Keywords

1 Introduction

1.1 Background and Motivation

Southern Africa has been identified as a hotspot for global change processes and biodiversity, whereby agricultural expansion is regarded as a key driving force for the declining species diversity (IPBES 2018). The projected doubling of the African human population by 2050 (as compared to 2010) and the climate change-induced increased frequency of extreme droughts underline the urgency of science-informed assessments in support of identifying sustainable land management options (IPCC 2019; Rötter et al. 2021). About 70% of the population of southern Africa relies on agriculture. Most of them are smallholders, of which about 94% depend on rainfed agriculture. Around 16% of the rural population has been characterized as “food insecure” during the last 5 years (Sikora et al. 2020).

In recent years, there has been growing attention and support for innovation for supporting sustainable agricultural production (e.g., Herrero et al. 2020) such as agro-ecology, sustainable intensification (SI), and climate-smart agriculture (Cassman and Grassini 2020; FAO 2010; Kuyah et al. 2021; Wilkus et al. 2021). Yet, there is some debate about which approaches should be applied in which contexts and to whose benefits. Site- and season-specific, knowledge-intensive agricultural management practices combined with advanced breeding tools hold promise to increase resource use efficiencies and crop performance considerably (e.g., Hoffmann et al. 2018; Hammer et al. 2020) and can be supported by digital technologies (Herrero et al. 2020; Von Braun et al. 2021).

There exist a considerable number of socioeconomic constraints that need to be overcome to create a fertile ground so that technological innovations can unfold (Gatzweiler and von Braun 2016). Economic weaknesses have been sharpened since the start of the COVID-19 pandemic with considerable negative consequences for food security (Savary et al. 2020). Limited access to land, water, other resources and markets for smallholder farmers has negatively affected rural livelihoods. Moreover, recent shifts of a considerable proportion of agricultural production and land use away from human food-related activities toward animal feed, timber, and biofuels in some regions have presented trade-offs between food security and energy needs. In other regions, land use change from agriculture toward mining, nature conservation, or settlements has reduced the agricultural production area.

Southern African savanna landscapes are composed of arable land, rangelands, and orchards/homegardens (Rötter et al. 2021). In this chapter, we focus on the potential of technological improvements on crop productivity and rural livelihoods of small-scale farmers who largely depend on the ecosystem services (ES) these three major land use types provide. Small-scale farmers in the region are highly diverse in terms of resource endowments such as land and water. The generally huge yield gaps (with yield levels at 20% of the potential), food insecurity, and shrinking land holdings call for radical changes in land use policies and management to avoid societal unrest growing in the future. In national plans on sustainable development, sustainable intensification (Cassman and Grassini 2020) of these systems, not surprisingly, has the highest policy priority (Sikora et al. 2020). It is seen as an important means to provide incentives to the younger farmer generation, boost agricultural development, and to set land aside for nature conservation.

A broad range of management interventions has been suggested for promoting sustainable intensification (e.g., Kuyah et al. 2021; Vanlauwe and Dobermann 2020; Wilkus et al. 2021), including cereal intercropping with legumes, conservation agriculture, agroforestry, site-specific fertilizer application and irrigation. Most experimental studies on testing such interventions have just looked at impacts on dry matter production and yield, but a few also looked at other ecosystem functions such as carbon sequestration and water and nutrient use efficiency. Yet, to date, no study has looked in an integrated manner at the complexity of smallholder systems with a broad range of interacting ES at the landscape level. The SALLnet project has that ambition (Rötter et al. 2021), and here we present a few of the results of such integrated analyses across different scale levels, from field via farm to landscape level.

1.2 Problem Statement and Objectives of the Chapter

A key question for many smallholder-dominated agricultural landscapes in southern Africa is: “how can the multiple Ecosystem Services (ES) be enhanced through sustainable land management interventions and enabling policies?” (Rötter et al. 2021).

Limpopo province in the northern-most corner of South Africa combines most of the global change threats, and is also featuring several of the typical land use changes that have been observed across southern Africa over the last few decades (Chap. 20). High population growth, severe land degradation, and high climate variability in conjunction with low agricultural productivity and poverty have already led to a decline of essential ES in Limpopo province such as provision of food and feed, nutrient cycling, and habitat quality (Hoffmann et al. 2020; Pfeiffer et al. 2019).

Against this background, the present chapter aims at investigating how agro-technology improvements could support SI in small-scale farming systems in Limpopo province, South Africa.To do this, we suggest an integrated crop model APSIM (for applications in Africa, see, e.g., Whitbread et al. 2010) and downstream socioeconomic modeling by means of agent-based modeling. These models are calibrated to Limpopo province by using data from small-scale farmer surveys, on-farm agronomic sampling, and long-term crop experiments. To demonstrate the respective impact analysis of agro-technology improvements within our modeling framework, we use the example of improved soil nutrient and irrigation practices in combination. We also discuss further agricultural technology improvements and innovations that could be likewise assessed going forward.

The remainder of the chapter is structured as follows: To provide a basis for our analysis, an overview of the small-scale farming sector in Limpopo province is presented in Sect. 23.2 by using the results of a large-scale survey conducted within the course of the SALLnet project. Section 23.3 then analyzes the current yield gaps and resource use efficiencies in small-scale farming systems in Limpopo province, on which basis improvements shall be worked out in the following. Accordingly, Sect. 23.4 first provides an overview of the methods and tools to be used to analyze the impacts of different technologies and innovations on small-scale farming systems. Subsequently, these are calibrated to the study region in Sect. 23.5. Finally, Sect. 23.6 provides a summary of the main findings and draws some implications.

2 Farm Household Characteristics: Small-Scale Subsistence Versus Emerging Farmers

Limpopo is one of the least developed provinces in South Africa and currently experiencing both strong population growth and a high poverty rate. A large share of the population (89%) is living in rural areas, and farming is the main occupation (Gyekye and Akinboade 2003; LDARD 2012). In order to understand the structure of the smallholder farming sector in Limpopo province of South Africa, five study areas were selected from Limpopo based on differences in climatic aridity, demography, and socioeconomic factors. The selected sites are located in rural areas of the Mopani district: Mafarana, Gavaza, Ga-Selwana, Makushane, and Ndengeza. A comprehensive small-scale farmer survey was conducted between April and July 2019 after pretesting in selected villages; the interviews were conducted in person with farm household heads or individuals who are responsible for farm management. Permission to access farmers was obtained from tribal authorities of each village. The purpose of the survey was to collect information on socioeconomic, demographic, farm, and household characteristics as well as information on resource endowment and agricultural activities during the 2018/2019 crop season. Using a purposive random sampling procedure, data were collected from 215 farm households across the five villages in Limpopo, of which three households had to be excluded afterwards due to incomplete information. Therefore, the final data set for the following analysis covered 212 households.

