You are on page 1of 30

Natural Hazards (2020) 104:2561–2590

https://doi.org/10.1007/s11069-020-04285-1

ORIGINAL PAPER

Spatial correlation of China’s agricultural greenhouse gas


emissions: a technology spillover perspective

Yanqiu He1   · Xueying Cheng1 · Fang Wang1 · Ya Cheng1

Received: 9 March 2020 / Accepted: 31 August 2020 / Published online: 14 September 2020
© Springer Nature B.V. 2020

Abstract
Global warming is a key issue that is related to the sustainable development of various
countries, and agricultural sectors are particularly vulnerable to the effects of climate
change and increasing climate variability. To obtain a better understanding of agricultural
greenhouse gas (GHG) emissions, the estimation method proposed by the Intergovern-
mental Panel on Climate Change was used to estimate agricultural GHG emissions in 31
provinces in China with respect to five factors: agricultural energy consumption, agricul-
tural farmland utilization, crop cultivation, ruminant feeding, and straw burning. To ana-
lyze emission reduction strategy interactions as well as the spillover of agricultural techni-
cal information between regions, we used the spatial Durbin model and further explored
the different channels of technology spillover. The results obtained were as follows: (1)
ruminant feeding and straw burning are the major sources of agricultural GHG emissions
in China; (2) emission reduction strategies interact in the various regions, and imitation
behaviors are increasing; (3) the correlation of agricultural GHG emission reduction in
the different regions in China is not only limited to direct imitation behaviors, and it also
reflects the spillover of technical information, i.e., agricultural technological progress plays
an important role in the regional linkages of agricultural GHG emissions; (4) a shortening
of the economic distance facilitates agricultural technology exchanges between regions.
Therefore, to reduce agricultural GHG emissions, it is recommended that all regions
should establish regional cooperative emission reduction mechanisms via agricultural tech-
nical cooperation.

Keywords  Agricultural GHG emissions · Technology spillovers · Strategic interaction ·


Spillover channels · Spatial Durbin model

1 Introduction

Global warming, which has become a major threat to human survival, is a phenomenon
that must be considered to ensure future economic and social development (Nelson et al.
2019). Carbon dioxide, a major greenhouse gas (GHG), is one of the primary causes

* Yanqiu He
linxiatingqiu@126.com
1
College of Management, Sichuan Agricultural University, Chengdu 611130, China

13
Vol.:(0123456789)
2562 Natural Hazards (2020) 104:2561–2590

of global warming. In 2018, there was a 1.7% increase in global energy consumption-
related ­CO2 emissions, reaching a historic high of 33.1 Gt. This increase was the highest
since 2013, and it was 70% higher than the average increase observed since 2010 (IEA
2019). Thus, developing energy-saving techniques and taking action to reduce GHG
emissions has become a common global goal. The agricultural sector is particularly vul-
nerable to the effects of climate change and increasing climate variability (Owusu and
Asumadu-Sarkodie 2017). This implies that it is necessary to quickly transform global
agriculture, forestry, and land use, so that climate crises can be avoided (IPCC 2019).
As a country with a large agricultural production sector and a major GHG emitter, Chi-
na’s agricultural GHG emissions account for 17% of the country’s total emissions (Li
et al. 2011). Therefore, effectively reducing GHG emissions in this country will not only
directly facilitate the realization of China’s “Total and Intensity Dual Control” emission
reduction goals, it will also significantly affect global GHG reduction. Consequently,
strategies by which agricultural GHG emissions can be reduced have become an impor-
tant research focus.
Due to the strengthening of existing economic ties between regions, it has become diffi-
cult to realize emission reduction by relying on the unilateral actions of individual regions,
implying that cooperative, regionally coordinated emission reduction actions are needed.
Therefore, the spatial correlation of GHG emissions has attracted wide attention, and it has
been established that GHG emissions have significant spatial spillover effects. The pos-
sibility of the existence of a complex network of association structures has also been sug-
gested (Meng et al. 2013; Sun et al. 2016; Yang et al. 2016; Wu et al. 2015). The regional
correlation of GHG emissions can be attributed to linkages between economic, techno-
logical, and policy factors as well as the similarity in the energy consumption behavior of
microsubjects and the imitation of microsubject environmental behavior (Cole et al. 2013;
Liu et al. 2014; Dong et al. 2016; Li et al. 2016; Marbuah and Amuakwa-Mensah 2017). It
can also be attributed to cross-regional output changes owing to changes in final demand
(Zhang 2017; Jiao et  al. 2018). Some scholars have also reported that there is a signifi-
cant imbalance in the spatial distribution of total agricultural GHG emissions, emission
intensity, and emission efficiency, and the tendency for the occurrence of agglomerations
has also been reported (Wang et al. 2019). Agricultural economic development, production
structure, technological innovation, labor force, and urbanization are all important factors
that affect the spatial distribution of agricultural GHG emissions (Ma, 2018; Wu and Feng
2019; Xiong et al. 2016; Lu et al. 2018).
Technology is regarded as the most effective strategy to enhance the efficiency of agri-
cultural production (Xiong et al. 2016) and reduce agricultural GHG emissions (Soni et al.,
2013; Liu et  al. 2019; Ismael et  al. 2018), and reportedly, technological progress can be
estimated in terms of R&D capital investment (Snyder et al. 2016; Zhao et al. 2018), pat-
ent development (Lu et al. 2019), energy intensity (Dyer et al. 2010), and total factor pro-
ductivity (Tian and Zhang 2017; Pierluigi et al. 2017; Yang et al. 2018; Li et al. 2017; Xu
et al. 2019; Rybaczewska-Błażejowska and Magdalena 2018). Additionally, studies on the
impact of technological progress on agricultural GHG emissions revealed that technology
plays an important role in the reduction of agricultural GHG emissions, and progress in
resource reuse and recycling technologies are both important strategies to control emis-
sions (Ismael et al. 2018; Fan and Wei 2016). It has also been reported that low technical
efficiency is the major cause of a low agricultural carbon use efficiency (Zhang et al. 2020;
Hu, 2018; Fei and Lin 2017). In some previous studies, the spatial spillover of agricultural
technology (Cheng et al. 2016; Yang et al. 2017) and its impact on agricultural GHG emis-
sions (Wang et al. 2019; Li et al. 2015) were analyzed, and based on the results obtained,

13
Natural Hazards (2020) 104:2561–2590 2563

the positive spatial spillover effect of agricultural technology, which easily spread to neigh-
boring areas and enhanced emission reduction, was very obvious.
In summary, regional associations of GHG emissions with respect to macroperspec-
tives, such as economy, technology, and policy, as well as microperspectives, such as cor-
porate behavior and final demand, have been elucidated in previous studies, which reported
that, as a key factor in the reduction of GHG emissions, technological advancement does
not only facilitate the reduction of local GHG emissions through spatial spillover effects,
it also facilitates GHG emission reduction in neighboring areas. However, these previous
studies had two shortcomings. On the one hand, different scholars used different indica-
tors to measure technological advancement based on their research needs, i.e., in previous
studies, no unified measurement standard was used, and only the level of technological
progress was considered, leaving out potential technological advancements. On the other
hand, discussions on the spatial spillover effects of technological progress were limited
to geographic distance, and no in-depth analysis of the possible channels for technology
spillover was considered. Therefore, in this study, using the Intergovernmental Panel on
Climate Change (IPCC) estimation method, we estimated the agricultural GHG emissions
in 31 provinces in China with respect to five factors: agricultural energy consumption,
agricultural farmland utilization, crop cultivation, ruminant feeding, and straw burning and
measured technological progress with respect to two factors: agricultural technology pro-
gress level and agricultural technology progress potential. Additionally, we also analyzed
the interaction of agricultural GHG emission reduction strategies and behaviors as well as
the spillover of agricultural technical information via the establishment of a spatial Durbin
model (SDM). Using different spatial weight matrices to examine the channels of technol-
ogy spillovers, our results provide a reference for energy conservation and emission reduc-
tion in the agricultural sector in China and other countries.

2 Data and methods

2.1 Data sources

In this study, the primary data used, which spanned the 2009–2017 period, corresponded
to 31 provinces of China. Hong Kong, Macao, and Taiwan were excluded because of
missing data. The activity data required for agricultural GHG emission estimation were
obtained from the China Energy Statistics Yearbook and China Rural Statistical Yearbook.
The input and output indicators for the estimation of agricultural technological progress
were obtained from the China Rural Statistical Yearbook and China Human Resources Sta-
tistical Yearbook. The variables for the establishment of the spatial Durbin model were
obtained from the China Rural Statistical Yearbook, China Human Resources Statistical
Yearbook, and manually from Baidu Library.

