different years, and the results rejected the null hypothesis at a significance level of 1%. Therefore, fixed effects analysis was used. The threshold quantity and values for panel threshold regression analysis are presented in

Table 5. Threshold quantity and values from 2020 to 2025.

Year	TS			CS
	Number of Thresholds	Threshold	p Value	Number of Thresholds	Threshold	p Value
2020	Single threshold	1.0016	0.0800	Double threshold	0.8920	0.0500
					1.3332	0.0233
2021	Single threshold	1.0017	0.0900	Double threshold	0.9342	0.0600
					1.4119	0.0400
2022	Single threshold	0.9866	0.0914	Double threshold	0.9651	0.0500
					1.5513	0.0220
2023	Absent	—	—	Double threshold	1.1750	0.0533
					1.5464	0.0633
2024	Absent	—	—	Double threshold	1.1971	0.0433
					1.5446	0.0100
2025	Absent	—	—	Double threshold	1.2982	0.0667
					1.5758	0.0000

.

The threshold value for the Industrial Structure Upgrade Index shows very little change and remains stable around 1. From 2023 to 2025, there is no threshold effect of ecological governance on the overall upgrade of the industrial structure. However, there is a clear dual threshold effect of ecological governance on industrial decarbonization, and the threshold value gradually increases. This indicates that there is a significant potential for optimizing the internal structure of the industry, and it requires progressive strengthening of ecological governance to promote internal low-carbon development.

The panel threshold regression coefficients for the Industrial Structure Upgrade Index during 2020–2022 are shown in

Table 6. TS threshold regression coefficient.

Variable	2020	2021	2022
ln Population	−0.0517	0.0052	-0.0105
ln Economy	−0.2649 *** 1	−0.7712 ***	−0.8238 ***
ln Finance	−0.1286 ***	0.0297 **	0.1440 ***
ln Tec-Edu	−0.1905 *	0.0352	0.0232
ln Business	0.1237 ***	0.2628 ***	0.2471 ***
ln Salary	0.6457 ***	0.8059 ***	0.8322 ***
Cons	−2.1282 ***	−4.4183 ***	−5.6944 ***
$\mathrm{E}\mathrm{G}\le \mathsf{\gamma }$	−0.3967 **	−0.0202 *	−0.0164 **
$\mathrm{E}\mathrm{G}>\mathsf{\gamma }$	0.0769 ***	0.0580 ***	0.0418 ***

¹ ***, **, and * represent passing significance tests at the levels of 1%, 5%, and 10%, respectively.

. Starting from 2023, there is no significant threshold effect; hence, no regression coefficients are available. When the intensity of ecological governance exceeds the threshold value, it has a promoting effect on the upgrade of the industrial structure, but this effect gradually weakens. The regression coefficient for 2020 is 0.0769, which decreases to 0.0580 in 2021, and further decreases to 0.0418 in 2022. The reason for this change is that in 2022, the national average Industrial Structure Upgrade Index has already reached 1.9480, which means that the proportion of the tertiary industry is close to double that of the secondary industry. Therefore, it is difficult to continue the transition to the tertiary industry.

Factors that are positively correlated with industrial structure upgrading are the commercial economy and human capital level. The commercial economy is represented by the total retail sales of social consumer goods, which mainly includes wholesale and retail trade, accommodation, and catering industries, all of which belong to the tertiary industry. Therefore, the development of the commercial economy is inevitably positively correlated with the upgrade of the industrial structure. Compared to the secondary industry, the tertiary industry has a greater demand for human resources. As its share increases, the demand for various resources such as manpower and materials also increases. A large number of talent resources will flow into the tertiary industry. The impact of population and investment in science and education on the upgrade of the industrial structure is not significant.

The panel threshold regression coefficients for the Industrial Decarbonization Index during 2020–2022 are shown in

Table 7. CS threshold regression coefficient.

Variable	2020	2021	2022	2023	2024	2025
ln Population	0.2744 *** 1	0.2725 ***	0.2487 ***	0.1960 ***	0.2935 ***	0.2173 ***
ln Economy	−0.0179	0.0014	0.0200	0.0623 **	0.0786 ***	0.0764 ***
ln Finance	−0.1423 ***	−0.1535 ***	−0.1348 ***	−0.1231 ***	−0.0951 ***	−0.1106 ***
ln Tec−Edu	0.1941 ***	0.2448 ***	0.2584 ***	0.2653 ***	0.2472 ***	0.2705 ***
ln Business	0.0407 **	0.0260 **	0.0134	0.0131	0.0117	0.0375 *
ln Salary	0.119 ***	0.1028 ***	0.0866 ***	0.0602 ***	0.0591 **	0.0519 **
Cons	−2.4933 ***	−2.8389 ***	−3.0405 ***	−3.0029 ***	−3.8523 ***	−3.5374 ***
$\mathrm{E}\mathrm{G}\le {\mathsf{\gamma }}_1$	0.2976 **	0.0171 *	0.0089	−0.0165	−0.0377 *	−0.0539 ***
${\mathsf{\gamma }}_1<\mathrm{E}\mathrm{G}\le {\mathsf{\gamma }}_2$	0.0993 ***	0.0843 ***	0.0763 ***	0.0797 ***	0.0606 ***	0.0767 ***
$\mathrm{E}\mathrm{G}>{\mathsf{\gamma }}_2$	0.1203	−0.0317	−0.0603 *	−0.0984 *	−0.1435 ***	−0.2197 ***

¹ ***, **, and * represent passing significance tests at the levels of 1%, 5%, and 10%, respectively.

