Energy Policy 172 (2023) 113338

Contents lists available at ScienceDirect

Energy Policy
journal homepage: [Link]/locate/enpol

Green technology innovation and CO2 emission in China: Evidence from a

spatial-temporal analysis and a nonlinear spatial durbin model
Huanyu Chen a, Jizheng Yi a, *, Aibin Chen a, Duanxiang Peng a, Jieqiong Yang b
a
College of Computer and Information Engineering, Central South University of Forestry and Technology, Changsha, 410004, China
b
College of Life Science and Technology, Central South University of Forestry and Technology, Changsha, 410004, China

A R T I C L E I N F O A B S T R A C T

Keywords: Based on the panel data of 30 provinces in China from 2007 to 2019, this paper investigates the impact of green
Green technology innovation technology innovation on carbon intensity. Firstly, this paper studies the dynamic evolution and temporal and
Carbon intensity spatial pattern of China’s provincial green technology innovation and carbon intensity. On this basis, the
Spatial durbin model (SDM)
nonlinear spatial Durbin model (SDM) is used to explore the impact of green technology innovation on carbon
Direct effect
Spillover effect
intensity, and the relevant variables are controlled. The results show that there is a significant spatial agglom­
eration phenomenon in China’s provincial green technology innovation level and carbon intensity. The direct
impact of green technological innovation on carbon intensity in local region shows a significant "inverted-U"
relationship, that is, when the level of green technological innovation is relatively low, green technological
innovation will promote carbon emissions, while when the level of green technological innovation reaches a
certain level, this promotion relationship will change into inhibition relationship. From the perspective of
interregional spillover effect, green technological innovation also has an “inverted-U′′ nonlinear impact on
carbon emissions in adjacent regions, that is, the impact of green technological innovation on carbon emissions
in adjacent regions is also promoted first and then restrained.

1. Introduction the 75th session of the United Nations General Assembly. China will
increase its national contribution and take more powerful policy mea­
Climate change and environmental degradation are major problems sures to achieve carbon peak by 2030 and carbon neutrality by 2060.
facing the world today, which have posed a serious threat to the sus­ In order to make the Chinese government to realize its commitment
tainable development of human society (Muhammad, 2019). Many to achieve carbon neutrality by 2060, it is very important to reduce
studies have pointed out that carbon emissions are the culprit of climate carbon emissions from all walks of life in China. The key to achieving the
change and environmental degradation (Adedoyin et al., 2020; Farooq carbon emissions reduction target is to achieve the dynamic balance
et al., 2019; Li et al., 2021a). The increase of carbon emissions will not between economic development and carbon emissions. Since Grossman
only lead to global warming, which will lead to a series of natural di­ and Krueger (1991) first proposed the Environmental Kuznets hypoth­
sasters such as glacier melting, flood, drought and sea-level rise (IPCC, esis, more and more researchers began to study the influencing factors of
2018), but also pose a direct risk to human health (Jacobson et al., carbon emissions. The influencing factors involved in the study include
2019). According to the statistics of the International Energy Agency, in economic growth, energy consumption, urbanization, financial devel­
2019, the world’s total carbon emissions reached 33,621.5 Mt, of which opment, foreign direct investment, industrial structure and so on. For
China’s total carbon emissions was 9919.1 Mt, accounting for 29.5% of instance, in the empirical study on China, Gu et al. (2019) found that in
the world’s total carbon emissions, which has exceeded the total carbon the process of technological progress, the impact of energy consumption
emissions of North America and Europe. Therefore, to control global on carbon emissions presents an “inverted-U′′ shape. When the progress
carbon emissions and curb the trend of global warming, the focus is to of energy technology reaches a certain critical level, the increase of
realize China’s low-carbon development. At the same time, as the energy consumption will produce the effect of carbon emission reduc­
world’s largest emitter of carbon dioxide and the leader of climate action tion, rather than increasing carbon emissions. Liu and Bae (2018) found
in the past decade (Umar et al., 2020), China also made a commitment in the positive impact of urbanization on carbon emissions by using

* Corresponding author.
E-mail address: kingkong148@[Link] (J. Yi).

[Link]
Received 5 April 2022; Received in revised form 25 October 2022; Accepted 10 November 2022
Available online 20 November 2022
0301-4215/© 2022 Elsevier Ltd. All rights reserved.
H. Chen et al. Energy Policy 172 (2023) 113338

autoregressive distribution lag (ARDL) technology, and pointed out that through inter provincial interaction, so as to provide theoretical basis
for every 1% increase in urbanization, carbon emissions will increase by and theoretical support for government departments and relevant in­
1%. Huang and Zhao (2018) found that China’s different financial stitutions to formulate policies on carbon emission control and climate
development indicators, such as financial scale, financial efficiency and change mitigation.
financialization, have different effects on trade carbon emissions, and The rest of the paper is organized as follows. Section 2 is a brief
there are regional differences. literature review. Section 3 mainly introduces the methods and data
Recently, there is a lot of evidence that green technology innovation used in this study. The results and discussions are presented in Section 4.
has become an important means to reduce global carbon emissions Section 5 summarizes conclusions and discusses related policy
(Nikzad and Sedigh, 2017; Zhang et al., 2016). Hashmi and Alam (2019) implications.
pointed out that in any economy, technological progress and innovation
will make a great contribution to economic growth and its ecosystem. 2. Literature review
Green energy and green technology innovation are considered to be
important methods to achieve low-carbon and sustainable development In recent decades, with people’s attention to climate change and
(Kumail et al., 2020; Dong et al., 2020). Braun and Wield (1994) put greenhouse gas emissions, technological innovation, as the key driving
forward the initial concept of green technology and believed that green force to achieve low-carbon development, has attracted extensive dis­
technology should include pollution control, ecological treatment, pu­ cussion among researchers. At present, research on innovation and
rification, recycling, monitoring and various evaluation technologies. carbon emissions can be roughly divided into two categories. One is the
However, the academic community has not yet reached an agreement on overall technology innovation and carbon emissions, and the other is the
the definition of “green innovation”. At present, any process or product specific technology innovation and carbon emissions. The following will
with innovation, novelty, value, resource conservation and environ­ summarize these two kinds of literature.
mental improvement is regarded as “green innovation” (Lin and Long,
2021). The International Energy Agency pointed out that among various 2.1. Research on technology innovation and CO2 emissions based on
ways to mitigate climate change, green technology is expected to overall indicators
become a leading factor. In the 450 scenarios proposed, it can theoret­
ically contribute to more than 60% of the carbon emissions reduction The existing research on the impact of overall technological inno­
targets (IEA, 2013). However, due to the different degree and speed of vation on the environment can be roughly divided into two categories.
green technology R & D and promotion in different regions, the actual One type of research focuses on R & D expenditure as an alternative
impact of green technology innovation may depend on the specific social indicator of technological innovation. For example, Ganda (2019) used
or economic environment (IEA, 2015). Therefore, according to China’s the panel data of OECD economies from 2000 to 2014 and used the
national conditions, understanding the relationship between China’s system generalized method of moments (GMM) to analyze, pointing out
green technology innovation and carbon emissions is very important to that there is a statistically significant negative correlation between R &
realize China’s low-carbon development and reduce global carbon D expenditure and carbon emissions. Churchill et al. (2019) used
emissions. nonparametric panel data model to study G7 countries and found that
However, some studies have debated the above-mentioned carbon- the relationship between R & D expenditure and carbon emission has
reducing effect of green technology innovation. Braungardt et al. (2016) different characteristics in different periods, and the relationship be­
pointed out that although green innovation is usually considered as a tween the two changes with time. Another kind of research takes patent
basic element of a green growth strategy, it may in turn lead to an in­ generation as a measure of innovation and pays attention to the impact
crease in energy consumption due to the existence of rebound effect. of patent generation on carbon emissions. For instance, using the patent
Weina et al. (2016b) found in a study in Italy that although green data of 30 provinces and 32 economic sectors in China, Li et al. (2021b)
technology has significantly improved environmental productivity, it found that in the two sectors with high and low energy intensity, the
has not yet significantly promoted carbon emission reduction. More­ relationship between patent production and carbon emissions showed
over, Lin and Ma (2022) also pointed out that in the case study of 264 an “inverted-U′′ relationship, and the same relationship also applies to
cities in China, the role of green technology in carbon emission reduc­ less developed provinces. For the more developed provinces, the oppo­
tion is not significant. In general, the existing literature on the rela­ site relationship is found. Dauda et al. (2021) also found an “Inverted-U′′
tionship between green technology innovation and carbon emissions has relationship between technological innovation and carbon emissions in
not reached a consensus. Taking China as an example, investigating the studies of several African countries using fixed effect models and
relationship between green technology innovation and carbon emission generalized moment estimates. Salman et al. (2019) used patent appli­
will also help to expand the relevant literature. cation data to point out that technological innovation significantly
On the basis of existing research, the contributions of this paper are reduced carbon emissions by improving energy efficiency in the study of
as follows. Firstly, this paper analyzes the dynamic evolution charac­ seven ASEAN countries through panel quantitative expression model.
teristics of green technology innovation and carbon emission by using Different from its conclusion, Samargandi (2017) applied the autore­
kernel function estimation method, which may help the government gressive distribution lag (ARDL) method and found that the contribution
grasp the time evolution trend of them and test the performance of low- of technological innovation represented by the number of Arab patent
carbon development and green innovation in the past. Second, the applications on carbon emission reduction is insignificant.
spatial distribution pattern and aggregation characteristics of carbon
emissions are obtained by using spatial visualization method and global 2.2. Research on technology innovation and CO2 emissions based on
and local Moran index, which may help the government to allocate specific indicators
carbon treatment resources effectively by accurately locating high
emission provinces. Third, this study uses the spatial econometric model In addition to the research on overall technological innovation and
to explore the impact of green technology innovation on carbon emis­ carbon emissions, more and more researchers begin to pay attention to
sions at the spatial level, which fills the gap in relevant fields. In addi­ the relationship between specific technological innovation and carbon
tion, it also introduces the spatial spillover effect in the research, which emissions, such as environmental innovation, energy innovation and
can obtain more carbon emission information and clarify the action green technology innovation. Most of these researches have used the
mechanism of different influencing factors. Through direct effect, indi­ data of specific patent as a measure of innovation and their research
rect effect, total effect and feedback effect, we can analyze which factors methods and achievements also have their own characteristics. Toe­
can amplify or reduce their impact on provincial carbon emission belmann and Wendler (2020) used the number of patents applied for