2.2 Methods

2.2.1 Agricultural GHG emissions

Five major categories of agricultural GHG emission sources, totaling 31 items, were con-
­ O2 emission from energy consumption, including the operation of agricultural
sidered. (1) C
machinery, farmland irrigation, and household consumption, (2) ­CO2 emission from farmland

13
2564 Natural Hazards (2020) 104:2561–2590

utilization, including the input of fertilizers, pesticides, and plastic mulch, (3) C
­ H4 release
from rice cultivation and N ­ 2O release from the production of other crops. (4) C­ H4 and N­ 2O
emissions from ruminant feeding, including intestinal fermentation and fecal management,
and (5) C ­ O2, ­CH4, and ­N2O emissions from straw burning. The emissions of each category
can be calculated as:


5

5
Ei = Ej = (ej × fj ) (1)
i=1 i=1

where Ei represents the total GHG emission corresponding to a given category, i  , Ej rep-
resents emissions from a source, j , belonging to the category, i  , and ej and fj represent the
activity data and emission factor of the source, j , respectively. All the emission factors are
provided in Appendix 1.
To facilitate the analysis, the estimated GHG emissions corresponding to N ­ 2O and C ­ H4
were converted into carbon equivalents as previously described (IPCC 2007) such that the
GHG effect caused by 1 t of N ­ 2O and 1 t of C
­ H4 was equivalent to that caused by 298 t of C
­ O2
(81.2727 t C) and 25 t of ­CO2 (6.8182 t C), respectively (Guan et al. 2018). Table 1 shows the
activity data used in the estimation of the agricultural GHG emissions.

2.2.2 Agricultural technological progress

To calculate China’s agricultural technological progress, the Malmquist index method, which
was first proposed by the Swedish statistician Malmquist (1953), was used, and in 1982, Caves
et al. (1982) proposed that total factor productivity can be expressed in terms of the Malmquist
index.
Firstly, the distance function was defined as:
( )
( ) yt x t
Dt xt , yt = et = ( ) (2)
ȳ t xt
( )
where xt and yt represent the input and output vectors within the period,t  , ȳ t xt represents
the maximum potential output of the input, xt , within the period, t  , with the production

Table 1  Description of activity data


Category Indicator Source

Energy consumption Amount of coal, coke, crude oil, gasoline, kerosene, diesel oil, fuel oil, China
electric power, and natural gas used in agricultural production Energy
Sta-
tistics
Year-
book
Farmland utilization Application amount of fertilizers, pesticides, and agricultural film, plow- China
ing area Rural
Crop planting Planting area of rice, wheat, corn, soybeans, and vegetables Statis-
tical
Ruminant feeding Annual average stock of cattle, horses, donkeys, mules, pigs, goats, and
Year-
sheep
book
Straw burning Yield of rice, wheat, corn, soybeans, cotton, and canola

13
Natural Hazards (2020) 104:2561–2590 2565

frontier, ­St as the technical premise. The distance function,


( ) Dt , is equivalent to the technical
( )
efficiency,et , which is the ratio of the actual output, yt xt and the maximum output, ȳ t xt
under the same input conditions.
According to Eq.  (2), the actual output corresponding to periods t and t + 1 can be
expressed as follows:
( ) ( ) ( )
yt xt = ȳ t xt × Dt xt , yt (3)

( ) ( ) ( )
yt+1 xt+1 = ȳ t+1 xt+1 × Dt+1 xt+1 , yt+1 (4)

Therefore, the output growth rate from period t to t + 1 is:


( ) ( ) ( )
yt+1 xt+1 Dt+1 xt+1, yt+1 ȳ t+1 xt+1
( ) = ( ) × ( ) (5)
yt x t Dt xt, yt ȳ t xt

Secondly, under the technical level corresponding to periods t and t + 1 , the productiv-
ity change from period t to t + 1 can be expressed as:
( )
Dt xt+1, yt+1
Mt = ( ) (6)
Dt xt, yt

( )
Dt+1 xt+1, yt+1
Mt+1 = ( ) (7)
Dt+1 xt, yt

Finally, the Malmquist index is defined as:


[ ( ) ( )]1
( ) Dct xt+1, yt+1 Dct+1 xt+1, yt+1 2
M xt+1, yt+1 ;xt, yt = ( ) × ( ) (8)
Dct xt, yt Dct+1 xt, yt

where xt and yt represent the input and output vectors corresponding to period t ,xt+1 and
yt+1 represent the input and output vectors corresponding to period t + 1 , and Dct and Dct+1
represent the distance functions with constant returns to scale within periods t and t + 1 ,
respectively.
As one of the important methods for the estimation of total factor productivity, the
Malmquist index measures the total factor productivity directly from the perspective of
inputs and outputs. Fare et al. (1994) decomposed the Malmquist index to obtain:

13
2566 Natural Hazards (2020) 104:2561–2590

[ ( ) ( )]1
( ) Dct xt+1, yt+1 Dct+1 xt+1, yt+1 2
M xt+1, yt+1 ;xt, yt = ( ) × ( )
Dct xt, yt Dct+1 xt, yt
[ c ( ) ( ) ( ) ] 21
(Dt+1 xt+1, yt+1 )2 Dct xt+1, yt+1 Dct xt, yt
= ( ) × c ( )× c ( )
(Dct xt, yt )2 Dt+1 xt+1, yt+1 Dt+1 xt, yt
( ) [ c( ) ( ) ] 21
Dct+1 xt+1, yt+1 Dt xt+1, yt+1 Dct xt, yt
= ( )2 × ( )× c ( )
Dct xt, yt Dct+1 xt+1, yt+1 Dt+1 xt, yt
( ) Dct+1 (xt+1, yt+1 ) [ ( ) ( ) ] 21
Dvt+1 xt+1, yt+1 Dvt+1 (xt+1, yt+1 ) Dct xt+1, yt+1 Dct xt, yt
× ( ) × × ( )× c ( )
Dvt xt, yt Dct (xt, yt ) Dct+1 xt+1, yt+1 Dt+1 xt, yt
Dvt (xt, yt )
= Pech × Se × Tch
(9)
where Pech represents the pure technology efficiency index with variable scale returns, Se
represents the scale efficiency index, and Tch represents the technological progress index.
To exclude the impact of economies of scale, the pure technical efficiency ( Pech ) was used
to estimate agricultural technological progress (Ma et al. 2011). The input and output indi-
cators for the estimation of agricultural technological progress are shown in Table 2.

2.2.3 Spatial correlation test

The global Moran’s I index was used to verify the global spatial distribution of clusters
as well as the local spatial distribution of clusters. The calculation of the Moran’s I
index is shown in Appendix 2. The value range of both the global Moran’s I and local
Moran’s I indices is [− 1, 1]. Global Moran’s I indices above zero signify a positive spa-
tial correlation, while indices below zero signify a negative spatial correlation. When
the index is close to 0, it implies the absence of any significant spatial effect. Addition-
ally, indices above zero indicate the formation of high–high and low–low patterns, while
those below zero indicate the formation of high–low and low–high patterns.

Table 2  Input and output indicators


Category type Indicators Source

input Agricultural machinery total power China Rural Statistical Yearbook


Agricultural fertilizer application volume
Total area of sown crops
Effective irrigated area
Number of agricultural workers China Human Resources Statistical Yearbook
Output Actual agricultural output(expected output) China Rural Statistical Yearbook
GHG emission (non-expected output) Formula 1

13
Natural Hazards (2020) 104:2561–2590 2567

2.2.4 Agricultural technology spatial spillover channels

Agricultural technology spillover channels, which may be geographic or economic, are


shown in Fig.  1. Firstly, transportation between regions that are geographically close
favors agricultural technology spillover to a greater extent, given that the flow of pro-
duction factors (human capital, technology, etc.) between regions is associated with a
lower cost, and this makes the spillover of technology and knowledge easier. Secondly,
regions with stronger economic ties have smaller gaps with respect to R&D technol-
ogy and technology absorption capabilities. Thus, technology spillovers are easier.
Contrarily, weaker economic ties hinder the exchange of technical knowledge between
regions, and this limits technology spillover. Therefore, economic ties should be taken
into account when considering technological spillover (Liu and Jia 2007; Yu et al. 2016;
Xu et  al. 2017). Thirdly, it is more likely for agricultural industries to agglomerate in
areas that are geographically or economically close. This facilitates resource sharing
and economies of scale (Zhang and Mu 2019), and most importantly, enhances the shar-
ing of technology.
According to the above analysis, different spatial weight matrices were used to
analyze possible channels for the spatial spillover of agricultural technology between
regions. Primarily, the geographic distance was set as the spatial weight matrix, which
was defined as:
{
−1∕d2 (i ≠ j)
wijd =
0 (i = j) (10)

where wijd represents the geographic distance weight, and d represents the spherical dis-
tance between the capitals of the two provinces under consideration.
Further, the economic distance weight, which was defined as follows, was used:

Channels for
agricultural
technology spillover

Geographic Economic
channels channels

Small gap in R&D


Low flowing cost of Agricultural industrial
technology and
production factors agglomeration
absorption capabilities

Fig. 1  Agricultural technology spillover channels

13
2568 Natural Hazards (2020) 104:2561–2590

{ 1
(i ≠ j)
wije = |GDPi −GDPj | (11)
0 (i = j)

where wije represents the economic distance weight, GDPi and GDPj represent the gross
regional products for regions i and j , respectively, and d represents the spherical distance
between the capitals of the two provinces under consideration.