. As mentioned earlier, there is a dual threshold effect of ecological governance on the internal decarbonization structure of the industry, with the threshold value gradually increasing. The regression coefficients for EG indicate that the greatest promoting effect occurs when the governance intensity is between the two threshold values. Although the marginal benefits of internal decarbonization diminish, it is necessary to continue strengthening low-carbon governance in the industry. The commercial economy is mainly related to the tertiary industry, so its relationship with the industrial decarbonization index is not significant.

The optimal range of ecological governance intensity for the coming years can be determined based on the threshold values and regression coefficients of TS and CS. TS has threshold values in 2021–2022, and when the intensity exceeds the threshold value, it has a promoting effect on the upgrade of the industrial structure. On the other hand, CS exhibits a dual threshold effect, and when the intensity value falls between the two threshold values, it promotes the internal decarbonization of the industry.

The optimal intensity range for 2022 is [0.9866, 1.5513]. When EG falls within this optimal range, it has a promoting effect of 0.0418 on TS and 0.0763 on CS. The optimal intensity range for 2023 is [1.1750, 1.5464], and when EG falls within this range, it has a promoting effect of 0.0797 on CS and no significant threshold effect on TS. The optimal intensity range for 2024 is [1.1971, 1.5446], and when EG falls within this range, it has a promoting effect of 0.0606 on CS and no significant threshold effect on TS. The optimal intensity range for 2025 is [1.2982, 1.5758], and when EG falls within this range, it has a promoting effect of 0.0767 on CS and no significant threshold effect on TS. The trend of the optimal range changes is shown in Figure 4, and it shows an overall increasing trend.

From the overall results, the scope for the overall upgrading of the industrial structure in the coming years is relatively small. Ecological governance primarily promotes the internal decarbonization of industries, and the optimal range for ecological governance intensity increases year by year. This indicates that in the future, it is necessary to continuously increase governance efforts, focusing on the energy-saving and emission-reduction effects in high-carbon industries and the growth models with increased benefits. At the same time, it is important to strengthen the introduction of emerging technologies, attract high-end talents, and reduce energy consumption in various industries. Given the relatively fixed overall structure at present, efforts should be made to optimize the internal structure as much as possible.

4. Conclusions

This paper combines indicator data prediction and threshold regression models to empirically examine the threshold effect of future ecological governance intensity on the upgrading of industrial structure. It proposes a new predictive method to forecast the trends of ecological governance intensity and the level of industrial structural transformation. Based on the predicted data, a threshold regression analysis was conducted on the level of industrial structure upgrading, and the scientific validity of this method was empirically validated, providing some reference for the development path of future ecological environmental management.

The main conclusions are as follows. Firstly, ecological governance intensity has a promoting effect on the upgrading of industrial structure, but this effect gradually diminishes, and by 2023, there is no significant threshold effect. In other words, the overall space for industrial structure upgrading in the future is relatively small. Secondly, ecological governance intensity exhibits a clear dual threshold effect on industrial decarbonization, and the threshold values are gradually increasing. This indicates that there is significant room for optimizing the internal structure of the industry, and it requires progressively strengthening ecological governance to promote the development of internal decarbonization.

For the upgrading of industrial structure, the diminishing effect of ecological governance intensity indicates that the proportion of the tertiary industry is already much higher than that of the secondary industry. Excessively promoting the development of the tertiary industry would squeeze the resources and development space of the secondary industry, which is not conducive to overall coordinated development. Regarding the industrial low-carbon index, ecological governance exhibits a dual threshold effect, with the maximum promoting effect occurring when the governance intensity is between the two threshold values.

Taking the above conclusions into consideration, the following specific recommendations are proposed. (1) It is necessary to control the development of the commercial economy and actively attract high-end talents. (2) Industrial enterprises should actively strive to achieve low-carbon, environmentally friendly production, and research on development processes. This can be accomplished by introducing emerging technologies to reduce pollutant emissions and energy consumption while accelerating the transition from high-carbon industries to low-carbon industries, for example, by developing efficient energy storage systems, applying clean combustion technology and carbon capture technology. (3) The government should enforce policies that compel companies to enhance their pollution treatment capabilities, reduce carbon emissions and other pollutant discharges, and compel companies to seek green and sustainable development paths. This will be beneficial for the long-term development of companies and the entire industry.

The timeliness of previous research is no longer able to meet the needs of the current ecological governance field, and it is difficult to verify whether the proposed viewpoints align with practical needs. This article presents a new predictive method and studies the trends in ecological governance intensity and the level of industrial structural transformation to reveal future development paths and provide a reference for the development of ecological environmental management. This method should have broad applicability in studying the trend of government decision-making effectiveness.

In future research, further exploration can be conducted on the impact of ecological governance intensity on different industries and their mechanisms. This would help formulate targeted ecological governance policies to promote the sustainable development and structural upgrading of different industries. Cross-country comparisons can be made on the ecological governance intensity and industrial structural upgrading in different countries and regions, exploring the successful experiences and lessons learned in industrial structural upgrading from various countries and regions. Drawing on lessons from international experiences can provide insights and references for the industrial structural upgrading and ecological governance in our country. Future research should also focus on the influence of social participation and governance mechanisms on ecological governance intensity and industrial structural upgrading. Specifically, exploring the roles and mechanisms of public participation, corporate social responsibility, non-governmental organizations, and other stakeholders in ecological governance and industrial upgrading can facilitate the formation of a multi-stakeholder governance model.