Comparison of agrochemicals allocation efficiency between greenhouse and open-field vegetables in China

The overuse of agrochemicals in greenhouse production system has aroused high concerns in China. Existing studies have investigated the status and determinants of agrochemicals overuse for greenhouse vegetables whereas little is known about whether the agrochemicals are allocated efficiently from economic perspective. We use a translog production function and the inputs and outputs data of vegetable production in 34 Chinese cities during 2004–2017 to calculate agrochemicals allocation efficiency for both open-field and greenhouse vegetables. We find that the agrochemicals are allocated inefficiently due to overuse for both open-field and greenhouse vegetables, whereas the overuse degree of chemical fertilizer used in greenhouse vegetables is lower than that of open-field vegetables during a growing duration. In addition, we also find that the application levels of agrochemicals for greenhouse vegetables per mu (15 mu = 1 hectare) are higher than those of open-field vegetables, but the application levels of agrochemicals for per kilogram greenhouse vegetables are significantly lower. We conclude that the overuse of agrochemicals in greenhouse production system may attribute to the year-round production of greenhouse vegetables induced by economic incentives. Therefore, reducing the number of production rotations may be an effective method to alleviate the overuse of agrochemicals in greenhouse vegetables.

www.nature.com/scientificreports/ addition, greenhouse production system also increases the potential health risks to both farmers and consumers due to exposing to the pollutants and pollutants residual in vegetables 8,18,19 . A large number of studies have investigated the status and determinants of agrochemicals overuse of greenhouse vegetables in China 11,[20][21][22][23][24] . More specifically, farmers often used up to about 5 times the greenhouse vegetables requirement in the case of N fertilizer in Beijing 11 . Hanafi 25 documented that lots of pesticides are used in greenhouse because of the diseases and insect pests infection caused by high temperature and humidity. The study by Yu et al. 12 found that the special microclimate environment in greenhouse, such as the strong evaporation of soil water and a lack of rainfall leaching, together with the inappropriate farming methods, including flood irrigation and excessive application of agrochemicals, have resulted in low fertilizer use efficiency. However, greenhouse could produce vegetables with less pesticide and fertilizer by adopting modern technology and management practices such as biological control, closed watering system, and carefully directed fertilization [26][27][28] .
Exploring how farmers allocate the agrochemicals plays core role in deeply understanding the overuse of agrochemicals and the pollution caused by agrochemicals. The quantities of agrochemicals application are determined by vegetable farmers whose targets are to maximize their profit given the technology and input prices. Therefore, we cannot make comparison of agrochemical use efficiency directly between greenhouse and open-field vegetables because greenhouse vegetables usually have higher sale price and yield than open-field vegetables 3,4 .
Following Hopper's 29 method, this paper aims at evaluating the allocation efficiency of agrochemicals used in greenhouse and open-field vegetables by calculating the marginal product values (MPVs) of agrochemicals and making comparison with the agrochemicals' prices. Farmers would not think they overuse the agrochemicals in production process if the value produced by additional unit agrochemical is larger than its corresponding cost. Therefore, compared to open-field vegetables, farmers could apply more agrochemicals to greenhouse vegetables because the decreased marginal value of agrochemicals could be offset by the higher product price. Based on the results of this paper, we would answer whether agrochemicals used in greenhouse vegetables are more but allocative efficient, just like the hypothesis of "poor but efficient" proven by Hopper 29 . The hypothesis of "poor but efficient"proven by Hopper means that although the farmers in India is very poor, but they allocate their input efficiently because the value of an input's marginal product equals to its price. Therefore, although Chinese greenhouse vegetables used more agrochemicals than open-field vegetables, the allocation efficiency of agrochemicals may be higher than that of open-field vegetables because of higher yield and sale prices of greenhouse vegetables. Given the high concerns of agrochemicals overused by greenhouse vegetables, uncovering the allocative efficiency of agrochemicals would provide significant implications for agricultural policy.