Table 23.1 presents a summary of selected descriptive statistics regarding crucial farmer and farm characteristics, including farm performance, resource management, socioeconomics, as well as external incentives (e.g., agricultural extension services, access to credits and markets). Accordingly, we found that the average farm household in the survey sample has a household head who is on average 66 years old. The share of female-headed households was the same as the national general household survey in 2019 with 48.8% (Statistics South Africa 2019). The average household in the survey owned 4.4 ha land, of which 70% is left fallow during the winter (dry season). We found considerable variation in farm size, especially regarding their cultivated area. In terms of production systems, the small-scale farms in the sample were mainly characterized by mixed crop-livestock production. Our survey showed that maize (Zea mays L.) is most important to ensure household food security and cultivated by nearly all farms. The secondary major crops are legumes such as peanuts (Arachis hypogaea), Bambara nuts (Vigna subterranea L.), and cowpea (Vigna unguiculata) which are produced by 59% of farms. Horticultural crops such as fruits (e.g., mango, banana) are grown by 32%, and vegetables (e.g., tomato, onion, cabbage, paprika) by 15% of the surveyed farms. Maize and legumes were mainly grown for household consumption but vegetables contributed to both household consumption and income generation. With regard to livestock farming, cattle provided the main source of livestock income while farms also kept goats, pigs as well as chickens. On average, 41% of agricultural income stemmed from crop sales and 25% from livestock sales. Moreover, the degree of commercialization for crops was 39% and for livestock was 6%, indicating the proportion of selling value of the total value of the production, based on market prices in 2019.

Table 23.1 Descriptive statistics of the small-scale farmer survey conducted in Limpopo in 2019
Full size table

Agricultural products were mainly traded on informal on-farm markets (58%). Only 17% of the local farmers had access to formal off-farm markets. Social grants including old age and child support grants played an important role on farm household incomes for most smallholders. According to Statistics South Africa (2019), around 59% of the households received grants as their main sources of income in Limpopo. Direct support from the government as well as extension services mainly occurred in the form of input supplies, mechanization, livestock health services, and training. The number of visits of extension services on average was 1.32 times in a year. The field preparation was usually carried out by a rented tractor or donkey. Nevertheless, among these farmers, only 6% had their own private tractor. In this respect, merely 10% of the respondents had access to formal credits but 37% of the farmers invested in the last 5 years mainly in equipment for irrigation, fences, and machinery. Besides household members working on their own farms, the permanent and seasonal employed labor worked amounted on average to 48.5 and 17.33 man-days per year (1 man-day = 8 h/person). Regarding irrigation, the most common source of water was tap water (41%) which was usually only available in the home garden next to their residential building. 34% of the sample was purely rain-dependent, while on average 9% and 16% of farmers had access to public water sources and private boreholes. Hence, 49% of the sample used primitive irrigation methods (e.g., buckets, furrows).

According to the collected information described above, the smallholder farmers in the sample were found to be highly heterogeneous in terms of farm and farmer characteristics, resource management as well as external incentives such as agricultural extension services, access to credits, and markets. Moreover, the heterogeneous groups of smallholder farmers were reliant on different forms of government interventions and agricultural policies, depending on the objective and characteristics of each group.

3 Yield Gaps and Current Resource Use Efficiencies in Small-Scale Farming Systems

In Sect. 23.3.1, we give a brief account of the different yield gaps (see Kassie et al. 2014) as well as of current water and nitrogen efficiencies for maize cultivated by smallholders. Based on literature and simulation results, we show the scope for closing or narrowing down the yield gap between actual farmer’s yields and potential yields that could be attained under irrigated or rainfed conditions with best management. Next, we look at the efficiency gains that might be obtained by distinct management interventions—restricting ourselves to water and nutrient management, whereby in the latter with focus on the macro-nutrients nitrogen (N) and phosphorus (P). Furthermore, Sect. 23.3.2 presents the results of efficiency analysis of current maize-based small-scale farming systems in five villages in Mopani district, Limpopo province. As a consequence of this, we discuss the potentials of a number of alternative management interventions in enhancing farm income and other ecosystem services in Sect. 23.3.3.

3.1 Current Resource Use Efficiencies for Different Small-Scale Farming Systems in the Region

Regarding crop production in South Africa, maize is the major staple crop and mostly grown by smallholders under rainfed conditions. The yield of maize in the study regions within Limpopo province is fairly low—for small-scale farmers ranging between 1 t ha−1 and 2 t ha−1. This is mainly due to manual farming techniques together with low input provision such as no or little fertilization, lack of quality seeds, and no irrigation (FAO 2010). The observed increase in water scarcity and land degradation, and particularly poor soil fertility through nutrient mining in most smallholder farming systems poses a serious threat to crop production in southern Africa (Vlek et al. 2020). Therefore, a logical agricultural intervention measure is to produce more grain yield per volume of water used (“more crop per drop”) and replenish the nutrient-depleted soils with mineral and organic fertilizer. Water use efficiency (WUE) needs to be increased through appropriate water, nutrient, and crop management interventions measures to sustainably raise agricultural productivity. In a water management context, the term WUE refers to crop production per unit of water used, with units such as kg grain ha−1 mm−1 or kg m−3 (Sadras et al. 2012).

The simulated WUE under different treatments was derived from simulated maize results for 35 years (1985–2020) at the experimental Syferkuil site within the Limpopo province (Table 23.2). The WUE varied between 0.1 kg m−3 and 1.39 kg m−3 among treatments and years. Those different, predefined fertilizer treatments in maize production could also be mirrored by the farmer behavior observed within the small-scale farmer survey as presented in the second section. Accordingly, the two lower levels of 0N:0P and 10N:5P were approximately applied by the vast majority of the small-scale farmers within the survey, who were primarily producing for self-subsistence purposes. 40N:30P represents the level used by those small-scale farmers, who already emerged to a more market-oriented role. The highest intensity treatment of 120N:60P again approximates the level of intensity commonly used by commercial farmers in maize production in this region.

Table 23.2 Average water use efficiency under different treatments (in kg grain DM yield/m3 water)
Full size table

The results showed that combined deficit irrigation and the 120N: 60P kg ha−1 fertilizer application gave the highest WUE value of 1.15 kg m−3 (average over 35 years). This finding is in agreement with the results of Kurwakumire et al. (2014) who found that WUE under rainfed conditions ranged from 0.038 kg m−3 to 0.113 kg m−3 (control), while it improved from 0.3 kg m−3 to 0.8 kg m−3 for crops receiving 120N:40P:60K fertilization.

NUE is here defined as the incremental maize yield per applied nutrient. The average NUE ranged from 23.67 to 45.51 kg grain yield/kg nutrient (Table 23.3). The highest NUE was obtained in the case of 40N: 30P (kg ha−1) fertilizer application under rainfed cultivation. The NUE values presented in Table 23.3 are within the range of values for a similar study reported by Ngome et al. (2013) for smallholder farmers on three dominant soil types of Western Kenya. The values are also largely in agreement with findings from a national soil fertility program in Kenya tailored to the needs of smallholder farmers in agro-ecological zones with medium to high agricultural potential (Smaling et al. 1992). In an early review, van Duivenboden (1992) found for West Africa similar NUE values as presented in Table 23.3.

