Factors Affecting Farmers’ Use of Inorganic Fertilizers and Manure in South Asia

45 Fertilizer, though one of the most essential inputs for increasing agricultural production, is a 46 leading cause of nitrous oxide emissions from agriculture, contributing significantly to global 47 warming. Therefore, understanding factors affecting farmers’ use of fertilizers is crucial to 48 develop strategies to improve its efficient use and to minimize its negative impacts. Using data 49 from 2,558 households across the Indo-Gangetic Plains in India, Nepal, and Bangladesh, this 50 study examines the factors affecting farmers’ use of organic and inorganic fertilizers for the 51 two most important cereal crops – rice and wheat. Together, these crops provide the bulk of 52 calories consumed in the region. As nitrogen (N) fertilizer is the major source of global 53 warming and other environmental effects, we also examine the factors contributing to its 54 overuse. We applied multiple regression models to understand the factors influencing the use 55 of inorganic fertilizer, Heckman models to understand the likelihood and intensity of manure 56 use, and a probit model to examine the over-use of N fertilizer. Our results indicate that various 57 socio-economic and geographical factors influence the use of inorganic fertilizer in rice and 58 wheat. Across the study sites, N fertilizer over-use is the highest in Haryana (India) and the 59 lowest in Nepal. Across all locations farmers reported a decline in manure application, 60 concomitant with a lack of awareness of the principles of appropriate fertilizer management 61 that can limit environmental externalities. Educational programs highlighting measures to 62 improving nutrient-use-efficiency and reducing the negative externalities of N fertilizer over- 63 use are proposed to address these problems.


Introduction 67
Achieving food security, addressing climate change, and halting environmental and natural 68 resource degradation are among the key challenges faced by the agricultural sector in efforts 69 to achieve sustainable development goals (SDGs 1 ) and the Paris Agreement to limit global 70 temperature increase to below 2 ºC (IPCC, 2014). Fertilizer use, particularly nitrogen (N), is 71 one of the important land management practices to increase crop production and improve soil 72 fertility. Thus, the use of soil fertility enhancing amendments to supply essential nutrients in 73 crop production is of clear importance. Along with the nutrient supply from soil organic matter, 74 crop residues, wet and dry deposition, and biological nitrogen fixation, synthetic fertilizer is a 75 primary source of essential nutrients in crop production. 76

77
The success of the Green Revolution (GR) in 1960s to increase food production and to reduce 78 hunger world-wide was made possible, partly due to increasing use of chemical fertilizer 79 (Erisman et al., 2008). However, excessive chemical fertilizer use during and post GR caused 80 a number of environmental and ecological problems such as soil acidification, degradation, 81 and water eutrophication, severely undermining the sustainability of agriculture (Lu and Tian, 82 2017). The loss of applied nutrients into the environment resulted in the fertilizer-induced 83 emission of nitrous oxide (N2O) from agricultural production, a major source anthropogenic 84 greenhouse gas emissions (Sutton et al., 2013). Around 60% of nitrogen pollution is estimated 85 to originate from crop production alone, particularly through nitrogen (N) fertilizer application 86 respectively. For defining over-application of N, we followed the findings from (Sapkota et al., 240 2018a) in Bihar and Haryana in India and adapted it for our study sites. According to their 241 study, wheat farmers who applied more than 140 kg N ha -1 suffered a yield penalty and had 242 high greenhouse gas emissions intensity. Similarly, rice yield leveled off with increased N 243 application beyond 100 kg N ha -1 . Based on (Sapkota et al., 2018a), we created two variables, 244 (1) 'over-use of N for rice' and (2) 'over-use of N for wheat'. The variable first is a binary 245 variable with value of one if the amount of nitrogen applied is more than 120 kg N ha -1 in rice 246 fields, and zero otherwise. Similarly, the variable 'over-use of N for wheat' is a binary variable 247 with value of one if the amount of nitrogen applied is more than 140 kg N ha -1 in wheat plots 248 and zero otherwise. Hence, we applied probit model combining all locations together (for 249 details, see (Wooldridge, 2010)). 250 251