There are at least three methods to measure the use efficiency of fertilizers and pesticides. First, the ratio of the minimum possible input in theory to the observed input is defined as use efficiencies [30][31][32][33] , which can be used to evaluate the overuse status of fertilizers or pesticides. Second, another indicator for use efficiency widely accepted by many countries is the fraction of input harvested as product 34 . Zhang et al. 35 converted the quantities of multiple pesticides with the same functions into an index and compared the index with the recommended range to identify the use efficiency. According to the aforementioned definitions, only the quantities of inputs and outputs are used in these two measurements which do not take the differentiated prices between greenhouse vegetables and open-field vegetables into consideration. The goal of a farmer is to maximize his profits based on market price of vegetables when he makes decisions on fertilizers or pesticides application. Therefore, this paper will calculate the marginal product values and makes the comparisons of the MPVs of chemical fertilizer, organic fertilizer and pesticide with their corresponding prices to identify the allocative efficiency of agrochemicals. The equality of one input's MPV and its price indicates that farmers allocate this input efficiently, and the sign and magnitude of an inequality mean the direction and severity of the overuse. This method is widely adopted by agricultural economists to study the input's allocative efficiency 29,[36][37][38][39] .
To calculate the MPVs of agrochemicals, we have to estimate the production function. To the extent that fertilizer and pesticide have different influential mechanism on vegetable production, thus treating the inputs symmetrically would result in biased estimation of production function [40][41][42] and lead to erroneous MPVs. Lichtenberg and Zilberman 40 , Zhang et al. 42 , Paris 43 , Chambers and Lichtenberg 44 , and Sun et al. 45 have revealed that pesticide plays a damage-abating role in production process whereas other inputs act as the productivity-increasing roles. Therefore, the specification of production function is revised in their studies based on the damage-abating role of pesticides. Guan et al. 41 furtherly pointed out that different categories of inputs contribute to crop yields through different process. Guan et al. 41 proposed a more general framework by specifying an asymmetric production function. They categorize all of the inputs in crop production into growth inputs and facilitating inputs by integrating both agronomic principles and agricultural economics concepts. Growth inputs are defined as those directly involved in the biological process of crop growth including land, seed, chemical fertilizers and organic fertilizers. Facilitating inputs are those to improve growth conditions including pesticides, capital and labors.
There are two central contributions of this paper. First, to best of our knowledge, this paper firstly evaluates the allocative efficiency of fertilizers and pesticides for both greenhouse vegetables and open-field vegetables by making comparisons of the input's MPV with its price in China. MPV accounts not only for the impacts of input on output, but also for the effects of differentiated prices between greenhouse and open-field vegetables on input application level. To the extent that evaluating the allocative efficiency of agrochemicals in both greenhouse and open-field vegetables could provide a new insight to understand the overuse of agrochemicals.
Second, based on the authoritative input-output data, this paper evaluates the allocative efficiency of agrochemicals used in both greenhouse and open-field vegetables during a growth duration to rule out the impacts of the continuous cropping model. Greenhouse production system is initially developed to provide regulated and controlled growing condition to produce crops in areas and seasons usually unsuitable for agricultural production. However, farmers may grow vegetables several times in greenhouse per year to harvest more outputs. The overuse of agrochemicals could be significantly reduced if farmers just take advantage of the function to provide controlled growing conditions without year-round production.

Method
This paper adopts the method developed by Guan et al. 41 to calculate MPVs of chemical fertilizer and pesticides.
