These phenomena make African smallholders running into poverty trap if the nutrient mining process is not reversed to improve farms’ livelihoods.

These facts illustrate the failure of the current farming systems and policy intervention approaches aiming at supporting livelihood of rural population and ensuring satisfactory living conditions for populations in sub-Saharan Africa. The adoption of proven technologies that can help improve agricultural production remains low (Kanyama-Phiri et al., 2017; Nord et al., 2022b). The low performance of current farm systems, experience and lessons learned (Nord et al., 2022a)showed there is a need for holistic approaches (Kanyama-Phiri et al., 2017)that could lead to more successful farming systems and policy interventions. To that end, agricultural research has a key role to play. It can help in creating a better understanding of the condition for successful intervention. Interventions that lead to transforming the current farming systems as well as identifying pathways for transition to farming system designs that create and catalyze the resilience of farmers’ livelihood.

Integrated system modeling offers the opportunity to better understand the issues farmers are facing and for identifying and testing potential solutions. For Stewart et al. (2019), integrated approach accounting for bio-physical and socio-economic factors is required for addressing soil fertility issue in Sub-Saharan African, and therefore for improving agricultural productivity and livelihood conditions of farmers. However, capturing farming systems heterogeneity constitutes an important step in integrated farming research and systems modelling (Le, 2005). Policy interventions wise, accounting for heterogeneity is crucial for ensuring the effectiveness of agricultural policies. Gassner et al.(2019)highlighted that developing agriculture in Africa goes through differentiating policy intervention based on smallholder heterogeneity. Indeed, smallholder farms in general (Tittonell et al., 2005; Chikowo et al., 2014) and those of West African drylands in specific, are characterized by their socio-economic and agro-ecological heterogeneity (Stewart et al., 2019). They exhibit different biophysical (Nord et al., 2022a) and socio-economic settings in relation to their livelihood endowment and orientation which change over time.

Objectives of the study

The overall objective of this study is to analyse agricultural livelihood heterogeneity in Pontieba for farming system design and integrated agricultural livelihood systems modeling. The specific objectives of the study are: (1) To identify main factors discriminating agricultural livelihoods in Pontieba (2) To identify main agricultural livelihood system types in the village of Pontieba (3) To characterize agricultural livelihood system types in the village of Pontieba

Conceptual framework

Households farms are characterized by their settings comprising biophysical resources (e.g. land, water and trees), economic resources (e.g. financial and infrastructures) and socio-demographic resources (e.g. labour, capabilities and networks). These settings vary from household-farm to household-farm defining thereby the heterogeneity of a given population in a given region or location. Therefore, this heterogeneity needs to be captured for successfully designing efficient and profitable, adaptive or resilient farming systems as well as effective policy interventions. The Sustainable Livelihood Framework (Chambers and Conway, 1991; Sconnes, 1998) offers the possibility to holistically apprehending the household-farm as it considers all the settings of the household-farm. These settings are grouped into five types of capital: human capital (demography, education of household members and their profession), natural capital (e.g. land holdings and tenure, planted trees), physical capital (e.g. agricultural equipment, transportation means, farming and household tools), financial capital (livestock, off-farm employment remittance) and social capital (e.g. networks and membership to organization/ association). The level of endowment in these capitals will define different livelihood strategies of household-farms. Our study therefore used the sustainable livelihood framework as a guide for collecting a multidimensional dataset used for identifying the agricultural livelihood system types in Pontieba.

Study site

The study was conducted in 2014 in Pontieba located 11° 7' 0" North and 3° 7' 0" W in the Ioba province, South-western Burkina Faso (Figure 1). Pontieba is part of the South-Sudan climatic zone with an average annual rainfall of 900 to 965 mm. The vegetation cover is savannah. The main soil type encountered in the village is leached ferruginous tropical soils, hardened in some locations. The main livelihood activities are subsistence rain-fed agriculture, animal husbandry, trade, handicraft, and traditional mining. Cereals and cotton are the main cultivated crops in the village. Livestock productions concerns mainly ruminants, pigs and poultry. Pontieba is situated at 7 km away from Dano, the main town of the province. The population of the village is 2,215 inhabitants at the last census in 2006 with growth rate of 2.5% and a population density of 71.4 inhabitants/km²  (INSD,  2009).  The   methodological   approach  in identifying agricultural livelihoods types used principal component analysis combined with cluster analysis and expert knowledge (key informants). Therefore, the choice of Pontieba for this study was guided by available background information and expert knowledge of the village.

