1 Introduction

The smallholder farming systems in sub-Saharan Africa (SSA) face many barriers to increased crop productivity such as low nutrient inputs, insufficient control of weeds, pests and diseases, and inadequate labor, and are generally referred to as low-input systems (e.g., Sheahan and Barrett 2017). Under the low-input systems, labor is often the major input and its availability is critical for timing operations during the production cycle; for example, labor shortage often leads to late planting whereby crops cannot take full advantage of the rains and to poor weed control leading to high competition for nutrients, water, and light, which leads to low yields (Giller et al. 2006) (Fig. 1).

Proposed technologies for sustainable intensification (SI) of crop production are often labor demanding, especially in the initial years of adoption. Labor shortages are frequent in small-holder agriculture in SSA and often aggravated by ill-health in diseases such as malaria and HIV/AIDS (Chapoto and Jayne 2008; Ajani and Ashagidigbi 2008) as well as by urbanization and a younger generation turning away from farming. This has hindered widespread uptake of technologies such as cut and carry green manures, moisture conservation strategies such as fanya juu, zaï, and tied ridges that entail substantial initial labor input or, e.g., increased labor needs for weed control (e.g., Nyamadzawo et al. 2013; Namara et al. 2003; Schuler et al. 2016; Grabowski and Kerr 2014; Gicheru 2016). Furthermore, small or no improvement in productivity may result if multiple limiting factors are addressed one at a time. Fertilizer application may, for example, increase yields marginally if weed management is delayed or carried out at low intensity (Chikoye et al. 2008; Rusinamhodzi et al. 2016), or fail to increased yields or even decrease yields by aggravating water shortage during dry spells (Rusinamhodzi et al. 2012). This may lead to abandonment and disrepute of the technologies as the investment of cash or labor does not pay off.

Fig. 1
figure 1

Weed management is generally one of the most labor-demanding tasks during the cropping cycle. a Farmers weeding by hand hoe, the most common weed control strategy on smallholder farms in SSA. b A farmer showing a typical field where time constraints precluded timely weeding (left) and the impact on the maize crop compared with timely weeding (right). The whole field had the same cropping history and fertilization

Smallholder farming households often have no or small economical margins and so face difficulties to invest labor and cash in improved technologies that will increase crop productivity noticeably only after several seasons. This is often so with Integrated Soil Fertility Management (ISFM)–based technologies that build soil fertility by combining organic and mineral nutrient sources (Rusinamhodzi et al. 2011; Nezomba et al. 2015) or the effects of legumes such as Desmodium spp. on Striga seed banks (Khan et al. 2008). However, benefits can be seen in one or two seasons if, for example, the technologies simultaneously address nutrient and short-term water shortages such as conservation agriculture applied in drier agro-ecological zones (Ransom 1990; Rusinamhodzi et al. 2011), or the push-pull cropping system which addresses maize stem borer, fall army worm and Striga while producing high-quality fodder (Vanlauwe et al. 2008; Midega et al. 2018). Intercropping with multipurpose legumes also shows promise due to the options these legumes provide at the farm level, in the short term on human and animal nutrition and in the longer term on soil fertility (Giller 2001). Intercropping systems often increase labor needs for weeding (Ransom 1989; Rusinamhodzi et al. 2012), though, and may decrease yields of, e.g., the companion cereal in the short term (Franke et al. 2017). Hence, additional strategies are needed to increase crop productivity while reducing labor input for vulnerable smallholder farmers, especially women farmers who constitute the majority (Palacios-Lopez et al. 2017). The underlying hypothesis of this study is that the short-term returns to labor are critical and may determine the fate of most promoted SI options on smallholder farms in SSA.

The objective of this study was to assess crop productivity change due to soil fertility, water, and weed management options in relation to labor demand on smallholder farms in sub-Saharan Africa through a systematic review. It is thus not intended to be another systematic review on productivity of SI options (e.g., effect of CA on yield) per se. Specifically, we sought to:

  1. a)

    Assess yield change relative to labor change relative to the most commonly used technologies (baseline) for SI in different agro-ecological conditions.

  2. b)

    Discuss whether the new technologies are beneficial, i.e., increase yield and reduce labor input for farmers sufficiently to warrant the risk of investment in the technologies.

