ICT-based agricultural advisory services and nitrogen management practices:A case study of wheat production in China

DING Ji-ping ,LI Jing-han ,LIU Jia-huan ,ZHANG Wei-feng ,JIA Xiang-ping

1 College of Economics &Management,Northwest A&F University,Yangling 712100,P.R.China

2 Knowledge Technology and Innovation Group,Wageningen University,Wageningen 6706KN,The Netherlands

3 College of Resources and Environmental Sciences,China Agricultural University,Beijing 100094,P.R.China

4 Agricultural Information Institute,Chinese Academy of Agricultural Sciences,Beijing 100081,P.R.China

Abstract Excessive use of nitrogen fertilizer in China and its adverse effects on agricultural production have been a national and global concern. In addition to massive public initiatives to promote sustainable farm practices,grass-rooted innovations are emerging in the niche,many of which take the forms of information and communication technologies (ICT) and digital services. This study examines the effects of ICT-based extension services provided by an entrepreneurial startup on adopting sustainable farming practices. We found no significant reduction in N-fertilizer use for wheat production. But the ICT-based services promoted farmers to adapt N-fertilizer use towards site-specific management. The business model of the entrepreneurial venture faces great challenges in becoming participatory and financially sustainable.

Keywords:venture capital,innovations,precision farming,private-public partnerships,nitrogen management

1.Introduction

Although the use of fertilizers has played an important role in achieving China’s food security,fertilizers have been applied at excessive rates and have led to serious environmental problems. Between 1980 and 2010,fertilizer use in China’s grain production increased by 438% while yields only increased by 170%,indicating the declining efficiency of fertilizer use (Cuiet al.2008;Powlsonet al.2014;Huanget al.2015). Numerous studies revealed unprecedented environmental problems and significant greenhouse gas (GHG) emissions associated with excessive use of nitrogen fertilizer(N-fertilizer) (Kahrlet al.2010;Zhanget al.2013;Wuet al.2021). For example,soil acidification has become a serious problem in recent decades,to which the acidification stemming from the high inputs of ammoniumbased N fertilizer has contributed about 60% (Guoet al.2010). The incidence of algal blooms in surface water has increased several-fold since the 1990s,and N-fertilizer use is responsible for 25-80% NH3volatilization caused by agriculture in China (Powlsonet al.2014). In 2010,total GHG emissions associated with N-fertilizer (both manufacture and use) represented 7% of total emissions from China (Zhanget al.2013).

Major efforts have been made by scientists to optimize nitrogen management through information and knowledge interventions. Farmers in developing countries rely mostly on their own experience,which is passed on by word-ofmouthvialocal social networks (Duveskoget al.2003).For example,a substantial percentage of Chinese farmers wrongly believe that using more fertilizer unconditionally increases yield (Jiaet al.2015). A study by Cuiet al.(2005) found that scientific nitrogen management based on soil analyses and improved synchronization between crop nitrogen demand and nitrogen supply can reduce 196 kg N ha-1without reducing wheat yields,saving 60%of N-fertilizer in China. Jiaet al.(2013) examined the effects of knowledge training on low-carbon farm practices and found that the in-season N-fertilizer use could be reduced by 20% without affecting maize yields in China.Huanget al.(2015) designed an experimental framework and compared the long-term effects of site-specific farm advisory services and traditional training approaches in China. Some studies discovered that,in addition to information and knowledge,demographic and contextual factors (e.g.,education,risk aversion,and access to credit) also determined farmers’ behaviors and decisionmaking with respect to sustainable farm practices (Sunding and Zilberman 2001;Silviaet al.2021). Their studies revealed major limitations of traditional knowledge training without farmers’ participation. By engaging farmers in field experiments and crop diagnoses,excessive N-fertilizer use can be reduced without adversely affecting crop yield,and the effects of learning continue in the long term.

The societal need to transform agriculture towards sustainability has engaged entrepreneurial innovations and digital technologies in China. The development of digitalization in rural China offers entrepreneurial finance and startups a broad opportunity to provide agro-technical services with the aid of information and communication technologies (ICTs). The Chinese government undertook promising actions to expedite informatization in the mid-1970s,but most public investments were spent on developing capacities in urban areas. Since the 1990s,vast investments have been made by the public and private sectors to increase the coverage of informatization infrastructure in rural China,including mobile phones,fixed line telephones,radio,television,and internet(Qianget al.2009). According to NBSC (2016) and CNNIC (2016),60% and more than 31% of the rural population owns a mobile phone and has access to the Internet,respectively. Between 2010 and 2016,a total of 262 startups obtained funding thanks to entrepreneurial finance,such as venture capital (VC) and private equity.Among them,38 focused on agro-technical services,and half were established in 2016,accounting for 27% of VC cases linked with food and agriculture in that year (Jia 2017).