2
H. Chen et al. Energy Policy 172 (2023) 113338

environmental patents as an indicator of environmental innovation. In 3. Methodology and data sources

the study of 27 European Union countries, it was found that environ­
mental innovation does help to reduce carbon emissions, while general 3.1. Data descriptions
innovation activities will not lead to emission reduction. Mongo et al.
(2021) used the autoregressive distribution lag model (ARDL) to study This paper constructs a provincial panel data of 30 provinces and
15 European countries and found that the impact of environmental municipalities in China from 2007 to 2019 (except Hong Kong, Macao,
innovation on carbon emissions has a rebound effect. In the long run, Taiwan and Tibet). We select carbon intensity as the explanatory vari­
environmental innovation tends to reduce carbon emissions, while in able to reflect the performance of China’s low-carbon economic devel­
the short term, the opposite effect was observed. Wang and Zhu (2020) opment to a certain extent. The calculation method of carbon intensity is
used the spatial econometric model to test whether energy technology shown in formula (1).
innovation is conducive to China’s carbon emission reduction. The re­
CE
sults showed that renewable energy technology innovation promotes CI = (1)
GDP
carbon emission reduction, while fossil energy technology innovation is
ineffective in reducing carbon emissions, and pointed out that the where CI represents the carbon intensity of a province, GDP represents
impact of energy technology innovation on carbon emissions is cross the real gross domestic product deflated with 2005 as the base year and
regional. CE represents the CO2 emissions from burning fossil fuels of that prov­
Like environmental innovation and energy technology innovation, ince which is calculated according to formula (2).
green technology innovation, as a specific technological innovation, is
also considered to be closely related to carbon emissions. However, ∑
8
CE = ECi × Ai × Ci × Oi × T (2)
there are some disputes on the role of green technology innovation in i=1
global carbon emission reduction and carbon neutrality. Some re­
searchers believe that there is a negative correlation between the two. where i represents the type of energy (i = 1, 2, …, 8 which include coal,
For example, by using the cross-sectional augmented autoregressive coke, crude oil, gasoline, kerosene, diesel, fuel oil and natural gas); EC
distribution lag (CS-ARDL) method, Yue et al. (2021) found that in represents the provincial energy consumption; A represents the average
Thailand, green technology innovation is conducive to reducing carbon low-order calorific value of an energy; C represents the carbon content
emissions and improving environmental quality. Qin et al. (2021) also per unit of calorific value of an energy; O represents the carbon oxida­
found the positive effect of green innovation on carbon emission control tion rate; T is the conversion coefficient between carbon and CO2, which
in the study of G7 countries, and pointed out the two-way causal rela­ is a constant and the value is 44/12. The values of the above variables
tionship between them. Paramati et al. (2021) pointed out that green come from the “general principles for the calculation of comprehensive
technology is the main factor to help OECD economies reduce carbon energy consumption (GB/t2589-2008)” and the “guidelines for the
emissions. In the study of Turkey, Shan et al. (2021) also found that preparation of provincial greenhouse gas inventories”.
green technology innovation will reduce carbon emissions based on We select green technology innovation as the core explanatory var­
STIRPAT model. Suki et al. (2022) used the unique method of bootstrap iable. The number of patent applications is not exactly equal to the level
ARDL and took Malaysia as the research background, and it is found that of technological innovation, but it can reflect the ability of technological
green innovation is negatively correlated with carbon emissions in both innovation to a certain extent. At the same time, there is a certain time
the short and long term. Some studies have found a mixed relationship lag in patents authorized due to monitoring, payment of annual fees, and
between the two. Xu et al. (2021) took 218 prefecture level cities in market environment (Lin and Long, 2021), while the number of patent
China as the research object, and found that green innovation and its sub applications promptly reflect green technology novation activities of the
categories have a positive impact on China’s carbon emission perfor­ year. Therefore, this study adopts number of green patent applications
mance by using two-way fixed effect model, instrumental variable to measure green technology innovation, which come from CNRDS
method, spatial econometric model and causal intermediary effect green patent-GPRD database.
model. It is also pointed out that green innovation significantly reduces Finally, in this research, economic development level, energy in­
and improves carbon emission performance through industrial structure tensity, population size and foreign direct investment are selected as
effect and foreign direct investment effect respectively. Shao et al. control variables. The original data of the various indexes is drawn from
(2021) found in N-11 countries that in the long run, green technology the official website of China Bureau of statistics, China Statistical
innovation has a significant negative impact on carbon emissions, while Yearbook, China Energy Statistical Yearbook and Wind database. The
in the short term, the correlation between the two is not significant. specific meaning of all the variables, their respective symbols and
Khattak et al. (2022) used the panel data of G7 countries from the first measurements are illustrated in Table 1.
quarter of 1990 to the fourth quarter of 2018 to investigate asymmetric
and periodic relationship between green and sustainable technological
innovation and carbon emissions. It is found that the adverse impact of Table 1
List of variables and descriptions.
green and sustainable technological innovation will lead to carbon
emissions in economic recession, while the positive impact of green and Symbol Variable Description Unit of measurement
sustainable technological innovation will reduce carbon emissions in CI Carbon Intensity Provincial CO2 Ton/10 thousand
economic prosperity. emissions/provincial GDP Yuan
To sum up, few studies take China’s provincial level as the research GTI Green Provincial number of Pieces
Technology green patent applications
object to explore the impact of green technological innovation on carbon
Innovation
emissions. At the same time, carbon emissions have obvious spatial ED Economic Provincial GDP per capita Yuan
nature, while most articles on the relationship between green techno­ Development
logical innovation and carbon emission often ignore the role of spatial Level
distance and geographical boundary in the research. Therefore, it is of ET Energy Intensity Provincial Energy Ton of standard coal
consumption/provincial equivalent/10
great significance to study whether green technology innovation in a GDP thousand Yuan
region contributes to carbon emissions reduction in adjacent regions. FDI Foreign Direct Provincial actual use of %
Investment foreign direct
investment/provincial
GDP