2.2.5 Impact of agricultural technological spatial spillover

To analyze the role of agricultural technology spillover on agricultural GHG emission


reduction, it is necessary to not only consider the spillover of agricultural technology pro-
gress among regions, it is also important to consider the interaction of the agricultural
GHG emission reduction strategies of the regions. Thus, the Durbin model, as defined
below, was chosen.
GHGInt = 𝜏n 𝛼 + 𝜌𝜔GHGInt + 𝛽tec Tecnt + 𝜃tec 𝜔Tecnt + 𝛽xnt + 𝜃𝜔xnt + 𝜇n + 𝜐t + 𝜀nt
( )
𝜀nt ∼ N 0, 𝜎 2 In
(12)
where GHGInt represents the intensity of the agricultural GHG emissions of the 31 prov-
inces for a period of 9 years, Tecnt represents agricultural technology progress, including
the pure technical efficiency ( Pech ) and agricultural technology progress potential ( Pch ),
xnt represents control variables, w represents the spatial weight matrix, 𝜀nt represents the
random error term, 𝜇n represents individual-fixed effects, and 𝜐t represents time-fixed
effects. The detailed variables are defined as follows (Table 3).
Additionally, to clearly show the impact of agricultural technological progress spillovers
on agricultural GHG emission reductions, direct and indirect effects were further decom-
posed. The derivation process is shown in Appendix 3. The analysis flowchart for this
study is shown in Fig. 2.

3 Empirical analysis

3.1 Calculation and analysis of agricultural GHG emissions and agricultural


technology

3.1.1 Agricultural GHG emissions

According to the methods described in Sect. 2, we estimated agricultural GHG emissions


in China for the 2009–2017 period and analyzed the total emissions, emission intensity
(Fig. 3), emission structure (Fig. 4), and the spatial distribution of the emissions (Fig. 5).
Figure  3 shows the dynamic trends of the total agricultural GHG emissions and the
emission intensity in China for the 2009–2017 period. The total amount of emissions
increased slightly from 15.4 million tons in 2009 to 17.5 million tons in 2017, with an
average annual growth rate of 1.6%. However, a significant decrease in emission intensity
was observed, with an average annual decrease of − 7.8%.
Figure  4 shows the sources of the agricultural GHG emissions in China as well as
their development trends. The average annual GHG emissions corresponding to ruminant

13
Natural Hazards (2020) 104:2561–2590

Table 3  Model variables
Independent variable Symbol Calculation Data sources

Agricultural technology progress Pech Agricultural pure technical efficiency Method in Sect. 2.2.2
Pch Number of agricultural universities and research structures Collected manually by Baidu Library
control variable Cul Cultivated area China Rural Statistical Yearbook
Emp Agricultural practitioners China Human Resources Statistical Yearbook
Cap Total power of agricultural machinery China Rural Statistical Yearbook
Str Ratio of crop production value to gross agricultural output value
2569

13
2570 Natural Hazards (2020) 104:2561–2590

Emission factor method of Malmquist index method

Agricultural
Agricultural GHG
technological

Data preparation
Energy Farmland
Input indicators

Planting Ruminant
Output
Straw

Descriptive statistical analysis

Descriptive analysis
Temporal and spatial characteristics and
structure of agricultural GHG emissions

Temporal and spatial characteristics of


agricultural technological progress

Moran's I Index

Spatial correlation

Agricultural GHG

Agricultural
Spatial analysis

technological

Spatial Durbin Model

Spillover effect of agricultural

Decomposition of
Spillover channels
spatial spillover effects

Discussions and Conclusions


8

Fig. 2  Analysis flowchart

feeding and straw burning, which are the major sources of agricultural GHG emissions,
were 40.9 and 39.3 million tons, accounting for 24.7 and 23.7% of the total emissions,
respectively. The growth trend of GHG emissions resulting from straw burning and
farmland utilization, which represent the focus of future agricultural GHG emission

13
Natural Hazards (2020) 104:2561–2590 2571

Total GHG emissions GHG emission intensity

Emission intensity Unit: t /


Total emissions Unit: 104t
180000 6.00
175000 5.00
170000
165000 4.00

million
160000 3.00
155000 2.00
150000
145000 1.00
140000 0.00
2009 2010 2011 2012 2013 2014 2015 2016 2017
Year

Fig. 3  Double-line graph of total agricultural GHG emissions and emission intensity in China for the 2009–
2017 period

50000
GHG from agricultural
45000 energy
GHG from farmland
40000 utilization
GHG emission Unit: 104t

GHG from ruminants feeding


35000
GHG from crop planting
30000

25000 GHG from Straw burning

20000 Trendline of agricultural


energy emissions
15000 Trendline of farmland
utilization emissions
10000
Trendline of ruminants
feeding Emission
5000
Trendline of crop planting
0 emissions
2009 2010 2011 2012 2013 2014 2015 2016 2017 Trendline of Straw burning
Year emissions

Fig. 4  The histogram of agricultural GHG emissions’ structure of China for the 2009–2017 period

reduction strategies, was the most obvious, with average annual growth rates of 2.8 and
2.2%, respectively.
Within the observation range, the regions with high agricultural GHG emissions were
found to be distributed across Heilongjiang, Jilin (northeast region), Hebei, Shandong
(northern coastal region), Inner Mongolia, Henan (middle Yellow River region), Hunan,
Hubei, Jiangxi, Anhui (middle Yangtze River region), Sichuan, Yunnan, Guangxi (west
region), Guangdong (southern coastal region), and Jiangsu (East Coast Region). This
spatial distribution could be attributed to the fact that the northeast regions and middle
Yangtze River regions are the areas where China’s agricultural production industry is most
concentrated. Inner Mongolia has the largest natural pastureland in China and a well-devel-
oped animal husbandry sector. Additionally, in 2018, the added value corresponding to
agriculture for Hebei, Henan, Shandong, Guangdong, Sichuan, and Jiangsu ranked among
the top six in China. Thus, these areas showed high agricultural GHG emissions. Most of
the southern coastal, eastern coastal, and western regions as well as Beijing and Tianjin

13
2572 Natural Hazards (2020) 104:2561–2590

Fig. 5  Spatial distribution of agricultural GHG emissions in China (unit: ­104 t)

showed low agricultural GHG emissions, because agriculture accounts for a relatively low
proportion of their economies.

3.1.2 Agricultural technology progress

Based on Fig.  6, the regional gap for pure technical efficiency showed a
“high–low–high–low” wave dynamic trend, but overall, the pure technology efficiency
increased gradually, indicating that China’s agricultural technology is advancing. The aver-
age pure technology efficiencies corresponding to 2013 and 2017 were higher, 1.45 and
1.57, respectively, and the lowest value, which was only 0.79, corresponded to 2015.

3.2 Spatial correlation analysis of agricultural GHG emissions and agricultural


technology

3.2.1 Agricultural GHG emissions

In this study, global Moran’s I indices of agricultural GHG emissions were calculated, and
global spatial aggregation characteristics were tested. The results are shown in Table 4.
The global Moran’s I indices of agricultural GHG emissions were all positive within the
data range for the 2009–2017 period and were significant at a 99% confidence level. These
results indicate that the spatial distribution of the agricultural GHG emissions is not random,
but has significant spatial agglomeration characteristics, i.e., low-value and low-value region
aggregations and high-value and high-value region aggregations. Therefore, each spatial

13
Natural Hazards (2020) 104:2561–2590 2573

Fig. 6  Boxplot of agricultural pure technology efficiency index in China for the 2009–2017 period

Table 4  Global Moran’s I index Year Moran’s I Z-statistics P value


of agricultural GHG emissions
for the 2009–2017 period
2009 0.269*** 3.486 0.000
2010 0.267*** 3.466 0.000
2011 0.268*** 3.477 0.000
2012 0.265*** 3.442 0.000
2013 0.253*** 3.296 0.000
2014 0.263*** 3.413 0.000
2015 0.255*** 3.318 0.000
2016 0.276*** 3.578 0.000
2017 0.289*** 3.720 0.000

***,** and * indicate variables were significant at 1%, 5% and 10%


statistical levels, respectively

region showed a clear spatial autocorrelation. The global Moran’s I indices were greater
than 0.25 for the entire study period, increasing from 0.269 in 2009 to 0.289 in 2017, estab-
lishing that China’s agricultural GHG emissions have relatively stable spatial characteristics.
To analyze the local aggregation of the agricultural GHG emissions, the local Moran’s I
indices were determined, and Moran scatter plots were plotted (Fig. 7).
In the Moran scatter plots, the coordinate points located in the first quadrant
[“high–high” agglomeration (H–H)] and third quadrant [“low–low” agglomeration (L–L)]
represent individual regions that are spatially positively correlated. The coordinate points
located in the second quadrant [“low–high” agglomeration (L–H)] and fourth quadrant
[“high-low” agglomeration (H–L)] represent individual regions that are spatially negatively
correlated. Over time, there were no obvious changes in the local aggregation characteris-
tics of the agricultural GHG emissions. Beijing, Tianjin, and Shanghai always showed an

13
2574 Natural Hazards (2020) 104:2561–2590

Moran scatterplot (Moran's I = 0.269)


lncarbon
1
6
13 14
12 23
30 22 20
21
26 31 195
4 2728
11
24 7 15
29 25
0 10 8
17
18
3
Wz

16

-1
9
2
1

-2
-3 -2 -1 0 1 2
z
(a) 2009

Moran scatterplot (Moran's I = 0.289)


lncarbon
1
14
6
12
30 13 20 23
22
21
26 4 19 31
29 1127 18
2428 7 5 15
8
0 25
10 3
17