While entrepreneurial innovations in digitalization have been a major source of transforming businesses (Van Campenhoutet al.2021),their presence and effects on sustainable farm practices are far from clear. Early research studied ICTs as a medium of dissemination of information and knowledge in agricultural development(Rao 2008;Aker 2011). Some anecdote studies concluded that the role of ICTs in different forms (radio,short message service (SMS),mobile phones,and the Internet) in agriculture is variable and may even be overstated (Jensen 2007). Most of these studies were conducted in development or social-purpose programs. A rigorous examination of emerging entrepreneurial startups and their impacts on sustainable farm practices is rare,especially in the context of the venture capital industry’s emerging interest in China’s agricultural transformation.

The overall goal of this study is to explore the effects of ICT-based advisory services on nitrogen management in wheat production in the North China Plain. In addition,the study explores the new forms of publicprivate partnerships and investigates the organizational characteristics of delivering sustainable farm services by entrepreneurial startups. The following section presents the research design,data collection method,and empirical models. Section 3 describes nitrogen-optimization technologies in the study,the ICT protocol used by the startup,and the estimation results. Sections 4 and 5 present the discussion and conclusions,respectively.

The study has particular value for the global debates about digital transformation and innovations in sustainable agriculture in developing countries. As an emerging concept and approach aimed at shifting farming systems towards resilience and adaptability,digital transformation seeks to promote synergies by “taking these objectives into consideration to inform decisions from the local to the global scales and over short and long timer horizon,to derive locally-acceptable solutions” (FAO 2021).Given the limitations of the public extension system and the top-down approaches,entrepreneurial innovations could be an alternative to public institutions. If proven successful,the new private model could shift both farming and extension systems toward more flexible and adaptable management. Experimental studies have found that ICT-based agricultural advisory services significantly affected farmers’ fertilizer-related decisions (Kaila and Tarp 2019;Cole and Fernando 2021). If sustainable approaches to agroecology do succeed in reducing nitrogen fertilizer use,the site-specific ICTs will become a technological solution that helps promote China’s food security sustainably.

2.Materials and methods

2.1.Context and research design

The research was conducted as part of a large-scale national research program into excessive fertilizer use in China. In 2015,the Ministry of Agriculture and Rural Affairs (formerly known as the Ministry of Agriculture)announced the “Zero-Growth Action Plan for Chemical Fertilizers and Pesticides” (hereafter referred to as Action Plan). The Action Plan was a national call to restrict the annual increase in the use of chemical fertilizers and pesticides to less than 1% between 2015 and 2019 and to achieve zero increase by 2020 for major agricultural crops,compared to the annual 3.9%increase in nitrogen use between 2000 and 2013 through China. In 2016,to ensure that scientific researches match national requirements,the Ministry of Science and Technology started a large-scale research program called“Researching and Developing Technologies for Reducing Chemical Fertilizer and Pesticide and Strengthening Efficiency” (hereafter,the National Research Program).Upon the proposal,the Ministry of Finance approved funding totaling US$ 363.6 million for the National Research Program for the period 2016-2018. The National Research Program comprises 49 projects from multiple perspectives,one of which focuses on the theme of “Informatization and Sustainable Management on Fertilizer” (hereafter,the Project).

The National Research Program highlighted innovations by engaging multiple stakeholders which involve scientists,public extension officers,and private sectors. The interdisciplinary group of scientists integrated expertise in agronomy,soil and environmental sciences from different institutions. To explore a new form of public-private partnership,the Project identified one entrepreneurial startup for partnering with private sectors. Receiving research funding,the scientists provided academic support to the entrepreneurial firm which developed information and communication facilities and delivered agroecological advisory services to farmers.Rather than charging farmers service fees,the startup’s revenues came mainly from selling agrochemicals to farmers. The startup hired local extension officers (as part-time consultants) in county agricultural bureaus for rapid outreach and local knowledge. Fig.1 illustrates the schematic flow chart of the organization.