For this method, we should firstly estimate the production function, and then use the estimated parametric of chemical fertilizer and pesticide to calculate MPVs. Both Cobb-Douglas form and Translog form are widely used as the specific form of the production function in empirical studies 32,37 . This paper uses the translog functional form because this form is very flexible and it's a second-order approximation of any production technology 10,32 . The translog functional form includes the linear terms, the quadratic terms and the interaction terms of all inputs. It is worth noting that the translog functional form treats all of the inputs symmetrically. To get more precise MPVs of fertilizers and pesticides, this paper adopts Guan et al. 's 41 method which integrates both agronomic principles and agricultural economics concepts by distinguishing between growth inputs and facilitating inputs in production function. The specification of translog production function used in this paper is as follows: where Y it refers to the vegetable output value per mu of the ith city in year t. L it , S it , F it and O it are growth inputs. They are orderly farmland cost per mu, seed cost per mu, chemical fertilizer quantity per mu and organic fertilizer cost per mu of the ith city in year t, respectively. The quadratic terms and the interaction terms of four growth inputs are followed. Output value and all of growth inputs are in logarithms.
P it , C it and Lb it are the facilitating inputs. They are orderly pesticides, capital and labor of the ith city in year t, respectively. In Eq. (1), we use different functional forms to reflect the different ways that growth inputs and facilitating inputs contribute to vegetable output. It is worth noting that the magnitude of outputs in Eq. (1) is mainly determined by growth inputs. The effects of facilitating inputs to vegetable output can be written as . This term can measure the impacts of growth conditions on vegetable production such as pests and disasters, and climate changes. Holding the quantity of growth inputs constant, the vegetable output achieves its maximum value if the growth conditions are optimal. Under non-optimal conditions, facilitating inputs are needed and the outputs will be downscaled.  46 , and Jin et al. 47 to study both Chinese livestock sector and crop sector. A three-stage stratified random sampling procedure is used to get the representative sample counties, villages and finally individual farms. NDRC collected the detailed inputs and output information of vegetable production on farm level, including vegetable output, seed cost, pesticide cost, the quantity and cost of chemical fertilizer (quantity of chemical fertilizer used here is the net quantity calculated after purification) and labor, organic fertilizer cost, land cost and other intermediate input costs per unit mu. To improve the accuracy of the data, bookkeeping method was used and farmers were required to record the inputs which were used several times in each growing duration every time.
However, NDRC does not release the farm level data. The data used in this paper are aggregated on city level. We use producer price index of agricultural products in corresponding province to eliminate inflation. There are 34 cities in our sample, divided into 6 regions, including Beijing, Tianjin, Shijiazhuang, Taiyuan and Hohhot in North China, Shenyang, Dalian, Changchun and Harbin in Northeast China, Shanghai, Nanjing, Hangzhou, Ningbo, Hefei, Fuzhou, Xiamen, Nanchang, Qingdao and Jinan in East China, Zhengzhou, Wuhan, Changsha,  48 . We show the cities and the regions in a geographical map in Fig. 1.

Results and discussion
Using large amounts of fertilizers and pesticides on per unit area is one of the dominated factors causing environment pollution in greenhouse production system. However, greenhouse vegetables have higher yield and higher sale price than their counterparts due to the off-season sales. Whether or not greenhouse vegetables have higher overuse degree of agrochemicals compared to open-field vegetables during a growing duration with consideration of both the vegetable's sale price and yield?  www.nature.com/scientificreports/ The status of fertilizer and pesticide application levels. Table 1 Table 1, indicating that greenhouse production system has higher input to output ratio of chemical fertilizer compared to that of open-field production system. However, different from chemical fertilizer, the costs of organic fertilizer per mu and per kg yield in greenhouse are both significantly higher than that in open-field (Figs. 4 and 5) during 2004-2017 with a slightly increase trend.