16
Wz

-1
9
1
2

-2
-3 -2 -1 0 1 2
z
(b) 2017

Fig. 7  Local Moran scatter plot of agricultural GHG emissions of 31 provinces in 2009 and 2017. Note:
1—Beijing, 2—Tianjin, 3—Hebei, 4—Shanxi, 5—Inner Mongolia, 6—Liaoning, 7—Jiling, 8—Heilongji-
ang 9—Shanghai, 10—Jiangsu, 11—Zhejiang, 12—Anhui, 13—Fujian, 14—Jiangxi, 15—Shandong,
16—Henan, 17—Hubei, 18—Hunan, 19—Guangdong, 20—Guangxi, 21—Hainan, 22—Chongqing,
23-Sichuan, 24—Guizhou, 25—Yunnan, 26—Tibet, 27—Shaanxi, 28—Gansu, 29—Qinghai, 30—Ningxia,
31—Xinjiang

L–L agglomeration. Heilongjiang, Liaoning, Jilin (northeast regions), Sichuan, Yunnan,


Xinjiang, Guangxi (western regions), Inner Mongolia, Anhui, Jiangxi (central regions),
Shandong, and Guangdong (coastal regions) showed an H–H agglomeration. Qinghai,

13
Natural Hazards (2020) 104:2561–2590 2575

Ningxia, Chongqing, Tibet, Shaanxi, Gansu (western region), Hainan, Fujian (Southern
Coastal Area), and Shanxi showed an L–H agglomeration, while Jiangsu, Hubei, Hebei,
and Henan showed an H–L agglomeration.
Although the Moran scatter plots could be used to divide the provinces into different
agglomerations, they could not be applied to distinguish between significant and insignifi-
cant agglomerations. To distinguish the four agglomeration types, H–H, L–L, H–L, and
H–L, in terms of significance at a significance level of 5%, so as to identify agricultural
GHG emission hot and cold spots, LISA agglomeration maps were drawn (Fig. 8).
Figure 8 shows that China’s agricultural GHG emissions have significant local spatial
correlations, and the provinces showing a significant agglomeration increased with time.
Beijing, Tianjin, and Shanghai, which are municipalities that are under the direct control of
the Central Government of China, with economies that are primarily focused on the devel-
opment of the tertiary industry, have always been identified as agricultural GHG emission
cold spots. On the other hand, Liaoning and Shandong in the north, and Sichuan, Guangxi,
and Guangdong in the south have always been identified as hot spots, and in 2017, Xin-
jiang appeared as a new hot spot. The hot spots were relatively scattered. H–L agglom-
eration regions corresponded to Henan and Hebei, which are densely populated provinces
with large agricultural productivity and many competitive agricultural products.

3.2.2 Agricultural technology progress

To test the spatial aggregation characteristics of agricultural technology progress, global


Moran’s I indices for agricultural technology progress were determined, and the results are pre-
sented in Table 5, which shows that within the data range (2009–2017), all the indices were
positive and significant at a 99% confidence level. This implies that there was a spatial spillo-
ver effect regarding agricultural technological progress. The global Moran’s I indices fluctuated
around 0.31, exceeding the global Moran indices corresponding to agricultural GHG emissions
and indicating that agricultural technology progress had a more obvious spatial variation.
The LISA agglomeration map corresponding to the average annual agricultural tech-
nology progress (Fig.  9) indicated that agricultural technology progress also showed an
obvious local agglomeration. Hot spots and cold spots showed a north–south differentia-
tion trend. Hot spots were found to be concentrated in central regions, such as Hunan and
Hubei, and southern coastal regions, such as Guangdong, Guangxi, and Hainan. On the
other hand, cold spots were found to be concentrated in a large number in the northern
regions, such as Shanxi, Hebei, Liaoning, Jilin, Heilongjiang, and Inner Mongolia. Shan-
dong and Ningxia were found to be located in the H–L agglomeration area, while Tianjin
and Yunnan were found to be in the L–H agglomeration area.

3.3 Spillover effect of agricultural technology progress

3.3.1 Model selection

To select spatial econometric models, as shown in Table  6, spatial relevance tests were
conducted. The LM lag (robust) and LM error (robust) showed significance at 1% and 10%
significance levels, respectively, and given that LM lag (robust) showed a higher level of
significance than LM error (robust) implies that the spatial lag model (SAR) performed
better than the spatial error model (SEM). However, the p values of the Wald lag, Wald
error, LR lag, and LR error tests were 0, 0, 0.011, and 0.009, respectively, indicating that

13
2576 Natural Hazards (2020) 104:2561–2590

Fig. 8  Lisa agglomeration map of agricultural GHG emissions in 2009 and 2017

13
Natural Hazards (2020) 104:2561–2590 2577

Table 5  Global Moran’s I index Year Moran’s I Z-statistics P value


of agricultural technology from
2009 to 2017
2009 0.383*** 4.005 0.000
2010 0.368*** 3.863 0.000
2011 0.298*** 3.229 0.001
2012 0.280*** 3.065 0.001
2013 0.274*** 3.001 0.001
2014 0.277*** 3.051 0.001
2015 0.269*** 2.941 0.002
2016 0.296*** 3.203 0.001
2017 0.333*** 3.546 0.001

***,** and * indicate variables were significant at 1%, 5% and 10%


statistical levels, respectively

Fig. 9  Lisa agglomeration map of annual average agricultural technology progress

the spatial Durbin model (SDM) could not be reduced to the SAR or the SEM. Thus, the
SDM was the most suitable model.
Likelihood ratio (LR) tests and a Hausman test were performed to investigate whether it is
necessary to consider spatial and time-period fixed effects. As shown in Table 7, the test and
p value of the Hausman test, at a confidence level of 99%, were 70.29 and zero, respectively,
rejecting the original hypothesis that fixed effect models are better than random effect mod-
els. Additionally, the LR test showed that individual-fixed effects were better than individual-
time double fixed effects. Thus, individual-fixed effects were finally chosen for the SDM.

13
2578 Natural Hazards (2020) 104:2561–2590

Table 6  The results of spatial Test Statistics P value


panel econometric model test
LM lag (robust) 6.69*** 0.010
LM error (robust) 3.15* 0.076
Wald lag 33.58*** 0.000
Wald error 41.99*** 0.000
LR lag 16.48** 0.011
LR error 17.10*** 0.009

***,** and * indicate variables were significant at 1%, 5% and 10%


statistical levels, respectively

Table 7  The results of spatial Test Statistics P value


and time-period fixed effects
model test
Hausman test 70.29 0.000
LR test 50.62*** 0.000

***,** and * indicate variables were significant at 1%, 5% and 10%


statistical levels, respectively

3.3.2 Analysis on the spillover channels of agricultural technology progress

The geographic distance weight ( wijd ) and the economic distance weight ( wije ) were used to test
the spatial spillover channels of agricultural technological progress, and the results are as follows:
Table  8 compares the estimation results obtained using the SDM, SAR, SEM, and the panel
model. In general, there was no significant difference in the fitting effects of the four models; how-
ever, with the SDM, it was possible to extract the spatial effects of independent variables, such as the
agricultural technology progress and the cultivated area. From the perspective of the spatial autore-
gression coefficient, the coefficients of the three spatial models all passed the 1% significance test,
and the spatial autoregression coefficient of the SDM was relatively large, indicating that it had a
better performance with respect to the extraction of the spatial effect. Therefore, with the SDM, the
interaction of emission reduction strategies and information spillover effect could be better analyzed.
The estimations made using the SDM showed that regardless of the weight type considered,
geographic or economic distance weight, the spatial autoregression coefficient,𝜌 , and the pure
technology efficiency spatial lag term coefficient,𝜃Pech , both attained the 99% confidence level,
indicating a spatial spillover of China’s agricultural technological progress. These results, thus,
confirmed an interaction of the agricultural GHG emission reduction strategies of the different
regions. However, under the economic distance weight, the absolute value of the spatial lag term
coefficient of pure technological efficiency increased by 0.6 percentage points, and its statistical
significance also increased by 0.2 percentage points. On the other hand, the spatial lag term coef-
ficient of the agricultural technological progress potential became significant, attaining a 90%
confidence level. This indicated that economic connections are a more important channel for the
spatial spillover of agricultural technology. Therefore, shortening the economic distance between
regions favored the spatial spillover of agricultural technology progress to a greater extent.
The spatial autoregression coefficient, 𝜌 , was positive, indicating the interaction of
interregional agricultural GHG emission reduction strategies and implying the existence of
strategy emulation and competition, i.e., the emission reduction in one region causes other
regions to reduce emissions.

