This study was carried out in 11 municipalities in four agriculture development centers (ADC) of Benin (Figure 1). The municipalities of Banikoara and Kandi in cotton crop zone (Northern Benin; ADC 2); Ouèssè, Glazoué, Dassa-Zoumè and Djidja in cashew, cotton and food crops zone (Center Benin; ADC 4); Agbangnizoun, Zogbodomè, Kloukanmè and Djakotomè in Fruit trees and food crops zone (South Benin; ADC 5) and Kétou in palm oil and food crops zone (South Benin; ADC 6). In cotton crop zone, cotton is the main crop growth. The food crops like maize, sorghum, rice, and cowpea are grown to profit of carry-over effect of fertilizer. Intensive cattle farming is also practice. This zone of Benin (Department of Alibori) is between 10˚31'19'' and 11˚45'00'' of latitude north and between 2˚3'00'' and 3˚16'30'' of East longitude [13]. Its climate is North-Sudanian [14], whose seasonal rhythm is influenced by the displacement of the intertropical front. Average annual rainfall is around 850 mm [13]. Soils are Tropical Ferruginous leached with concretions on sand-clay sediments. A wide variety of crops is fund in cashew, cotton and food crops zone. Cotton and cashew are main crops of this zone. However, maize, rice, cassava, cowpea are also mostly grown and intensive cattle farming. This zone has a subequatorial climate influenced by the southern Sudanian area. The average annual rainfall is around 1100 mm. In this zone there are Tropical Ferruginous soils, and Vertisols [15] [16] [17]. South Benin (Palm oil and food crops, fruit trees and food crops and fishing and market gardening zones) where food crops are mainly grown is characterized by Sudano-Guinean or subequatorial climate [15] [16] [17]. There are very rich loam-clay soils (valleys) and Ferrallitic soils. The annual rainfall generally recorded in the area varies between 800 mm to 1400 mm. These municipalities were chosen on the basis of rice cultivation by farmers and they also belong to research and development sites of National Institute for Agriculture Research in Benin.

The number of respondents was determined using the normal approximation of the binomial distribution proposed by Dagnelie [18] below:

N i = [ ( U 1 − α / 2 ) 2 × p i ( 1 − p i ) ] / d 2 ; where N_i is the number of farmers investigated who grow rainfed rice, U_1−α/2 is the value of the normal random variable for a probability of 1−α/2, α the type I error. For α = 5%, the probability 1−α/2 = 0.975 and we have U_1−α/2 = 1.96; p_i is the proportion of people engaged in rainfed rice cultivation in the study areas and d the estimation margin error, which is 5% for this study. A_i, the percentage of farmers engaged in rainfed rice grow in a municipality, A_i = Ac_i/At_i (Ac_i, the number of farmer of a municipality engaged in rainfed rice and At_i the total number of farmer of all the municipalities). The sample size of each municipality would be: n = A i ∗ N i . In each municipality, farmers were identified using simple random sampling. Thus, in total, 763 rainfed rice farmers were surveyed from June to December 2017 (Table 1).

Data collected include socio-economic characteristics of farmers, agricultural practices of rainfed rice farmers. Composite soil samples were taken from 277rainfed rice fields of farmers and were analyzed at laboratory of Soil Science, Water and Environment in National Institute of Agricultural Research of Benin (NIARB) for the determination of physicochemical parameters. Rice paddy yield were collected where composite samples were taken.

Composite soil samples were collected at a depth of 0 - 20 cm in the different rainfed rice fields. The elementary samples were collected randomly in each rainfed rice field. At least, 16 elementary samples were collected per field as recommended by Schvartz et al. [19]. These elementary samples were collected to form a composite sample per field. In total, 277 composite soil samples were collected from 25, 26, 28, 27, 29, 23, 24, 25, 22, 25, 23 farmer’s fields in Banikoara, Kandi, Ouèssè, Glazoué, Dassa-Zoumè, Djidja, Agbangnizoun, Zogbodomè, Kloukanmè, Djakotomè and Kétou respectively. The soil pH (H₂O, 1:1), organic carbon, and total soil nitrogen contents, available Bray-1 phosphorous, exchangeable potassium and exchangeable capacity cation were analyzed according to the methods described by Dagbenonbakin [20].

