Impact of urbanization on pollution-related agricultural input intensity in Hubei, China.

Agricultural input intensity increases significantly during the rapid urbanization in China, which has contributed to the increasingly serious non-point pollution. Using the vector autoregression (VAR) model, this study analyzes the impact of urbanization on pollution-related agricultural input intensity in Hubei, China. Results of an impulse response function analysis reveal that pesticide use intensity continues to rise following shocks from the urban population proportion and the secondary and tertiary industry proportion. Responses of chemical fertilizer intensity first decrease and then increase subjected to the shocks from the urban population proportion and secondary and tertiary industry proportion. The intensity of agricultural plastic film use first increases and then decreases when receives the shocks from the urban population proportion which is the opposite to the response to the shock from the secondary and tertiary industry proportion. In addition, the responses of pesticide use intensity, chemical fertilizer use intensity and agricultural plastic film use intensity trend decrease following their own shocks after positive initial responses. The variance decomposition results demonstrate that the shocks due to pesticide use intensity, chemical fertilizer use intensity and agricultural plastic film use intensity generally explain the largest proportion of their own variation over the 10-year horizon. However, an increase in the urban population proportion plays a critical role in determining the variations of pesticide use intensity in late periods, it account for 56.88% the variations in the tenth period. And the contribution of the urban population proportion to the variations in agricultural plastic film use intensity increases consistently, it account for 33.74% of the variations in the tenth period. Therefore, the hidden drivers of these phenomena need to be further understood regarding the relationships between urbanization and diffuse pollution from agricultural production.

Introduction

Since the open-door reform in the late 1970s, urbanization has accelerated rapidly in China due to rural–urban migration and administration adjustment (Shen et al., 2015, Goldstein, 1990, Chan, 1994). China's urbanization is characterized by a remarkable rate of economic growth, which has greatly enhanced the well-being and economic stability. However, the unprecedented scale of urbanization in China has also brought serious challenges, including economic disparity and the concentration of wealth, higher levels of pollution, an urban population explosion, high costs of living and a lack of basic public services (Chang, 2002, Economy, 2007, Ravallion et al., 2007, Zheng et al., 2007). This urban transformation poses great challenges to the sustainable development (Valipour, 2014, Valipour, 2015a).

Agricultural pollution is one of the urgent national priorities in China considering the wide-ranging occurrence and serious consequences (Saysel et al., 2002, Su et al., 2011). Rapid urbanization promotes excessive farmland conversion and the rapid development of highly polluting industries, which pose a serious threat to the eco-environment in China (Su et al., 2014b). Agricultural pollution consistently emerges in step with economic growth (Fischer et al., 2012). It is argued that China's agricultural pollution is closely linked to the large inputs of chemical fertilizers, pesticides and plastic films (Ma et al., 2014, Smith and Siciliano, 2015, Yang et al., 2014). It is estimated that China has become the largest user of agricultural chemicals in the world (Huang et al., 2002). Although Chinese government has taken actions through technology advancement and policy intervention, it is still in the struggle to find the best and most effective solutions to control the pollution-related agricultural input (chemical fertilizers, pesticides, agricultural plastic film) (Valipour, 2015b, Valipour et al., 2015). Besides, the booming urbanization should accelerate the agricultural land conversion in the rural areas, and the farmers may increase the pollution-related input intensity to maintain and enhance productivity. Consequently, in order to tackle with the agricultural pollution in the rapid urbanization owing to the great pollution-related agricultural input intensity, it requires elaborate efforts to quantify the relationship between urbanization and pollution-related agricultural input intensity.

Pollution-related agricultural input intensity represents the amount of agricultural inputs per unit of land (Björklund et al., 1999, Renetzeder et al., 2010). Many factors can lead to the variations in the pollution-related agricultural input intensity. Brasselle et al. (2002) reported that assurance and realizability effects, which are associated with land tenure security, provide great incentives for investments in agricultural land. Phimister and Roberts (2006) found that off-farm work has significant impacts on input intensity; and the intensity of fertilizer use declines as the amount of off-farm labor increases. Udry (1996) reported the gender-related differences in sub-Saharan Africa, which found that the intensity of manure use and labor inputs in plots farmed by women is lower than in similar plots farmed by men. Jorgenson (2007) reported that pesticide input intensity in 40 less-developed countries rose in correspondence with the growth of direct foreign investment. Chibwana et al. (2014) found that the Farm Input Subsidy Program (FISP) in Malawi, whose purpose is to stimulate the production of food crops, has increased fertilizer use intensity among smallholder farmers. However, the relationship between urbanization and pollution-related agricultural input intensity has not yet been clearly defined.