Table 23.3 Average nutrient use efficiency (NUE) under rainfed cultivation (in kg grain DM yield/kg nutrient)
Full size table
Fig. 23.1
A box plot of yield versus 12 treatments. The highest interquartile range is in R N 120. The highest median is with D N 120 and F N 120.

Box plots of simulated yield among the treatments at the experimental Syferkuil location. Yields are simulated from 1985–2020 (R, D, and F stand for rainfed, deficit, and full irrigation, respectively; N0, N10, N40, and N120 stand for the following N:P combinations: 0N: 0P; 10N: 5P; 40N: 30P; and 120N: 60P, kg ha−1)

Full size image

The simulated maize yields presented in Fig. 23.1 show that maize yield increased with increases in the rate of fertilizer and irrigation applications at the experimental Syferkuil site. However, the increase in fertilizer rates had a stronger effect on maize yield improvement than the irrigation treatment. Increasing the N:P fertilization rate alone increased average annual maize yield of RN40 to 3.12 t ha−1 (i.e., 144%) and FN120 to 4.75 t ha−1 (i.e., 271%) compared to that of RN0 which was 1.3 t ha−1, respectively. Considering irrigation alone, treatments of DN0 and FN0 increased maize yields only by 10.5% and 21.9%, respectively, compared to RN0. The combination of fertilization and irrigation application gave the highest average annual yields for the DN120 and FN120 treatments, achieving 5.4 t ha−1 and 5.7 t ha−1, respectively. Note, that in these treatments it is further assumed that other nutrients are always in ample supply and the crop is well protected from pests and diseases.

Fig. 23.2
2 parts. A, a line graph of grain yield versus years plots 2 increasing lines for commercial and small scale maize yields and 4 dots for Ndengeza, Ga selwana, Mafarana, and Gabaza in 2020. B, a bar graph of grain yield versus fertilizer application in different combinations. It marks the yield gap.

Maize yield gap between commercial and small-scale farmers—averaged for South Africa—colored dots show farmers yields in four villages of Mopani district in 2020 (a); Simulated average maize yields under different treatments in the Syferkuil, SA for period 1999–2020 (b) data source maize yield series: https://www.sagis.org.za/historic%20hectares%20and%20production%20info.html

Full size image

Figure 23.2a illustrates the maize yield gap over the last 30 years between yields actually obtained by large-scale commercial farmers and those of small-scale farmers in South Africa. It shows that although maize yields fluctuated over time for both groups of farmers, the yield trend for commercial farmers significantly increased over the last 30 years. However, the maize yields for small-scale farmers did not exceed the level of 1.5 t ha−1. Based on the simulated maize yields at Syferkuil (Fig. 23.2b), yields for small-scale farmers can be increased considerably by applying different treatments combining fertilizer and irrigation. Yet, the highest yield increases that could be obtained come at very poor water and nutrient use efficiencies—and thus high losses to the environment. The different yield gaps illustrated in Fig. 23.2a, b indicate the most effective interventions for raising yields.

3.2 Results of Efficiency Analysis of Current Maize-Based Small-Scale Farming Systems

Maize dominates the South African food system, being both the vital dietary staple crop and feed grain. Therefore, it is the most prevalent agricultural crop for small-scale farmers in rural areas (Obi and Ayodeji 2020). As the majority of small-scale farmers cultivate maize mainly for subsistence purposes, the levels of production and supply of maize from small-scale farming are important indications of food security. Over the past century, South African maize production experienced some significant changes (Greyling and Pardey 2019). Small-scale farmers played an important role in producing maize in the country in previous years.

In addition to environmental stress, agricultural production in general and maize cultivation in particular in southern Africa are both confronted with several macro- and micro-structural constraints (Mpandeli and Maponya 2014). Some of these constraints are inefficient policies and extension system support, limited access to agricultural credits which results in reduced investment in the agricultural sector, inadequate infrastructural facilities such as transportation and communication, along with the lack of proper economic incentives (e.g., Baloyi et al. 2012). These constraints result in yield reduction and harvest failures in recent years (Hove and Kambanje 2019), which exacerbates food insecurity and poverty, especially among smallholder farmers who often practice subsistence farming.

Despite government support and various strategies implemented to improve the productivity of the agricultural sector in South Africa in recent years (e.g., FAO 2017), small-scale farmers still perform considerably below their potential production capacity (Baloyi et al. 2012). Improving agricultural productivity and, thus, crop yields can generally be achieved by more efficient use of available farm resource endowments and by adopting new technologies (e.g., Ali et al. 2019).

For agricultural productivity growth, the concept of technical efficiency (TE) plays a major role and is widely discussed in the literature. It provides information on the performance of the farmers and their potential to improve productivity and efficiency by utilizing existing farm resources and technologies. In addition, technical efficiency analysis allows identifying the main factors affecting the efficiency level of farms. According to various studies in the literature, improvement in the efficiency levels of agricultural production, which is the main component of total factor productivity (TFP) growth, can be regarded as key in alleviating food insecurity in developing economies (Ogundari 2014).

To date, to our best knowledge, less attention has been paid to the analysis of efficiencies of small-scale maize farmers in South Africa and in particular the Limpopo province. To address this research gap, the aim of the present analysis is to evaluate the TE of smallholder maize farmers and the potential factors that lead to the deviations from the common production frontier. Accordingly, a parametric efficiency analysis [a single-step Stochastic Frontier Model (SFA)] considering a dual heteroskedastic production frontier is applied to 190 households cross-sectional survey data in 2019 from five selected study areas in the Limpopo province in South Africa. The model is designed to estimate TE of small-scale maize farmers by adopting a holistic approach of analysis that considers production inputs, perceived risks, and socioeconomic/socio-demographic characteristics to examine their joint influences on productivity and efficiency of the maize production among smallholders in the Limpopo province of South Africa. Moreover, by identifying the determinants of TE, the results of this analysis will offer policy implications to improve maize production, as well as to tackle food insecurity and poverty.

Following the model specification tests,Footnote 1 a single-step SF model with a Cobb–Douglas production function is implemented to estimate the stochastic frontier production function and the inefficiency function simultaneously using maximum likelihood estimation. Moreover, we considered heteroscedasticity in both error components (ui and vi), following the twofold heteroskedastic model of Hadri (1999). This implies that both variances of the inefficiency and idiosyncratic error terms (\( {\sigma}_{u_i}^2 \) and \( {\sigma}_{v_i}^2 \)) are depending on the explanatory variables with the half-normal distribution of inefficiency term \( {\sigma}_{u_i}^2 \) and independently normal distribution of \( {\sigma}_{v_i}^2 \). For a more detailed description of the methodology, see Yazdan Bakhsh et al. (2022).