Variables and hypotheses 252 253
We have eight dependent variables in our analysis: urea applied to rice and wheat, respectively, 254 DAP applied to rice and wheat, respectively, decision to use manure in rice and wheat, 255 respectively, as well as the quantity of manure applied in rice, and wheat, respectively. The use 256 of manure in rice and wheat are binary variables, while others are continuous. 257

Household characteristics 266
Household characteristics including education, age, and gender of household head, family size, 267 and migration can influence technology adoption decisions. Educated individuals are assumed 268 to be able to more easily acquire new information and are more likely to adopt (Chowdhury et 269 al., 2014). Past studies indicate that they are more likely to use chemical fertilizer (Omamo et 270 al., 2002;Takeshima et al., 2016). Further, elderly farmers apply more fertilizer relative to 271 manure, as its use requires less labor compared with manure application. Conversely, with 272 longer experience in agricultural management and benefits of organic matter in improving soil 273 fertility, older farmers have also been observed to prefer manure (Waithaka et al., 2007). Rural 274 out-migration reduces the availability of household members to perform farm tasks; however, 275 it also increases access to alternative income streams through remittances that can assist in 276 purchasing fertilizer. 277 278

Economic and social capital 279
Economic capital consists of land, livestock, farm assets, household endowments and off-farm 280 income sources, whereas social capital can include membership in village institutions such as 281 farmer cooperatives/clubs. To capture the effect of wealth on the use of fertilizer and manure, 282 we constructed household asset index (AI) using principal component analysis (

Farm land characteristics 302
To control for the potential effects of land attributes on fertilizer/manure use, we included 303 farmers' tenure status, soil fertility, soil depth, land slope, irrigation status, and distance to plot 304 from homestead in the analysis. Distant plots require increased transaction costs due to the 305 price of purchasing transport for inputs. Fields far from the household are also more difficult 306 to monitor. Therefore, input use was hypothesized to be inversely related to distance from the 307 household to the field. Farmers want to supply adequate inputs to fields that they can more 308 closely monitor and intervene in with management to achieve greater productivity. Fields that 309 farmers perceive as being less fertile may also receive more manure compared to fertile ones 310 because manure releases nutrients slowly and can improve soil physical and chemical 311 properties over time (Waithaka et al., 2007). Considerable differences were observed in the application of urea and DAP across study sites 319 (Table 2). The average amount of urea applied to rice was 315 kg ha -1 in Haryana (India), while 320 it is only 205 kg ha -1 in Nepal. The average amount of DAP applied in rice was highest in 321 Haryana (130 kg ha -1 ), followed by Bihar (95 kg ha -1 ), Bangladesh (65 kg ha -1 ), and then Nepal 322 (62 kg ha -1 ). The average amount of urea applied in rice is much higher in Bangladesh 323 compared to Bihar and Nepal. Generally, average rates of urea and DAP applied in both rice 324 and wheat were lowest in study sites in Nepal when compared with India and Bangladesh. 325 326 Unlike urea and DAP, farmers did not apply manure to all fields in the survey sample. In Bihar, 327 (India) and Nepal, 47% rice plots received manure, while 26% received manure additions in 328 Bangladesh. Manure was applied in 24% of plots cultivated to wheat in Nepal. Across all 329 locations, farmers in focus groups indicated that their use of manure is decreasing. They 330 reported that educated young household members are less interested in carrying manure to plots 331 and as a result, use of chemical fertilizer is increasing over time. In Haryana, the average 332 amounts of manure use in rice and wheat fields were 1,899 and 1,680 kg ha -1 , respectively 333 while they were 925 and 1,250 kg ha -1 in rice and wheat fields in Bihar. 334 The factors influencing the amount of urea used in rice are, in most cases, similar to the ones 354 affecting DAP (Table 3) Farmers applied less urea and DAP to fields they indicated were more inherently fertile in 383 comparison to those they deemed as fertile. Irrigation was positively associated with fertilizer 384 rates in Bangladesh and Nepal, but not in Bihar. Farmers with secured tenure also tended to 385 apply higher rates of fertilizer compared to those who rented-land. Most farmers are found to 386 use of Urea and DAP as complementary inputs in rice farming. We also observed that factors 387 affecting the amount of urea and DAP applied to wheat (Table 4) are generally similar to that 388 of rice in all countries (see Table 3 Wald chi-square tests for the Heckman two-step models were significant at 1% level (Tables 5  397 and 6), indicating the validity of the models in explaining observed differences in farmers' 398 decision to adopt manure in rice and wheat. The Inverse Mills Ratio was highly significant in 399 all models, implying that manure use intensity depends on the likelihood of farmers to decide 400 to apply manure. Sets of the observed factors appear to affect the likelihood of farmers' choice 401 to utilize manure. These factors differed from the ways they influence the rates of manure used 402 among the subset of farmers who chose to apply manure. Furthermore, some factors appear to 403 have contradictory effects on the choice to apply manure vis-à-vis its application rate. As 404 several factors influencing the adoption and intensity of adoption of manure in rice ( The likelihood of manure application increased with land size, but the rate of application was 420 inversely related. Farmers reported that the young and educated generation tend not to prefer 421 labour-intensive approaches such as manure application. Livestock ownership, the major 422 source of manure, was unsurprisingly positively associated with both likelihood and intensity 423 of manure use in all sites. Access to credit had variable effects manure use: no impact was 424 observed in Haryana, while negative and positive impacts on were found in Bihar relative to 425