The primary method to analyze the linear relationships between variables is ordinary least squares (OLS). However, OLS fails to take into account the endogeneity. Therefore, the general assumption for OLS is that the independent variable should not be correlated with the error term. Violating this assumption, when applying OLS on time series data, the endogeneity problem would causes bias in coefficients estimation. In this study, there may be a loop of causality between urbanization variables and the intensity variables of the pollution-related agricultural inputs which is one important cause of endogeneity. The VAR model proposed by Sims (1980) allows for the potential endogeneity in the multiple time series (Phillips, 1995). It is widely used to analyze linear interdependencies among multiple time series. The VAR model is a theory-free method for assessing the relationships between variables. It provides a flexible and computationally tractable method to capture the effects of structural shocks between variables by treating all of the selected variables as endogenous (Kumar et al., 2012). In this regard, the VAR model is appropriate for analyzing the temporal variations in pollution-related agricultural input intensity caused by urbanization. In this study, I specifically attempt to (1) develop the VAR model for analyzing the impact of urbanization on pollution-related agricultural input intensity; (2) discuss the dynamic relationships between urbanization and agricultural pollution-related input intensity in Hubei, China.

Study area

Hubei Province, located in the middle part of the Yangtze River, is a major agricultural production base in Central China (Fig. 1 ). The dominant climate in Hubei is humid subtropical climate. Hubei has a long growing season which couples with high rainfall for agricultural crops. The western portion of Hubei province is covered by mountains and hills, and the middle and the southern portion of Hubei province is mainly covered by plain.

Fig. 1

Location of Hubei, China.

To feed a growing population, agriculture in Hubei has gradually transformed from a traditional to an intensive mode. The intensive agriculture employs modern technology, but is in charged by small peasant households under a household-responsibility system. The peasant households are fancy to increase the agricultural inputs (pesticides, fertilizers, etc.) in pursuit of high yields, putting stress on the eco-environmental quality. It exemplifies the worsening non-point pollution associated with agricultural chemical inputs in China. Hubei province was therefore selected to analyze the impact of urbanization on pollution-related agricultural input intensity.

Materials and methodology

Data collection

The selected urbanization variables include the urban population proportion and the secondary and tertiary industry proportion (Chan and Hu, 2003, Zhang and Lin, 2012, Aunan and Wang, 2014). Pesticides, chemical fertilizers and agricultural plastic film represent the pollution-related agricultural inputs (Ma et al., 2014, Smith and Siciliano, 2015, Yang et al., 2014, Guang-Rong et al., 2014, Wang et al., 2015). The pollution-related agricultural input intensity is measured as the ratio of the amount of the input to the agricultural land area (Table 1 ). Data for the selected variables 1992–2013 were obtained from the Hubei Bureau of Statistics. Descriptive statistics are shown in Table 2 .

Table 1

Definitions of variables used in the VAR model.

Variable	Symbol	Definition	Unit
Urban population proportion	UP	Ratio of urban population to total population	–
Secondary and tertiary industry proportion	STP	Ratio of gross product of secondary and tertiary industry to gross regional domestic product	–
Pesticide use intensity	PUI	Pesticide use per unit of agricultural land	kg/h m2
Chemical fertilizer use intensity	CUI	Chemical fertilizer use per unit of agricultural land	kg/h m2
Agricultural plastic film use intensity	AUI	Agricultural plastic film use per unit of agricultural land	kg/h m2

Table 2

Statistical description of selected variables.

Variable	Obs.	Min.	Max.	Mean	Std. dev.
UP	22	0.281	0.545	0.404	0.081
STP	22	0.705	0.874	0.805	0.061
PUI	22	16.416	42.119	33.995	7.438
CUI	22	482.498	1055.636	841.238	170.970
AUI	22	10.739	22.292	16.899	2.677

VAR model

The VAR model is used to capture the dynamic impact of random disturbances on variables under different scenarios (Khalid and Kawai, 2003, Xu and Lin, 2015, Tan and Lu, 2015), and the each variable in the VAR model has an equation which explains its evolution based on its own lags and the lags of the other variables (Sims, 1980). The time series analysis using the VAR model is based on the assumption that the time series of variables can be considered to be weakly stationary or approximately stationary. Therefore unit-root test should be conducted to check the time series. If the original time series is not stationary, the cointegration test should be done to identify whether one time series has a long-run equilibrium. Meanwhile, the prior assumption for selecting VAR model is that the lagged pattern of the relationships exists among endogenous variables. So the Granger causality test which is a widely applied method for investigating whether one time series contains information that can help predict another is conducted (Granger, 1969).