Table 23.4 Maximum likelihood estimates of the stochastic frontier analysis for maize production of small-scale farmers in Limpopo, including the stochastic frontier function, the technical inefficiency function, and the production variability function
Full size table

Table 23.4 represents the summary of the results of the analysis. The dependent variable is maize output in kilograms. Considered production inputs include area under maize cultivation in ha, quantity of seeds and fertilizers in kg, pesticide in number of applications per year, labor used for maize cultivation in man-days unit, preparation costs (machinery and animal capital) in currency unit (ZAR), as well as the dummy of irrigating the maize fields. Furthermore, for the inefficiency component, the explanatory variables include: age, gender, and education level of household head, off-farm income, social grants, extension service support, credit access, member of the agricultural organizations, access to off-farm markets, having agricultural training, owning cattle, and total cultivated land. There is a wide variation in output and input use between farmers among selected villages. In this regard, the heterogeneity in production technologies is addressed by considering the geographical location of each village in the selected province, with the assumption that technology is homogeneous within the same geographical location but different across them. Therefore, dummies of location were included in the production frontier as a control variable. Since drought and pest were the two key uncertainties that the selected farmers perceived as the main reasons of the maize harvest failure in this year compared to previous years, we considered the dummies of farmers’ perceived risks (failure in crop production due to drought and pest) in both production frontier and idiosyncratic error, investigating their influences on both level and variability of the maize production.

In the following, the results of the three functional components of the analysis, as can be seen in Table 23.4, are briefly described:

Production Function

The results of the stochastic frontier model indicate the main determinants of the productivity level of the respective small-scale farmers in maize production. Since the output and input variables in the Cobb–Douglas production frontier model are in logarithmic form, the estimated coefficients can be directly interpreted as the partial production elasticities. The positive signs of the coefficients for almost all included variables follow the monotonicity property of the production function, whereas labor shows a negative sign, which is however not statistically significant. Irrigation exhibits the greatest coefficient among the all production factors (0.62), indicating that smallholder farmers with access to irrigation were found to have substantially higher maize yields compared to those cultivating under rainfed conditions. Moreover, pesticide application (0.52), maize area (0.34), fertilizer (0.19), and seed (0.17) are also important technological factors. Furthermore, farmers’ perceived risks with regard to drought and pest have significantly negative effects on maize productivity.

Technical Inefficiency

The results of the technical inefficiency model indicate the main drivers of inefficiency of small-scale maize farmers. Since the dependent variable in the technical inefficiency part of the model is the technical inefficiency, a negative sign of the coefficients indicates the negative effect on the technical inefficiency, or in other words, a positive effect on technical efficiency (Kumbhakar et al. 2015). Accordingly, the results indicate that gender (male farmer), age of less than 74 years, off-farm income, access to credits, social grants, member of agricultural organizations, extension services supports, access to markets, owning cattle, and agricultural training have positive effects on TE of the small-scale maize farmers. Education shows a negative but not significant effect on efficiency. On the contrary, the overall cultivated acreage of a farmer has a significantly negative effect on TE of smallholder maize farmers.

The estimated coefficients of the technical inefficiency cannot be directly interpreted from the model, due to the non-linear relationship between E(ui) and each of the explanatory variables. Therefore, the marginal effects are calculated to investigate the magnitude of the exogenous factors on inefficiency (Kumbhakar et al. 2015). Based on the results, having access to credits, being a member of an agricultural organization, and receiving training in agricultural practices significantly reduce production inefficiency within the investigated sample of small-scale farmers. The effect translates into achieving higher and more stable output. Following training in agricultural practices with 69%, having access to credits has a statistically significant effect on technical inefficiency, so that the level of technical inefficiency can be reduced, on average, by 66% with 1% increase in accessibility of credits. This is followed by the factors organization members (56%), social grants (35%) and market access (21%).

Production Variability

The results of the production variability model indicate the determinants of the variability in maize production. Here, the perceived risks for pests and droughts, as well as an interaction term out of both variables were included in the analysis. The results indicate that drought and pests can be seen as the main drivers for production risk in small-scale farming in Limpopo, especially in regard to maize.

Following the estimation of production frontier and technical inefficiency model and identifying the main determinants of technical inefficiency, we additionally investigated the technical efficiency scores of the smallholder maize farmers at both farm and regional level (Table 23.5). The results reveal that the mean estimated technical efficiency is 0.67, indicating that, on average, the smallholder maize farms produce 67% of potential output at given input levels and technology. The mean scores are relatively similar across the selected regions. This result suggests that there is opportunity to improve maize production by using current inputs and technology.

Table 23.5 Estimated technical efficiency (TE) in the selected regions
Full size table

Individual efficiency levels range from 0.07 to 0.99 and vary due to farm-specific characteristics (e.g., financial and agricultural supports, management practices, production specialization, etc.). According to the comparison of the main characteristics of the farmers within the 10% upper and lower efficiency levels, efficiency is strongly influenced by agricultural training, membership in agricultural organization, and number of visits by extension officers. In terms of farm characteristics, the farmers within the 10% upper levels of efficiency have access to more land and cultivated area than the farmers with the 10% lowest efficiency level. Besides maize as their staple crop, the farmers with the 10% highest level of efficiency focus more on vegetables (e.g., chili, green pepper, okra, cabbage, spinach, tomato, and onion) and legumes (e.g., green bean, peanut, cowpeas, and Bambara nuts) and more cattle than the lowest levels of efficiency. Moreover, these farmers apply more endowments such as fertilizer, pesticide, and irrigation on their field.

3.3 Potential of Different Types of Technological Improvements

In spite of the fact that climate risks have been shown to be among the main sources of variability and uncertainty in crop production, quantified information is still scarce for sub-Saharan Africa and far from perfect. This concerns information on climate-induced yield variability (Rötter and Van Keulen 1997), climatic yield potential and its variability for a given location, and the magnitude of different types of yield gaps (e.g., Kassie et al. 2014). There is a need for localized yield gap analysis that can indicate the scope for yield increases through well-targeted management interventions and/or introduction of new technologies such as new breeds, micro-dosing of fertilizer, or gravitational drip irrigation.

Yield gap analysis in the first place involves quantifying the differences between simulated potential yield and actual farmers’ yield level and identifying those factors most responsible for the yield differences (Van Ittersum et al. 2013). Factors that are most common in causing yield gaps include water stress, nutrient deficiency, pests, and diseases through sub-optimal agronomic management (Hoffmann et al. 2018; Kassie et al. 2014). The benchmarks for calculating yield gaps are yields under optimum management which are potential yield attainable under irrigation (Yp) or yield attainable under water-limited or rainfed conditions (Yw). These benchmarks are most commonly quantified by using crop simulation models (Van Ittersum et al. 2013; Hoffmann et al. 2018). The potential yield is limited by climate conditions (temperature, solar radiation, CO2 concentration) and plant genetic characteristics. The water-limited yield, which is also known as water-limited yield potential, is additionally influenced by precipitation and soil water characteristics.

Here we illustrate the current low agricultural productivity with a few examples from Limpopo, one of the least developed provinces in South Africa compounded by a high population growth rate and a relatively high poverty level of the rural population. About 89% of Limpopo’s population (5.8 million) lives in rural areas with farming as their main occupation. Most smallholder farmers have limited resource endowments and are still producing for subsistence purposes (see Sect. 23.2). These farmers are particularly vulnerable to climate variability and other environmental stresses.