3
H. Chen et al. Energy Policy 172 (2023) 113338

3.2. The kernel density estimator 3.3.2. Local Moran index

Although global Moran’s I can well reflect the overall spatial ag­
Kernel density estimation is a nonparametric method used to esti­ gregation of variables, local Moran’s I can describe the spatial distri­
mate the probability density function which can clearly reflect the dis­ bution characteristics and interrelationships of each province
tribution of a random variable (Yang et al., 2022; Yang et al., 2021). By respectively. Therefore, we explore the local spatial autocorrelation of
using kernel density function estimation, this paper reveals the dynamic each province by constructing a scatterplot of local Moran’s I. The local
evolution and characteristics of green technology innovation level and Moran’s I scatterplot is displayed in the form of a coordinate graph. The
carbon intensity in 30 provinces in China. If x1 , x2 , …, xn is n sample first quadrant and the third quadrant represent high-high clustering and
points of independent identically distributed F, and if its probability low-low clustering, respectively, which is a positive spatial correlation.
density is f (x), the kernel density estimation of any point x is shown in The second quadrant and the fourth quadrant are represented as low-
formula (3). high clustering and high-low clustering, respectively, that is, they are
(x − x ) spatially negatively correlated. The calculation of local Moran’s I is
1 ∑n
f̂n (x) = K
i
(3) shown in formula (7).
nh i=1 h
∑n

∑n (Xi − X)2
where, n denotes the number of samples, and n = 30 represents 30 Ii =
(Xi − X) ( ) 2
Wij Xj − X , s = i=1 (7)
provinces in China. H is the bandwidth, which is determined by mini­ s2 j=1j∕
=i
n
mizing the mean squared error. K (•) is the kernel function. Here, we use
Gaussian kernel function as the estimation method, as shown in formula where, Ii is Local Moran index, and the meanings of other symbols are
(4). the same as those in formula (5).
( )
1 x2
K(x) = √̅̅̅̅̅ exp − (4)
2π 2 3.4. The spatial econometric model

If the research units have obvious spatial correlation characteristics,

3.3. Spatial auto-correlation analysis
the interaction force between the research units needs to be considered
(Wang et al., 2018). Since both explanatory variables and explained
The purpose of spatial autocorrelation analysis is to determine
variables may have spatial correlation characteristics, we need to
whether a variable is spatially correlated and whether the variable is
introduce spatial econometric models for empirical research. According
spatially clustered or discrete. The commonly used spatial autocorrela­
to different spatial effects, spatial econometric models are divided into
tion indexes are Moran’s I, gear’s C and so on. This study selects global
spatial lag model (SLM), spatial error model (SEM) and spatial Durbin
Moran’s I and local Moran’s I in Moran’s I spatial autocorrelation co­
model (SDM) (Yang et al., 2022). SDM, which is the combination of
efficient to explore the spatial aggregation and dispersion characteristics
spatial lag model and spatial error model, not only considers the spatial
of provincial carbon intensity in China.
lag of the research unit, but also introduces the lag term of error into
independent variables, and it can degenerate into spatial lag model or
3.3.1. Global Moran index
spatial error model under certain conditions. SDM could be generally
In order to test the spatial aggregation effect of carbon intensity, we
expressed as formula (8).
constructed a global Moran’s I for global testing. Global Moran’s I, as
shown in formula (5), reflects the overall spatial connection of China’s Yit = ρWYit + Xit α + WXit θ + φi + δt + εit (8)
provincial carbon intensity.
where, Yit is the explained variable; Xit denotes the explanatory variable;
∑
n ∑
n ( )
Wij (Xi − X) Xj − X ρ and θ represent the spatial regressive coefficients; φi and δt represent
n
I=∑
n ∑ •
i=1 j=1
(5) the spatial effect and time effect and εit denotes random error term. W is
n ∑
n
Wij (Xi − X)2 the spatial weight matrix based on geographical distance, which could
i=1 j=1 i=1
be expressed by formula (9).
⎧ ⎧
⎪ 1 i∕
= j and neighboring ⎪
⎪
⎨ ⎨ 1,i ∕
⎪ =j
(6) (9)
2
Wij = 0 i∕
= j and not neighboring Wij = dij
⎪
⎪ ⎪
⎪
⎩ ⎩ 0, i = j
0 i=j

where, I is Global Moran’s index; n represents the number of provinces where, dij is the surface distance between the provincial capital cities of
in China, with a value of 30; Xi and Xj represents the carbon intensity of region i and region j. After obtaining the spatial weight matrix by taking
province i and province j respectively; X represents the mean of carbon the square of the reciprocal of the ground distance, it is also necessary to
intensity. Wij is space weight matrix, which uses geographical adjacency standardize the weight matrix. According to the selected variables and
relation to assign value, as shown in formula (6). The assignment rule is equation (8), the specific spatial econometric model can be rewritten as
as follows: when the space between province i and province j has a part equation (10). All variables are presented in logarithmic forms to
of common boundary, the value is 1, otherwise the value is 0. The range eliminate possible heteroscedasticity.
of Moran’s I is [-1, 1]. At the significance level, the greater the absolute
ln CIit = ρWlnCIit + α0 + α1 ln GTIit + α2 ln EDit + α3 ln ETit + α4 ln Pit
value of Moran’s I, the stronger the spatial correlation of carbon in­
tensity. If Moran’s I is significantly positive, it indicates that there is a + α5 ln FDIit + WlnGTIit θ1 + WlnEDit θ2 + WlnETit θ3 + WlnPit θ4
positive spatial correlation, that is, provinces with high (low) carbon + WlnFDIit θ5 + εit
intensity show obvious spatial aggregation. If Moran’s I is significantly (10)
negative, it indicates that there is a negative spatial correlation, that is,
there are significant spatial differences in carbon intensity in adjacent where i = 1, 2, ⋯, 30 represents the i-th province and t = 1, 2, ⋯, 13,
provinces. In addition, if Moran’s I is close to zero, it indicates that there which represents the year from 2007 to 2019. Aiming at the possible
is no spatial autocorrelation of carbon intensity, and the distribution of nonlinear relationship between carbon intensity and green technology
carbon intensity is random in Chinese provinces. innovation, the quadratic term of green technology innovation is

4
H. Chen et al. Energy Policy 172 (2023) 113338

introduced on the basis of equation (10), and equation (11) is obtained. 4. Results and discussion

ln CIit = ρWlnCIit + α0 + α1 ln GTIit + α2 (ln GTIit )2 +α2 ln EDit + α3 ln ETit + α4 ln Pit + α5 ln FDIit +WlnGTIit θ1 +WlnEDit θ2 + WlnETit θ3 +WlnPit θ4 +WlnFDIit θ5
+ εit
(11)