2 Materials and methods

2.1 Database compilation

We retrieved a total of 28 published papers (Table 1) and 631 data pairs; 59% of these were derived from studies carried out in farmers’ fields. The published papers were retrieved from online search engines such as Scopus, Google, Google Scholar, and Web of Science. Crop economic yield and the related labor input data were obtained from studies that included a baseline crop management treatment and an improved SI option under rain-fed conditions established in semi-arid and sub-humid environments with a major focus on SSA. Data had to be from studies with randomized plots and at least three replications. Data from farmers’ field trials were also included where each farmer represented one replicate provided that the same trial was established in at least three farms in a location. Our search was comprehensive including the following keywords and their combinations: sustainable intensification, intercropping, conservation agriculture, integrated soil fertility management (ISFM), labor input, labor productivity, reduced tillage, no-tillage, tillage system, crop/maize/corn/sorghum/millet/cassava yield, sub-humid, semi-arid, rain-fed, weed management, water management, profitability, field experiment, and smallholder. Labor input was the key search term and only studies that reported both yield and labor were retrieved. We also contacted authors to get clarifications and additional meta-data for some of the publications. We collected information on moderators such as climate (temperature and rainfall), altitude, soil texture of the experimental site, cropping systems, tillage management, weed control methods, rate of fertilizer, or soil amendment applied, as reported by the primary authors in the publications or in correspondence. These factors were considered to have a significant influence on the magnitude of change of yield. Rainfall was categorized using long-term mean annual precipitation reported by primary authors to form mean annual precipitation (MAP) classes as low (< 600 mm), medium (600–1000 mm), and high (> 1000 mm) based on FAO guidelines (Fischer et al. 2001). Soil type texture was categorized as clay, clay loam, sandy, loamy, and sandy loam (Brown 2003).

Table 1 List of data sources, study locations, tillage and cropping systems, durations of the studies, soils, mean annual precipitation (MAP) class, and number of data pairs generated. Tillage systems: B = basins; C = conventional; NT = no-till; R = ridging. Cropping systems: I = intercrop; R = rotation; S = sole crop

Data were collated in MS Excel in the form of treatment mean (\( \overline{X} \)), its standard deviation (δ), and the number of replicates (n) for yield as mentioned in the experimental design. However, labor data were often presented as a single value without any statistics. Once collated, we assessed the number of studies and observations present for different crops and technologies. Observations from the same study in the dataset were considered independent data points and included separately if they were from different sites, seasons, main crop, or companion crop to reduce problems with non-independence of data points. Due to data scarcity on most crops, only maize was retained as a focus crop; tillage technologies were ridging, basins, no-tillage, and conventional tillage (generally the baseline technology); and cropping systems were intercropping compared with sole-cropping or rotation compared with continuous monocropping (Table 2). There were very few data from studies that in a stepwise manner addressed multiple production constraints such as low nutrient and water availability and abundance of pests and weeds; hence, evaluation of such effects was not attempted.

Table 2 Terminology used

2.2 Calculations

2.2.1 Meta-analysis of maize grain yield

The response ratio (RR) was used as the effect size and was calculated for each observation of maize grain yield as the natural log of the proportional change between the mean of yield for the intervention (Ysi) treatment (e.g., intercropping, N fertilizers, or conservation agriculture) and the mean of the control (Yc) group (Rosenburg et al. 2000; Hedges et al. 1999). We chose the RR because of different units used by the primary authors and that none of the observations had zero values. The RR was calculated using the equation:

$$ \mathrm{RR}=\ln \left(\frac{Y_{si}}{Y_c}\right) $$

We used a linear regression approach to measure funnel plot asymmetry on the natural logarithm scale of the response ratios as measure of bias in the dataset as suggested by Egger et al. (1997). The influence of moderator variables such as mean annual precipitation (MAP), soil types, tillage type, and cropping system on the yield effect size was computed by meta-regression using the metaphor r-package (Viechtbauer 2010) in R-Studio ver. 1.1.456. The studies were assigned weights based on the inverse of the variance:

$$ {\mathrm{weight}}_i= 1/\left({\mathrm{variance}}_{\mathrm{i}}\right)=1/{{\mathrm{SD}}_{\mathrm{i}}}^2 $$

The weights took into account study variances of treatment effects and the residual of between-study heterogeneity using the procedure described by Borenstein et al. (2009). For each treatment group, we computed an effect size estimate, comparing each pair of treatment and control group. In some cases, the data from the control group was repeatedly used to calculate the effect size estimates, leading to correlated estimates. To address the non-independence of data points, we used the random effects model and also adjusted the effect size variances in cases sharing a common control treatment using the procedure described by Gleser and Olkin (2009). Mean effect size for each moderator was calculated with bias-corrected 95% confidence intervals. Labor data were often from a single measurement and did not contain any error statistics and was thus excluded from bias and meta-regression analyses.