13
Table 8  Model estimation results
Variables SDM SAR SEM Panel model
wijd wije wijd wije wijd wije
Natural Hazards (2020) 104:2561–2590

ln(Pech) ( 𝛽Pech) 0.017 (0.89) 0.016 (0.83) − 0.018 (− 1.23) − 0.038** (− 2.64) − 0.003 (− 0.15) − 0.038** (− 2.61) − 0.041*** (− 2.70)
ln(Cul) ( 𝛽Cul) − 0.244*** (− 3.44) − 0.219*** (− 2.89) − 0.169* (− 2.4) − 0.126* (− 1.71) − 0.232*** (− 3.2) − 0.126* (− 1.73) − 0.129 (− 1.63)
ln(Cap) ( 𝛽Cap) − 0.051 (− 1.21) − 0.034 (− 0.81) − 0.076** (− 2) − 0.111* (− 2.80) − 0.083** (− 2.02) − 0.112** (− 2.85) − 0.115*** (− 2.72)
ln(Emp) ( 𝛽Emp) − 0.018 (− 0.15) − 0.029 (− 0.18) − 0.012 (− 0.1) 0.080 (0.60) − 0.082 (− 0.65) 0.077 (0.58) 0.067 (0.47)
ln(Str) ( 𝛽Str) − 0.300*** (− 2.80) − 0.328*** (− 3.02) − 0.280** (− 2.54) − 0.358** (− 3.13) − 0.232* (− 2.14) − 0.361** (− 3.14) − 0.372*** (− 3.01)
ln(Pch) ( 𝛽Pch) 0.005 (0.58) 0.008 (0.98) 0.017*** (0.53) 0.021 (0.63) 0.003 (0.39) 0.021 (0.61) 0.020 (0.55)
w *ln(Pech) ( 𝛽Pech) − 0.072*** (− 2.85) − 0.078*** (− 3.17)
w *ln(Cul) ( 𝛽Cul) 0.323** (2.42) 0.225* (1.84)
w *ln(Cap) ( 𝛽Cap) 0.070 (0.87) 0.003 (0.06)
w *ln(Emp) ( 𝛽Emp) 0.605* (1.81) 0.227 (0.91)
w *ln(Str) ( 𝛽Str) − 0.501** (− 2.01) − 0.311 (− 1.47)
w *ln(Pch) ( 𝛽Pch) 0.009 (0.58) 0.031* (1.95)
rho 0.317*** (3.74) 0.238*** (3.08) 0.393*** (5.12) 0.006*** (11.73) 0.006*** (11.63) 0.007***(11.77)

***, **, and * mean variables were significant at 1%, 5%, and 10% statistical levels, respectively
2579

13
2580 Natural Hazards (2020) 104:2561–2590

The pure technology efficiency spatial lag term coefficient, 𝜃Pech , was negative, indi-
cating that the spatial spillover of agricultural technological progress had a positive effect
on the regional correlation of agricultural GHG emissions. This implies that technologi-
cal spillover may exhibit a trickle-down effect, whereby high-quality resources, emission
reduction technologies, and experiences are transferred from one region to other regions.
Inter-regional spillover did not only mimic emission reductions; it also supported emission
reduction through technology spillovers. When the economic distance weight was used, the
spatial lag term coefficient of agricultural technological progress potential, 𝜃Pch , attained a
90% confidence level; however, it was not significant when the geographic distance weight
was used. This observation also shows that strengthening economic links favors the diffu-
sion of agricultural technology between regions.

3.3.3 Decomposition of the spatial spillover effects of agricultural technological


progress

The Durbin model’s regression coefficients were affected by the spatial matrix and could
not accurately describe the influence of independent variables, so that the effects of inde-
pendent variables were decomposed into direct effects and indirect effects. The results are
as follows.
Table 9 shows that the use of economic distance weight increased the statistical signifi-
cance of the total effect and the indirect effects of pure technical efficiency by 0.4 and 0.1
percentage points, respectively. The agricultural technological progress potential changed
from insignificant to significant, and the total effect and indirect effects passed the signifi-
cance test at 5% and 10% significance levels, respectively. This implies that shortening the
economic distance actually facilitates agricultural technology exchanges between regions.
On the one hand, with the development of internet informatization, the channels for infor-
mation transmission have been further widened, and the limitations of geographic distance
on technology transfer have been gradually decreased. On the other hand, when the eco-
nomic gap between two regions is too wide, the absorptive capacity of the backward region
is lower; thus, the spread of technology and knowledge is limited. However, a smaller eco-
nomic gap between the two regions facilitates the effective introduction, acceptance, and
implementation of advanced technologies.
From the impact of pure agricultural technological progress, the direct effect of the
independent variables was positive and passed the significance test at a 10% significance
level, indicating that an increase in agricultural technology by 1 percentage point leads
to an increase in local agricultural GHG emissions by 0.014 percentage points. In addi-
tion to improving energy efficiency, technological progress also promotes economic scale
expansion, increases energy input, and generates new energy demand; thus, it partially or
completely offsets saved energy, resulting in a phenomenon known as the rebound effect
(Xuan and Zhou 2011; Yang et al. 2017). Additionally, both the threshold effect and the
time lag effect influence technological progress. This implies that the complete applica-
tion of new technologies, from R&D to implementation, can take several years (time lag),
while different regional characteristics can determine whether the technology will be opti-
mally utilized (threshold). The indirect effect of the independent variable was negative and
passed the significance test at a 1% significance level, indicating that for one percentage
point of agricultural technology progress, GHG emissions in neighboring areas decreased
by 0.095 percentage points. This can be explained by the trickle-down effect (Zhang 2014;
Zhang et  al. 2012; Chakrabarti 2016). On the one hand, the strong radiant power of the

13
Natural Hazards (2020) 104:2561–2590

Table 9  Decomposition of spatial effect


Independent variable wijd wije
Total effect Direct effect Indirect effect Total effect Direct effect Indirect effect

ln(Pech) − 0.082*** (− 2.83) 0.014 * (1.44) − 0.095*** (− 3.07) − 0.081*** (− 3.29) 0.012 (0.64) − 0.093*** (− 3.29)
ln(Pch) 0.022 (0.76) 0.006 (0.70) 0.164 (0.67) 0.053** (1.96) 0.011 (1.24) 0.043* (1.92)
Ln (Cul) 0.107 (0.51) − 0.234*** (− 3.38) 0.341* (1.79) − 0.001 (− 0.001) − 0.212*** (− 1.24) − 0.212 (− 0.02)
ln(Cap) 0.043 (0.37) 0.042 (1.05) 0.086 (0.77) − 0.027 (− 0.31) − 0.029 (− 0.71) 0.002 (0.02)
ln(Emp) 0.849* (1.64) 0.013 (1.79) 0.838* (1.79) 0.257 (0.72) − 0.012 (− 0.09) 0.269 (0.87)
ln(Str) − 1.195*** (− 3.14) − 0.332*** (− 3.17) − 0.863** (− 2.52) − 0.849*** (− 2.96) − 0.352*** (− 3.29) − 0.498** (− 2.03)
2581

13
2582 Natural Hazards (2020) 104:2561–2590

central area diffused the advanced agricultural production technology as well as the emis-
sion reduction technology to neighboring areas, promoting agricultural productivity and
emission reductions in neighboring areas. On the other hand, for the neighboring areas, the
agricultural production environments were relatively similar, and the technology was found
to be highly adaptable. As learning and imitation effects spread, better technology spillover
could be achieved.
The total, direct, and indirect effects of the agricultural industrial structure all passed the
significance test at a 5% significance level, and the coefficients were all negative, indicat-
ing that industrial structure is associated with obvious spatial effects, leading to a spatial
spillover of GHG emission reduction. One region achieves emission reductions through the
upgrading of their agricultural industrial structure, and the neighboring regions will then
also adjust and upgrade their industrial structure per the imitation effect, resulting in a spa-
tial spillover effect of increasing GHG emission reduction.
Employment in the agricultural sector had a positive indirect impact on agricultural
GHG emission reduction in neighboring areas and passed the significance test at a 10%
significance level. This indicates that the regional spillover effect of agricultural GHG
emission reduction may be generated via the transfer of agricultural labor. Additionally, the
direct and indirect effects of cultivated land passed the significance test with at 1 and 10%
significance levels, respectively; however, offset effects, which caused the total effect to be
insignificant, were observed.
The capital investment coefficient was not significant. While it may be applied to all
aspects of agricultural production, its impact on GHG emission reduction was low when it
was invested in projects that have little correlation with GHG emissions, such as improving
agricultural infrastructure. Additionally, its impact on agricultural GHG emissions can be
both positive and negative. The purchase of advanced machinery and the improvement of
equipment for efficient agricultural production reduces GHG emissions. On the other hand,
higher production also increases agricultural GHG emissions. Thus, the offset between
these positive and negative effects makes its impact on GHG emissions uncertain.