[Insert here
Bangladesh and Nepal. Factors such as off-farm income, market access, training and 426 membership in village institutions also increased the use of manure in both crops. 427 Farmers are also less likely to apply manure they perceived of as being highly saline. although application over multiple seasons may be required for desirable impacts (Ding et al., 432 2020). Those who did apply manure also used lower rates. Similar is the findings for plots that 433 are located farther from homestead. In contrast, farmers were more likely to adopt manure in 434 fields they perceived of having greater soil depth. In Bangladesh and Nepal, irrigation was also 435 positively associated with both the likelihood and rate of manure use. Compared to rented 436 fields, farmers with tenure also had greater rates of manure application. 437 438

Factors affecting the over-use of N fertilizer 439
A total of 33.6% of rice fields in our data were reported by farmers as receiving more than 120 440 kg N ha -1 and 18.5% of the wheat fields received more than 140 kg N ha -1 (see appendix: Tables 441 A1 and A2). Though the critical value of N that is used here for defining over-use of N may 442 vary across sites, we believe it to be rational here as our objective is to examine the factors 443 contributing to the over-use of N. Moreover, we estimated the models by changing the critical 444 value of N for rice as more than 100 kg N ha -1 and for wheat 120 kg N ha -1 , the variables 445 significantly affecting the over-use of N in rice and wheat remained almost same. Therefore, we analyzed the factors affecting over-use of fertilizer by combining data from all locations 447 (Table 7). 448 [Insert here Table 7] 449 Compared to FHHs, MHHs appear to be more likely to over-use N. Household heads with 450 secondary and higher educational levels also appear to have a higher likelihood to over-use of 451 N (significant at 5% and 1% level). Land size, off-farm income, and wealth were also 452 significantly and positively associated with the over-use of N. 453 454 Households that are farther from market are perhaps unsurprisingly less likely to over-use N 455 fertilizer. Training on fertilizer management is negatively associated with the N over-use. 456 Farmers are not likely to over-use N fertilizer in plots they consider as having higher inherent 457 fertility levels, or that have deeper soil profiles. Over-use of N fertilizer does appear to increase 458 with access to irrigation, a consistent findings as observed in the states with successful GR. 459 Across the study locations, Nepal has lowest percentage of over-use of N in both rice and 460 wheat. All other countries conversely have a greater likelihood of overusing N. 461 462 Overall, a wide variation is observed across the study sites regarding the over-use of N fertilizer 463 in rice and wheat. The findings indicate that training on fertilizer management is more crucial 464 than formal education to reduce the over-use of N. This also calls for a new study focusing on 465 the institutional analysis including the information and support that farmers get from extension 466 agents to regularly update the fertilizer need for their farm, information on soil testing and 467 nutrient adjustment as per requirement, and increased access to nutrient management tools. 468 469