Four cropping systems (cluster) emerged from multiple correspondence analyses and ascending hierarchical classification results (Figure 2). Tillage method, crop residues use, rice varieties, crops rotation, seeds origin, organic fertilizers use and seeding mode are variables strongly link to cropping system (p < 0.0001). First cluster (system 1) includes farmers who practiced manual tillage, used organic fertilizer, left crop residues in fields and practiced direct seeding of local variety obtained from last harvest. Second cluster (system 2) includes farmers who practiced mechanical tillage; nursery of improved varieties obtained from research center, used only inorganic fertilizer, left crop residues in fields and practiced crop rotation. Third cluster (system 3) represents farmers who practiced manual tillage; direct seeding of improved varieties obtained from research

center, did not use fertilizer and burnt crop residues. Fourth cluster (system 4) represents farmers who practiced harness tillage; direct seeding of local variety obtained from local markets or last harvest, did not use fertilizer and used crop residues as animal feed. However, 34.82% 19.1%, 23.04% and 23.04% of farmers adopted system 1, system 2, system 3 and system 4 respectively. These cropping system influenced rainfed rice yield. In fact, plant residues management and ploughing method determine rice yield [29]. The positive effect of rice crop residues incorporation into soils on rice yields has been reported previously by Zingore et al. [30]. Machine tillage improved permeability; percolation as well as water retention capacity that increase rice yield [29]. Farmers of Center Benin mostly adopted the system 1 and system 3, while those of southern and northern Benin adopted the system 2 and system 4 respectively (Figure 3). The area used for rainfed rice by farmers who practiced the system 3 is higher than that used by farmers for other systems (Table 2). Inorganic fertilizer use is not variable link to rainfed rice cropping system. Then, inorganic fertilizer’s adoption was not determined by rice cropping systems.

Inorganic fertilizer use intensity ranged from 12.6 to 248.9 kg·ha⁻¹ with a mean of 74.68 kg·ha⁻¹. The results of double hurdle model of factors influencing adoption and use intensity of inorganic fertilizer by rainfed rice farmers (Table 3) showed many factors determine inorganic fertilizer adoption (hurdle 1) and use intensity (hurdle 2). Inorganic fertilizer technology adoption depends significantly of age, educational status, Household size, contribution of rice to household income and organic fertilizer use. However, use intensity of inorganic fertilizer is significantly influenced by the age of farmer, training of farmer on soil fertility, filed areas under rice crop, contribution of rice to household income, years of farming experience, inorganic fertilizers availability and organic fertilizer

Means followed by the same alphabetical letters and for the same variables are not significantly different (p > 0.05) according to the Student Newman-Keuls test.

Note: *, **, *** indicates that the corresponding coefficients are significant at the 5%, 1%, and 1‰ levels respectively.

use. The age of farmers was significant (p = 0.004) and influenced negatively inorganic fertilizer technology adoption. The result implied that older rainfed rice farmers were less likely to adopt inorganic fertilizer. This result was similar with those found by [31]. However, the age of farmers was also significant (p = 0.006) and positively affected by inorganic fertilizer intensity use. In fact, older farmers were more risk averse and were assessed the attributes of a technology than younger farmers [31]. However, they were more financially able to intensify inorganic fertilizer use than young farmers. Educational status of farmers was significant (p = 0.008) and was positively related to adoption, suggesting that the more educated farmers were more likely to adopt inorganic fertilizer than, the less illiterate farmers. This result corroborated with those observed by Chianu and Tsujii [32] showing that the level of education of the household head affected positively and significantly the farmers’ decision to adopt or not inorganic fertilizer in the savannas of northern Nigeria.