The mathematical formula for a pth order VAR model is expressed aswhere Y is a K  × 1 vector of the variables representing urbanization and pollution-related agricultural input intensity; Y is the pth lag of Y; Φ is a K  ×  K vector of the coefficient matrix; X is a T  × 1 vector of exogenous variables; H is a K  ×  T vector of the coefficient matrix, and ɛ is a K  × 1 vector of the error term, which satisfies N(0, σ 2).

Results and discussion

General trend of urbanization and agricultural pollution-related input intensity

Fig. 2(a) illustrates the trend of increasing urbanization in Hubei. The proportion of the urban population grew from 0.293 in 1992 to 0.594 in 2013 at a rate of 2.99% per year. However, the urban population proportion decreased between 1996 and 1999. Urban population grew more slowly than the rural population under the strict birth control policy of “one child for one family”. The secondary and tertiary industry proportion grew by 21.18% between 1992 and 2013, with a 0.92% annual average rate of increase. In addition, a decreasing trend in the secondary and tertiary industry proportion was observed from 1993 to 1997 due to the soft landing policy engineered by the Chinese government to slow high inflation in the summer of 1993. Therefore, the development of secondary and tertiary industry in Hubei shifted from fast growth to slow growth.

Fig. 2

Temporal trend of urbanization (a) and pollution-related agricultural input intensity (b) in Hubei from 1992 to 2013.

Fig. 2(b) shows an increasing trend in the pollution-related agricultural input intensity in Hubei. The annual growth in the intensity of pesticide use was rapid at the beginning of the 1990s but then slowed. Chemical fertilizer use intensity grew faster than pesticide use intensity and agricultural plastic film use intensity. The intensity of chemical fertilizer use doubled during this period. Agricultural plastic film use intensity grew at a speed of 2.35%. In additional, there was a considerable increase of agricultural plastic film intensity in 2003. The primary cause was that the price of agricultural plastic film decreased because China's economy suffered a serious downturn in 2003 during the outbreak of severe acute respiratory syndrome (SARS) (Keogh-Brown and Smith, 2008). The low price of agricultural plastic film stimulated the increases of agricultural plastic film intensity in Hubei.

The Mann–Kendall's test was applied to statistically detect the general trend (increase, decrease, or no significant change). The significant increasing trends were detected for urbanization and agricultural pollution-related input intensity since the values of UFk of UP, STP, PUI, CUI and AUI are positive, and the calculated probabilities are greater than the 95% significance level (Table 3 ).

Table 3

Mann–Kendall's test (UFk).

	UP	STP	PUI	CUI	AUI
1992	0.000*	0.000*	0.000*	0.000*	0.000*
1993	1.000*	1.000*	1.000*	1.000*	−1.000*
1994	−0.522*	−0.522*	1.567*	1.567*	0.522*
1995	0.679*	−0.679*	2.038*	2.038*	0.679*
1996	1.470*	−0.490*	1.960*	2.449*	0.490*
1997	1.691*	0.188*	2.067*	2.818*	0.564*
1998	1.952*	1.051*	1.952*	3.154*	0.150*
1999	2.227*	1.732*	2.227*	2.969*	0.000*
2000	2.711*	2.294*	2.711*	2.711*	0.417*
2001	3.130*	2.773*	2.952*	2.594*	0.984*
2002	3.503*	3.192*	3.192*	3.036*	1.635*
2003	3.840*	3.566*	3.017*	3.429*	2.194*
2004	4.149*	3.539*	3.416*	3.783*	2.562*
2005	4.434*	3.887*	3.668*	4.106*	2.792*
2006	4.701*	4.206*	4.008*	4.404*	2.920*
2007	4.952*	4.502*	4.322*	4.682*	3.242*
2008	5.190*	4.614*	4.614*	4.943*	3.460*
2009	5.417*	4.886*	4.735*	5.189*	3.750*
2010	5.633*	5.143*	5.003*	5.423*	4.023*
2011	5.840*	5.386*	4.996*	5.645*	4.283*
2012	6.039*	5.617*	4.892*	5.737*	4.530*
2013	6.232*	5.837*	4.765*	5.781*	4.709*

Notes:

Significance at 95% confidence interval.