Fig.1 A schematic flow chart explaining the business model of the startup. Dashed lines indicate activities with farmers’ reaction;solid lines indicate activities without farmers’ reaction.

The present study was conducted as an independent evaluation of the effectiveness and organizational capacity of ICT-based farmer services provided by the entrepreneurial venture. After several months of communication among the multiple stakeholders of the Project,the scientists and the startup agreed to share information and data to make the research a pilot and case study. A light-touch approach was used as a precondition for implementing the evaluation throughout the lifetime of the project;researchers undertook independent observations and limited their interventions in company’s business to the minimum,which ensured a distance between the evaluator and the Project and avoided incurring additional costs and resources to the Project.

The case study was conducted in Shandong Province,where the entrepreneurial startup provided agroecological services. The startup obtained seed capital from a venture cash capital investor in 2015. The founding partners of the startup had mixed career backgrounds in fertilizer manufacture and information services such as social media and informatics. The partnerships between the Project and the startup combined the complementary capacities and resources of multiple stakeholders.Farmer services were provided in Binzhou,a northern prefecture in Shandong Provice,which has an annual rainfall of 567.7 mm. The local annual mean temperature and sunshine hours reach 12.3°C and 2 704.2 hours,respectively. Meanwhile,winter wheat and summer maize are planted in rotation in the case study area. Winter wheat is routinely planted at the end of September (after the maize harvest) and harvested in the following June.

Previous research found excessive nitrogen fertilizer use and lack of appropriate knowledge about sustainable fertilizer use in the region. For example,based on field experiments lasting several years,soil scientists argued that 128 kg N ha-1could be saved in the winter wheat cropping season (Norseet al.2012). Similar evidence has also been provided by other researchers,supporting a significant reduction in fertilizer use,or alternatively,soil acidification and surface water algal blooms (Vitouseket al.2009). Despite robust scientific evidence and active policy advocacy,inappropriate nitrogen management,which suggests that more fertilizer leads to more yield,persisted and was deeply rooted in farmers’ perceptions;the public extension system was dysfunctional due to low accountability (Jiaet al.2015). For example,only 28% of farmers were able to read the nutrient labeling on fertilizer packaging,and 44% of farmers wrongly believed “the more fertilizer,the better” in Shandong Province (Jiaet al.2015).

In this study,farmers were advised to reduce overall N-fertilizer use and to optimize nitrogen use at variable rates according to the crop growth stage and site-specific variability. A limit of 180-220 kg ha-1was advised for total N-fertilizer use in the region. The rate 180-220 kg N ha-1application was established based on the wheat requirements for achieving maximum economic yield and the soil nutrient uptake. According to the handbook of fertilization of major cash crops in China,only 2.3 kg N was needed to produce 100 kg wheat grain yield (Zhanget al.2009). In the North China Plain,yield of wheat in the wheat-maize doubled crop rotation were 7-10 t ha-1,with an annual optimum N application of 180-220 kg N ha-1.This variable rate of N-fertilizer use and precision farming should avoid any surplus N reaching the environment and improve both yield and nitrogen use efficiency (Matsonet al.1998;Tilmanet al.2002).

In addition,the startup advised farmers to adjust N-fertilizer application at different stages for a better match between crop nitrogen demand and the nitrogen supplied by fertilizer and by the soil nutrient. In principle,the farmers were advised to apply N-fertilizer at a ratio of 1:1,namely 100 kg N ha-1at the basal stage of the growing season,and another 100 kg N ha-1as a top dressing. The N management strategy at the basal stage was based on soil test. Prior to the 2016-2017 cropping season (i.e.,shortly after the maize harvest in September),86 sample soil plots at 0-20 cm soil layer were randomly taken from 45 farmers before planting and after the harvest. The soil samples were used to measure soil nutrients such as total N content,organic matter content,Olsen-P,NH4OAc-K,and pH level. Excessive use of N-fertilizer in the early stage is neither economically justified nor productive because young crops have limited capacity to uptake N,which is attributed to the lack of an extensive root system in the early growth stage (Sinclair and Rufty 2012). The latter 100 N kg ha-1strategy applies N fertilizer once or twice and the optimum N rates at the top dressing stage is based on the actual tiller number and crop population,which was affected by the growing period climate and soil nutrient. Specifically,three randomly selected 0.5 m wheat of single row at each sample plot were counted by scientists to determine the total tiller number. The plot with the greatest tiller number (900 m-2) requires N fertilizer to be applied twice:first during the reviving stage (52.5 kg N ha-1) and again during the jointing stage (60 kg N ha-1). An average of 112.5 kg N ha-1was used at the erecting stage in total tiller number with a range from 900 to 1 200 m-2. Across the plots with the total tiller number ranging from 1 200 to 1 500 m-2,an average of 112.5 kg N ha-1was applied at the jointing stage (Chenet al.2016).