4.1. The dynamic evolution of Chinese provincial carbon intensity and

green technology innovation
Although SDM considers the spatial effects of independent variables
and dependent variables, its estimation results can not represent the We estimate the kernel density every other year from 2007 to 2019 to
marginal effects of independent variables on dependent variables (Liu describe the evolution characteristics of the dynamic distribution of
and Song, 2020). In order to judge the spatial spillover effect of green China’s inter provincial carbon intensity and green technology innova­
technology innovation on carbon intensity more effectively, in accor­ tion. Figs. 1 and 2 respectively show the kernel density estimation re­
dance the partial differential method proposed by LeSage and Pace sults of China’s provincial carbon intensity and green technology
(2010) and the spatial extension made by Elhorst (2013), this paper uses innovation for the selected seven years. The curves of different colors
formula (12) to decompose the impact of green technology innovation represent different years.
on carbon intensity: From Fig. 1, we can see the dynamic evolution characteristics of
China’s provincial carbon intensity. First of all, from 2007 to 2019, the
Yit = (I − ρW)− 1 (Xit α + WXit θ) + (I − ρW)− 1 φi + (I − ρW)− 1 δt kernel density curve and its center of carbon intensity moved to the left
+ (I − ρW)− 1 εit (12) of the coordinate axis, indicating that China’s overall carbon intensity
has been decreasing. In addition, with the increase of years, the peak
Using the partial derivative of the k-th explanatory variable, the
value of carbon intensity kernel density curve is also growing while
partial matrix shown in Equation (13) will be obtained.
shifting to the left, and the curve of low-carbon intensity area changes
⎡ ⎤
⎡ ⎤ from wide and blunt to narrow and sharp, which indicates that more and
∂Y1 ∂Y1 αk ω12 γk ⋯ ω1n γk
⎢ ⋯ ⎥ more provinces in China have realized the transformation from medium
[ ] ⎢ ∂X1k ∂Xnk ⎥ ⎢ ⎥
∂Y ∂Y ⎢
⎢
⎥
⎥
⎢
⎢ ω21 γk αt ⋯ ω2n γk ⎥
⎥ and high carbon intensity to low-carbon intensity. At the same time, the
× = ⎢ ⋮ ⋱ ⋮ ⎥ = (1 − ρW)− 1 ⎢ ⎥ regional differences of carbon intensity in different regions of China are
∂Xik ∂Xnt ⎢ ⎥ ⎢ ⋮ ⋮ ⋱ ⋮ ⎥
⎢ ∂Yn ∂Yn ⎥ ⎣ ⎦
⎣ ⋯ ⎦ also shrinking, and the polarization phenomenon has been improved to
∂X1k ∂Xnk ωn1 γk ωn2 γk ⋯ αk a certain extent. In general, although some provinces still have high
(13) carbon intensity, most provinces have controlled their carbon intensity
at a low level by 2019, and China’s low-carbon development shows a
where ωij is the element of the matrix W. The direct effect is defined as good trend.
the average of the diagonal elements of the matrix, representing the From Fig. 2, we can see the dynamic evolution characteristics of
effect of the k-th independent variable on the dependent variable of local provincial green innovation level in China. First, on the whole, the
provinces; while the average of the non-diagonal elements is defined as kernel density curve and its center of green technology innovation level
the indirect effect, representing the effect of the k-th independent vari­ moved to the right of the coordinate axis from 2007 to 2019, indicating
able on the dependent variable of other provinces, or the effect of the kth that China’s overall green technology innovation level has been
independent variable of other provinces on the dependent variable of continuously improving. In addition, from 2007 to 2013, the level of
the local province. The total effect of the k-th variable is the sum of the green technology innovation in most provinces was very low. After
direct and indirect effect of the k-th variable.

Fig. 2. Kernel density plot of China’s provincial green technology innovation in

selected years. (For interpretation of the references to color in this figure
Fig. 1. Kernel density plot of China’s provincial carbon intensity in
legend, the reader is referred to the Web version of this article.)
selected years.

5
H. Chen et al. Energy Policy 172 (2023) 113338

2013, the level of green technology innovation in most provinces has Fig. 3. According to the spatial distribution map of carbon intensity in
been improved to a certain extent. At the same time, with the increase of the past four years, we can draw the following conclusions. First, on the
years, the peak value of the kernel density curve continues to decline whole, China’s overall carbon intensity showed a downward trend from
and the width of the curve continues to lengthen, and the peak value has 2007 to 2019. The carbon intensity of northern provinces was higher
not been concentrated in specific regions since 2015, which indicates and that of southern provinces was lower. Second, the carbon intensity
that the regional differences in the level of green technology innovation of the eastern and southeastern coastal areas is generally lower than that
in different regions of China are increasing. The level of green tech­ of other areas. Third, the carbon intensity of Ningxia, Shanxi, Inner
nology innovation in some provinces has improved significantly, while Mongolia and Xinjiang has been at a high level, and the improvement is
that in some provinces has improved slowly. Overall, the level of green slow. Fourth, compared with 2007, the carbon intensity of most prov­
innovation among provinces in China is on the rise, but the gap between inces in central and South China with high economic growth rate has
provinces is widening and the polarization phenomenon is serious. dropped to a lower level from 2019. Fifth, the carbon intensity of
provinces at all levels shows certain spatial aggregation characteristics.
The quartile map of China’s provincial green technology innovation
4.2. The spatial and temporal patterns of Chinese provincial carbon level is shown in Fig. 4. According to the spatial distribution map of
intensity and green technology innovation green technology innovation in the past four years, we can draw the
following conclusions. First of all, on the whole, China’s overall level of
In order to intuitively show the temporal and spatial distribution of green technology innovation showed an upward trend from 2007 to
China’s inter provincial carbon intensity and green technology innova­ 2019. The level of green technology innovation in the eastern and
tion level and analyze the aggregation of them in spatial distribution, central regions is high, while the level of green technology innovation in
this study uses the quartile map to make a preliminary analysis of the the western region is low. Second, the level of green technology inno­
data from 2007 to 2019. Due to space constraints, we only list the vation in the eastern coastal areas is generally higher than that in other
quartile map of 2007, 2011, 2015 and 2019. We divide the carbon in­ areas. Third, the green technology innovation of Ningxia, Gansu,
tensity and green technology innovation level into four levels. Each level Qinghai, Xinjiang and Inner Mongolia has been at a low level, and the
is displayed in a different color. The darker the color, the higher the improvement is slow. Fourth, compared with 2007, the level of green
carbon intensity and green technology innovation level. The first level technology innovation in most of the more developed eastern and
represents the lowest level of carbon intensity and green technology southern provinces has increased significantly from 2019. The more
innovation, and the fourth level represents the highest level of them. developed regions tend to pay more attention to green technology
The quartile map of China’s provincial carbon intensity is shown in

Fig. 3. The quartile maps of China’s carbon intensity in selected years.

6
H. Chen et al. Energy Policy 172 (2023) 113338

Fig. 4. The quartile maps of China’s green technology innovation in selected years. (For interpretation of the references to color in this figure legend, the reader is
referred to the Web version of this article.)

innovation. Fifth, the overall level of green technology innovation in all significant. Therefore, this paper uses global Moran index and local
provinces shows spatial correlation and heterogeneity. Moran scatter diagram to reveal the spatial agglomeration effect of
carbon intensity. The global Moran’s I of China’s carbon intensity from
4.3. The spatial autocorrelation analysis of Chinese provincial carbon 2007 to 2019 is shown in Table 2. The results in the table show that the
intensity global Moran’s I of China’s carbon intensity passed the significance test
at the significance level of 1% from 2007 to 2019, which indicatesthat
Although some spatial aggregation characteristics can be observed China’s provincial carbon intensity is not randomly distributed in these
from the above-mentioned carbon intensity spatial distribution quartile years, but shows significant spatial aggregation characteristics. At the
map, it is still uncertain whether these spatial characteristics are same time, the global Moran’s I of carbon intensity from 2007 to 2019 is
positive, which indicates that there is a significant positive autocorre­
lation in China’s provincial carbon intensity in these years. In addition,
Table 2
except for 2007, the global Moran’s I fluctuated between 0.396 and
Global Moran’s I of China’s carbon intensity and its statistical test.
0.465 in other years. The global Moran’s I fluctuated little, indicating
Year Global Moran’s I Z-value P-value
that China’s provincial carbon intensity showed a stable spatial aggre­
2007 0.356 3.619 0.000*** gation effect. Therefore, when studying the influencing factors of
2008 0.444 4.397 0.000*** China’s carbon intensity, we must consider the spatial effect in order to
2009 0.420 4.170 0.000***
2010 0.465 4.596 0.000***
get the empirical results consistent with the reality.
2011 0.464 4.672 0.000*** The global Moran’s I reveals the overall characteristics of China’s
2012 0.452 4.563 0.000*** provincial carbon intensity, and the local Moran’s I scatter chart can be
2013 0.432 4.384 0.000*** used to reveal the local evolution characteristics and differences of
2014 0.432 4.389 0.000***
China’s provincial carbon intensity. Due to space reasons, we only
2015 0.420 4.286 0.000***
2016 0.416 4.209 0.000*** provide the local Moran scatter diagram of 2007, 2011, 2015 and 2019
2017 0.396 4.099 0.000*** in Fig. 5. In the figure, the abscissa represents the normalized carbon
2018 0.398 4.140 0.000*** intensity, while the longitudinal axis represents the spatial lag value.
2019 0.397 4.141 0.000*** The first quadrant includes provinces with high carbon intensity and
Note: ***, ** and * indicate that coefficients are statistically significant at 1%, surrounded by high carbon intensity provinces, which is called High-
5% and 10%, respectively.

7
H. Chen et al. Energy Policy 172 (2023) 113338

Fig. 5. Local Moran’s I scatter plots of China’s province carbon intensity in 2007, 2011, 2015 and 2019.