A large number of studies have proved that about 40-60% of crop yield is attributable to chemical fertilizer 49,50 , and organic fertilizer is mainly used to maintain soil fertility and improve soil structure 51 . Yan and Gong 52 documented that the nutrient use efficiency of chemical fertilizer was higher than that of organic fertilizer which cannot improve crop yield in the current season but would benefit crop yield in the long term. Our study finds that both chemical and organic fertilizers applied in greenhouse vegetable system are higher than in open-field system, which may contribute to the higher yield in greenhouse. Although Ti et al. 53 had stated that fertilizer (nitrogen) use efficiency in greenhouse system is lower than in open-field system, this study proves that greenhouse vegetable system could consume less chemical fertilizer than open-field system to harvest the same yield in reality. One possible explanation is that greenhouse system could provide regulated and controlled growing conditions for vegetables. Therefore, compared to open-field vegetables, greenhouse vegetables could have longer harvest duration, produce more outputs and higher input to output ratio. Figures 2, 3, 4 and 5 also shows the chemical fertilizer and organic fertilizer consumption across six regions in China. This study finds that higher amounts of chemical fertilizer are applied per mu in Central China for both greenhouse and open-field vegetable systems, while Northeast China use the lowest amounts of chemical Table 1. Application levels of fertilizer and pesticide. *,**,*** indicate statistically significance at the 10%, 5% and 1%, respectively. 1 hectare = 15 mu. Quantity of chemical fertilizer used here is the net quantity calculated after purification. a We keep 5 decimal places due to the small value.

Vegetable
Open-field Greenhouse Differences ). Previous studies analyzed the regional disparity in fertilizer input of main crops in China and draw similar conclusion: the input intensity showed a decreasing trend from east to west 54,55 .
In addition, Figs. 6 and 7 depict pesticide costs for tomato and cucumber in different regions and over years. Greenhouse production system is prone to invasions of pests and pathogens due to the characteristics of high humidity, high temperature and continuous cropping in greenhouse environment 25 Table 1. Significant differences also exist in pesticide costs across regions. More specifically, Central China consumes relative high levels of pesticide for both greenhouse and open-field vegetable systems.
To summary, farmers use more chemical fertilizers and pesticides on per mu whereas use less on per kg greenhouse vegetables compared to open-field vegetables. We may conclude that the input to output ratio of agrochemicals for greenhouse vegetables is not lower than that of open-field vegetables during one growth duration with yield into consideration. However, why do the pollution caused by agrochemicals in greenhouse system always arise people's concerns? There are two possible explanations: First, greenhouse system could be used to produce vegetables all year round without rest for soil due to the regulated and controlled conditions. Second, although the input to output ratio of agrochemicals is high for greenhouse system, the pollution caused by agrochemicals may be more serious due to the special microclimate environment of high temperature and humidity. In addition, the sale price of greenhouse vegetables during off season is significantly higher. Therefore, farmers have economics incentives to produce more vegetables by using more agrochemicals, which may result in pollution.  Table 2 reports the estimators of the asymmetric translog production function. We use GMM technique and take the inputs of last year as instrumental variable to avoid endogenous problems when we estimate the production function. Among the growth inputs, only the quadratic term of seed contributes to the growth of cucumber outputs. This is because selecting the input variables is a trade-off between applying more inputs to get more useful information and increasing the risk of multicollinearity 30,32 . However, the linear terms of land, seed, organic fertilizer, and some of the interaction terms have statistically significant impacts on tomato outputs. Among the facilitating inputs, both pesticide and labor have significant impacts on cucumber outputs, and both capital and labor have impacts on tomato outputs. The positive coefficients for facilitating inputs indicate that farmers are likely to harvest less output if they apply more facilitating inputs in production process under non-optimal conditions. Hopper 29 firstly compared the MPVs of agricultural inputs with their prices to support the hypothesis that Indian farmers are "poor but efficient": Indian farmers allocate their input efficiently because the additional output value produced by using one more unit of input equals to the input's price. Based on the estimated parametric in Table 2, MPVs of fertilizers and pesticides are calculated by using Eqs. (2) and (3), and the results are reported in Table 3. The null hypothesis that MPVs of chemical fertilizer equal to zero is not rejected at 10% significance level indicating overuse of chemical fertilizer for both greenhouse and open-field tomatoes. However, we find that one kg chemical fertilizer applied on greenhouse cucumber could produce output value of 4.052 RMB which is statistically significant different from zero and insignificant from its price. This implies that farmers allocate chemical fertilizer efficiently on greenhouse cucumber. However, farmers overuse chemical fertilizer on open-field cucumber. www.nature.com/scientificreports/ Based on the results reported in Table 2, pesticide has no significant effects on tomato output, thus, the MPV of pesticide used in tomato production regardless of greenhouse or open-field is zero. This implies that pesticide is overused for both greenhouse and open-field tomatoes. Similarly, one RMB increase in pesticide costs will negatively affect the output values implying overuse for both open-field and greenhouse cucumbers. It is consistent with the damage-abating role of pesticide. In other words, the more pesticides used the lower the yield due to the severe pests and diseases. The difference in pesticide costs between tomato and cucumber is expected. The special smell of tomato plants makes it more resistant to diseases and insect pests than cucumber plants.