The dynamic relationships between urbanization and agricultural pollution-related input intensity

Unit-root test

The VAR model is inappropriate for the analysis of non-stationary time series because the parameter estimation would be biased or unreliable. To make a time series stationary, differencing can be performed to remove any trend or seasonality. Fig. 2 shows an obvious increasing trend in the urbanization and agricultural input intensity variables related to pollution in Hubei from 1992 to 2013. A unit-root test was thus performed to test whether the time series of the variables were non-stationary. The Augmented Dickey–Fuller (ADF) Test was selected to test whether these variables contain unit roots (Dickey and Fuller, 1979), and the results are presented in Table 4 . The null hypotheses that five variables contain a unit root cannot be rejected, implying that differences must be taken for all of the variables. The results show that the null hypotheses are rejected, and all variables are in stationary time series in the first difference. Thus, the time series of the urbanization variables and the intensity variables of the pollution-related agricultural inputs can be tested for cointegration.

Table 4

ADF unit root tests.

	ADF statistic	P-value	Conclusion		ADF statistic	P-value	Conclusion
UP	−1.129	0.684	Non-stationary	ΔUP	−4.626	0.002***	Stationary
STP	−0.320	0.906	Non-stationary	ΔSTP	−3.617	0.015**	Stationary
PUI	0.387	0.977	Non-stationary	ΔPUI	−5.160	0.001***	Stationary
CUI	1.389	0.998	Non-stationary	ΔCUI	−2.740	0.085*	Stationary
AUI	−1.581	0.474	Non-stationary	ΔAUI	−6.316	0.000***	Stationary

Notes: Δ indicates the first difference operator.

Significance at 0.1.

Significance at 0.05.

Significance at 0.01.

The number of maximum number of lags has been set at 4. All unit root tests regressions include an intercept, and exclude a time trend.

Cointegration test

The cointegration test is designed to find a long-run equilibrium in variables that are returned to being stationary by differencing. Imposing inappropriate cointegration relationships can lead to uncertainty in the estimations and bias the impulse response analysis stemming from the VAR. The Johansen cointegration test proposed by Johansen and Juselius (1990) is generally applicable to test for cointegration among multiple time series because it allows for the estimation of more than one cointegrated relationship.

The lag length of the Johansen cointegration test is frequently determined by the lag length of the VAR model. The VAR models that are established for assessing the lag lengths have the same structure as the VAR models that are established for impulse response function analysis. But these VAR models have not been perfectly proved the feasibility for impulse response function analysis in the current analysis procedure. They only play a role in selecting the lag lengths. The alternative statistical criteria for the selection of the lag length of the VAR model under consideration are the sequentially modified LR test statistic (LR), the final prediction error (FPE), the Akaike information criterion (AIC), the Schwarz information criterion (SC), the Hannan–Quinn information criterion (HQ), etc. (Zhang and Cheng, 2009). The lag length selection results are presented in Table 5 . Ivanov and Kilian (2005) reported the robustness of the simulation results to the lag order, the number of variables and the number of observations. The number of observations affects the accuracy of alternative statistical criteria, but the differences in accuracy are small given the number of observations. Consequently, in order to improve the robustness, the lag length selected by the most criteria is the optimal number of lags in this study. It appears that the optimal number of lags in the VAR model for analyzing the relationships among UP, STP and PUI is 1. The optimal number for analyzing the relationships among UP, STP and CUI is 2, and the optimal number for analyzing the relationships among UP, STP and AUI is 1. The Johansen co-integration tests using the selected lag lengths were then performed to examine the long-run equilibria of the variables. Table 6 shows that all of the null hypotheses of no cointegration between the urbanization and the intensity variables of the pollution-related agricultural inputs can be rejected.

Table 5

VAR lag length selection criteria.