2.2.2 Yield-labor relations

We sought to assess whether the interventions for crop improvement would lessen the labor burden of labor-constrained smallholders. For better clarity, the change in yield or labor input was assessed as a percentage of the baseline. The percentage response in yield (ΔY) or labor input (ΔL) to intervention compared with the baseline technology was calculated using the equation:

$$ \Delta Y\%\mathrm{or}\ \Delta L\%=\frac{{\overline{X}}_{si}-{\overline{X}}_c}{{\overline{X}}_c}\times 100 $$

Where \( {\overline{X}}_{si} \) is the means of yield or labor input for the intervention treatment (cropping and tillage technology) and \( {\overline{X}}_c \) is the means of yield or labor input of the control group, respectively.

These ratios were presented graphically as a percentage change for easier interpretation. We also calculated the ratio between yield response and labor response (i.e., ΔYL). Our objective was to assess changes (labor and yield) of intervention crop production systems in relation to the baseline which was normally the most common production system in a locality. Labor input was disaggregated by the cropping calendar activities when such data were available, and finally, total labor input was also considered. The influence of moderator variables such as mean annual precipitation (MAP), tillage type, and cropping system on the % yield change was assessed. Initially, the analysis included observations from all relevant studies that tested technologies that could be classified as SI. Hence, these also included studies where, e.g., fertilizer or pest control technologies were tested. To specifically assess the yield-labor relations of the tillage technologies and cropping systems per se, the next step included only a subset of paired observations with the same management in other respects than the tillage or cropping system technology which was compared with its respective baseline treatment. The only exception to this was that different weeding technologies were accepted and the effect of that also taken into consideration.

3 Results and discussion

3.1 Assessment of bias and maize yield moderators

The contour funnel plot (Fig. 2) using maize grain yield RR was fairly symmetrical around 0, and the Egger’s regression test (z = − 1.46, p = 0.18) for asymmetry did not suggest any publication bias in the dataset but data was heterogeneous. Thus, publication bias was not a major concern in our analyses.

Fig. 2
figure 2

Contour-enhanced funnel plot to assess bias in experiments on the effect of sustainable intensification options on crop productivity over baseline production options. The shaded regions represent the 90, 95, and the 99% confidence intervals

All the MAP categories were significant on the maize yield responses (Table 3). When MAP was lower than 600 mm per year, results suggest that the yield change will be negative (RR = − 0.185) but increased with increased rainfall up to 0.256 when MAP was above 1000 mm per year. In terms of cropping, it is possible to increase maize yield employing any of sole monocropping, intercropping, or rotation systems. However, intercropping was found to be the most beneficial (RR = 0.250) and that benefit was significant. Crop rotation led to the least benefit. Effect size was also affected significantly by soil texture; results showing more benefits were obtained with clay soils (RR = 0.49), followed by sandy loam soils (RR = 0.272), whereas the sandy and loam soils gave negative responses. Ridges (RR = 0.516) and no-tillage (RR = 0.249) resulted in the largest yield benefits for the tillage systems considered. The planting basins did not result in any yield advantage over the baseline tillage system compared (Table 3).

Table 3 Meta-regression and sub-group analysis of studies reporting maize grain yield of sustainable intensification options against baseline scenarios

3.2 Distribution of yield and labor changes

Absolute maize yields varied widely between studies and seasons with a range of an almost complete failure due to drought or flooding, to 8.1 t ha−1 in the baseline systems, and to 9.0 t ha−1 in the interventions with a median of 2.2 t ha−1 and 2.7 t ha−1, respectively. Many different non-maize crops were reported in the intercropped or rotational systems, often with few observations per crop; hence, the analysis was restricted to the maize compartment. Labor was reported in a variety of time and monetary units, precluding compilation of absolute values. The number of studies and observations differed strongly between the tested systems which affected the statistical strength of the analysis. For example, almost all data for planting basins were derived from a study carried out in one location (Rusinamhodzi et al. 2012), and also ridging was presented in a few studies.

Maize yield responses and the concomitant total labor response from paired observations exhibited great variability across the studies. About 50% of the intervention cases recorded 0% or negative yield change when compared with the baseline production system (Fig. 3). The average yield change was + 41% above the baseline although yield changes of up to 10 times were recorded but in very few cases. The average labor change was 0% with a uniform distribution of cases below and above zero (Fig. 3). Results show that new interventions could reduce labor input by as much as − 75% but may also increase it by double (+ 100%) in some cases. A positive change for yield indicates an advantage, for labor input, this means more labor is needed for such a technology and is an undesirable outcome under labor limitation. With opportunities to hire labor, an increase in labor demand can be offset by a relatively larger yield increase and/or high output prices. However, a complete economic evaluation including all possible scenarios of labor input is outside the scope of this paper.