4 Discussion

1. Only few previous studies have focused on the association of agricultural GHG emis-
sions between regions. Using different spatial models (Cole et al. 2013; Dong et al.
2016; Li et al. 2016; Marbuah and Amuakwa-Mensah 2017) or inter-regional input–out-
put tables (Meng et al. 2013; Liu et al. 2014; Zhang 2017; Jiao et al. 2018), it has been
confirmed that GHG emissions have significant spatial spillover effects. Additionally,
using social network analysis, the possibility of a complex network of correlations was
confirmed (Sun et al. 2016; Yang et al. 2016). Yet, the conclusions made from these
studies were relatively coarse scale, and specific industries were not analyzed. In this
study, we explored the spatial correlation of agricultural GHG emissions from two per-
spectives: emission reduction strategic imitation and technological information spillover.
It was observed that strategic imitation behaviors are on the rise, and technological
information spillover, which contributes to cooperation in the field of agriculture, has
a positive impact on agricultural GHG emission reduction.
2. Most previous studies only focused on specific sources of agricultural GHG emissions.
For example, Sovik and Klove (2007) measured N ­ 2O and C
­ H4 emissions from a con-
structed wetland in southeastern Norway and found that ­N2O and ­CH4 had an approxi-

13
Natural Hazards (2020) 104:2561–2590 2583

mately equal potential to contribute to global warming. In a study on the changes in


GHG emissions, as a result of the conversion of agricultural land, Arevalo et al. (2011)
observed that converting agricultural land to fast-growing short-rotation woody crops
has the potential to mitigate future climate change. Mantoam et al. (2016) focused on
carbon dioxide emissions resulting from the use of agricultural machinery, and based
on the results obtained, the hypothesis that more powerful tractors would require less
energy and emit less GHG per functional unit (mass and power) was proven. Addition-
ally, Sun et al. (2020) proposed that establishing and driving a region-specific model
with high-resolution data can improve the estimation accuracy of ­CH4 emissions from
paddy fields. We focused on the various sources of agricultural GHG emissions, aiming
to comprehensively and accurately measure agricultural GHG emissions, and calculated
agricultural GHG emissions with respect to five factors: agricultural energy, farmland
utilization, crop planting, ruminant feeding, and straw burning. This helps us grasp the
structure of agricultural GHG emissions and identify key areas for emission reduction.
We found that ruminant feeding and straw burning are the main sources of agricultural
GHG emissions in China. However, the growth trend of the GHG emissions from straw
burning and farmland utilization was the most obvious.
3. In previous studies, different researchers used different indicators to measure techno-
logical progress based on their research needs. Thus, there is no unified measurement
standard, and in most of these previous studies, only the technological progress level
was considered, while technological progress potential was neglected. Agricultural total
factor productivity has been extensively used to measure agricultural technological
progress (Tian and Zhang 2017; Pierluigi et al. 2017; Yang et al. 2018; Li et al. 2017;
Xu et al. 2019; Rybaczewska-Błażejowska and Magdalena 2018), and in few previ-
ous studies, agricultural technological progress was measured from the perspectives
of R&D capital investment (Snyder et al. 2016; Zhao et al. 2018), patent development
(Lu et al. 2019), energy intensity (Dyer et al. 2010), and carbon productivity (Cheng
et al. 2016). However, in this study, in addition to the agricultural technology progress
level, agricultural technology progress potential was also considered, and a more com-
prehensive measurement of technological progress was performed. Thus, we found that
the agricultural technology level can spillover through both geographic and economic
channels, while agricultural technology progress potential can only spillover through
economic channels.
4. In previous studies, agricultural technology spillovers have been examined from the
perspective of geographic distance, and the reasons for inter-regional agricultural GHG
linkages have been explained from the perspective of technology spillovers; however,
the possible channels of technology spillover were not examined in depth. Cheng et al.
(2016) used the agricultural carbon productivity to measure technological progress and
discussed the spatial correlation of agricultural carbon productivity based on the rook
adjacent weight. The results obtained showed that technological progress has significant
positive spatial autocorrelation and agglomeration characteristics. Yang and Li (2017)
also found that agricultural frontier technological progress and technical efficiency have
positive spatial spillover effects, based on the spatial adjacency relationship. Wang et al.
(2019) investigated the impact of food production technical efficiency on agricultural
GHG emissions by establishing a spatial error model. The results obtained showed that
technical efficiency significantly suppressed emissions. Additionally, via the establish-
ment of a spatial Durbin model, Li et al. (2015) observed that an increase in agricultural
science and technology input has an emission reduction effect on neighboring areas. In
addition to the analysis of regional GHG associations from the perspective of agricul-

13
2584 Natural Hazards (2020) 104:2561–2590

tural technology spillovers, in this study, the channels for agricultural technology spillo-
ver were further explored. Using geographic and economic weights, it was observed
that the spatial spillover effect of agricultural technological progress and agricultural
technological progress potential is more obvious when the economic weight is used.
This implies that shortening the economic distance will actually facilitate agricultural
technology exchanges between regions. This provides a reference for the exploitation
of more agricultural technology spillover channels in the future.

5 Conclusions

Using 2007–2017 data on 31 provinces in China, in this study, the interaction of agricul-
tural emission reduction behaviors, and the spillover effects of agricultural technological
progress in different regions of China were explored, and possible channels for techno-
logical spillover were discussed. The conclusions arrived at are as follows: (1) Ruminant
feeding and straw burning are the major sources of agricultural GHG emissions in China;
however, the growth trend of GHG emissions from straw burning and farmland utilization
is the most obvious. Thus, these two emission sources should be the focus of agricultural
GHG emission reduction strategies in the future. (2) The spatial correlation of agricultural
GHG emission reduction in various regions of China is not only limited to the interaction
of emission reduction strategies, and it also reflects the spillover of technical information.
On the one hand, owing to the constraints of China’s environmental regulations and eco-
nomic assessment, both economic and environmental competition exists between regions
in China. The adjustment in GHG emissions in one region will lead to the same in other
regions (increase or decrease), so as to meet national standards. Additionally, due to learn-
ing and imitation effects, the spatial spillover of technological progress will benefit more
regions with respect to emission reduction. (3) The investigation of the possible chan-
nels of spatial agricultural technology spillover using different spatial weighting matrices
showed that regions with strong economic ties are more likely to experience technology
spillovers, especially technological progress potential, which only spills over through the
economic weight. Therefore, strengthening economic ties between regions and advancing
the technical cooperation between them is an important strategy to reduce emissions.
Our policy recommendations are as follows: (1) Regional collaborative emission reduc-
tion mechanisms should be established and regional cooperation should be strengthened. This
will enhance the agriculture-related economic links between regions and will enable a bet-
ter allocation of agricultural resources among regions, thereby improving the conditions for
regional cooperation on agricultural GHG emission reduction. Additionally, the strategy of
“rich neighbors” should be advocated, regional barriers to emissions reduction eliminated, and
the sharing of energy savings and emission reduction experiences promoted through techni-
cal or financial cooperation. (2) To make full use of the demonstration and learning effect,
technology sharing platforms should be established to maximize the diffusion of regional agri-
cultural technology improvements. Additionally, for possible polarization effects, technology
transfer and control should be improved. (3) The structure of the agricultural industry should
be adjusted and optimized, i.e., plantation, animal husbandry, and fishery sectors should be
optimized according to local conditions.
This study can be extended by considering two avenues for future research: (1) Agriculture
is not only an important source of GHG emissions, and it also plays a role in GHG absorption
via two major channels: crop carbon sinks (Tian et al. 2015) and forest carbon sinks (Zhang

13
Natural Hazards (2020) 104:2561–2590 2585

and Pang 2016). Therefore, the dual characteristics of the agricultural industry can be consid-
ered in future studies. (2) The inclusion of the impact of national environmental policies on
regional cooperation in reducing emissions into the GHG emission analysis framework should
also be investigated.

Author contributions  YH, XC, FW wrote and revised the paper; YC revised the paper.

Funding  This research was funded by a National Natural Science Foundation Youth Project (Grant number
71704127).

Data availability  Data and material are available.

Code availability  Code is available.

Compliance with ethical standards 


Conflicts of interest  The authors declare no conflicts of interest.

Appendix 1: Agricultural GHG emission coefficient

See Table 10.
The energy consumption emission coefficient uses China’s specific value and is calculated
as follows.
( )
12
c × o × 44 × 1000
fi =
109
(13)
ji

where fi is the carbon dioxide emission factor of the ith fossil energy,ci and o is the default
carbon content and the oxidation factor of the ith fossil energy. Assuming that the fossil
energy is completely burned, the oxidation factor is 1, ji is the average low calorific value
of the ith fossil energy (data source: China Energy Statistics Yearbook), 109 ∕j is the mass
of the ith fossil energy that generate1 trillion joules of heat, 12 and 44 are the mass of car-
bon and carbon dioxide (Tables 11, 12, 13, 14, 15).
The calculation formula for the emission factor is as follows:
Ei = 0.5 × EL + 0.5 × CA (14)
where Ei is the emission coefficient of the ith province, EL is electricity marginal emission
coefficient, CA is capacity marginal emission coefficient, data come from China Develop-
ment and Reform Commission.