High clustering type provinces. The second quadrant includes low-

Table 3
carbon intensity provinces, but the provinces surrounded by high-
The result of LM test, Hausman test, LR test and Wald test.
carbon intensity provinces are called Low-High clustering type prov­
inces; The third quadrant shows the provinces with low-carbon intensity Variable Statistics P-value

and surrounded by low-carbon intensity provinces, which is called Low- LM_lag 62.647 0.000***
Low clustering type provinces; The fourth quadrant represents the LM_err 40.266 0.000***
Robust_LM_lag 28.738 0.000***
provinces with high carbon intensity but surrounded by low carbon
Robust_LM_err 25.317 0.000***
intensity provinces, which is called High-Low clustering type provinces. LR_lag 33.641 0.000***
It can be seen from the figure that most provinces are High-High clus­ LR_err 43.623 0.000***
tering type provinces or Low-Low clustering type provinces in 2007, Wald_lag 35.236 0.000***
2011, 2015 and 2019. Moreover, these two types of provinces accounted Wald_err 42.574 0.000***
Hausman 30.67 0.000***
for 76.7% in 2007, 80% in 2011, 86.7% in 2015 and 83.3% in 2019.
Overall, from 2007 to 2019, the spatial aggregation of provincial carbon Note: ***, ** and * indicate that coefficients are statistically significant at 1%,
intensity in China has been significantly improved. At the same time, 5% and 10%, respectively.
high clustering type provinces include: Inner Mongolia, Hebei, Liaoning,
Ningxia, Shanxi, Xinjiang, Heilongjiang, Gansu. These provinces remain spatial Durbin model, we also need to test the robustness of the model. In
basically unchanged in the first quadrant, indicating that the agglom­ this study, LR test and Wald test were used to verify whether SDM can
eration is relatively stable, and the provinces in other quadrants also degenerate into SLM or SEM. The test results show that no matter which
remain basically unchanged, and the fluctuation with years is relatively test method is adopted, the null hypothesis that SDM will degenerate
small. Therefore, it can confirm the hypothesis that China’s provincial into SLM or SEM is significantly rejected at the level of 1%. Therefore,
carbon intensity has a stable spatial agglomeration effect. the spatial Durbin model constructed in this study is very reasonable.
The above statistical test results can be seen in Table 3.
4.4. Results for spatial panel model regression
4.4.2. Results of spatial durbin estimation
4.4.1. Spatial model selection According to the model selected and constructed above and the
Through spatial correlation analysis, we found that the provincial selected explanatory variables and explained variables, the regression
carbon intensity in China has stable spatial aggregation characteristics. results in Table 4 can be obtained. Table 4 not only shows the regression
In order to further determine the spatial correlation of the model, it is results of Durbin model, but also lists the results of mixed OLS regression
necessary to perform an LM test on the model. The null hypothesis of the for comparison. In Table 4, columns (2) and (3) show the regression
LM test is that there is no spatial lag term and spatial error term. The test results of linear and nonlinear mixed OLS models respectively, and
results show that both LM test and robust LM test reject the null hy­ columns (4) and (5) show the estimation results of linear and nonlinear
pothesis at the significance level of 1%. Therefore, when we study the SDM respectively. By comparing the regression results of the two mixed
relationship between green technology innovation and carbon intensity, OLS models, we find that the R2 of the nonlinear regression model is
the traditional econometric model without considering spatial effect will larger, and the Akaike information criterion (AIC) and Bayesian infor­
not be well applicable. At this time, it is necessary to introduce spatial mation criterion (BIC) are smaller. This shows that the fitting effect of
econometric model for regression analysis. Then, in order to make the the nonlinear mixed regression model is significantly better than the
model more robust, we conducted a Hausman test on the panel model to linear mixed regression model, and there is a nonlinear relationship
select a more suitable model from the fixed effect and random effect between green technology innovation and carbon intensity. However,
models. The results of Hausmann test showed that the null hypothesis the mixed OLS model ignores the spatial spillover effect of green tech­
was rejected at the significance level of 1%, so the fixed effect model is nology innovation on carbon intensity. By comparing the regression
more suitable for this study. Furthermore, before the regression of results of two dynamic SDM, we find that the R2 and Log-L of nonlinear

8
H. Chen et al. Energy Policy 172 (2023) 113338

Table 4 effect and total effect (Wang et al., 2021), with the results shown in
The results of model regression. Table 5. Due to the existence of feedback effect, the regression coeffi­
Variables Mixed OLS model Dynamic SDM cient of direct effect of SDM is inconsistent with those of the corre­
sponding variables in Table 4.
linear nonlinear linear nonlinear
From the direct effect, the regression coefficient of green technology
lnGTI 0.039(1.57) 0.212*** 0.056*** 0.101*** innovation level is significantly positive at the significance level of 1%,
(3.87) (4.32) (3.99)
(lnGTI)2 – − 0.012*** – − 0.003***
indicating that green technology innovation has a positive role in pro­
(-3.54) (-6.72) moting regional carbon emissions. The results show that when the level
lnED − 0.004 0.004(0.08) 0.029(0.44) − 0.028(-0.38) of green technology innovation increases by 1%, the regional carbon
(-0.10) intensity increases by 0.104%. In addition, the significant negative
lnP − 0.016 − 0.035(-1.02) − 0.330 − 0.361(-0.42)
quadratic coefficient indicates that the “inverted U′′ shaped effect does
(-0.47) (-2.98)
lnFDI 0.051*** 0.055*** 0.007(0.97) 0.012*** exist, that is, when the level of green innovation is low, it has a signif­
(3.75) (4.03) (3.27) icant positive impact on carbon emissions, while when the level of green
lnET 1.283*** 1.272*** 1.207*** 1.152*** technology innovation is high, carbon emissions gradually decline. This
(35.78) (35.82) (27.66) (23.40) may be because the process of green technology patent from application,
cons 0.867(1.59) 0.375(0.68) – –
W × lnGTI – – 0.011(0.43) 0.116***
authorization to production and use is very long. Green technology
(2.93) usually goes through industrialization and market-oriented develop­
W× – – – − 0.006*** ment from application to realizing its specific beneficial role (Lin and
(lnGTI)2 (-4.37) Ma, 2022). In the case of low level of green technology innovation, the
W × lnED − 0.046 − 0.142(-1.16)
– –
marketization degree of green technology is not high, and the market
(-0.44)
W × lnP – – 0.282(1.36) 0.262(1.27) cannot provide sufficient incentives for the popularization and appli­
W × lnFDI – – 0.035**(2.26) 0.049*** cation of green technology, so that enterprises do not pay high attention
(3.07) to green technology. In this way, the promotion and application of green
W × lnET – – − 0.129 − 0.182* technology are limited, and makes the beneficial role of green technol­
(-0.98) (-1.76)
R2 0.809 0.856 0.860 0.891
ogy cannot be brought into play timely and effectively. At the same time,
Log-L – – 544.678 575.962 in the process of green technology research and development, resources
sigma2_e – – 0.004*** 0.014*** are inevitably consumed, which leads to the generation of carbon
(13.94) (18.52) emissions. In the case of high level of green technology innovation, the
AIC − 9.139 − 19.676 − 1065.354 − 1125.923
marketization of green technology has reached a certain degree, and the
BIC 14.658 8.087 − 1017.760 − 1071.468
attention and recognition of green technology by the government and
Note: ***, ** and * indicate that coefficients are statistically significant at 1%, enterprises have reached a certain level, so that once the green tech­
5% and 10%, respectively and t value is in brackets. nology patent is authorized, it can be quickly supported by the gov­
ernment finance and enterprises. This kind of support enables the rapid
SDM are significantly larger than those of linear SDM, and the AIC and promotion and application of green technologies, timely and effectively
BIC of nonlinear SDM are also significantly smaller than those of linear playing their beneficial role, thus reducing carbon emissions and
SDM. To sum up, the results of nonlinear SDM regression are better than reducing environmental pollution. For other control variables in the
other regression results. Next, we will focus on the results of nonlinear model, the direct effect regression coefficients of the two variables
SDM regression. (foreign direct investment and energy intensity) are also significantly
From the results of the nonlinear SDM regression in column (5) of positive at the significance level is 1%.
Table 4, we can draw the following conclusions. First, foreign direct In terms of indirect effects, green technology innovation has a sig­
investment and energy intensity in a province are significantly positively nificant spillover effect on carbon intensity, and also presents a signifi­
correlated with local carbon emissions, but there is no significant cor­ cant “inverted U′′ phenomenon. The main reason may be the existence of
relation between local population, economic growth and carbon in­ agglomeration effect. When the local green innovation technology is
tensity. Second, foreign direct investment in the vicinity still continuously promoted and applied and the innovation level of green
significantly promotes local carbon emissions, while energy intensity in technology is continuously improved, the green technology and various
adjacent areas is negatively correlated with local carbon emissions. At R & D resources in the surrounding areas will be attracted to the region.
the same time, the spatial spillover effect of population and economic Therefore, the loss of green technology and R & D resources in the
growth in adjacent areas is not significant. Third, the primary and sec­ surrounding areas will inevitably affect their green innovation activities,
ondary coefficients of green technology innovation are significant at the increasing carbon emissions. Nevertheless, when the local green tech­
significant level of 1%, and the coefficients are positive and negative nology innovation level reaches a certain height, the agglomeration ef­
respectively. fect will become a feedback effect. This feedback effect will drive the
Because the spatial lag term is included in the model, the estimation development of green technology innovation activities, improve the
results can not directly reflect the marginal effect of variables. There­ green technology innovation level, and give full play to the role of green
fore, partial differential method is used to measure direct effect, indirect technology in carbon emission reduction in the surrounding areas.
Among other control variables, the indirect effect coefficient of only one
Table 5 variable (foreign direct investment) is significantly positive at the level
Estimation of direct effects, indirect effects and total effects. of 1%, which indicates that foreign direct investment has promoted the
carbon emissions of adjacent areas.
Variables Direct effects Indirect effects Total effects
Total effect is the sum of direct effect and indirect effect. In addition
lnGTI 0.104*** 0.141*** 0.245*** to the level of economic development and population size, the total ef­
(lnGTI)2 − 0.003*** − 0.007*** − 0.009***
lnED − 0.035 − 0.162 − 0.197
fect of the regression coefficient of other explanatory variables at the 1%
lnP − 0.356 0.253 − 0.103 significant level is significantly positive. The negative coefficient of the
lnFDI 0.015*** 0.058*** 0.073*** quadratic term of green technological innovation means that there is an
lnET 1.150*** − 0.069 1.081*** inflection point, that is, the highest point of the “inverted-U′′ curve.
Note: ***, ** and * indicate that coefficients are statistically significant at 1%, Before the inflection point, green technology innovation played a
5% and 10%, respectively. negative role in carbon emission reduction. However, after the inflection