While there is some debate surrounding the meaning of Sustainable Intensification (SI) (Garnett et al. 2013), few would refute the necessity of yield increases for SI under typical smallholder conditions. Regardless of the socioeconomic scenario, few farmers would adopt technological innovations without thorough explanations or demonstration, and most will expect immediate benefits (Vanlauwe and Dobermann 2020). The spatial and time-wise variability, which is characteristic of the systems in question, mean that uniform “best practice” is usually not a solution (Rurinda et al. 2020). Crop simulation studies can be used to explore the potential benefits of different SI options and assess the ecosystem services of different management interventions (e.g., Hoffmann et al. 2020). Reviews, such as the recent one performed by Kuyah et al. (2021) can help to select promising options from technology packages that have shown to be successful in different smallholder environments of Africa.

Two undisputed problems for many smallholders in SSA are: (1) the general degradation of the soil (Vanlauwe et al. 2014) and (2) climate variability and change (Rötter and Van Keulen 1997; IPCC 2019). The latter include increased frequency of extreme weather events, such as heat-waves and prolonged droughts (IPCC 2019) and within-season dry spells (Rurinda et al. 2020). Such conditions are particularly treacherous for low input systems, as they exacerbate pest and disease problems (see, e.g., Rötter et al. 2018), such as the fall armyworm in crops like maize, which recently has been common in the study region. Poor soil and residue management also leads to low soil organic carbon (SOC) levels and nutrient mining, which is often aggravated in drier and more marginal soil regions (Lal 2004). Vanlauwe and Dobermann (2020) propose for sub-Saharan Africa that SI cannot happen without the use of fertilizer, and that its use needs to be combined with other agronomic improvements. In addition, the low-income nature of smallholder systems means that many cannot afford to waste fertilizer in seasons with unfavorable weather.

Using crop model APSIM (for applications in Africa, see, e.g., Whitbread et al. 2010), we set-up simulation experiments for different management interventions meant to show agronomic pathways of how to help subsistence farmers to move onto the next step of the “ladder” of farming (see. Table 23.6). Two sites were chosen which are contrasting in terms of soil parameters as well as climate conditions—Gabaza is the more fertile and more humid site compared to the semi-arid Selwane site. Planting density and cultivar choice was as recommended by extensions officers for the selected villages in Mopani district. Simulation results for the main rainy season (with weather data from 1998 to 2019, i.e. 20 seasons: October–May) for each site are presented in Fig. 23.3. The treatments (thermo mechanical treatment (TMT)), as detailed in Table 23.6, were based on the current management (status quo) as observed throughout ground truthing field trips to the villages studied in 2019. The so-called status quo technology consists of no input in terms of fertilizer or irrigation. Further TMTs explore urea N application at sowing under rainfed conditions (25 and 50 kg ha−1), as well as the implementation of deficit irrigation (based on rainwater harvesting).

Table 23.6 Technology packages (or treatments) defined for the simulation runs
Full size table
Fig. 23.3
A grouped box plot of yield versus simulated maize yields for Gabaza and Selwane. The longest interquartile range is with T M T 50 N.

Simulated maize yields for Gabaza and Selwane over 20 seasons [1998–2019; source: NASA POWER Data (https://power.larc.nasa.gov/)]. For scenario details, see Table 7. Maize cultivars are represented by shades of green and mean values by solid black points

Full size image

Results of this simulation experiments showed that overall, maize yields were, as expected, estimated to be distinctly higher at the more humid and fertile site, Gabaza, where, on average, maize yields were reaching well above 3 t ha−1 with the best available combination of technologies, whereas at Selwana the maize yield ceiling has not been exceeding 2 t ha−1. The deficit irrigation treatment reduced yield variability. Cultivar trends were similar in both sites, with the lowest yields from landrace Katumani, followed by SC601, and the hybrid H 614.

4 Tools Required to Model Impacts of Different Technological Innovations on Sustainable Agricultural Land-Use (at Multiple Scales)

In this section, the challenges of developing a framework for analyzing sustainable land management scenarios are firstly described in Sect. 23.4.1. Based on this, the used modeling techniques of agro-ecosystems (crop) modeling as well as agent-based modeling are described in general in Sects. 23.4.2 and 23.4.3.

4.1 The Challenge of Developing a Framework for Analyzing Sustainable Land Management Scenarios

Establishing a framework that helps analyze sustainable land use and management options at multiple scales is multifaceted and complex as discussed by Rötter et al. (2021). In regions with a high climate variability and diverse agro-ecological conditions, site- and context-specific interventions are increasingly needed, but these must consider other connected land uses and the resources they share. A suitable strategy demands a multidisciplinary approach. This is particularly the case for modeling multiple ecosystem functions and services, results of which require careful interpretation by a multidisciplinary team of scientists together with local experts (see Rötter et al. 2016).

A typical case that can exemplify a complex land management scenario framework for dry savannas in southern Africa is the crop-livestock system as found in the villages studied in the Mopani district of Limpopo province. From an arable crop production perspective, once the main maize crop has been harvested, leaving the maize biomass and stubble in the field would be one way to ensure the soil is not left bare throughout the dry season and soil integrity is maintained through biomass decomposition. Depending on the water balance of the soil and access to seed, there may even be an opportunity to implement the cultivation of a legume cover crop during this part of the season to add fresh biomass and fix nitrogen. It is only when looking into the livestock aspect of the broader landscape system that key limitations become clear. Post maize harvest, livestock herders graze their cattle on the remaining maize biomass for dry-season feed, removing any soil cover and chance of crop residue nutrient cycling and soil organic carbon development. This is one of the use cases examined by the SALLnet project (see, e.g., Rötter et al. 2021) where a well-designed, combined use of various ecosystem modeling approaches can play an important role in the development of sustainable land use management interventions. In our example of the villages in Mopani district, the crop model APSIM has been combined with the rangeland vegetation model A Dynamic Global Vegetation Model (aDGVM2) (e.g., Hoffmann et al. 2018 and Pfeiffer et al. 2019, respectively).

While developing a usable land management framework is certainly a challenge, there are a number of suitable modeling components, existing data and tools upon which it can be based. With the digitalization and open-access nature of data, such new datasets as digital soil map (e.g., iSDA) and weather information (e.g., NASA Power, CHIRPS) are becoming increasingly accessible and reliable, as are the techniques and technologies to utilize these. For example, via the Internet of Things (IoT) and growth of smartphone ownership the world over. But, it needs pulling together, be managed by multidisciplinary teams and networks of scientists and other key stakeholders if it is going to be used to help implement sustainable land management options for current and future landscapes. This type of novel, multi-scale, and integrated approach promoted by the SALLnet project for southern African savanna landscapes is further described by Rötter et al. (2021).