Farmer focus group findings 470
Our discussions with farmers in the study villages confirmed the complex interactions of 471 variables that conditions decisions on fertilizer use. Although many farmers indicated that they 472 could not clearly understand or 'separate' the impact of a single production factor such as 473 fertilizer to distil its implication for overall crop productivity, they did however clearly indicate 474 that they understand the importance of fertilizer but lack knowledge on its balanced application. 475 Therefore, they are more likely to go for over-application of N, mainly from urea, ignoring 476 other nutrients. 477 478 Farmers reported that government subsidy plays important role in fertilizer use. In all South 479 Asian countries exempting Sri Lanka, government subsidy is higher on urea than on other 480 fertilizers. In focus groups, farmers indicated that this has a strong influence on their 481 application of fertilizer N at rates that are often higher than required. Together, the uneven 482 subsidy structure also works as disincentives in using recommended rates of secondary and 483 micronutrients. This is in line with the findings of (Sapkota et al., 2018a). show that they prefer using off-farm income to purchase agricultural inputs rather than credit. 510 Thus, public policy that create off-farm income generation opportunities are likely to be of use 511 particularly for resource poor and smallholder farm familieswhen increasing fertilizer use 512 are the goal. Conversely, where farmers routinely over-apply fertilizers, or practice imbalanced 513 application, more complex policy and development interventions may be needed, including 514 educational programs, direct training, and behavioral nudging methods that may encourage 515 more rational use. Similarly, few farmers in the study locations appeared to have sophisticated 516 knowledge on the negative effects of fertilizer over-use. Thus, training on fertilizer 517 management, along with the dissemination of knowledge on negative externalities of its 518 inappropriate use is likely to be important in preventing unsound over-use. 519 520 Another crucial finding is that application of manure is decreasing in all locations. Perhaps 521 surprisingly, where farmers have higher levels of education, our data suggest that they are more likely to opt out of manure application to rice and wheat fields. Combined with evidence on 523 decreasing concentrations of soil organic matter in our study countries, this is arguably 524 concerning, and indicates the need for appropriate educational and extension programs, 525 methods to offset the high labor demand and costs of manure application, possibly through 526 scale-appropriate machinery options to encourage educated and young farmers to use manure 527 when and where it is most needed to maintain long-term soil fertility. This study is based on survey methods involving interviewing farmers to answer questions 549 about their socioeconomic and farming activities. Like all socioeconomic surveys (or any data 550 collection that involves collecting data from family or community representatives) Institutional 551 Research Ethics Committee (IREC) of International Maize and Wheat Improvement Centre 552 (CIMMYT) classified it as low risk and approved the study. Entire research methods were 553 performed in accordance with the relevant guidelines and regulations issued by CIMMYT 554 institutional research ethics committee (IREC). 555 556 Consent to Participate 557 Each questionnaire of this study had a front page section that required informed consent for 558 interview before the interview could proceed. Interviewers were trained and under instructions 559 to read aloud the consent statement to each interviewee before the interview could advance. 560 Participants were informed that they were under no obligations to answer any questions or they 561 could stop the interview at any time, without giving any reasons and ask that any partial data 562 recorded be removed from the records. This way the survey was consistent with CIMMYT-563 IREC policies and those generally applied in low-risk social science research. 564 565 Consent to Publish 566 Before the interview could proceed, we obtained informed consent from the participant that the 567 data will be analyzed and a paper will be published.  Availability of data and materials 599 Authors do not have the right to share the data. However, it will be made available to the reader 600 upon request.      1576 Note: Standard errors in parenthesis. *, **, and ** refer to 10%, 5%, and 1% level of significance, respectively. 'na' refers to not applicable. 977 Note: Standard errors in parenthesis. *, **, and ** refer to 10%, 5%, and 1% level of significance, respectively. 'na' refers to not applicable. 740 Note: Standard errors in parenthesis. *, **, and ** refer to 10%, 5%, and 1% level of significance, respectively. 'na' refers to not applicable. Distance to plot -0.27*** -350.18*** -0.29*** -311.01*** -0.33*** -260.14*** (0.09) (120.21) (0.10) (94.21) (0.08) (91.08)