Variables	Lag	LR	FPE	AIC	SC	HQ
UP STP PUI	0	–	0.000	−2.037	−1.888	−2.012
	1	60.854*	1.19 × 10−6*	−5.147*	−4.550*	−5.046*
	2	8.536	0.000	−4.911	−3.867	−4.734
	3	2.793	0.000	−4.274	−2.783	−4.022
UP STP CUI	0	–	0.009	3.824	3.973	3.849
	1	82.258*	0.000	−0.713	−0.116*	−0.612
	2	13.342	9.31 × 10−5*	−0.877*	0.167	−0.701*
	3	6.010	0.000	−0.598	0.893	−0.346
UP STP AUI	0	–	0.000	−3.646	−3.496	−3.620
	1	56.025*	3.29 × 10−7*	−6.433*	−5.837*	−6.332*
	2	5.451	0.000	−5.940	−4.896	−5.763
	3	6.980	0.000	−5.768	−4.277	−5.516

Lag length selected by the criterion.

Table 6

Johansen cointegration tests.

Variables	Hypothesized no. of CE(s)	Eigenvalue	Trace statistic	0.05 critical value	P-value
UP STP PUI	None	0.542	25.088	24.276	0.040**
	At most 1	0.310	8.691	12.321	0.188
	At most 2	0.042	0.909	4.130	0.394
UP STP CUI	None	0.652	39.260	24.276	0.000**
	At most 1	0.558	18.145	12.321	0.005**
	At most 2	0.087	1.822	4.130	0.208
UP STP AUI	None	0.564	30.323	24.276	0.008**
	At most 1	0.355	12.911	12.321	0.040**
	At most 2	0.162	3.701	4.130	0.065

Rejection of the hypothesis at the significance level of 0.05.

Granger causality test

According to the results of the ADF unit root tests and the Johansen cointegration tests, the time series of the variables are suitable for Granger causality tests. The selected lags are 1 and 2 in the Granger causality tests for each pair of variables, and the results are presented in Table 7 . I detected one-way causalities running from UP to PUI and from UP to AUI and a two-way causality running from UP to CUI. Meanwhile, there is a one-way causal relationship running from STP to PUI and two-way causal relationships running from STP to CUI and from STP to AUI. The findings imply that urbanization Granger-causes movement in the pollution-related agricultural input intensity in Hubei.

Table 7

Granger causality test.

Lag	Null hypothesis	F-statistic	P-value	Decision
1	PUI UP	0.016	0.902	Fail to reject null
	UP PUI	8.434	0.010***	Reject null
2	PUI UP	1.301	0.301	Fail to reject null
	UP PUI	2.859	0.089*	Reject null
1	CUI UP	1.948	0.180	Fail to reject null
	UP CUI	1.433	0.247	Fail to reject null
2	CUI UP	4.122	0.037**	Reject null
	UP CUI	3.972	0.041**	Reject null
1	AUI UP	2.094	0.165	Fail to reject null
	UP AUI	10.006	0.005***	Reject null
2	AUI UP	1.474	0.260	Fail to reject null
	UP AUI	4.062	0.039**	Reject null
1	PUI STP	0.653	0.430	Fail to reject null
	STP PUI	5.320	0.033**	Reject null
2	PUI STP	1.531	0.248	Fail to reject null
	STP PUI	2.800	0.093*	Reject null
1	CUI STP	0.760	0.395	Fail to reject null
	STP CUI	3.907	0.064*	Reject null
2	CUI STP	3.166	0.071*	Reject null
	STP CUI	4.460	0.030**	Reject null
1	AUI STP	10.164	0.005***	Reject null
	STP AUI	4.029	0.060*	Reject null
2	AUI STP	6.131	0.011**	Reject null
	STP AUI	2.843	0.090*	Reject null

Rejection of the null hypothesis at the significance level of 0.1.

Rejection of the null hypothesis at the significance level of 0.05.

Rejection of the null hypothesis at the significance level of 0.01.

The symbol indicates “does not granger cause”.