The SMS-based messaging tools were used for communication. Scientists analyzed all the data and rapidly transferred the results to the startup whose staff further calibrated the parameters according to the feedback from the local extension agents. Finally,after combining site-specific information (climate,soil analyses,seed variety,crop rotation,etc.),the startup drew up a set of guidelines for sustainable practices and sent them to the contacted farmers. Both SMS and messaging applications installed on smartphones were planned,but the company ultimately used SMS-based messaging to communicate with all the farmers.1The startup first developed an applet and mini-program on WeChat,a popular social media free messaging application in China.However,a survey showed that not every farmer had a smartphone and some were still using conventional mobile phones.The benefiting farmers are hereafter referred to as ‘treated’ farmers. Throughout the wheat growing season in 2016-2017,the startup sent a total of eight messages to each of the treated farmers with detailed advice on how to conduct sustainable farm management. Together with the advice,the startup also sent the results of the diagnosis -either site-specific or community-based -to the farmers,including the results of soil analyses,crop population,leaf analysis,and weather.A local extension agent was identified in the messages in case farmers had additional inquiries or requests. The details of technical regulations and in-field guidance are summarized in Table 1.

2.2.Survey and data collection

The sampling strategy aimed at a trade-off between a‘light-touch’ intervention that would capture the natural course of the pilot project and statistical legitimacy.Based on the agreement between the research team and the startup,in this case study,researchers presented themselves as independent observers (but not advisors).The startup refused the researchers’ proposal of randomly assigning the treatment (i.e.,the provision of ICT-based advisory services) as being used in several other studies with random control trials (Guoet al.2010) or framed interventions (Huanget al.2015). The startup insisted on consulting local extension agents to identify “best suited”farms whose size is relatively large in the selected village and who will be potential customers of the startup in the future,meaning the ‘treatment’ suffered from potential selection bias. However,thanks to the agreement,the startup provided the research team with a complete list of the farmers they were advising (or about to advise);these farmers were identified as “treated” in the analysis.On average,the startup provided advisory services for two treated farmers in each village. Within each farmer,two plots were randomly selected. In total,there were 45 treated farmers in 25 village communities that were attached to nine townships in four counties,in Binzhou City,Shandong Province. To have a control group in the same village and neighboring communities,this study randomly selected eight farmers per village who used the same wheat-maize crop rotation system and received no direct farming services from the startup. Therefore,the research design was framed under a with-without analysis,although the sample selection bias remained a concern.

To address potential estimation bias due to the nonrandom sampling and to control for time-invariant factors,this study conducted two-round surveys and designed a before-after analytical framework based on the same respondents. The baseline survey was conducted in mid-November 2016,through a face-to-face questionnairebased interview. The main aim of the baseline survey was to investigate farm management during the 2015-2016 wheat growing season (i.e.,before the startup pilot program began). In addition to the general information,all surveyed farmers were asked about details concerning inputs (especially fertilizer use) and yield from their own randomly selected two plots. One year later,at the end of the same cropping season,the research team returned to the same farms and conducted a follow-up survey.While the majority of survey questions were identical to those used in the baseline survey,the follow-up survey asked the farmers in the treated group,who had received short messages and site-specific agroecological services,questions concerning their awareness,perception,and feedback linked with the services. For example,if a farmer confirmed reception of an SMS from the startup,we asked whether he (or she) had read the whole message and had integrated the information and knowledge into their wheat production management.We also asked whether the treated farmers had ever contacted the technician identified in the SMS as being available to answer additional inquiries or provide additional support.

Based on the information collected during the two surveys,this study constructed a longitudinal dataset at the household and plot level. The final dataset consisted of 161 farmers and 612 plots and contained detailed information concerning the farmers’ wheat production management.2Some farmers refused (or were unable) to take part in the follow-up survey,and this led to an average attrition rate of nearly 30%. To ensure that the dataset was balanced in both the baseline and intervention period,we removed the baseline data corresponding to the samples missing in the follow-up survey.At the end,a total of 116 plots belonging to 31 farmers received site-specific field production management services from the startup through ICT-based technologies,accounting for 19.0% (116 of 612) of the plots sampled in the present study.