Fig. 3
figure 3

The density distribution of the response ratios (RR) for a crop yield and b labor input. The dashed line is at RR = 0, referred as the line of no effect

3.3 Yield change from sustainable intensification options

The average maize grain yield change when using planting basins under sole-cropping in the 600–1000 mm rainfall category was − 12% but was + 71% in the only observation in the > 1000 mm rainfall regime (Fig. 4). There were no data on planting basins associated with intercropping and rotation. In the no-till system, the highest yield increase (+ 20%) was recorded under intercropping in the > 1000 mm rainfall category. The rotation option under no-till led to a large increase (+ 41%) in yield. Under the ridging system, all the tested cropping options under all the rainfall categories increased yield. Sole cropping recorded + 443% under > 1000 mm and rotation systems all gave large increases of + 143% and + 55% for the 600–1000 mm and > 1000 mm rainfall categories respectively (Fig. 4).

Fig. 4
figure 4

Maize productivity change in each of the three intervention tillage systems × cropping system combination in three mean annual precipitation (MAP) classes. The full set of cropping systems was not present in all the MAP classes and tillage combinations

Conventional tillage was most often the baseline practice when other tillage options were introduced as interventions for intensification. However, in many instances, conventional tillage was also associated with other interventions such as improved variety, fertilizer application, or intercropping, and this may have increased average yields of this tillage option in the comparison with the other tillage systems when observations from all the publications were included in the analysis. Similarly, sole cropping was also frequently associated with other interventions but was always the baseline for comparing the other cropping systems. It is clear from our study that intervention options were often introduced as a single technology. While such an approach may be desirable with complex technologies and farming systems, it can also limit the possible benefits due to the absence of the potential synergistic effects as reported previously (Sime and Aune 2016; Chivenge et al. 2011; Vanlauwe et al. 2011). Studies employing the component-omission approach and evaluating all the possible combination of practices in the same study may help assess the maximum technology performance across multiple locations (Thierfelder et al. 2015).

Overall, the reduced tillage practices resulted in small yield benefits above the conventional tillage treatment and productivity was strongly influenced by mean annual precipitation (MAP). Similar results have also been reported from long-term experiments, e.g., Kapusta et al. (1996) reported no significant differences in yield between no-tillage and conventional plowing on poorly drained soils after 20 years of continuous practices. Even the planting basins and ridges which are designed to retain moisture generally showed increased yields at higher rainfall. This is because, under rain-fed conditions, climatic conditions have a greater influence on yield than any other management decisions as has also been reported by Dam et al. (2005). Lal (1997) also observed that maize yield depended more on the amount of rainfall received and its distribution during the season than tillage management. This observation clearly shows that interventions solely based on tillage management are unlikely to increase yield unless combined with other practices and underlines the importance of rainfall (amount × distribution) in the locations from which data were obtained.

Intercropping reduced the yield of maize under low rainfall (< 600 mm), but yields were at par or increased slightly over sole cropping when rainfall was considered sufficient and non-limiting (> 1000 mm). This could be attributed to less competition for moisture between maize and the companion legume crop under high rainfall conditions than under drier conditions (Kermah et al. 2017). The effect of rotation based on this dataset was unclear mainly due to the absence of sufficient detail on how the legume crop residues were managed. However, it is well established that crop rotation is one of the tenets of good agronomic practices and can lead to substantial yield increases of the succeeding cereal through both residual nutrition and control of pests and diseases (Franke et al. 2017).

The studies considered for this review generally did not have different levels of nitrogen (N) within a study, but application rates varied across studies. This meant that the effects of increasing N were confounded as many other factors varied considerably from one location to the other. However, in conservation agriculture-based SI options, the wide C:N ratio and the relatively large amounts of readily decomposable carbon compounds may lead to N immobilization (Cadisch and Giller 1997), thus high N inputs may be required when poor-quality crop residues are used as mulch, especially in the short term. The lack of significant yield benefits from the no-till treatment above the conventional tillage may suggest that the no-till treatment needed more N fertilizer as has been suggested (Vanlauwe et al. 2014). This is because the paired treatments considered in this study often received the same amount of fertilizer between the conventional and the no-till treatment.

3.4 Labor input dynamics

Planting basins preparation increased average labor input by + 702% over conventional tillage, whereas no-till reduced land preparation significantly by − 83%, and the ridging system increased land preparation labor input by + 19%. The ridging systems reduced significantly the labor needed for weeding. Sole cropping with ridging reduced labor input for weeding by − 44%, whereas ridging × rotation combination reduced weeding labor by − 32%. No-till similarly reduced average labor input for weeding for rotation and sole cropping (− 82% and − 90%, respectively). In the intercrop option under no-till, average weeding labor input increased by + 18%. The largest and significant increase in labor input of + 35% for weeding was recorded under planting basins. The no-till systems reduced significantly the total labor input. Sole cropping with no-till reduced labor input by − 25% whereas intercropping with no-till reduced total labor input by − 27%. The planting basins increased total labor input significantly (+ 81%) but the ridging system only increased total labor marginally by + 9% in the sole cropping system.