9
H. Chen et al. Energy Policy 172 (2023) 113338

point, the negative role showed an obvious downward trend, and green Table 7
technology began to play its due beneficial role. The overall effect also The estimated results of robustness test based on lagged instrumental variables.
had the phenomenon of “inverted-U”. Variables Statistics Direct Indirect Total
effects effects effects
4.5. Robustness test lnGTI 0.158***(3.12) 0.153*** 0.145*** 0.298***
(lnGTI)2 − 0.005*** − 0.005*** − 0.009*** − 0.014***
4.5.1. Differences in the spatial weight matrix (-3.34)
W × lnGTI 0.136**(2.29)
Spatial weight matrix is the core element of spatial econometric – – –
W × (lnGTI)2 − 0.007*** – – –
model, and different spatial weight matrices may produce different (-4.53)
estimation results (Zhao and Wang, 2022). Therefore, we use the spatial Control Yes Yes Yes Yes
weight matrix based on economic distance to replace the geographical variables
distance spatial weight matrix to verify the robustness of the model. The R2 0.873 – – –
Log-L 554.631
results of the robustness tests of the spatial Durbin model are presented
– – –

in Table 6. From column (1) in Table 6, we can see that the estimation Note: ***, ** and * indicate that coefficients are statistically significant at 1%, 5%
coefficient and significance of green technology innovation and its and 10%, respectively and t value is in brackets.
square term in the robustness test are similar to those of the original
nonlinear Durbin model, and the goodness of fit (R2) value is also
Table 8
similar. At the same time, from column (2–4) in Table 6, we can also see
The estimated results of robustness test based on dynamic SDM.
that the coefficients and significance of direct effects, indirect effects and
total effects of green technology innovation and its square terms in the Variables Statistics Direct Indirect Total
effects effects effects
robustness test are also similar to the effect decomposition results of the
original nonlinear spatial Durbin model. Therefore, after replacing the lnGTI 0.126***(3.71) 0.125*** 0.123*** 0.248***
(lnGTI)2 − 0.003*** − 0.004*** − 0.008*** − 0.012***
spatial weight matrix, the estimation results of main variables do not
(-3.42)
change significantly, and we can consider these estimates to be reliable. W × lnGTI 0.133***(2.97) – – –
W × (lnGTI)2 − 0.008** – – –
4.5.2. Lagged instrumental variable (-2.35)
Because of the inverse causality, the omission of variables and the Control Yes Yes Yes Yes
variables
existence of measurement errors, endogenous problems inevitably arise.
R2 0.866 – – –
The lagged endogenous variables are often used as an instrumental Log-L 543.328 – – –
variable to alleviate endogeneity in empirical studies (Zhao et al., 2022).
Note: ***, ** and * indicate that coefficients are statistically significant at 1%, 5%
Therefore, this paper takes the lagged green technology innovation as
and 10%, respectively and t value is in brackets.
the instrumental variable to study the relationship between green
technology innovation and carbon emissions. On the one hand, the
lagged variable has a strong correlation with the current value of green variable, it is useful to add it as another control variable to check the
technology innovation, and has an impact on carbon emissions through robustness of the results. Table 8 exhibits the results of the robustness
the current value. On the other hand, lagged green technology innova­ check. We can see that, at the level of sign and significance, the co­
tion variable will not be affected by current carbon emissions. Therefore, efficients of the main explanatory variables green technology innovation
it is feasible to take the lagged green technology innovation as the and the square term are roughly the same as those of the benchmark
instrumental variable of green technology innovation. The regression regression, which also confirms the robustness of the model.
results are shown in Table 7. The coefficients of green technology
innovation and its square terms in robustness test are similar to those of 5. Conclusions and policy implications
original nonlinear SDM, indicating that the benchmark regression re­
sults are robust. This study aims to explore the impact of green technology innovation
on China’s provincial carbon emissions. Under this general goal, this
4.5.3. Dynamic spatial durbin model study first explores the dynamic evolution and temporal and spatial
Finally, we use dynamic SDM to test the robustness of the results; in pattern of China’s provincial carbon intensity and green technology
other words, we add the lagged dependent variable as the control var­ innovation, and then uses the nonlinear SDM to explore the impact of
iable into the regression process and regress again. Although it is green technology innovation on carbon intensity. The following con­
controversial to use the lagged dependent variable as the independent clusions are drawn. First, from 2007 to 2019, overall carbon intensity in
China continued to decline and overall level of green technology inno­
vation in China continued to rise. At the same time, the regional dif­
Table 6
ferences of carbon intensity in different regions of China are also
The estimated results of robustness test based on replacement space weight
matrix.
shrinking, and the polarization phenomenon has been improved to a
certain extent. However, the provincial gap of China’s green innovation
Variables Statistics Direct Indirect Total
level is widening, and the polarization phenomenon is serious. Second,
effects effects effects
China’s provincial carbon intensity have significant autocorrelation and
lnGTI 0.134***(3.50) 0.136*** 0.142*** 0.279*** aggregation in space. Third, the regression results of SDM reveal the
(lnGTI)2 − 0.006*** − 0.005*** − 0.008*** − 0.013***
(-3.26)
direct effect and spillover effect of green technology innovation. In
W × lnGTI 0.125***(3.33) – – – terms of direct effects, green technological innovation has an “inverted-
W × (lnGTI)2 − 0.008*** – – – U′′ nonlinear impact on local carbon intensity. Only when green tech­
(-4.73) nological innovation reaches a certain level can it play a positive role in
Control Yes Yes Yes Yes
improving the environment and curbing carbon emissions. In terms of
variables
R2 0.885 – – – spillover effect, green technological innovation also has an “inverted-U′′
Log-L 563.347 – – – nonlinear impact on carbon emissions in adjacent regions. The impact of
local green technology innovation on carbon emissions in adjacent areas
Note: ***, ** and * indicate that coefficients are statistically significant at 1%, 5%
and 10%, respectively and t value is in brackets. is also promoted first and then restrained. In addition, energy intensity