When piloting farmer services in the villages,the startup targeted ‘larger’ farms who were more responsive to ICTbased advisories. Table 2 summarizes the characteristics of both the treated and control farmers before the treated farmers received advisory services. Meanwhile,the twosamplet-tests were used to estimate the difference inmeans of farm characteristics between the treated and control farmers. The land cultivated by treated farmers was on average 4.5 ha. This was five times bigger than that cultivated by farmers in the control group. The household head of the treated farmers is younger than that of control farm.Other than the farm size and the age of household head,no significant differences in demographic characteristics were identified between the treated and control farmers in the baseline survey. In other words,for the pilot stage of the ICT-based advisory services of agroecological fertilizer management,the startup focused on larger farms.

2.3.Model specification

Many factors that were not controlled by the study could influence farmer’s N-fertilizer use,hence a multivariate econometric model was used to estimate the impacts of site-specific services on farmers’ N-fertilizer use:

DeltaNijk=β0+β1ICT-based serviceijk+β2Farm sizei+β3ICT-based serviceijk×Farm sizei+β4HHi+μijk

whereDeltaNijkis the outcome variable of interest and is measured in six ways,which represents the difference between N-fertilizer used in wheat production in two cropping seasons,2015/2016 and 2016/2017. First,theDeltaNijkare three binary variables,which equal 1 when the farmeriused less N-fertilizer in plotjin 2016/2017 than that in 2015/2016,otherwise 0. That is,the reduction of nitrogen fertilizer use in the cropping season (k=1),at the basal stage (k=2),and at the top dressing stage (k=3). Second,in addition to the binary measures,three continuous variables were used:total N-fertilizer use per hectare in the cropping season (k=1),at the basal stage (k=2),and at the top dressing stage (k=3). The intervention of site-specific advisory services facilitated by ICTs were delivered to farmerion two plotsj(j=1,2) under wheat.

The key independent variable of interest on the right side of the equation,ICTbased serviceijk,is a binary variable. Itequals 1 when farmiand plotkreceived the site-specific services from the startup,otherwise 0. The ICT-based advisory services included soil analyses,crop population,and SMS-based communication and advisory services throughout the wheat production season. To explore the effects of land consolidation and farm size on N-fertilizer use,the model included the variable Farm sizei. The interaction term between ICT-based serviceijkand Farm sizeiwas used to examine the heterogenous treatment effects of the advisory service on N-fertilizer use. The model also included the vector HHito control for additional farmer-level characteristics. To test whether N-fertilizer use in the baseline period affected the reduction in the following year,we included rational use of N-fertilizer in the baseline period in the estimation. It is a binary variable,1 when N-fertilizer use -be it overall or at different growth stages -in the baseline year by farmerifell within the optimal range;otherwise 0. For other control variables,age and education were measured in years for the household head,and share of off-farm labor was measured in percentage. Township dummy variables were used to control for territorial variations. The termμijkis the specific stochastic disturbance and is assumed to be subject to identical independent distribution. A logit model and a fixed effects model were employed to estimate the above equation;the results are presented in Section 3.2. The descriptive statistics for all variables are presented in Table 3.

Table 2 Farm characteristics by research design in 20161)

3.Results

3.1.Description analysis

The problem of excessive N-fertilizer use appeared to be less significant than that reported in previous research.Table 4 presents N-fertilizer use and the mean difference of N management between the treated and control farms.It shows that farmers applied N-fertilizer twice on average and there was no significant difference between the large treated farms and the control small farms. Prior to the entrepreneurial pilot project,the treated farmers used 265.6 kg N ha-1for wheat production (i.e.,less than 30%excess). For the control farmers,the excess was higher(90 kg N ha-1,about 45% of the recommended amount).Other studies conducted in Shandong reported an even higher rate of overuse of N-fertilizer in wheat production,such as 365 kg N ha-1in 2003 (Cui 2005) and 357-380 kg N ha-1between 2008 and 2010 (Jiaet al.2015).