4.2 Agro-Ecosystems Modeling

Agro-ecosystems models or crop simulation models (CSMs) are powerful tools to assess the risk of producing a given crop in a particular soil–climate regime and to assist in management decisions that minimize the risks to crop production. CSMs integrate knowledge from different disciplines and provide researchers with capabilities for conducting simulation experiments to either complement actual field experiments with additional outputs, link ecophysiological understanding to genetics for target crop improvement (Cooper et al. 2021), or to extrapolate experimental results in time and space (e.g., Rötter and van Keulen 1997). A pre-condition is that models have been well calibrated and validated using pluriannual and multi-locational field data. For sub-Saharan Africa, there are only very few generic crop simulation models (for the main cereals and legumes) that have been evaluated thoroughly under different agro-ecological conditions and with comprehensive and complete soil-weather-crop data sets (Rötter and Van Keulen 1997; Whitbread et al. 2010); The most widely applied and tested among these crop models are APSIM, Decision Support System for Agrotechnology Transfer (DSSAT), and World Food Studies (WOFOST) (see Rötter et al. 2018).

The essence of these three dynamic, process-based crop simulation models (CSMs) is that they can take interactions between genotype (G), environment (E), and management practices (M) into account (Rötter et al. 2015; Cooper et al. 2021). Working on a daily time step, these models are driven by daily values of the most important meteorological variables for plant growth and yield formation such as temperature, global radiation and precipitation and are supplied by a range of soil and crop parameters (and initial system conditions) supplying information on soil physical and chemical properties and major characteristics/traits regarding physiological properties of different crops/crop cultivars and their canopy architecture. Last, but not least, usually management data on material inputs (water, fertilizer; biocides) and their timing can be specified (see Fig. 23.4). The three models APSIM, DSSAT, and WOFOST differ in the original purpose of their development that is reflected in the type/detail and number of input and output data (see, e.g., Pirttioja et al. 2015).

Fig. 23.4
A flow diagram. Management input data, weather data, genetics of species, ecotype and cultivar, and soil profile data leads to cropping system model. Its output is water balance and productivity, nutrient balance and productivity, biomass, yield, and soil carbon sequestration.

Schematic of a crop simulation modeling platform: input and output

Full size image

CSMs have been applied in various ways to support plant breeding, e.g. to design crop ideotypes for different environments aimed at minimizing resource use per unit of dry matter produced (Rötter et al. 2015), or to estimate yield potential of crop ideotypes designed for current and projected future climates (e.g., Senapati and Semenov 2020).

The strength of process-based models (see Fig. 23.4) in ex ante evaluation of new technologies and in aiding ideotype breeding is because of their capability to describe causal relationships between crop growth and environmental and management factors driving them, and hence, to quantify the interactions between genotype (G), environment (E), and management (M). The models’ limitations relate to the accuracy of the process descriptions and uncertainties related to their parameters. Together these affect their usability for ideotype design and assessment of the effects of improved or new agro-technologies on biomass, final yield, water use, and emissions.

4.3 Agent-Based Economic Modeling

To integrate all biophysical and socioeconomic descriptive data and results gathered so far and link them with behavioral economics aspects of the respective decision-makers, i.e. the small-scale farmers, an agent-based farm household model (ABFHM) has been developed and is presented in this section. The model determines improved decisions under different future scenarios at farm level and for policy impact assessment at regional level.

The ABFHM presented in this section build upon the model in Feil et al. (2013). They develop and validate a generic numerical agent-based real options model that describes the simultaneous investment decisions of all firms, including their respective interactions, within a market under uncertainty. The uncertainty enters the model as one or more exogenous stochastic variables that represent uncertain exogenous demand, for instance. Based on this, the firms make investment and production decisions in each production period, which again form the sectoral supply of the respective product at the aggregate level. Accordingly, the endogenous equilibrium price process for the produced commodity can be derived directly. Hence, their model does not rely on some restrictive preconditions of other real options applications, which merely focus on one myopic firm and take market prices as exogenously given and deterministic. The numerical model is solved by a combination of stochastic simulations and genetic algorithms (GA) (for the detailed description of the numerical optimization procedure, cf. Feil et al. 2013). We adjust and enhance this generic model for application to the object of research at hand, i.e. small-scale farming in Limpopo. The basic structure of the ABFHM is illustrated in Fig. 23.5 and will be briefly explained in the following.

Fig. 23.5
A flow chart of text boxes. Stochastic processes for uncertain variables and farm related data production costs, adoption costs for technology packages, and more, and policy interventions lead to text box for A B F H M which leads to text box for results.

Basic structure of the agent-based farm household model (ABFHM)

Full size image

Consider a number of N homogenous and risk-neutral competing small-scale farms, each having repeatedly the opportunity to adapt predefined technology packages up to their maximum acreage Xcap, on which they cultivate maize in this example, either now or at a later point with in the period under consideration T. This technology package could, for instance, be the application of full irrigation combined with fertilizer application of 120:60 kg N:P per ha, which would cause additional investment and operational costs, on the one hand, but increasing maize yields and hence income on the other hand. Size, adaptation costs, and production are assumed to be proportional, i.e. there are no economies of scale. The production capacity of a farm n in t, resulting in a production output \( {X}_t^n \), can be adjusted via technology adoption just once in a period, resulting in an additional production output \( {\varDelta X}_{t+\varDelta t}^n \) in the following period. If it is assumed that the adoption costs are sunk in total, there are no possibilities to reverse the technology adoption, i.e. the associated investment costs are perfectly irreversible. However, in every period the production output declines corresponding to a geometric depreciation rate. Then production follows:

$$ {X}_{t+\varDelta t}^n={X}_t^n\cdotp \left(1-\lambda \right)+{\varDelta X}_{t+\varDelta t}^n $$
(23.1)

The stochastic demand process μt and the price elasticity η (e.g., for maize) are assumed to be known. Prices result from the reactions of all market participants on the exogenous stochastic demand process and, hence, need to be determined endogenously within the model. Without loss of generality, the relationship between market supply Xt and price Pt is defined by an isoelastic demand function according to Eq. (23.2):

$$ {P}_t=D\left({X}_t,{\mu}_t\right)={\left(\frac{\mu_t}{X_t}\right)}^{\varPi }\ \textrm{with}\ \varPi =-\frac{1}{\eta } $$
(23.2)

Within the model, perfect competition is assumed. Accordingly, the farms are assumed to have rational expectations and complete information regarding the development of demand and the technology adoption behavior of all competitors. Because of this, it should be expected that in equilibrium all farms have the same critical price which triggers the adoption, in the following called the trigger price. However to derive this equilibrium by means of the GA approach, the competing firms need to interact, which they do by defining their (at first different) trigger prices. This interaction of the firms equals a second price sealed bid auction in which each farm can sell its product if it asks less or equal the market price. To derive the adoption decisions of the farms, it is assumed that firms with lower trigger prices have a stronger tendency to adopt. Thus, all farms can be ranked by their trigger prices, starting with the lowest, i.e. \( \acute{P}^n\le \acute{P}^{n+1} \). Consequently, firm n + 1 does not adopt if firm n has not already completely adopted the respective technology package on all of their available acreage. Likewise, if firm n has fully adopted the TP, firm n − 1 has fully adopted it also. Moreover, in every period t, a marginal (or last) firm exists which adopts to the extent that its trigger price equals the expected product price of the next period. For the size of investment of a firm \( \overset{\sim }{n} \) in t, corresponding to its additional production output in t + ∆t, follows:

$$ {\varDelta X}_{t+\varDelta t}^{\tilde{n}}\left({\overline{P}}^{\tilde{n}}\right)=\max $$
(23.3)