3.5 Labor productivity of intensification options

Relations between the yield response and labor input revealed the existence of four situations defined by high yield and low labor demand (Fig. 5). Top left quadrant (HYLL) is characterized by observations of higher yields and a reduction in labor input as a result of the intervention. Top right quadrant (HYHL) represents observations of interventions that increased both yield and labor input. The bottom left quadrant (LYLL) shows the observations of interventions that reduced labor input and yield. The bottom right quadrant (LYHL) shows observations where the interventions increased labor input but decreased yield.

Fig. 5
figure 5

The relation between percentage change (Δ) in total labor input and yield for the tested tillage technologies compared with conventionally tilled and cropping systems compared with sole monocropping (n = 410). The red, dashed lines indicate lines of no-effect. In case of labor response ratio, it is the line where the intervention has the same labor input as the baseline technology

The review of the literature showed that the metrics related to labor input are very different across studies. While returns to labor are widely used, it does not clearly distinguish situations of high labor input and high returns against low input and low returns—i.e., may show the same ratio. In the future, a form of standardization may be needed. Under real conditions of farms, however, activities are not performed in a systematic way and aggregating such labor input is often problematic. For example, farmers may weed a plot over a week with different people involved (young, able, and aged) and during different times of the day, and capturing fully such dynamics if complex. Future studies reflecting on-farm situations need a minimum plot size (beyond those used at experimental stations) and all people involved as well as activities fully characterized.

3.5.1 Labor productivity in tillage systems

Our review of studies on tillage systems provided a total of 95 (land preparation labor), 118 (weeding labor), and 138 (total labor) paired observations of tillage technologies compared with a baseline technology and where other management factors were the same except that weed management technology sometimes differed. This baseline was most often conventional flat tillage (albeit with different tools) but no-till technology was also sometimes compared with ridged tillage where this dominated in the area. There were very few paired observations from low-rainfall conditions. Observations of basins were also few, with paired observations only from under intermediate rainfall. However, data indicate that basins require more labor for land preparation and weeding than the conventional baseline (avg + 700% and + 35%, respectively), placing this technology in the two higher-labor demand quadrants (Fig. 6). Both situations of yield increase or decrease were shown, with an average of − 9% compared with the baseline. The ratio between the change in yield and labor inputs (ΔYL) thus averaged 0.16 and 0.66 when considering labor for land preparation and weeding, respectively, indicating that labor demand increased more than yield in the reported studies. The only report on total labor use corresponded to ΔYL of 0.95.

Fig. 6
figure 6

The relation between change (Δ) in a labor for land preparation, b weeding labor, and c total labor input and yield response for planting basins, no-till systems with herbicides or manual weeding, and ridge systems given as % of the conventionally tilled reference point within the same study. The red, dashed lines indicate lines of no-effect. Data are presented across the MAP categories. Numbers of observations for land preparation/weeding/total labor are 33/33/1 for basins, 17/17/59 for no-till with herbicides, 29/45/62 for no-till with manual weeding, and 16/23/16 for ridging

In contrast, just over 80% of the observations of ridged systems compared with the conventionally tilled were found in the top left quadrant (Fig. 6) for all labor categories. All studies reported lower labor use than in the conventional baseline; labor use for land preparation, weeding, and in total decreased on an average of − 24%, − 48%, and − 33%. Average yield increase was largest (+ 72%) under intermediate rainfall conditions and lower under high rainfall (+ 10%); observations were very few under low rainfall. The ΔYL across rainfall conditions were 2.0, 3.4, and 2.2 for land preparation, weeding, and in total, respectively, indicating that yield increased more than labor demand.

No-till compared with the baselines showed higher variation and was found in all quadrants (Fig. 6), the pattern being strongly affected by the weed management technology. On average, no-till increased weeding labor by + 109% and total labor by + 14% when manual weeding was done but decreased weeding labor by − 45% and total labor by − 29% when herbicides were used. However, labor for land preparation decreased by an average of − 56% with no consistent difference between studies with herbicides and manual weeding. Yield responses to these technologies varied between MAP categories; under high rainfall conditions, the yield decreased significantly (avg − 12%) under manual weeding but increased significantly when herbicides were used (avg + 11%); under intermediate rainfall, it was not significantly changed under manual weeding but increased (avg + 63%) when herbicides were used. The average ΔYL across rainfall conditions were thus 1.7, 2.6, and 1.8 for land preparation, weeding, and in total, respectively, when herbicides were used and were 3.3, 0.6, and 0.9 with manual weeding only.