Overall use of N-fertilizer by both the treated and control farmers for wheat production declined,and the reduction was greater on the treated farms. As shown in Table 4,about 65.52% of the treated farmers reduced their overall N-fertilizer use,while the reduction was slightly less (60.34%) on the control farms. It should be noted that total N-fertilizer use on the treated farms(265.60 kg ha-1) was lower than that on the control farms(290.80 kg ha-1) in the baseline year,possibly because of the difference in the size of the farms. Considering an already lower level of N-fertilizer use by the treated farms(relative to the control farms),the reduction of 41.66 kg ha-1(223.94-265.60 kg ha-1) -equivalent to 15.7% saving-is significant compared with the equivalent reduction of 11.3% on the smaller control farms.

The behavioral changes in nitrogen management on the treated and control farms became clearer when the N-fertilizer use was broken down into different wheat growth stages. As shown in Table 4,farmers in the study area used more nitrogen before the top dressing stage in the baseline year. For example,on the treated farms,about 40% (105.15 of 265.60 kg ha-1) of N-fertilizer was used in the early stage and the remaining 60% was used after jointing in the baseline year. The proportion did not differ significantly on the control farms in the baseline year. In the treatment year,the rates of N-fertilizer used at different growth stages approached the recommended rates on the treated farms,109.81 and 114.13 kg ha-1,respectively. The ICT-based services thus appeared to be effective. On the control farms,the rates of N-fertilizer use before and after jointing remained 117.04 and 140.89 kg ha-1,respectively.

Table 3 Descriptive statistics on all the variables

Table 4 N-fertilizer use by treated and control farmers for wheat production in the 2015-2016 and 2016-2017 cropping seasons1)

3.2.Multivariate analysis

Multivariate analysis shows robust evidence of farmers’adaptive and partial compliance with the ICT-based advisory. The estimated coefficient of the variable ICT service was 0.21 (38-17%),and it was positive and statistically significant (Table 5,column 1). All other factors being constant,farmers who received services from the startup had a 38% higher probability of reducing N-fertilizer use than the control farmers. When further examining N-fertilizer use at different growth stages (i.e.,at the basal stage in column 2 and at the top dressing stage in column 3),the effects were also significant,indicating behavioral changes in nitrogen management that are more site-specific and adapted to local characteristics and variations. Despite behavior changes towards sitespecific and adaptive N-fertilizer management,it is interesting that the overall and by-stage N-fertilizer use did not reduce significantly (Table 6). This indicates a partial adoption of ICT-based farming advisory.

The effects on individual farmers’ wheat production pipelined by the startup were heterogeneous according to farm size. The estimated coefficient on the interaction of ICT services with farm size was negative and statistically significant,implying thatceterisparibusone additional hectare in farm size decreased the probability of reducing N-fertilizer use by 17% (Table 5,column 1,row 3). In other words,the probability of N-fertilizer reduction on large farms was lower than that on small farms. It is partly because the larger farms had already used lower rates of N fertilizer than small farmers because of their more professional farm management decisions in the baseline period (Huanget al.2012).

Improved nitrogen management had a positive effect on wheat production. As seen in Table 7,the average yield obtained by the treated farmers was 7.38 t ha-1in the baseline year,which increased to 7.84 t ha-1(i.e.,a 6.2%increase in yield) in the 2016-2017 cropping season. Similar effects were observed for the control farmers but the magnitude of the increase (3.0%) was slightly lower than that obtained by the treated farmers. Given that the ICT-based advisory services did not significantly reduce the overall nitrogen use,the positive effects on yield for the treated farmers were mostly attributed to variable-rate N-fertilizer use at different crop growth stages and site-specific management.

Table 5 Estimated reduction in N-fertilizer use in 2015-2016 and 2016-2017 wheat production in Shandong,China (Logit model)1)

Table 6 Estimated farmers’ use of N-fertilizer for wheat production in the 2015-2016 and 2016-2017 in Shandong,China (Fixed-effects models)1)

Table 7 Farm wheat yields (t ha-1) in the 2015-2016 and 2016-2017 cropping seasons1)

The survey asked the treated farms about their communication and interactions with the startup or with the extension agents in the county identified in the messages. Only 5% of the treated farmers reported a strong motivation and follow-up contact for additional inquiries concerning problems or solutions.A similar finding was reported in a recent study on Chinese farmers’ compliance with agricultural production standards (Dinget al.2019). All surveyed farmers were also asked about their information exchange with other farmers in the villages. The reported rates of such communications were surprisingly low. In fact,most farmers obtained farming advisory from agro-inputs dealers who may oversell the use of fertilizers.