The goal of the model is to identify the optimal trigger prices of the farms, which can be expected to be (nearly) identical in equilibrium according to the above assumptions. For this, an objective function needs to be established which determines the investment behavior of the agents in the model. Each farm’s adoption decisions aim to maximize the expected net present value (NPV) of the future cash flows \( {F}_0^n \), in the real options terminology also called option value, by choosing its farm-specific trigger price \( \acute{P}^n \):

$$ \underset{\overline{p}n}{\max}\left\{{F}_0^n\left(\overline{p}n\right)\right\}=\underset{\overline{p}n}{\max}\left\{{\sum}_{t=0}^T\left(-k+{P}_t\right)\cdotp {X}_t^n\left({\overline{P}}_t^n\right)\cdotp {e}^{-r\cdotp t}\right\} $$
(23.4)

k denotes the total costs of technology adoption per output unit and period, which are composed of the capital cost of the initial investment outlay I (e.g., for irrigation machinery) and all other relevant costs c (e.g., material costs for fertilizer, labor costs for running the irrigation system):

$$ {k}_t^{\prime }=I\cdotp \left\{{e}^{r\cdotp \varDelta t}-\left(1-\lambda \right)\right\}+c $$
(23.5)

The numerical model is solved by a combination of stochastic simulations and genetic algorithms (GA). Furthermore, policy interventions like investment subsidies, price regulations, or water use rights can be flexibly integrated into the model as needed. For the detailed description of the numerical optimization procedure, cf. Feil et al. (2013).

5 Assessing Effects of Selected Technology Improvements on Ecosystem Services

A number of promising, alternative agricultural management interventions in support of SI have been identified during the course of the SALLnet project. Apart from improved soil nutrient and water management practices, a whole range of new technologies could be very useful to increase agricultural productivity in a sustainable manner, including new plant seeds and breeds more resilient to climate variability and anticipated changes, mechanization and utilization of digital tools. However, while a few of such agricultural innovations were found among the most advanced farms in Limpopo, the vast majority of farms surveyed in the villages during the 2019 campaign showed only very basic management interventions—mostly relating to soil fertility and water management.

Given that the status quo in terms of crop management can generally be considered as minimal, in the following section, we concentrate on improving those technologies that, according to Vanlauwe and Dobermann (2020) are considered as the most crucial building blocks for lifting subsistence farmers to the first step out of poverty and of increasing productivity, i.e. fertilizer application and water management.

5.1 Example of Integrated Analysis of the Effects of Selected Improved Technologies and Agricultural Innovations on Rural Livelihoods and Other Ecosystem Services at Community Level

In Table 23.7, the assumptions used for the simulations by means of the ABFHM as described in Sect. 23.4.3 are listed. Three technology packages (TP) for maize cultivation are investigated following Sect. 23.3.3. TP 0 represents the status quo treatment of no fertilizer application and rainfed irrigation, TP 1 combines deficit irrigation with a fertilizer application of 40 kg N and 30 kg P per ha, and TP 2 combines full irrigation with a fertilizer application of 120 kg N and 60 kg P per ha. Therefore, irrigation technology has to be invested in and implemented for TP 1 and 2, which causes capital costs.

Table 23.7 Parameters used in the agent-based farm household model
Full size table

In the ABFHM, both TP 1 and TP 2 are each applied to 100 small-scale farmers (i.e., the agents), who are assumed to be in close proximity to each other in one region. According to the classification conducted in Sect. 23.2, at the start of the ABFHM simulations these small-scale farmers represent subsistence-oriented farmers with a sole focus on maize farming and, so far, no use of irrigation technology and fertilizer (i.e., they all start with TP 0). Without loss of generality, these farmers are assumed to have a standardized acreage of one ha for maize cultivation. Based on this, they make decisions in every consecutive production period starting with year one, whether or not they adopt TP 1 or TP 2, depending on which TP of both is exogenously available. On the one hand, this would imply higher expected yields and therefore additional expected revenues, as farmers would sell their excess maize corn on a local market at equilibrium market prices as derived within the ABFHM (cf. Sect. 23.4.3). On the other hand, this would also entail additional costs, in fact the investment and running costs of the irrigation machinery as well as the costs for purchasing and applying fertilizer.

The results of the ABFHM simulations are presented in Table 23.8. Accordingly, for each TP 1 and TP 2, separate model runs are conducted over infinite time, approximated by a discrete time period in the numerical model of 30 years. For both these runs, the average resulting maize corn trigger prices of the farms for adopting the respective TP are provided in the third column. Furthermore, the share of farms out of the sample of 100 farms that adopt the respective TP over time within the respective ABFHM run (column four) is shown. Moreover, the resulting average maize corn yield within the overall sample of 100 farms (column five) as well as the resulting overall increase in maize corn production at community, respectively, village level (column six), is also displayed.

Table 23.8 Results of the agent-based farm household model
Full size table

Accordingly, the average maize corn trigger price of the farms for adopting TP 1 is 4.30 and for TP 2 is 3.10 ZAR per kg, respectively. Comparing this to a considerably higher observed actual average maize corn price of about 9.00 ZAR per kg in the villages at the time of the survey in early 2019 (cf. Sect. 23.2), it becomes obvious that it could be economically worthwhile for the small-scale farmers to adopt the described TP’s already now. This, of course, would require the technology to be available to small-scale farmers in the region in general, that is water access, irrigation machinery, and fertilizer, at the assumed conditions, which in reality is often not the case in the investigated villages (cf. section two). The lower average trigger price for TP 2 compared to TP 1 indicates that it is economically worthwhile for the farmers to adopt TP 2 sooner. Based on this, 28% of small-scale farmers would adopt TP 1 and even 56% would adopt TP 2 over time. The resulting increases in regional maize corn supply would be 43% in the case of TP 1 and as much as 175% in the case of TP 2 in the long run.

5.2 Discussion of Other Promising Land Management Interventions

The SALLnet farm survey campaign in five villages during 2019 revealed for some farmers a few other management interventions. Some farmers had started diversifying crops with the relatively drought-tolerant sorghum and the bambara groundnut legume—as had been promoted by local extension service providers. Farm machinery consisted of poorly maintained moldboard plows that were few in number and therefore shared by many. Instead of site-and season-specific crop cultivar selection, farmers were generally unknowledgeable about the cultivars they planted and tended to simply make do with what was available. This was largely dependent on what was saved from the previous harvest, what was available in local agricultural supply stores, and in some instances what was given to them by the University of Limpopo. Hence, clearly, there is a wide range of promising options that can be used to go beyond those currently found at the study sites and beyond moderate fertilizer application in combination with deficit irrigation.