Land preparation through conventional moldboard tillage is labor demanding and for smallholder farmers often possible only during a short time window in connection with the onset of the rains or shortly after harvest of the preceding crop. Access to draught power is generally needed for moldboard tillage and thus often not within the reach for the poorest households or for farmers in geographical areas lacking draught animals. Many farmers therefore have to submit to the back-breaking drudgery of working the land with hand tools. Also, for farmers who can afford to hire draught animals, land preparation and planting may be delayed due to low availability of animals. Weeding is another labor demanding field activity and often uses the largest share of farm labor over the growing season (Vissoh et al. 2004). If staggered planting is practiced, e.g., in order to spread risks under fluctuating and uncertain rainfall, high labor demand for land preparation may entail delayed weeding, thus seriously affecting crop competition for water, nutrients, and light. Substandard weed management has been estimated to cause yield losses of 25–100% in smallholder farming systems in SSA (van Rijn 2000). Hence, revised technologies which decrease labor demand for these tasks are needed.

According to the reviewed studies, the ΔYL in the range 1.0–6.2 suggests that ridging is likely to be a relatively secure technology for smallholder farmers while offering potentially decreased labor needs. Weeding labor was most clearly decreased by ridging, both when carried out by hand hoe (Mafongoya and Jiri 2015) and by mechanical animal-drawn tools for ridging (Riches et al. 1997) as compared with manual hand tools in a conventional, flat tillage system. Furthermore, the ridging in Riches et al. (1997) was achieved as part of the weeding thus sparing labor for ridging during land preparation. Although the number of studies and observations in the ridging category was small, the average ΔYL of 3.4 for weeding labor suggests that in the smallholder farming setting, the mechanical weeding achieved by re-ridging can help achieve sufficiently low weed pressure to increase crop yield per invested labor hour substantially even if not always increasing yields per unit area. Decreased labor use and increased net benefits were also found by Shetto et al. (2000) and Shetto and Kwiligwa (1993) when testing animal-drawn weeding tools combined with in-row hand weeding, but careful land preparation was necessary to make the technology successful. Ridging tools are generally constructed for tractor or animal draught power, hence limiting the availability to many smallholder farmers. However, two-wheel tractors may be more within the reach for better-off farmers and farmer groups, although the availability of spare parts and service skills are among factors that may hinder uptake (Bymolt and Zaal 2015). Also, tools managed by human traction, such as hand ridgers, may be used by two persons to create smaller ridges (e.g., Mishra et al. 2016).

Riches et al. (1997) reported increased soil water content in ridged system, presumably because of higher water-capture and infiltration in the resulting micro-topography. The study suggests that under low-rainfall conditions even non-tied ridges along the contour can enhance water infiltration and crop yields while saving especially weeding labor. Larger increases in soil water in tied ridge systems have been found in arid and semi-arid regions (e.g., Brhane et al. 2006; Hulugalle 1990), but we were not able to find studies that also reported labor use and met the pre-determined criteria for inclusion; hence, their relation between labor use and yield change could not be tested.

The data for planting basins, derived almost exclusively from one study carried out under intermediate rainfall (Rusinamhodzi 2015), showed high labor demand. This was especially so for land preparation but also for weeding, in both cases associated with the use of manual labor. The yields did not increase correspondingly in this study, as revealed by the ΔYL being consistently below 1. High labor demand has also been an impediment to adoption of the zaï system (Schuler et al. 2016), being feasible only where labor opportunity costs are low, land can be worked with reasonable effort before the onset of rains, and/or used for higher-value crops. However, mechanization of zaï construction in Burkina Faso using cattle traction decreased labor use from an average of more than 400 h ha−1 to approximately 30 h ha−1 (Barro et al. 2005). The observations included here showed an average yield decrease of − 12% in the basin systems compared with the conventionally tilled baseline system. Restricting the use of this type of planting pits to drier sites and/or soils with sufficiently high infiltration capacity to avoid waterlogging in the pits may increase the yield response to positive such as seen by, e.g., Amede et al. (2011), Roose et al. (1999), and Kabore and Reij (2004). Paired with the mechanization of a planting pit construction shown by Barro et al. (2005), this suggests that planting basins may be more beneficial in dry sites than in the study sites included here. However, the independence of draught power of manually dug planting basins which forms part of their promotion (Rusinamhodzi 2015; Andersson et al. 2011; Andersson and Giller 2012) is then irrelevant. Furthermore, in such dry environments, limitations in access to draught animals and their physical condition before the rains start may be a hindrance (Mabuza et al. 2013) unless the scarification performed as part of the mechanization can be done before the dry season.