Financial sustainability is at the heart of the challenges facing the entrepreneurial innovations we studied. The ICT-based advisory services potentially produce economic benefits of 90 US$ per hectare (i.e.,about 40 CNY per Chinese mu,1 mu=1/15 ha) by saving the cost of fertilizer and increasing yield.Given the small average farm size,farmers’interest and incentives to participate in lowcarbon practices are limited;the average profit obtained by applying nitrogen optimizationtechnologies would be only 264 CNY per farm,equivalent to 40 US$ for the whole wheat cropping season.3The average farm size was 0.44 ha in the local area in 2016.The financial sheet obtained from the startup further indicated that the operating cost of the project was 98 US$ per farm for the site-specific advisory services,including personnel,utilities,R&D and overhead costs. After adding the costs relating to soil analyses,crop population and other sitespecific diagnoses,which were financed by the National Research Program,the unit cost of delivering the ICTbased advisory services totaled almost 150 US$ per farm. These costs are almost twice higher than other community-based and participatory approaches such as Farmer Field Schools (FFS).4The unit cost of FFS was estimated at 80 US$ per farm for an integrated farm management with a thematical highlight of nitrogen management in China (Guo et al. 2015).ICT-based agroecological services thus seem to be a “heavy-touch” and resourceintensive solution to agricultural sustainability.

4.Discussion

This study confirms the heterogeneous effects of ICTbased services by farm size. The entrepreneurial venture was inclined to target ‘larger’ farms. However,large farms were ‘risk-averse’ than small farmers when being advised to significantly reduce N-fertilizer use. They may pilot new technologies first on one or two individual plots and then scale up the practices after the effects are verified after several seasons. This is in agreement with the results of Ma and Zheng (2021),who reported the heterogeneous impacts of smartphone-based information use. Large farmers were found to be ‘shrewd’ in N-fertilizer use given greater cost relating to fertilizer use;they are more interested in and participatory in adopting efficiencyenhancing technologies,such as nitrogen optimization.

The findings indicate a dynamic process of adopting sustainable practices in agriculture,from partial adoptions to a full behavior change. The results show limited impact on full adoption of the recommendation of sitespecific N-fertilizer. As found in a few studies,technology adoption is a deliberate decision that demands attention,and occasionally those intentions are lost in the shuffle,buried by everyday distractions (Larochelleet al.2019;Carrión-Yaguanaet al.2020). Text messaging provides new avenues for farmers to get information and reminders,and it can improve and increase the adoption rate of technology practices (Stocwellet al.2012).However,the complex concept of site-specific and timevariant N-fertilizer use is not easily conveyedviaICT(Steinkeet al.2021). The effects of ICT-based services on farmers’ decision-making are often intermediate,and the process is dynamic and social,involving adaptive learning and trust (Akeret al.2016).

Our analysis of the effectiveness of ICT-based farmer services and financial sustainability calls for deeper thinking and debate about the organizational identity of startups that combines the interest of the public and private sectors. The startup registered itself as a private company providing agribusiness and technological services;venture capital was the primary external resource of capital. While the startup is profit driven and mainly responsible to the shareholders and equity investors,the agroecological services generated significant environmental value and moderate developmental effects (for smallholder farmers). This blending of profit-driven business and social mission goes beyond the conventional concept of socially responsible behavior in the private sector (such as corporate social responsibility). As Dacinet al.(2011) indicated,from the global perspective the concept of social entrepreneurship,an emerging phenomenon and new financial tools,are facilitating entrepreneurial innovations that address persistent social and environmental problems. Social entrepreneurship,by definition,is primarily mission driven and adopts hybrid business strategies by “doing good and doing well”. The startup we studied in this case does not seem to fit this definition. The Chinese government has made tremendous efforts to encourage entrepreneurial innovations (called Mass Entrepreneurship and Mass Innovations) to unleash the growth potential of the economy. While most initiatives are implemented in the business sectors,the government could consider promoting and incentivizing more private capital to invest in entrepreneurial innovations tackling social and environmental problems. In the era of digital revolution in agriculture,a new governance form of public-private entrepreneurship is up and coming (Qianget al.2009).