10
H. Chen et al. Energy Policy 172 (2023) 113338

and foreign direct investment have significantly increased China’s pro­ to different regions or corresponding indicators to further investigate the
vincial carbon emissions. Based on the above empirical evidence, we put regional heterogeneity of the impact of green technology innovation on
forward the following policy recommendations. First of all, the Chinese carbon emissions in China. Finally, while considering the spatial effect,
government should seriously consider the differences in economic we can also introduce other models such as the intermediary effect
growth patterns, economic structure and industrial structure in different model and use the compound model to deepen the current research.
regions, fully understand the regional agglomeration and imbalance of
China’s inter provincial carbon emissions, and locate provinces with Funding
sustained high carbon emissions, such as Ningxia, Shanxi, Inner
Mongolia and Xinjiang. For historical high carbon emission provinces, This work was supported in part by the Hunan Provincial Natural
special and effective measures will be introduced to deal with their Science Foundation of China (Grant no. 2022JJ31022), the Under­
backward or energy intensive economic growth model and achieve graduate Education Reform Project of Hunan Province (Grant no. HNJG-
carbon emission reduction in high carbon emission provinces. 2021-0532).
Second, due to the difference between the economic development
model and the degree of development, there are great differences in the Data availiability statements
level of green technology innovation among Chinese provinces. For
some eastern developed regions, green technology innovation activities The datasets generated during and/or analyzed during the current
are relatively active, and their green technology innovation level may study are available in the Chinese Research Data Services Platform
have approached or exceeded the turning point of the “inverted-U′′ ([Link] and China National Bureau of Statistics
shaped curve, which has played a positive role in carbon emission ([Link]
reduction and environmental improvement. For the underdeveloped
areas in the west, the level of green technology innovation is still low, Author contribution
and its level of green technology innovation may still be on the left side
of the “inverted U-shaped” curve, which has a negative impact on Huanyu Chen: Conceptualization, Methodology, Empirical analysis,
environmental improvement. Generally speaking, government agencies Writing - original draft, Writing – review & editing. Jizheng Yi:
should further adjust government functions, improve the degree of Conceptualization, Methodology, Resources, Funding acquisition, Su­
cooperation between the government and the market, and speed up the pervision, Project administration, Writing – review & editing. Aibin
application and marketization of existing mature green technologies. In Chen: Resources, Supervision, Writing – review & editing. Duanxiang
addition, it is also very important to increase the government’s support Peng: Validation, Data curation, Writing – review & editing. Jieqiong
for green technology innovation, so that the government’s scientific and Yang: Validation, Data curation, Writing – review & editing.
technological fiscal expenditure is more inclined to subsidies for the R &
D, promotion and application of green technology. At the same time, Ethics approval and consent to participate
governments at all levels should also strengthen the publicity of resi­
dents’ low-carbon life and improve residents’ willingness to pay for Not applicable.
green technology. Through these measures, we will strive to make the
green technology innovation level of each province on the right side of Consent for publication
the “inverted-U′′ shaped curve, so as to give full play to the positive role
of green technology. Not applicable.
Third, establish the concept of mutual benefit between regions,
strengthen exchanges and cooperation between regions, especially the Declaration of competing interest
innovation cooperation and transaction of green technology between
regions, and promote the cross regional circulation and application of The authors declare that they have no known competing financial
innovative green technology. While continuously developing green interests or personal relationships that could have appeared to influence
technology, regions with high level of green technology innovation the work reported in this paper.
should also cultivate the sharing consciousness of all walks of life in the
region, formulate green technology innovation sharing scheme, realize Data availability
the effective overflow of green technology innovation ability, coordinate
the green technology innovation ability of all regions, and make regions I have shared the link to my data.
with low level of green technology innovation realize the rapid devel­
opment of green technology innovation, so as to realize carbon emission Acknowledgements
reduction.
However, there are still some limitations to this study. First of all, in We would like to thank professor Zhenghui Li for his modification
terms of research objects, this study only examined the impact of green opinions of the paper and the anonymous reviewers for their valuable
technology innovation at the provincial level on carbon intensity in comments and insightful suggestions.
China, but in studies in other countries and regions, the relationship
between the two has not been further confirmed. Secondly, this study is References
mainly based on the overall sample data of China’s provinces, ignoring
the possible regional heterogeneity. Third, this paper only considers the Adedoyin, F., Ozturk, I., Abubakar, I., Kumeka, T., Folarin, O., Bekun, F.V., 2020.
Structural breaks in CO2 emissions: are they caused by climate change protests or
spatial spillover effect of green technology innovation on carbon emis­
other factors? J. Environ. Manag. 266, 110628 [Link]
sions, and does not analyze and confirm the mechanism of impact and jenvman.2020.110628.
possible transmission mechanism in detail. In view of the above limi­ Braun, E., Wield, D., 1994. Regulation as a means for the social control of technology.
tations of the research, possible future work arrangements are as fol­ Technol. Anal. Strat. Manag. 6 (3), 259–272. [Link]
09537329408524171.
lows. First, we can expand cross national and cross regional research by Braungardt, S., Elsland, R., Eichhammer, W., 2016. The environmental impact of eco-
collecting more comprehensive data from other countries or regions to innovations: the case of EU residential electricity use. Environ. Econ. Pol. Stud. 18
explore the relationship between green technology innovation and (2), 213–228.
Churchill, S.A., Inekwe, J., Smyth, R., Zhang, X., 2019. R&D intensity and carbon
carbon emissions in different countries and regions. Second, the samples emissions in the G7: 1870-2014. Energy Econ. 80, 30–37. [Link]
of China’s provinces can be divided into different sub samples according eneco.2018.12.020.