The analysis of organic fertilizer shows that one RMB increase in organic fertilizer will produce output value of more than 10 RMB for both greenhouse and open-field tomato or cucumber, indicating underuse of organic fertilizers. Therefore, increasing the application of organic fertilizer will contribute greatly to the vegetable output. However, it is worth emphasizing that fertilizer with manure and especially animal feces might result in high concentrations of heavy metals and antibiotic in soils 8,18 . Higher inputs of organic fertilizer in greenhouse lead to greater potential threat to environment and human health.
To summary, MPVs of chemical fertilizer and pesticide are significantly lower than their prices for both greenhouse vegetables and open-field vegetables excluding those of chemical fertilizer applied on greenhouse cucumber. These findings reflect that chemical fertilizer and pesticide are generally overused in vegetable sector with sale price of vegetables into consideration. However, the higher MPVs of chemical fertilizer for greenhouse www.nature.com/scientificreports/ vegetables imply that greenhouse vegetables do not have higher overuse degrees of chemical fertilizer than open-field vegetables. The results in this part furtherly document that greenhouse vegetables do not have lower allocation efficiency for chemical fertilizers than open-field vegetables with vegetables' sale price and yield into consideration. Therefore, the serious pollution caused by agrochemicals in greenhouse vegetables may attribute to the year-round production model because farmers usually take greenhouse production system as an important method to increase outputs by growing vegetables several times in greenhouse per year. The overuse degree and environment pollutions could be significantly alleviated in greenhouse production system if farmers just take advantage of the function to provide controlled growing conditions during off-season rather than year-round production.
One possible policy implication is that reducing the number of production rotations may be an effective method to alleviate the overuse of agrochemicals used in greenhouse vegetables. To produce more and more vegetables, farmers not only use the greenhouse to provide the regulated and controlled growing conditions in off-season but also produce vegetables in the other seasons. We found that more and more farmers use greenhouse to grow only one season of tomato per year in off-season to protect the soil quality in the south of Jiangsu province. In addition, to balance environment protection and vegetable production, modern farming www.nature.com/scientificreports/ www.nature.com/scientificreports/ and management practices should be introduced to vegetable industry, such as biological control, rational and precise fertilization, which are much easier to implement in greenhouse. We also found that farmers replaced chemical fertilizer and pesticide with organic fertilizer and biological control methods to protect the soil and improve the quality of products in the south of Jiangsu province. Table 2. Estimation results of the asymmetric translog production function. Growth inputs include land, seed, chemical fertilizer and organic fertilizer. Facilitating inputs include pesticide, capital and labor. Standard errors are shown in parentheses. *,**,*** indicate statistical significance at the 10%, 5% and 1%, respectively.  Table 3. Marginal product values and the prices of fertilizer and pesticide. One more unit of agrochemical does not produce any vegetables. NA refers to no value. a The input does not significantly affects the output, therefore, the MPV is zero.