Table 10  GHG emission Agricultural energy Emission Agricultural energy Emission


coefficient of main agricultural coefficient coefficient
energy consumption (unit:
kg(CO2)/kg)
Coal 1.47 Kerosene 4.55
Coke 2.96 Fuel oil 4.62
Oil 4.38 Diesel oil 4.60
Gasoline 4.69 Natural gas 0.29

13
2586 Natural Hazards (2020) 104:2561–2590

Table 11  GHG emission coefficient of farmland utilization. Data source: Tian et al. (2012)
Farmland utilization Emission coefficient Farmland utilization Emission coefficient

fertilizer 3.28 kg(CO2)/kg Agricultural film 18.99 kg(CO2)/kg


pesticide 18.09 kg(CO2)/kg Plowing 12 kg(CO2)/ha

Table 12  GHG emission coefficient of major ruminants. Data source: Min and Hu (2012)
Ruminant Emission coefficient of Feces management
intestinal fermentation
Emission coefficient of ­CH4 Emission coefficient of N
­ 2O

cattle 55.93 kg(CH4)/head year 6.67 kg(CH4)/head year 1.34 kg(N2O)/head year


horse 18 kg(CH4)/head year 1.6 kg(CH4)/head year 1.39 kg(N2O)/head year
donkey 10 kg(CH4)/head year 0.9 kg(CH4)/head year 1.39 kg(N2O)/head year
mule 10 kg(CH4)/head year 0.9 kg(CH4)/head year 1.39 kg(N2O)/head year
pig 1 kg(CH4)/head year 3.8 kg(CH4)/head year 0.5 kg(N2O)/head year
goat 5 kg(CH4)/head year 0.2 kg(CH4)/head year 0.3 kg(N2O)/head year
sheep 5 kg(CH4)/head year 0.2 kg(CH4)/head year 0.3 kg(N2O)/head year

Table 13  GHG emission Straw burning Emission coefficient


coefficient of straw burning.
Data source: Liu et al. (2011) kg(CO2)/kg g(CH4)/kg g(N2O)/kg

Rice 0.66 20 0.1


Wheat 0.59 20 0.1
Rape 0.80 30 0.1
Soybean 0.54 30 0.1
Corn 0.62 30 0.1
Cotton 0.46 20 0.1

Table 14  Emission coefficient of rice cultivation in China (unit: kg(CH4)/m2). Data source: Wang et  al.
(1998)
Area Emission Area Emission Area Emission Area Emission
coefficient coefficient coefficient coefficient

Beijing 13.23 Shanghai 31.26 Hubei 38.20 Yunnan 05.70


Tianjin 11.34 Jiangsu 32.40 Hunan 35.00 Tibet 06.83
Hebei 15.33 Zhejiang 35.60 Guangdong 41.20 Shanxi 12.51
Shanxi 06.62 Anhui 31.90 Guangxi 36.40 Gansu 06.83
Inner Mongolia 08.93 Fujian 34.60 Hainan 38.40 Qinghai 00.00
Liaoning 09.24 Jiangxi 42.20 Sichuan 16.90 Ningxia 07.35
Jilin 05.57 Shandong 21.00 Chongqing 16.90 Xinjiang 10.50
Heilongjiang 08.31 Henan 17.85 Guizhou 16.10 – –

13
Natural Hazards (2020) 104:2561–2590 2587

Table 15  Emission coefficient of electric power in China (unit: kg(CO2)/kwh)


Area Emission Area Emission Area Emission Area Emission
coefficient coefficient coefficient coefficient

Beijing 1.99 Shanghai 0.71 Hubei 2.96 Yunnan 1.69


Tianjin 1.73 Jiangsu 6.68 Hunan 1.93 Tibet 0.86
Hebei 10.76 Zhejiang 2.80 Guangdong 9.99 Shanxi 4.19
Shanxi 4.12 Anhui 2.39 Guangxi 3.04 Gansu 4.58
Inner Mongolia 5.01 Fujian 2.97 Hainan 1.47 Qinghai 0.25
Liaoning 3.88 Jiangxi 1.28 Sichuan 1.55 Ningxia 1.87
Jilin 1.59 Shandong 11.01 Chongqing 0.34 Xinjiang 13.21
Heilongjiang 5.23 Henan 8.86 Guizhou 0.77 – –

Appendix 2: equations of Moran’s I

The globe Moran’s I is defined as follows:


∑n ∑n � �� �
i=1 𝜔ij yi − ȳ yj − ȳ
j=1
I= ∑ ∑n (15)
s2 ni=1 i=1 𝜔ij

1 ∑( )
n
s2 = y − ȳ (16)
n i=1 i

∑n
y
ȳ = i=1 i
(17)
n
The local Moran’s I is defined as follows:

n
Ii = zi 𝜔ij zj (18)
j=1

yi − ȳ
zi = (19)
s
where I and Ii are global Moran’s I index and local Moran’s I index, yi and yj are the agri-
cultural GHG emissions of provinces and cities i and j, ȳ is the average agricultural GHG
emissions, 𝜔ij is row i, column j of spatial weight matrix, s2 is variance of agricultural
GHG emissions, and n is the number of provinces and cities.

Appendix 3: Decomposition of the total effect of the Durbin model

For spatial econometric models, the variation of the independent variable not only affects
itself through the spatial multiplier effect, but also indirectly affects other regions. The Spa-
tial Durbin Panel Data Model can be rearranged as follows:

13
2588 Natural Hazards (2020) 104:2561–2590

( ) ( )
ynt = (I − 𝜌w)−1 𝛽xnt + 𝜃wxnt + (I − 𝜌w)−1 𝜏n 𝛼 + 𝜇n + 𝜐t + 𝜀nt (20)

Decompose the total effect into direct and indirect effects through partial derivatives.
Result is as follows:

⎡ 𝜕E(y1 ) 𝜕E(y1 ) ⎤
𝜕E(y1 )
⎢ 𝜕 (x1k ) 𝜕 (x2k )
⋯ ⎥
𝜕 (xnk )
⎢ 𝜕E(y2 ) 𝜕E(y2 ) ⎥
𝜕E(y2 )
𝜕E(yl ) 𝜕E(yn )
⋯ = ⎢ 𝜕 (x1k ) 𝜕 (x2k ) 𝜕 (xnk ) ⎥
𝜕 (xik ) 𝜕 (xnk ) ⎢ ⋮ ⋱ ⋮ ⎥
⎢ 𝜕E(yn ) 𝜕E(yn ) 𝜕E(yn )

⎢ ⋯ ⎥ (21)
⎣ 𝜕 (x1k ) 𝜕 (x2k ) 𝜕 (nnk ) ⎦

⎡ 𝛽1k 𝜔12 𝜃k ⋯ 𝜔1n 𝜃k ⎤


−1 ⎢
𝜔 𝜃 𝛽2k 𝜔2n 𝜃k ⎥
= (I − 𝜌𝜔) ⎢ 21 k ⎥
⋮ ⋱ ⋮
⎢ ⎥
⎣ 𝜔n1 𝜃k 𝜔n2 𝜃k ⋯ 𝛽nk ⎦

where yi is agricultural GHG emission intensity of provinces i, xik is kth independent vari-
able of province i, 𝜔ij is the ith row and the jth column of spatial weight matrix. The direct
effect is defined as the sum of the diagonal elements in the matrix (formula (9)) and means
the impact of independent variables on dependent variables in the same area, including
feedback effects. The indirect effect is defined as the average of all the elements other
than the diagonal elements, also known as the spatial spillover effect. It is used to measure
the influence of independent variables of “adjacent” areas on the dependent variables in
another region. The total impact is the sum of the direct impacts and indirect impacts.

References
Arevalo C, Bhatti J, Chang S, Sidders D (2011) Land use change effects on ecosystem carbon balance: from
agricultural to hybrid poplar plantation. Agric Ecosyst Environ 141(3–4):342–349
Caves DW, Christensen LR, Diewert WE (1982) The economic theory of index numbers and the measure-
ment of input, output, and productivity. Econometrica 50(6):1393–1414
Chakrabarti AS (2016) Stochastic Lotka-Volterra equations: a model of lagged diffusion of technology in an
interconnected world. Physica AStat Mech Appl 442:214–223
Cheng L, Zhang J, Tian Y, Zhou X (2016) The spatial variation characteristics and dependency of agricul-
tural carbon productivity in China. Res Sci 38(2):276–289
Cole MA, Elliott RJR, Okubo T, Zhou Y (2013) The carbon dioxide emissions of firms: a spatial analysis. J
Environ Econ Manage 65(2):290–309
Dong F, Long R, Li Z, Dai Y (2016) Analysis of carbon emission intensity, urbanization and energy
mix: evidence from China. Nat Hazards 82(2):1375–1391
Dyer JA, Kulshreshtha SN, McConkey BG, Desjardins RL (2010) An assessment of fossil fuel energy
use and CO2 emissions from farm field operations using a regional level crop and land use database
for Canada. Energy 35(5):2261–2269
Fan C, Wei T (2016) Effectiveness of integrated low-carbon technologies: evidence from a pilot agricul-
tural experiment in Shanghai. Int J Clim Change Strateg Manag 8(5):758–776
Fare R, Grosskopf S, Lovell CAK (1994) Production Frontiers. Dissertation, University of Cambridge
Fei R, Lin B (2017) Technology gap and ­CO2 emission reduction potential by technical efficiency meas-
ures: a meta-frontier modeling for the Chinese agricultural sector. Ecol Indic 73:653–661
Guan X, Zhang J, Wu X, Cheng L (2018) The shadow prices of carbon emissions in China’s planting
industry. Sustainability 10(3):753
Hu Z (2018) Technological progress, technical efficiency and agricultural carbon emissions in China.
Esat Chin Wcon Manag 32(6):100–105