Impulse response function analysis

Granger causality cannot comprehensively capture the relationships between the urbanization variables and the intensity variables of the pollution-related agricultural inputs. The VAR model was applied to analyze the impact of urbanization on pollution-related agricultural input intensity using variables that are connected by long-run equilibria. Generally, impulse response function analysis is used to track the dynamic effects on endogenous variables when the VAR model receives a one-time residual shock. The preliminary step in impulse response function analysis is the estimation of several three-variable VAR specifications for urbanization and pollution-related agricultural input intensity. The criteria presented in Table 5 were employed to establish the VAR model. I assume that the VAR model with three variables including UP, STP and PUI using 1 lag is VARa, the VAR model with three variables including UP, STP and CUI using 2 lags is VARb, and the VAR model with three variables including UP, STP and AUI using 1 lag is VARc. To test the fit of the developed VAR specifications, stability tests are performed to examine whether further implementation of the VAR model will yield an invalid conclusion. The stability test can be actualized by calculating the inverse roots of the AR characteristic polynomial. On condition that all the inverse roots lie inside the circle, the estimated VAR model satisfies the condition for stability. Fig. 3 shows that all of the inverse roots of the VARa, VARb and VARc models lie inside the circle. Therefore, the results of the impulse response analysis based on the VAR model established in this study are valid.

Fig. 3

Inverse roots of VARa model (a), VARb model (b) and VARc model (c).

Fig. 4 plots the response of pollution-related agricultural input intensity variables, which consist of PUI, CUI and AUI returning to a one-time positive shock from changes in urbanization. The data in Fig. 4(a)–(c) were analyzed by the VARa model, and Fig. 4(a) shows the response of pesticide use intensity to the proportion of the urban population. The positive response of pesticide use intensity continues to rise after the shock and reaches its highest level in the seventh period; it then remains relatively stable for 4 years. The increasing urban population proportion represents the shift in the population from rural to urban areas (You, 2015). The population reduction in the rural areas implies labor transfer from agriculture to industry and service employment in urban areas. Due to the decreases in agricultural labor, the peasant households have to increase pesticide use per unit of agricultural land to prevent, destroy, or repel pests and to protect plants from damage caused by weeds and plant diseases in Hubei (Ntow et al., 2006).

Fig. 4

Impulse response function analysis.

Fig. 4(b) shows the response of pesticide use intensity to the secondary and tertiary industry proportion. The shock from the secondary and tertiary industry proportion causes an initial reduction in pesticide use intensity at a rate of 1.25%, but the response trend returns back to zero after 4 years. However, the increase in pesticide use intensity is relatively weak and never exceeds 0.40%. The development of secondary and tertiary industry provides opportunities to earn wealth and comfort and raises the standard of living (Zheng et al., 2007). People, especially those in medium-high to high income groups, begin to pursue healthier lifestyles and pay more attention to the factors that may harm their health. For example, organic food that is produced with reduced pesticide use, which coincides with lower instances of pesticide residues, is becoming more and more popular. People are willing to pay more for organic than conventional food to avoid potential health risks. Peasant households catering to this current trend decrease pesticide use intensity on their agricultural lands. However, they may find that the reduction in pesticide use intensity may have serious negative impacts on agricultural production in Hubei and have no choice but to increase pesticide use to obtain higher incomes.

Fig. 4(c) shows the response of pesticide use intensity to pesticide use intensity. After an initial shock, the response peaks at 2.77% in the first period, but the positive effect of pesticide use intensity continues to decay and becomes negative after 8 years. This reveals that pesticide use intensity has the momentum to grow continually in the early period, but there are limits to its growth as pesticide use intensity is particularly high in Hubei.

The data in Fig. 4(d)–(f) were analyzed by the VARb model. Fig. 4(d) shows the response of chemical fertilizer use intensity to a one-time positive shock from the urban population proportion. The response is 1.95% in the first period, and it reaches its lowest point after 1 year. However, chemical fertilizer use intensity then swings from a reduction to an increase over time and peaks in the fifth period. The subsequent response is positive until the tenth period. Currently, machines cannot be widely used to spread chemical fertilizer in Hubei, and the farmers in many areas are primarily the elderly, children and women, who spread chemical fertilizer by hand (Liu et al., 2012). During urbanization, young labor forces leave from the rural areas into the urban cities. Therefore, the remaining old and female labor tends to apply more chemical fertilizer (Aregay and Minjuan, 2012). However, most peasant households do not have enough labor to spread chemical fertilizer by hand, so use intensity is decreased after a short time. In contrast, off-farm employment, which reduces livelihood risks in peasant households, has a positive effect on the rise in fixed agricultural investment (Takahashi and Otsuka, 2009). As a result, the peasant households in Hubei increase their chemical fertilizer use intensity for a long time into the future.