For many regions in the semi-arid tropics, including the Limpopo province, serious water-resource constraints are increasing and set to continue. Agricultural management needs to buffer such constraints, but not all interventions can be implemented in every farming context. In terms of integrated soil-nutrient management, it must be recognized that soils in the region are poor and low in soil organic matter (SOM) and therefore water retention capacity. This in turn makes soils and their resources vulnerable to erosion. In South Africa, more than 60% of the soils are low in SOM, highly degraded and, furthermore, low in terms of productivity due to nutrient mining and erosion (Vlek et al. 2020). In addition, climate change effects such as warmer and drier conditions throughout most of South Africa are expected to increase the rates at which soil C mineralization occurs, along with associated losses in soil ecosystem functions. As taken into account through our modeling experiments, the application of mineral fertilizers (especially N and P fertilizer) is the major soil-nutrient/fertility replenishing option, which can lead to substantial yield gains—as shown by our analyses. However, in order to sustain any yield gains, the application of mineral fertilizer must be complemented with conservation agriculture (CA) techniques, organic inputs, and good agronomic practices (see Thierfelder et al. 2017). Moreover, meaningful soil management must focus on restoring soil health through stopping and reversing nutrient depletion and organic matter decline (Swanepoel et al. 2018). A few considerations when working toward such goals:

  • Conservation agriculture (CA) is knowledge intensive. It not only does it require specialist skills, but it often necessitates special seeds (e.g., cover crops) and machinery (e.g., seed drills). While certainly key to sustainable farming systems design, CA is challenging in the smallholder context (see Baudron et al. 2015).

  • Constraints to soil amendments arise from the scarcity of organic material due to multiple claims on biomass, i.e. crop residues, wood litter, cattle manure. For example, crop residues in the field, including stubble, can be used as animal feed or left to decompose and partially replenish the soil.

  • In marginal, food insecure, high-risk environments, farmer advisory services and new technologies are often lacking. This makes it difficult to employ appropriate risk-management strategies that enable site- and season-specific input adjustments and the design of efficient cropping systems. This brings us to the reality that the benefits of digitalization have not yet reached the smallholder context, despite already benefiting larger scale producers and agribusiness (see Baumüller and Kah 2020).

Kuyah et al. (2021) and Wilkus et al. (2021) described a whole range of promising options, i.e., innovative agronomic practices for SI in Sub-Saharan Africa (SSA), including doubled-up legume intercropping, CA, agroforestry practices, push-pull technology, etc. While these all can boost production, there are additional options for pushing small-scale production levels toward those of commercial operations and increasing the efficiency of the farming systems in the medium to long term (Baudron et al. 2015). Here are a just a few suggestions:

Crop Improvement Programs

Acceleration of crop improvement is increasingly difficult in highly variable climates of already risk-prone systems—as found in Limpopo. Basis for any targeted breeding program taking shifts in future environmental conditions into account, is ex ante evaluation of genotype-by-environment-by-management (GxExM) interactions, as supported by CSMs. To make progress, crop improvement strategies need to develop crop varieties (G) and agronomic practices (M) specifically adapted to current and future local conditions (E) (Cooper et al. 2021). In a review on the bottlenecks of developing climate-resilient maize for the stress-prone (sub-) tropics, Cairns and Prasanna (2018) present the benefits of tailor-made genotypes for well-defined target environments. One major hurdle in this context has been data availability (e.g., lack of high-resolution soil and climate data). However, progress is being made, exemplified through the release of online and open-access databases such as the iSDAsoil (see: https://www.isda-africa.com), a mapping system for SSA at a resolution of 30 m, including 24 billion locations across Africa. Such resources could contribute toward site-specific recommendations for farmers in the near future.

Mechanization of Farming Practices

Farmers of Limpopo region shared tractors and plows and were therefore limited in terms of when they could prepare the fields and sow, often leading to poorly-timed cultivation with yield penalties. The region is also impacted by El Nino-Southern Oscilliation (ENSO) events, with marked inter-annual differences in soil moisture supply (Moeletsi et al. 2011). Future improvement for these systems is therefore likely to be centered around timing, i.e. the ability of farmers to adapt quickly to climate conditions. The use of scale-relevant, affordable machinery in general could enable farmers to react in a more agile manner, for example through implementing seedbed management independently without having to wait for the availability of shared tractors and plows.

Digital Tools

While digital tools can help connect agricultural stakeholders and spread information, also the digitalization of the agricultural systems themselves is also high on the research and development agenda. The German development organization “Welthungerhilfe,” for example, shows how a mobile phone-based app “AgriShare” helps connect farmers in Zimbabwe without assets with those who can supply it, such as commercial or private hiring services (Welthungerhilfe 2018). This includes services for production, processing, and transportation.

6 Synthesis and Outlook

In this chapter, we explored how, in the face of increasing climatic risks and resource limitations, improved agro-technologies could support SI in smallholder farming systems in Limpopo province, South Africa. We used the example of improved soil nutrient and irrigation practices in combination to perform an integrated bio-economic analysis of the potential benefits for smallholder farmers and some essential ESs they rely on. From the perspective of technical feasibility, economic viability, and environmental benefits the selected technology packages clearly show that they can support SI pathways for small-scale farmers. However, for implementing such technological improvements successfully at a larger (e.g., landscape and regional) scale in the near (short- to medium term) future, establishment of the required mandatory foundations for this will be necessary. That is, enabling policies as well as dissemination and training of farmers in new applying technologies.

Policy Recommendations

Solid foundations can be laid by adjusting national and local policies and resource management regulations at higher aggregation levels—in accordance with the scale targeted at. To achieve this, concerted actions for SI of small-scale farming systems will be required involving politicians, extension services, up- and downstream agribusinesses, and other supporting institutions. This includes in the first instance the gradual establishment of an appropriate infrastructure in the respective regions regarding transport routes to allow the access of input and output markets, specific packaging needs (for fertilizer) of small-scale farmers as well as effective and knowledge-oriented and competent extension services.

Research Challenges

Regarding crop sciences and agro-ecosystems modeling there are a number of future research challenges. A very urgent one is to accelerate breeding of climate-resilient crop cultivars as needed under progressive climate change. For developing these, next-generation breeding technologies such as high-density genotyping coupled to high throughput (precision) phenotyping and the use of crop simulation models is required. These methods can help untangle GxExM interactions and provide farmers with suitable genotypes, along with tailor-made management recommendations, informed by in silico experimentation (Hammer et al. 2020). Yet, an essential pre-requisite will be to improve agro-ecosystems for this task. That can be realized in various manners, e.g. modifying modeling routines for better capturing relevant processes at sub-daily scale, e.g. transpiration under low air humidity, collecting new phenotypic data to incorporate accurate information about genetic diversity, or linking genomic to ecophysiological information to improve model calibration (Rötter et al. 2015).