Household-farm sampling and surveys

The household-farms were randomly sampled from a list of households of the village provided by local leaders. We sampled and surveyed 108 household-farms from 316 in the village, meaning 34% of the village’s household-farms. The total population size of the village was 2,215 inhabitants. The surveys were conducted at the end of the cropping season 2013/2014, one to two months after harvest, in January and February 2014. The data was collected through face to face interview with the head of the household-farm helped by other key members of the household-farm. Field visits were organized with each farmer to measure the area of the farm’s plots with GPS units and record geographic coordinates. The questionnaires were guided by the sustainable livelihood framework covered mainly household characterization (e.g. demography, education and profession), farm lands inventory and land tenure, agricultural and farm tools inventory, crop and livestock production, off-farm income and remittance. The proximity of households from permanent roads was extracted from map reading.

Identification of household-farm types

The identification of the agricultural livelihood system types in Pontieba combined multivariate analysis and expert knowledge. The methodological flowchart is shown in Figure 2. The multivariate analysis consisted in two steps. The first step used Principal Components Analysis (PCA) for identifying the main factors that discriminate household-farms. The collected multidimensional dataset was prepared by selecting main variables per capital in reference to the sustainable livelihood framework (Table 1). The PCA was run with the varimax option and only Principal Components (PC) with Eigen values of at least 1 (>=1) were considered. The second step consisted in K-mean cluster analysis (K-CA). The key variables contributing most to the factors loadings (Loadings>=0.6) from the PCA results were used. The knee method was employed to decide on the optimal number of clusters. ANOVA was used to characterize identified agricultural livelihood system types and the results were confronted to expert knowledge.

Testing the heterogeneity amongst the identified agricultural livelihood systems

The heterogeneity amongst the different agricultural livelihood systems in the Pontieba was tested. Analysis of variance (ANOVA) was used to detect differences amongst the ALS. According to whether  the  equal   variance   across  groups is  assumed  or  not, different post-hoc tests are used to decide on the groups’ heterogeneity. For deciding on the post-hoc test to use, the Levene’s test of variance equality was first run. This test indicates if the null hypothesis of equal variance across the different groups can be rejected. When the p-value of the Levene’s test is lower than the chosen threshold p-value (0.05), the null hypothesis is rejected and the equal variance is not assumed. The least square difference (LSD) test was used when the Levene’s test of equal variance indicates that there is equal variance. When the Levene’s test suggested that the equal variance cannot be assumed; the Games-Howell test was used instead of the LSD. Two main indicators were used for testing the difference amongst ALS; the yield performance of main crops and the land use choice through the land area allocated to each land use type (crop).

Characteristics of farming settings in Pontieba

Table 2 presents the characteristics of farming settings in Pontieba. It shows that households have an average size of 7 members dominantly headed by males, as females’ heads are 7%. Around 60% of household’s heads were illiterate reflecting the low literacy rate in the country and particularly in rural areas. This situation is a potential constraint to the adoption  of  good  practices/innovations susceptible of improving farms’ livelihood. The networking amongst farmers appeared to be relatively low as up to 58% of farmers did not belong to farmer or credit organization (Table 2). Farms possessed less than 1 ha of land per person (0.55 ha/person). Land pressure is likely to increase dramatically in the village in short term giving the population growth rate of 2.5% (INSD, 2009). The cropping system is subsistence based with 55% of household’s farmed land allocated to basic cereals (sorghum, millet and maize). The level of equipment among farmers was very low: The average number of bullocks for land ploughing is only 0.15 per farm. The ratio livestock to land is low (0.45) suggesting a low potential for crop-livestock integration found to be the most promising way for affordably improving sustainable soil nutrient management in the region.