Fig. 4(e) shows the response of chemical fertilizer use intensity to a one-time positive shock from the secondary and tertiary industry proportion, which causes an initial reduction in chemical fertilizer use intensity by approximately 11.29%. The response in chemical fertilizer use intensity reaches its nadir after 1 year, and then continues to rise and peaks at 18.63% in the sixth period. It subsequently decreases rapidly in the last 4 years. The development of secondary and tertiary industry increases the amount of land required for construction (Su et al., 2014a, Su et al., 2014c), and the process of agricultural land conversion is driven by the local government to finance urban infrastructure and transportation (Tan et al., 2009, Zhou et al., 2015). The expansion of construction land increases the insecurity of the agricultural land because it is held publically under a household responsibility system. Insecure land tenure means that the peasant households have no incentive to invest in the agricultural land, such as through the use of organic manure and phosphate fertilizer (Li et al., 1998). So the growth of secondary and tertiary industry reduces the use of chemical fertilizer per unit of agricultural land. However, long-term land use will improve land tenure security. Chemical fertilizer use intensity therefore presents a rising trend in the long term as the peasant households in Hubei increase their investment.

Fig. 4(f) shows the response in chemical fertilizer use intensity to a one-time positive shock from chemical fertilizer use intensity. The shock causes an initial increase in chemical fertilizer use intensity at a rate of 24.92%, but the response declines until the fifth period before starting to tame off. Finally, chemical fertilizer use intensity increases by 15.33% after 5 years. This response demonstrates that the intensity of chemical fertilizer use has momentum to keep growing, which is accompanied by a decreasing trend in the early period when the use of chemical fertilizer begins to increase in Hubei. However, peasant households tend to control the rate of growth of chemical fertilizer use intensity due to existing high levels of fertilizer use and rising prices (Huang et al., 2013).

The data in Fig. 4(g)–(i) were analyzed by the VARc model, and Fig. 4(g) shows the response of agricultural plastic film use intensity to a one-time positive shock from the urban population proportion. The urban population proportion has a positive impact on the spread of agricultural plastic film use, and the response peaks at 0.63% in the second period; thereafter, the response weakens. The growth of the urban population proportion changes the plantation pattern in China as people consume more vegetables, melons and fruits (Hao et al., 2011). Peasant households can enhance the output of these crash crops with extensive coverage by agricultural plastic film. The application of agricultural plastic film increases as a consequence.

Fig. 4(h) shows the response of agricultural plastic film use intensity to a one-time positive shock from the secondary and tertiary industry proportion. The secondary and tertiary industry proportion has a negative impact on the spread of agricultural plastic film use. There is an initial reduction of approximately 0.21% in chemical fertilizer use intensity, but it rises to approximately zero in subsequent periods. The main reason for this is that the development of secondary and tertiary industry in China raises the prices of agricultural chemical inputs, such as agricultural plastic film (Huang et al., 2013). The rising production costs impose an additional strain on peasant households, so they choose to reduce their use of agricultural plastic film.

Fig. 4(i) shows the response of agricultural plastic film use intensity to a one-time positive shock from agricultural plastic film use intensity. The shock has a positive impact on agricultural plastic film use intensity, which peaks at 1.69% in the first period. However, it decreases rapidly after 1 year to become stable at a rate of approximately 0.12% thereafter. This result reveals that the rapid growth in agricultural plastic film use does not persist. While large quantities of agricultural plastic film have been used on agricultural land to increase crop yields and improve water retention efficiency in Hubei, large amounts of agricultural plastic film residue have been generated and have not degraded. Plastic film residues seriously reduce crop yields because they damage the physical structure of the soil and lead to poor water flow (Wang et al., 2015). Accordingly, the use of agricultural plastic film cannot continue to increase in Hubei.

Variance decomposition

Variance decomposition is applied to account for how much of the variance in the forecast error in one variable can be explained by innovations stemming from the other variables in the VAR model (Pesaran and Shin, 1998, Su et al., 2014a). It can be used to measure the proportions of the variations in one intensity variable of the pollution-related agricultural inputs with respect to the whole that is caused by its own shock and the shocks from the urbanization variables. Fig. 5 illustrates the variance decompositions of PUI, CUI and AUI due to UP, STP and the shocks from themselves. The data in Fig. 5(a)–(c) were analyzed by the VARa model, and they show the contributions of the different shocks from the urban population proportion, the secondary and tertiary industry proportion and pesticide use intensity to the variations in pesticide use intensity over the 10-year period. The results reveal that 82.93% of the forecast error variance in pesticide use intensity is explained by its own shock while the shocks from the urban population proportion and the secondary and tertiary industry proportion contribute to 0.12% and 16.95% of the variations in pesticide use intensity after 1 year, respectively. However, the contribution of its own shock continues to decline, and it reaches 35.72% in the tenth period. Meanwhile, the contributions of the urban population proportion and the secondary and tertiary industry proportion continue to rise. The urban population proportion whose contribution reaches 56.88% in the tenth period plays a critical role in the variations in pesticide use intensity.