No-till was, as expected, saving labor during land preparation in the included studies. However, we found no relevant publications with no-till in areas with < 600 mm precipitation where labor use was documented along with other data needed. Nevertheless, the labor savings indicated can be a great advantage especially in areas where the growing season is short and the availability of labor and draught animals is low at the time of the onset of rains and planting (Mupangwa et al. 2017), in particular where soil is hard and difficult to work before the rains (Rusinamhodzi 2015), or the draught animals are in poor condition due to low feed availability during the dry season (Rufino et al. 2011; Mabuza et al. 2013) or unavailable to female-headed households, e.g., due to customary gender roles (Palacios-Lopez et al. 2017).

Whereas no-till consistently increased labor demand for weeding (87% and 44% for medium and high rainfall conditions, respectively) when compared with ridging, the data indicated that no-till compared with conventional tillage increased labor demand for weeding under medium rainfall conditions (avg + 126%) but decreased it under high rainfall conditions (avg − 36%). This apparent difference between MAP classes was traced back to differences in weed management technology; studies performed under medium rainfall were predominantly using only manual weeding in both no-till and the baseline systems (37 out of 38 observations that reported weeding labor) whereas studies performed under high rainfall conditions often applied herbicides in the no-till treatments but only manual weeding in the baseline (14 out of 24 observations). A decisive factor for the outcome in terms of labor use and ΔYL was thus the use of herbicides and illustrates a shift of input type from labor to cash for herbicide purchase. The choice of weed management technology apparently did not only affect labor use but also the yields. Although the density of weeds was usually not reported, the tendency to lower the yield in the manually weeded no-till systems (in spite of the larger number of work days) than in no-till systems with herbicides suggests that the manual labor invested was insufficient to control the weeds. This may be a contributing cause for the low yield response of the no-till technology found under high rainfall conditions, in addition to previously suggested poor aeration at increased soil water contents (Rusinamhodzi et al. 2012) or low N availability (Vanlauwe et al. 2014).

3.5.2 Labor productivity in cropping systems

Studies specifically investigating different cropping systems provided a total of 119 (land preparation labor), 269 (weeding labor), and 272 (total labor) observations of manually weeded rotations or intercropping contrasted with sole cropping. The effect of intercropping on labor demand and yield was highly variable and observations fell into all quadrants for weeding and total labor (Fig. 7). On average, labor demand for weeding and in total were similar in intercropped and sole-cropped systems (+ 4% and − 5%, respectively). The reported labor use for land preparation was, for many observations, the same for intervention and baseline, though, and only plot-wise yields were reported. Yields were on average similar to the baseline (− 4%). The resulting average ΔYL were 0.78, 0.90, and 1.1 for land preparation, weeding, and in total, respectively.

Fig. 7
figure 7

The relation between change (Δ) in a labor for land preparation, b weeding labor, and c total labor input for manually weeded intercropping and crop rotation given as % of the monocropped reference point within the same study. The red, dashed lines indicate lines of no-effect. Data are presented across the MAP categories. Numbers of observations for land preparation/weeding/total labor are 23/167/170 for intercrop and 96/102/102 for rotation

The relation of maize yield and labor use in rotational systems was also highly variable but a dominance of observations in the top right quadrant showed increased or similar labor use compared to monocropped systems (+ 13%, − 1%, and + 7% for land preparation, weeding, and in total) (Fig. 7). Eighty-nine percent (89%) of the observations of these systems showed increased maize yields (avg + 66%) compared with continuous maize monocropping. The average ΔYL were 1.4, 1.7, and 1.6 for land preparation, weeding, and in total, respectively.

Despite the lower net area devoted to maize in the intercropped systems, average maize yield was similar to that in the sole-cropped baseline systems, while the system also produced a companion crop. Variation was large, though, reflecting differences in, e.g., rainfall, crop combinations, and crop management in the intervention and baseline treatments. However, although the maize ΔYL frequently fell below 1 there were also some observations with higher maize ΔYL. This suggests that some intercropping systems may have the potential to increase crop productivity even at decreased labor inputs if tailored to the local agroecology, but that under other conditions in intercropping may bring a yield penalty to the main crop. The acceptable yield penalty will depend on, e.g., the companion crop’s yield and market price (if sold) and if it has multiple uses as food and feed and for building soil fertility.

In this review, the variation in tested rotational crops and the variability in labor demand and yield response was large. However, the ΔYL was more consistently above 1 suggesting beneficial relations between yield and labor demand. Furthermore, literature has shown that rotational systems may increase maize yield significantly (Franke et al. 2017) especially where the alternate crop was green manure (Ojiem et al. 2014). However, even though production systems with green manures may be attractive in terms of yield per unit labor and land, adoption of such systems requires that farmers can invest in a season of green manure crop before reaping the benefits, since they may not contribute to immediate household nutrition. Grain legume rotations, on the other hand, have shown variable effects on the accumulated maize grain yield compared with maize monocropping (Rao and Mathuva 2000; Ojiem et al. 2014). The productivity and value of the rotation crops for home consumption, cash or soil fertility management, and the economic context of the farmers will thus strongly influence the attraction of the rotations.