13
Natural Hazards (2020) 104:2561–2590 2589

International Energy Agency (IEA) (2019) Global energy & CO2 status report 2019. https​://www.iea.
org/repor​ts/globa​l-energ​y-co2-statu​s-repor​t-2019. Accessed 26 Mar 2019
IPCC (2007) Climate change 2007: the physical science basis: Working Group I Contribution to the
Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge Univer-
sity Press, New York, p 2007
IPCC (2019) Climate Change and Land. https​://www.ipcc.ch/srccl​/2019 Accessed 15 Aug 2019
Ismael M, Srouji F, Boutabba MA (2018) Agricultural technologies and carbon emissions: evidence
from Jordanian economy. Environ Sci Pollut Res 25(11):10867–10877
Jiao J, Yang Y, Bai Y (2018) The impact of inter-industry R&D technology spillover on carbon emission
in China. Nat Hazards 91(3):913–929
Li B, Zhang J, Li H (2011) Research on spatial-temporal characteristics and affecting factors decomposi-
tion of agricultural carbon emission in China. China Popul Resour Envrion 21(08):80–86
Li Q, Li C, Xiao X, Wu H (2015) The spatial effects of agricultural carbon emissions in China: based on
spatial Durbin model. J Arid Land Resour Environ 29(4):30–35
Li L, Hong X, Tang D, Na M (2016) GHG emissions, economic growth and urbanization: a spatial
approach. Sustainability 8(5):462
Li N, Jiang Y, Yu Z, Shang L (2017) Analysis of agriculture total-factor energy efficiency in China based
on DEA and Malmquist indices. Energy Procedia 142:2397–2402
Liu X, Jia R (2007) How science and technology of Beijing regions is diffused in China—what technol-
ogy market can tell us. Sci Manag Sci Technol 12:32–38
Liu L, Jiang J, Zong L (2011) Research on greenhouse gas emission inventory from agricultural residue
combustion: a case study of Jiangsu Province. Envrion Sci 32(5):1242–1248
Liu Y, Xiao H, Zikhali P, Lv Y (2014) Carbon emissions in China: a spatial econometric analysis at the
regional level. Sustainability 6(9):6005–6023
Liu X, Yu Y, Luan S (2019) Empirical study on the decomposition of carbon emission Factors in agri-
cultural energy consumption. IOP Conf Ser Earth Environ Sci 252:042045–042051
Lu X, Kuang B, Li J, Han J, Zhang Z (2018) Dynamic evolution of regional discrepancies in carbon
emissions from agricultural land utilization: evidence from Chinese provincial data. Sustainability
10(2):552
Lu N, Wa W, Wang M, Zang C, Lu H (2019) Breakthrough low-carbon technology innovation and car-
bon emissions: direct and spatial spillover effect. Chin Popul Resour Environ 29(05):30–39
Ma D (2018) Spatial heterogeneity and influencing factors of agricultural energy carbon emis-
sion efficiency in China—an empirical research of spatial panel data model. Resour Dev Mark
34(12):1693–1765
Ma H, Huang D, Yao H (2011) Total-factor energy efficiency analysis of three major economic regions
in China: based on super-DEA and Malmquist. China Popul Res Environ 21(11):38–43
Malmquist S (1953) Index numbers and indifference surfaces. Traba de Estad de Investig Operat
4(2):209–242
Mantoam EJ, Romanelli TL, Gimenez LM (2016) Energy demand and greenhouse gases emissions in
the life cycle of tractors. Biosyst Eng 151:158–170
Marbuah G, Amuakwa-Mensah F (2017) Spatial analysis of emissions in Sweden. Energy Econ
68:383–394
Meng B, Xue J, Feng K, Guan D, Fu X (2013) China’s inter-regional spillover of carbon emissions and
domestic supply chains. Energy Policy 61:1305–1321
Min J, Hu H (2012) Calculation of greenhouse gases emission from agricultural production in China.
China Popul Res Environ 22(7):21–27
Nelson AR, Gennaro DA, Ignacio A (2019) Global warming and warning. Clinics 74:e1219
Owusu PA, Asumadu-Sarkodie S (2017) Is there a causal effect between agricultural production and car-
bon dioxide emissions in Ghana? Environ Eng Res 22(1):40–54
Pierluigi T, Pier PM, Giovanni Z, Donatella V, Irene P (2017) A non-parametric bootstrap-data envel-
opment analysis approach for environmental policy planning and management of agricultural effi-
ciency in EU countries. Ecol Indic 83:132–143
Rybaczewska-Błażejowska M, Magdalena G (2018) Eco-efficiency evaluation of agricultural production
in the EU-28. Sustainability 10:4544
Snyder CS, Davison EA, Smith P, Venterea RT (2016) Research into dynamic lag effect of R&D
input on economic growth based on the vector auto-regression model. J Comput Theor Nanosci
13:6787–6796
Soni P, Taewichit C, Salokhe VM (2013) Energy consumption and ­CO2 emissions in rainfed agricultural
production systems of Northeast Thailand. Agric Syst 116:25–36

13
2590 Natural Hazards (2020) 104:2561–2590

Sovik AK, Klove B (2007) Emission of N2O and CH4 from a constructed wet land in southeastern Nor-
way. Sci Total Environ 380:28–37
Sun Y, Liu H, Liu C, Cui R (2016) Research on spatial association of provinces carbon emissions and its
effects in China. Shanghai Econ Res 2:82–92
Sun J, Wang M, Xu X, Cheng K, Yue Q, Pan G (2020) Re-estimating methane emissions from Chinese
paddy fields based on a regional empirical model and high-spatial-resolution data. Environ Pollut
265:115017
Tian Y, Zhang J (2017) Research on carbon emission from China’s agriculture, low carbon agricultural
productivity and their coordination. J Chin Agric Univ 22(05):208–213
Tian Y, Zhang J, Li B (2012) Agricultural carbon emissions in China: calculation, spatial-temporal com-
parison and decoupling effects. Res Sci 34(11):2097–2105
Tian Y, Zhang J, Wu X, Cheng L (2015) Research on dynamic change and regional differences of Chi-
na’s planting industry carbon sink surplus. J Nat Res 30(11):1885–1895
Wang M, Li J, Zheng X (1998) Methane emission and mechanisms of methane production oxidation,
transportation in the rice fields. Atmos Sci 4:3–5
Wang J, Zhang Y, Tian Y (2019) Influencing factors and spatial spillover of agricultural carbon emis-
sions in major grain producing areas in China. J South Chin Agric 50(07):1632–1639
Wu Y, Feng K (2019) Spatial-temporal differentiation features and correlation effects of provincial agri-
cultural carbon emissions in China. Environ Sci Technol 42(3):180–190
Wu X, Zhang J, Cheng L, Tian Y (2015) Potential of agricultural carbon emission reduction in China’s
provinces and its spatial correlation characteristics—spatial Durbin model based on spatial weight
matrix. Chin Popul Resour Resour 25(06):53–61
Xiong C, Yang D, Xia F, Huo J (2016) Changes in agricultural carbon emissions and factors that influence
agricultural carbon emissions based on different stages in Xinjiang, China. Sci Rep 6(1):36912–36921
Xu Y, Liu Y, He Y (2017) Trans-regional technology transfer pattern and policy implications of Beijing
universities and research institutes. Sci Res Manag 38(S1):444–452
Xu X, Huang X, Huang J, Gao X, Chen L (2019) Spatial-temporal characteristics of agriculture green
total factor productivity in China, 1998–2016: based on more sophisticated calculations of carbon
emissions. Int J Environ Res Publ Health 16(20):3932
Xuan Y, Zhou S (2011) Technology Innovation, return effect and energy efficiency in China’s industrial
industry. Financ Trade Econ 1:116–121
Yang L, Li Z (2017) Technology advance and the carbon dioxide emission in China—empirical research
based on the rebound effect. Energy Policy 101:150–161
Yang G, Wu Q, Tu Y (2016) Researches of China’s regional carbon emission spatial correlation and its
determinants: based on the method of social network analysis. Bus Econ Manag 4:56–68
Yang Y, Lin W, Zhang L (2017) Agricultural technology progress, technical efficiency and grain produc-
tion-empirical analysis from China’s provincial panel data. J Agric Technol 5:46–56
Yang Z, Wang D, Du T, Zhang A, Zhou Y (2018) Total -factor energy efficiency in China’s agricultural
sector: trends, disparities and potentials. Energy 11(4):853
Yu Y, Gu X, Chen Y (2016) Research on the gravity model, boundary effect and the Inter-provincial technol-
ogy transfer: based on the panel data of Chinese provinces from 2009 to 2013. Soft Sci 30(07):15–18
Zhang H (2014) Does environmental regulations enhance carbon emissions performance? Interpretation
from the perspective of spatial spillover. Econ Manag 36(12):166–175
Zhang Y (2017) Interregional carbon emission spillover-feedback effects in China. Energy Policy
100:138–148
Zhang Z, Mu Y (2019) Can industrial agglomeration improve agricultural carbon productivity? Chin
Popul Resour Environ 29(07):57–65
Zhang Y, Pang J (2016) The research on economic accounting of forest carbon sinks and its balance sheet
establishment. Stat Res 33(11):71–76
Zhang J, Qing Y, Zhang L, Lu C (2012) The spatio-temporal differentiation of C­ O2 emissions in cities along
the Lower Yellow River. Geogr Res 31(12):2229–2239
Zhang H, Guo S, Qian Y, Liu Y, Lu C (2020) Dynamic analysis of agricultural carbon emissions efficiency
in Chinese provinces along the Belt and Road. PLoS ONE 15(2):e0228223
Zhao X, Wang X, Guo Q (2018) Research on the relationship between regional innovation capability and
coordinated development of agricultural economy and ecological environment—spatial econometric
analysis based on provincial panel. Sci Technol Progress Policy 35(07):35–42

Publisher’s Note  Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

13

You might also like