Fig. 5

Variance decomposition.

The data in Fig. 5(d)–(f) were analyzed by the VARb model, and these figures show the contributions of different shocks in the urban population proportion, the secondary and tertiary industry proportion and chemical fertilizer use intensity to the variations in chemical fertilizer use intensity over the 10-year period. The shocks from the urban population proportion and the secondary and tertiary industry proportion account for 0.51% and 16.95% of the variations in chemical fertilizer use intensity, respectively. It suggests that the impacts of these two factors on chemical fertilizer use intensity are less important compared with pesticide use intensity. Although the variations in the forecast error variance due to the urban population proportion jumps from approximately 0.51% to 26.09%, the largest proportion of the variations in chemical fertilizer use intensity is explained by its own shock in the tenth year as well as before.

The data in Fig. 5(g)–(i) were analyzed by the VARc model, and these figures show the contribution of different shocks from the urban population proportion, the secondary and tertiary industry proportion and the intensity of agricultural plastic film use in the variation in agricultural plastic film use intensity over the 10-year period. The initial contributions of the urban population proportion and the secondary and tertiary industry proportion are negligible; its own shock explains 98.39% of the variations in agricultural plastic film use intensity in the first period. The contribution of the urban population proportion consistently gains strength so that the urban population proportion accounts for 33.74% of the variations in the tenth period, and the contribution of the intensity of agricultural plastic film use accounts for more than 60.00% of the variations in the tenth period. However, the shocks from the secondary and tertiary industry proportion do not explain more than 4% of the variations in the intensity of agricultural plastic film use over the 10-year period.

Conclusions

This study applied the VAR model to analyze the impact of urbanization on pollution-related agricultural input intensity in Hubei, China. Specifically, the methodology included the unit-root test, the cointegration test, the Granger causality test, the impulse response function analysis and variance decomposition. The urban population proportion and the secondary and tertiary industry proportion were used as the urbanization variables, and pesticide use intensity, chemical fertilizer use intensity and agricultural plastic film use intensity were the variables indicating pollution-related agricultural input intensity.

Results of the impulse response function analysis using the VAR model indicate that the response of pesticide use intensity continues to rise after shocks from the urban population proportion and the secondary and tertiary industry proportion, although the initial responses are positive and negative, respectively. The initial response of chemical fertilizer use intensity following shocks from the urban population proportion and the secondary and tertiary industry proportion are positive and negative, respectively; these responses first decrease and then increase. The response of agricultural plastic film use intensity first increases and then decreases after shocks from the urban population proportion, but the opposite response to the shock from the secondary and tertiary industry proportion is observed. In addition, the responses of pesticide use intensity, chemical fertilizer use intensity and agricultural plastic film use intensity show a downward trend following their own shocks after initial positive responses. Based on variance decomposition using the VAR model, most of the forecast error variance in pesticide use intensity is explained by its own shock during the initial period, but an increase in the proportion of the urban population plays an important role in explaining the variation in pesticide use intensity during the last period. The shock from chemical fertilizer use intensity explained the largest proportion of its variations over the 10-year period, and the shock from agricultural plastic film use intensity also explains 98.39% of its variations in the first period. Afterward, the contribution of the urban population proportion gains strength, but the contribution of the secondary and tertiary industry proportion remains negligible.

Results of this study reveal that specific policies should be proposed to control the rapid growth of pollution-related agricultural input intensity as the Chinese government accelerates urbanization. The policies should be designed to address the characteristics of the impacts on pollution-related agricultural input intensity, which are caused by increasing urbanization. The hidden reasons behind the phenomena analyzed in this study should be further addressed to better coordinate the relationships between urbanization and increasing pollution from agricultural production. Moreover, increasing agricultural outputs should be realized with little or even no growth in the pollution-related agricultural input intensity.