This anecdotal study projects a trajectory of reconfiguring China’s farming system towards resilience and sustainability in the digital era. While the farming system is being transformed from the private to the public sphere,external stakeholders have become increasingly involved in the internal farm operations,the decision making and knowledge exchange (Noe and Alr?e 2011).The development of ICTs and the Internet of Things(IOTs) can transform agriculture by translating tacit knowledge that was not previously communicated into codified information. However,technological innovations such as ICTs or IOTs should not be applied based on the desire for “control-and-command”,but rather on the desire to mobilize a collective and distributed intelligence more efficiently than the discipline-based individual approaches (Menard and Casabianca 2011). In this exemplary case,the entrepreneurial startup networks the national research capacities and individual farmers through ICTs. Nevertheless,the engagement of individual farmers was not participatory,nor was local extension capacities involved. Such a business model does not fit agroecological approaches that seek to reach locally acceptable and adaptable solutions. The advancement of communication and digital technologies needs to make the system ‘smart’ responsive and adaptive to the disturbance caused by promoting learning and sociotechnical governance,rather than providing a linear way of centralized decision-making by technical authorities,be they public or private.

What does China learn from the Project on reforming its national research support on agricultural sustainability and digitalization? A total of US$ 363.6 million was budgeted and delivered through 49 projects of the National Research Program throughout 2016 and 2018 to promote applied research and technology development dealing with excessive agrochemicals use in agriculture. However,very lean resources were invested in searching for ‘innovations’ from societal and management perspectives. This study and numerous other studies suggest that sectoral innovations and sustainability transitions only occur when the ‘social’ (e.g.,human advisory and field facilitators) and ‘technical’ (e.g.,digital services) aspects of innovations can collaboratively identify a diverse set of system weaknesses (Jacobsson and Bergek 2011). The reconfiguration of China’s farming systems will be impossible if the process is independent of the reform of the socio-technical regime and the inspiration of a culture conducive to scientistled entrepreneurship. After all,putting technological innovations into practice requires experiment,outreach,adaption,and collaborations,and such a process needs to include a range of actors regardless of their technological capacities (Jacobsson and Bergek 2011).

The present study has some limitations. Firstly,this study did not investigate the mechanism that influences farmers’ behavior change,such as knowledge about fertilizer. If farmers are “told” to adopt fertilizer practices without increasing internalized knowledge,the sustainability of their input use may still be a concern.Secondly,although fixed effect (FE) model was used,the endogeneity problem was not completely overcome because of the surveyed sample. However,unlike some previous studies of randomizing farmer services(Huanget al.2015),this study took a different approach of action research and did (and could) not introduce an experimental research design. Observing the real behavior and strategies of businesses was part of the research objectives. Thirdly,surveyed details of the cost of fertilizer and dissemination (such as labor and opportunity cost) were not sufficient for additional analysis. The research interest of economic impacts associated with reduced fertilizer use can be referred to some other studies such as Cole and Fernando (2021).Lastly,it is likely that conducting survey may have an impact on farmers’ nitrogen fertilizer use behavior (socalled,Hawthorne effect). Although such a distortion would be similar for both treated and control farmers,it might affect the rates of N-fertilizer.

5.Conclusion

Excessive fertilizer use has been a significant concern in China’s agricultural sustainability,and ICT opens wider opportunities to address associated problems. This case study is the first to document the operation of an entrepreneurial startup financed by venture capital. It examines the effectiveness of ICT-based farmer services for nitrogen management in wheat production in the North China Plain. This study found genuine changes in sustainable farm practices through ICT-facilitated services.Nevertheless,the service provider showed a biased interest by serving ‘large’ farms as they opt for pioneering new technologies and are more entrepreneurial. While the smallholder farms seemed to be benefiting from reducing N-fertilizer use and adopting nitrogen optimization in wheat production,they may not be the targeting users for the business entrepreneurs. Farmer services of sustainable practices by high-tech startups may not suit the self-sufficient or semi-self-sufficient farmers. Additional institutions and public extension services should be complementary,and entrepreneurial innovations may not be the mainstream or backbones for China to transform its agriculture towards sustainability.

Acknowledgements

We are grateful to the members of the survey team who worked hard to collect the data. We acknowledge financial support from the National Natural Science Foundation of China (72003148),the Agricultural Science and Technology Innovation Program of the Chinese Academy of Agricultural Sciences (CAASASTIP-2016-AII),the Chinese Universities Scientific Fund (2452020072),the National Key Research and Development Program of China (2016YFD0201303).

Declaration of competing interest

The authors declare that they have no conflict of interest.