Inorganic fertilizer adoption was significantly (p = 0.034) influenced by farmer household size. This result implied that farmers with high household size adopt more inorganic fertilizer than farmers with less household size. Indeed, inorganic fertilizer technology adoption requires more workers that can be provided by memberships of the farmer household. Training of farmer on soil fertility is also significant (p < 0.0001) and positively liked with inorganic fertilizer use intensity. The farmers who participated to soil fertility management training at least one time ago use more fertilizer (96.45 ± 25.8 kg·ha⁻¹) than those who never participated (35.12 ± 11.56 kg·ha⁻¹). However, inorganic fertilizer use intensity was negatively related to the area under rainfed rice crop (p = 0.023). The implication of the result was that, as farm size increases, farmers were more likely to use less inorganic fertilizer amount per hectare. The result confirmed those found by Edward et al. [31] about the negative relation between area under cultivation and inorganic fertilizer adoption. Contribution of rice to household income is significant (p = 0.015) and positively linked to inorganic fertilizer adoption. It’s also significantly (p = 0.003) linked to the use intensity of inorganic fertilizer. In fact, farmers who earned a substantial income from rice cultivation are predisposed to adopt new technologies to increase their production. They also made more effort to comply with good fertility management practices and a higher use of inorganic fertilizers. Inorganic fertilizer use intensity was significantly (p = 0.007) related to years of farming experience. Farmers, who have been growing rainfed rice for a long time, use more inorganic fertilizer than those who started to grow rainfed rice. This result implied that farmers with experience in inorganic fertilizer technology easily intensify the amount used especially with the decline of soil fertility. Those farmers perceived more the negative effect of declining soil fertility so they use more inorganic fertilizer.

Inorganic fertilizers availability was significant (p = 0.006) and positively related to its use intensity. Farmers used more inorganic fertilizer when the latter is available and accessible to them. Likewise, organic fertilizer use was positively (p = 0.034) linked to inorganic fertilizer adoption and negatively (p = 0.005) linked to its use intensity. This result suggested that higher quantity of organic manure applied, reduced the amount of inorganic fertilizer used. This was probably because farmers understand the complementary relationship derivable from the combined application of organic manure and inorganic fertilizer. This corroborates the finding of Chianu and Tsujii [32]. These different results showed that inorganic fertilizer technology was not a variable determining the soil fertility management by rainfed rice farmers. However, low yields were recorded in rice cropping systems. Policies to promote the adoption of inorganic fertilizers should target young rice farmers with at least primary school education level in order to hope that this technology will be adopted. Local rice markets need to be stimulated to increase local production of rice. This would lead farmers to increase plantings areas which was a determining factor in the adoption and intense use of inorganic fertilizer. The development of local rice markets would increase farmers’ incomes, which was also determinant in the adoption and use intensity of inorganic fertilizers.

The regression tree (Figure 4) showed that nitrogen, phosphorous, potassium, organic carbon, pH water, and exchanging capacity cation determined the rainfed rice yield in traditional cropping system. Nitrogen was the root of the tree so it was the main soil parameter that determines rainfed rice yield. Lower content of nitrogen (<0.054%) induces lower rice yield even others nutrient. This result was in accordance with the finding of Atsuko et al. [29] who showed that nitrogen application rate strongly determines the rice yield in southern-central Benin. Nitrogen is the most important variable that limits rice yield [33]. The phosphorous impact on rice yield was also reported by Dakshina-Murthy et al. [34] who observed increasing of rice yield with phosphorous supplying. Phosphorous application induced increasing of rice yield and quality [10]. Then, rice cultivation in Benin need to supply phosphorous because in tropical Africa soil, phosphorous availability of plant was the lowest due to susceptibility losses by leaching, run off (on sandy textured soils) and fixation by soil inorganic colloids [35]. Potassium influence on rice yield was reported by

many authors [36] [37]. Soil acidity induces high amounts of exchangeable aluminum (Al) that was a severe constraint to crop productivity because it reduces the availability of many nutrients (e.g. P, Ca, Mg, Zn) and inhibits root growth of most plants and limits nutrient to uptake [38]. The implication of the result of regression tree (Figure 4) was to increase rice yield, farmers need to use inorganic fertilizer in adequate amount. However, many of them did not adopt inorganic fertilizer technology (23%) and among farmers who adopted this technology, they used less than recommended rate of inorganic fertilizers (275 kg·ha⁻¹) which are 200 kg·ha⁻¹ N₁₅P₁₅K₁₅ and 75 kg·ha⁻¹ of urea 46% [39]. This situation leads lower rice yield in traditional cropping systems.