11
H. Chen et al. Energy Policy 172 (2023) 113338

Dauda, L., Long, X., Mensah, C.N., Salman, M., Boamah, K.B., Ampon-Wireko, S., Muhammad, B., 2019. Energy consumption, CO2 emissions and economic growth in
Dogbe, C.S.K., 2021. Innovation, trade openness and CO2 emissions in selected developed, emerging and Middle East and North Africa countries. Energy 179,
countries in Africa. J. Clean. Prod. 281, 125143 [Link] 232–245. [Link]
jclepro.2020.125143. Nikzad, R., Sedigh, G., 2017. Greenhouse gas emissions and green technologies in
Dong, K., Dong, X., Jiang, Q., 2020. How renewable energy consumption lower global Canada. Environ. Dev. 24, 99–108. [Link]
CO2 emissions? Evidence from countries with different income levels. World Econ. Paramati, S.R., Mo, D., Huang, R., 2021. The role of financial deepening and green
43 (6), 1665–1698. [Link] technology on carbon emissions: evidence from major OECD economies. Finance
Elhorst, J.P., 2013. Spatial Econometrics: from Cross-Sectional Data to Spatial Panel. Res. Lett. 41, 101794 [Link]
Springer, Berlin. [Link] Qin, L., Kirikkaleli, D., Hou, Y., Miao, X., Tufail, M., 2021. Carbon neutrality target for
Farooq, M.U., Shahzad, U., Sarwar, S., ZaiJun, L., 2019. The impact of carbon emission G7 economies: examining the role of environmental policy, green innovation and
and forest activities on health outcomes: empirical evidence from China. Environ. composite risk index. J. Environ. Manag. 295, 113119 [Link]
Sci. Pollut. Control Ser. 26 (13), 12894–12906. [Link] jenvman.2021.113119.
019-04779-x. Salman, M., Long, X., Dauda, L., Mensah, C.N., Muhammad, S., 2019. Different impacts
Ganda, F., 2019. The impact of innovation and technology investments on carbon of export and import on carbon emissions across 7 ASEAN countries: a panel quantile
emissions in selected organization for economic Co-operation and development regression approach. Sci. Total Environ. 686, 1019–1029. [Link]
countries. J. Clean. Prod. 217, 469–483. [Link] [Link].2019.06.019.
jclepro.2019.01.235. Samargandi, N., 2017. Sector value addition, technology and CO2 emissions in Saudi
Gu, W., Zhao, X., Yan, X., Wang, C., Li, Q., 2019. Energy technological progress, energy Arabia. Renew. Sustain. Energy Rev. 78, 868–877. [Link]
consumption, and CO2 emissions: empirical evidence from China. J. Clean. Prod. rser.2017.04.056.
236, 117666 [Link] Shan, S., Genç, S.Y., Kamran, H.W., Dinca, G., 2021. Role of green technology innovation
Grossman, G.M., Krueger, A.B., 1991. Environmental Impacts of a North American Free and renewable energy in carbon neutrality: a sustainable investigation from Turkey.
Trade Agreement. [Link] (NBER working paper, No. w J. Environ. Manag. 294, 113004 [Link]
3914). Shao, X., Zhong, Y., Liu, W., Li, R.Y.M., 2021. Modeling the effect of green technology
Hashmi, R., Alam, K., 2019. Dynamic relationship among environmental regulation, innovation and renewable energy on carbon neutrality in N-11 countries? Evidence
innovation, CO2 emissions, population, and economic growth in OECD countries: a from advance panel estimations. J. Environ. Manag. 296, 113189 [Link]
panel investigation. J. Clean. Prod. 231, 1100–1109. [Link] 10.1016/[Link].2021.113189.
jclepro.2019.05.325. Suki, N.M., Suki, N.M., Afshan, S., Sharif, A., Meo, M.S., 2022. The paradigms of
Huang, L., Zhao, X., 2018. Impact of financial development on trade-embodied carbon technological innovation and renewables as a panacea for sustainable development:
dioxide emissions: evidence from 30 provinces in China. J. Clean. Prod. 198, a pathway of going green. Renew. Energy 181, 1431–1439. [Link]
721–736. [Link] 10.1016/[Link].2021.09.121.
International Energy Agency (IEA), 2013. World Energy Outlook Special Report 2013: Toebelmann, D., Wendler, T., 2020. The impact of environmental innovation on carbon
Redrawing the Energy Climate Map. IEA Publications, Paris, France. dioxide emissions. J. Clean. Prod. 244, 118787 [Link]
International Energy Agency (IEA), 2015. Energy Technology Perspectives 2015: jclepro.2019.118787.
Mobilising Innovation to Accelerate Climate Action. IEA Publications, Paris, France. Umar, M., Ji, X., Kirikkaleli, D., Xu, Q., 2020. COP21 Roadmap: do innovation, financial
IPCC, 2018. Global Warming of 1.5 Degrees. Retrieved from. [Link] development, and transportation infrastructure matter for environmental
/sr15/pdf/sr15 _spm_final.pdf. sustainability in China? J. Environ. Manag. 271, 111026 [Link]
Jacobson, T.A., Kler, J.S., Hernke, M.T., Braun, R.K., Meyer, K.C., Funk, W.E., 2019. jenvman.2020.111026.
Direct human health risks of increased atmospheric carbon dioxide. Nat. Sustain. 2 Wang, B., Sun, Y., Wang, Z., 2018. Agglomeration effect of CO2 emissions and emissions
(8), 691–701. [Link] reduction effect of technology: a spatial econometric perspective based on China’s
Khattak, S.I., Ahmad, M., ul Haq, Z., Shaofu, G., Hang, J., 2022. On the goals of province-level data. J. Clean. Prod. 204, 96–106. [Link]
sustainable production and the conditions of environmental sustainability: does jclepro.2018.08.243.
cyclical innovation in green and sustainable technologies determine carbon dioxide Wang, H., Cui, H., Zhao, Q., 2021. Effect of green technology innovation on green total
emissions in G-7 economies. Sustain. Prod. Consum. 29, 406–420. [Link] factor productivity in China: evidence from spatial durbin model analysis. J. Clean.
10.1016/[Link].2021.10.022. Prod. 288, 125624 [Link]
Kumail, T., Ali, W., Sadiq, F., Wu, D., Aburumman, A., 2020. Dynamic linkages between Wang, Z., Zhu, Y., 2020. Do energy technology innovations contribute to CO2 emissions
tourism, technology and CO2 emissions in Pakistan. Anatolia 31 (3), 436–448. abatement? A spatial perspective. Sci. Total Environ. 726, 138574 [Link]
[Link] 10.1016/[Link].2020.138574.
LeSage, J.P., Pace, R.K., 2010. Spatial econometric models. In: Handbook of Applied Weina, D., Gilli, M., Mazzanti, M., Nicolli, F., 2016b. Green inventions and greenhouse
Spatial Analysis. Springer Berlin Heidelberg, pp. 355–376. [Link] gas emission dynamics: a close examination of provincial Italian data. Environ. Econ.
978-3-642-03647-7_18. Pol. Stud. 18, 247–263. [Link]
Lin, B., Ma, R., 2022. Green technology innovations, urban innovation environment and Xu, L., Fan, M., Yang, L., Shao, S., 2021. Heterogeneous green innovations and carbon
CO2 emission reduction in China: fresh evidence from a partially linear functional- emission performance: evidence at China’s city level. Energy Econ. 99, 105269
coefficient panel model. Technol. Forecast. Soc. Change 176, 121434. [Link] [Link]
org/10.1016/[Link].2021.121434. Yang, K., Yi, J., Chen, A., Liu, J., Chen, W., 2021. ConDinet++: full-scale fusion network
Lin, H., Long, C.X., 2021. Do discretion criteria for patent administrative law based on conditional dilated convolution to extract roads from remote sensing
enforcement encourage innovation among firms? China Economic Quarterly images. Geosci. Rem. Sens. Lett. IEEE 19, 1–5. [Link]
International 1 (2), 160–175. [Link] LGRS.2021.3093101.
Liu, H., Song, Y., 2020. Financial development and carbon emissions in China since the Yang, K., Yi, J., Chen, A., Liu, J., Chen, W., Jin, Z., 2022. ConvPatchTrans: a script
recent world financial crisis: evidence from a spatial-temporal analysis and a spatial identification network with global and local semantics deeply integrated. Eng. Appl.
Durbin model. Sci. Total Environ. 715, 136771 [Link] Artif. Intell. 113, 104916 [Link]
scitotenv.2020.136771. Yue, X.G., Liao, Y., Zheng, S., Shao, X., Gao, J., 2021. The role of green innovation and
Liu, X., Bae, J., 2018. Urbanization and industrialization impact of CO2 emissions in tourism towards carbon neutrality in Thailand: evidence from bootstrap ADRL
China. J. Clean. Prod. 172, 178–186. [Link] approach. J. Environ. Manag. 292, 112778 [Link]
jclepro.2017.10.156. jenvman.2021.112778.
Li, W., Chien, F., Hsu, C.C., Zhang, Y., Nawaz, M.A., Iqbal, S., Mohsin, M., 2021a. Nexus Zhang, J., Terrones, M., Park, C.R., Mukherjee, R., Monthioux, M., Koratkar, N., Kim, Y.
between energy poverty and energy efficiency: estimating the long-run dynamics. S., Hurt, R., Frackowiak, E., Enoki, T., 2016. Carbon science in 2016: status,
Resour. Pol. 72, 102063 [Link] challenges and perspectives. Carbon 98, 708–732. [Link]
Li, W., Elheddad, M., Doytch, N., 2021b. The impact of innovation on environmental carbon.2015.11.060.
quality: evidence for the non-linear relationship of patents and CO2 emissions in Zhao, C., Wang, B., 2022. How does new-type urbanization affect air pollution?
China. J. Environ. Manag. 292, 112781 [Link] Empirical evidence based on spatial spillover effect and spatial Durbin model.
jenvman.2021.112781. Environ. Int., 107304 [Link]
Mongo, M., Belaïd, F., Ramdani, B., 2021. The effects of environmental innovations on Zhao, M., Lv, L., Wu, J., Wang, S., Zhang, N., Bai, Z., Luo, H., 2022. Total factor
CO2 emissions: empirical evidence from Europe. Environ. Sci. Pol. 118, 1–9. https:// productivity of high coal-consuming industries and provincial coal consumption:
[Link]/10.1016/[Link].2020.12.004. based on the dynamic spatial Durbin model. Energy 251, 123917. [Link]
10.1016/[Link].2022.123917.

12