Main factors discriminating agricultural livelihood system types in Pontieba

The PCA results revealed 10 factors with total Eigen values of at least 1 (Table 3). The 10 factors beard 80.87% of initial total variance. Using the rotated component matrix, the factors were named after variables with greater loadings and most correlated to the factors as shown in Table 4. The most discriminating factors of household-farms in Pontieba, with at least 10% of initial total variance, were PC1, PC2 and PC3 which were highly correlated with natural capital (H_HOLDINGS with loadings b=0.88 and H_TLUCP with loadings b=0.92), human capital (H_LABOUR with loadings b=0.90). The PC1 was named Land PC as it was highly associated with land variables. This PC is a measure of land abundance as the loading was positive. The PC2 was named Livestock PC for it was highly associated with livestock, indicating it is a measure of livestock endowment because the loading was positive. As for the PC3, it was named Labour PC for it is  highly  associated  with labour, hinting that the PC is a measure of labour resource abundance given that the loading was positive. These three factors represented 16, 12 and 11% of initial total variance, respectively. Other discriminating factors were PC4 to PC8 which carried less than 10% of initial total variance (6-8%) each. The PC4 was most correlated with human capital (H_LABAGE and H_MEANAGE with loadings b=0.91 for both) with positive loading. This PC reflects the aging of human resource. The PC4 was therefore named Age PC. It carried 8% of initial total variance. The PC5 and PC6 were most correlated with Financial capital (H_GROSSINCCP with loadings b=0.75 for PC5 and H_SHCOTINC with loadings b= -0.90 for PC6). The PC5 carried 8% of initial total variance. It has a  positive loading and indicated the level of income per capita. It can be seen as a measure of the wealth of farm members. The PC5 was then named   Income   PC.   The    PC6     was    highly associated with share of cotton income but has a negative loading. This means the PC reflects low income from cotton. It was  named Cotton PC and carried 7% of initial total variance.

The PC7 and the PC8 were most correlated with    natural    capital    and    financial     capital, respectively. PC7 was most correlated with H_SHMFCRP (with loadings b=0.91). It reflects the orientation toward marketable food crop production since the loadings is positive. It was therefore named marketable food crops PC. It carried 6% of initial total variance. The PC8 was most correlated with H_SHNFINC (with loadings b=0.83). This PC was named Off-farm income PC. It carried 6% of initial total variance. The last two PCs, PC9 and PC10 were both most correlated to human capital. PC9 was most correlated with H_HEDUYR (with loadings b=0.75). It was named education PC and carried only 4% of initial total variance. This PC reflects the level of education. As for PC10, it was most correlated with H_DEPEND (with loadings b=0.85). This last PC reflects the level of dependency within the farm and was therefore named dependency PC. It carried 4% of initial total variance.

Agricultural Livelihood System types in Pontieba

The typology analysis results revealed three agricultural livelihood system types in the village of Pontieba. These agricultural livelihood system types were characterized using a radar diagram showing  the  livelihood  dimension structure built from standardized values of key variables (Figure 3), in addition to the income composition and the livelihood orientation. The Table 5 shows keys variables for which the three Agricultural Livelihood System type were found significantly different at 5% using ANOVA.

Agricultural Livelihood System type I: Poor-income, landless and subsistence-based farms

The agricultural livelihood system type I (Poor-income, landless and subsistence-based farms) represented 40% of the study sample. This agricultural livelihood system type had the lowest asset endowment. It had in average 2.67 ha of total land holdings, meaning 0.47 ha per person. The livelihood orientation was subsistence-based as income from basic cereals (sorghum, millet and maize) formed 32.47% of annual gross income with 60.85% of cultivated land dedicated to these cereals. Only 10.74% of cultivated lands were allocated to cotton which is the main local and regional cash crop. These farms also have low labour and less transportation. They have the lowest annual gross income. Only 46,152 FCFA per person was found (USD 93.35/person). Livelihoods of these farms can be considered to be vulnerable as the annual income per person is below the national poverty line estimated to be 108,454 FCAF (USD 219.36/person/year).

Agricultural Livelihood System type II: Medium-income, high-dependency, cotton-and livestock-oriented

The Agricultural Livelihood System type II (Medium-income, high-dependency, cotton-and livestock-oriented) represented 40% of the study. This farm type had the highest dependency ratio (0.37). The livelihood orientation is market-turned. In effect, around 20% of the cultivated land is allocated to cotton cropping. In addition, the contribution of basic cereals income to the annual gross income (18.33%) is lower than in the case of Agricultural Livelihood System type I. It also had a better endowment in livestock than the farm type I. The number of Tropical Livestock Unit (TLU) per capita was 0.23. The labour endowment and transportation were not significantly different for farm types I and II. The Agricultural Livelihood System type II showed a medium annual income estimated to 101,295 FCFA/person, equivalent  to USD 204.88/person/year. This amount is nearly the poverty line in Burkina Faso, USD 219.36/person/year.