3.6 Implication for SI in sub-Saharan Africa

The perennial challenges such as low fertilizer use, poor markets, labor shortage, erratic rainfall, multiple uses of crop residues, small land sizes, and poor soils on smallholder farms in SSA remain active (Giller et al. 2009). The results reported here have shown clearly that no-till systems can reduce labor input but at a cost of herbicide purchase—meaning that this option could be relevant to relatively wealthy farmers who can afford this initial investment demand. On the other hand, planting basins require very high labor input especially on land preparation but do not entail large capital expenditure (Nyamangara et al. 2014; Rusinamhodzi 2015). In situations of good soil structure such as clay soils, the basins can be maintained for several seasons thus reducing the labor input for land preparation during subsequent seasons. The cost of herbicides can also potentially diminish across seasons, as has been proven that their consistent and correct use result in long-term decrease in weed densities across seasons (Muoni et al. 2013; Muoni et al. 2014; Nyamangara et al. 2014), making basins attractive in the long term under agro-ecological conditions where they give a positive response (i.e., especially in low-rainfall areas).

Literature suggests that the labor demand increases in intercrops compared with sole-cropped systems (e.g., Rusinamhodzi et al. 2012). Our review did not conclusively support this but the variation in the data was large suggesting that confounding factors may have masked the effect. Furthermore, we specifically focused on the maize crop because of the variation in companion crops; including the companion crop in the analysis will increase the labor demand but also the total yield. While labor requirements can be very important for labor-constrained farmers, intercropping systems are more relevant to meet farmers’ goals especially in light of population growth and the concomitant decrease in farm sizes, as well as dietary needs of the households. In intercropping systems, increased weeding time has been associated more with hindered movements and the need to take care of crops with different growth rates than with increased weed densities (Rusinamhodzi et al. 2012) but this analysis indicates that this is not always the case. Furthermore, long-term use of fast spreading legume cover crops may actually result in little need for weeding (Mhlanga et al. 2015), and more labor will just be required for planting and harvesting activities to take care of the different planting and harvest dates whereas the total annual labor demand may be reduced.

Overall, the relationships between yield and labor responses have shown the development pathways needed to achieve SI. The data revealed four scenarios: (HYLL) high yield returns—low labor demand, (HYHL) high yield returns—high labor demand, (LYLL) low yield returns—low labor demand, and (LYHL) low yield returns—high labor demand. Data seemed to be fairly spread among the four quadrants. To address these particular situations, several investments are needed to complement the tested technologies: HYHL requires investments in mechanization and herbicides to reduce labor input while maintaining productivity, LYLL requires an increase in fertilizer inputs, improved water and weed and pest management and use of improved varieties. Scenario LYLL requires the combination of, e.g., increased fertilizer inputs, improved water management, improved varieties, and mechanization to reduce labor input, suggesting that the tested technologies were not suitable under the agro-ecological conditions of the studies. HYLL is the most desirable scenario showing SI although additional savings of labor and/or increased yields through attention to crop-limiting factors such as nutrients, water availability, and crop varieties can further improve the yield-to-labor relations. Published studies addressing several constraints in a step-wise manner (e.g., Sime et al. 2015; Sime and Aune 2016) clearly illustrate the benefits that can be gained.

4 Conclusions

The relationship between yield and labor input revealed four scenarios: high yield returns—low labor demand (HYLL), high yield returns—high labor demand (HYHL), low yield returns—low labor demand (LYLL), and low yield returns—high labor demand (LYHL). These relationships were significantly affected by rainfall regimes and thus can vary from one season to the next even though the same management intensity can be maintained. The analysis has shown that options can increase the yield-to-labor ratio but conclusions regarding specific practices are not strong due to the low availability of strongly linked yield and labor data, which calls for more original studies. Such studies should preferably report labor data disaggregated by cropping calendar activities to allow analysis of gains or losses in labor productivity for different tasks and at different times of the growing season, and also of potential implications regarding, e.g., gender workload and resource endowment. Going forward, there is also a need to evaluate labor use response rates where more than one limiting factor, e.g., both nutrient and water management, are addressed in a stepwise manner. Though net benefits were observed even in the absence of mechanization or herbicide (ridging, intercropping, and rotations), the greater benefit may be seen with a combination which may also sustain adoption by farmers. To optimize the benefits, these practices should be combined with moisture management, improved varieties, nutrient inputs, and pest control as determined by the local agro-ecological and socioeconomic setting.