Agricultural Livelihood System type III: Better-off income, land-and labour-rich, cotton-and livestock-oriented

The third farm type, Agricultural Livelihood System type III (Better-off income, land-and labour-rich, cotton-and livestock-oriented) represented the best endowed and wealthiest farm type out of the three. It had the highest labour endowment (7 workers), the highest land holdings (4.25 ha) and the highest number of transportation (4). This last setting might play an important role in facilitating the farmer access to market, to other farmers and villages, and thereby increases his exposure to innovations and opportunities. This farm type is also market- turned like in the case of Agricultural Livelihood System type II. The land area dedicated to cotton cropping was around 23% of cultivated. As for the livestock endowment, it was 0.35 TLU per person. The values for cotton and livestock as well as the contribution  of   cereal   income  to  annual  gross income (19.91%) were higher than in the case of Agricultural Livelihood System type I, but were not significantly different from Agricultural Livelihood System type II. The Agricultural Livelihood System type III was the only one farm type with annual income above the poverty line in Burkina Faso. This annual income was 144,428 FCFA/person (USD 292.12/person).

Agricultural livelihood systems heterogeneity in Pontieba

The results of the ANOVA test are summarized in Table 6. The ANOVA test showed that there were significant differences amongst Agricultural Livelihood Systems for cotton and Maize land use. Indeed, the cotton land area of ALS 3 was larger than the two other ALS while for maize; the ALS 3 had the largest farmed area compared to ALS 1. No significant difference was found amongst ALS for sorghum land use. The analysis of the yield performance also revealed heterogeneity across ALS. The ALS 1 had the highest cotton yield. Though the p-value of the ANOVA test indicated a significant difference amongst ALS1, the Levene’s test rejected the null hypothesis of equal variance.

The post-hoc test under this condition of non-equal variance across ALS revealed no significant difference amongst ALS. The different ALS was found significantly different for the sorghum which is to the most cultivated cereal in the study region. The ALS 3 showed the highest yield. These results demonstrated that clusters identified in Pontieba are functional clusters regarding land use and crop yield. The performance and land choice analyses of the three Agricultural livelihood systems will be further analysed in subsequent work.

By combining principal component analysis and cluster analysis, the present study succeeded in clearly identifying main discriminating factors among smallholder farms in the village of Pontieba, Ioba province in south-western Burkina Faso. Variables discriminating agricultural livelihoods in Pontieba are labor, labor age, education and dependency (human asset); land holdings and livestock (natural asset); annual gross income and non-fam income (financial asset), and production orientation (cotton and marketable food crops production). The study identified three agricultural livelihood system types in the village of Pontieba: Poor-income, landless and subsistence-based farms, medium-income, high-dependency, cotton-and livestock-oriented farms, and better-off income, land-and labour-rich, cotton-and livestock-oriented farms. This farm types showed different livelihood settings and production performance. Therefore, the failure to consider farm heterogeneity in a location hampers the effectiveness of interventions aiming at improving rural livelihood. Accounting for farms’ heterogeneity is keys to farming design studies, in particular for integrated farming systems modeling and analysis seeking to propose innovative solutions for adaptive, sustainable and resilient agricultural livelihoods in a changing biophysical and socio-economic context. The results of this study can be used for policy intervention.

Further studies should test the framework of the current study by performing behavioral analysis in terms of land use decision making, yield function of main crops and livestock for identified different agricultural livelihoods systems. 

The study was conducted in one village with a population of 306 households. There a need to conduct a similar study at a larger geographical area and with a larger population size.

The authors have  no  relevant  financial  or  non-financial interests to disclose

The study was undertaken as part of, and funded by, the CGIAR Research Program on Dryland Systems (CRP-DS) led by the International Center for Agricultural Research in the Dry Areas (ICARDA). It was continued to be supported by the CGIAR Research Program on Grain Legumes and Dry Cereals (CRP-GLDC). The opinions expressed here belong to the authors, and do not necessarily reflect those of CRP-DS, CRP-GLDC, ICARDA, or CGIAR.