Three databases (Scopus, Web of Science and ScienceDirect) were used to search for published peer-reviewed literature on cereal–legume intercrop systems from 1980–2022, using the PRISMA methodology (Figure 1). The search terms used were (“intercrop*” OR “mixture*” OR “multicrop*”) AND (“water use*” OR “water use efficiency” OR “water productivity”) and (“Kidney bean” OR “Common bean” OR “Lima bean” OR “Adzuki bean” OR “Mung bean” OR “Black gram” OR “Scarlet runner bean” OR “Ricebean” OR “Moth bean” OR “Tepary bean” OR “Horse bean” OR “Broad bean” OR “Field bean” OR “Garden pea” OR “Pea” OR “Protein pea” OR “Chickpea” OR “Cowpea” OR “Pigeon pea” OR “Lentil” OR “Bambara groundnut” OR “Vetch” OR “Lupins” OR “Lablab” OR “Jack bean” OR “sword bean” OR “Winged bean” OR “Velvet bean” OR “Yam bean”) and (maize OR rice OR wheat OR durum OR barley OR sorghum OR millets OR oats OR triticale OR rye OR fonio OR teff OR “Wild rice” OR spelt OR einkorn OR emmer OR kamut OR “Canary grass” OR quinoa OR amaranth OR buckwheat OR kaniwa OR pitseed OR goosefoot).

The initial search retrieved 1,333 articles; after that, articles were screened for duplicates, and 692 remained. We retained articles written in English and excluded those not written in English. From the remaining 662 articles, titles and abstracts were examined to check whether studies were field experiments comprising grain cereal and legume intercrop systems. We excluded all articles not considering grain cereal and legume intercrop systems. Following the screening, 54 abstracts remained. Full-length articles were only considered if they measured plant parameters included yield (Y) and water use {WU -[evapotranspiration (mm), rainfall (mm) received and/or water used (m³)]} and the experiments presented yield (in kg ha^–1, t ha^–1, and g m^–2) values for monocrop treatments for both intercrop component crops. This was to allow yield and nutrient yield comparisons between monocrops and intercrops to be compared. Articles that did not meet these criteria were excluded. Overall, 9 articles met the inclusion criteria (refer to Supplementary Information 1, Table 1 for the PRISMA flowchart). Since the work was largely exploratory and, given the limited number of retained articles, we did not consider bias. A glossary of key terms has been provided in Supplementary Information 2.

Multiple data records were extracted from each publication based on the number of experiments and appropriate treatments within experiments in an article. We extracted site-specific data (geographic coordinates and mean annual rainfall (mm) (Supplementary Information 1, Table 2), management data (plant densities, fertilizer and water management), yield, water applied as well as water productivity (Supplementary Information 1, Table 3). Data were directly extracted from published tables or digitized graphs using WebPlotDigitizer (29). All grain yield data were presented as t ha^–1. Only treatment mean values were extracted regardless of the number of replications.

The nutrient yield of intercrop systems was calculated based on an equation adapted from (34) as follows

where NY_C and NY_s are the nutrient yield for cereal (C) in the monocrop system and corresponding intercrop system (S), Y_C is the grain yield of the cereal in the monocrop system (C), Y_IC is the grain yield of the cereal component (C) in intercrop system (S), Y_IL is the grain yield of the legume component L in intercrop system S, NC_C is the nutrient concentration of the cereal, and NC_L is the nutrient concentration of the legume component.

Nutritional water productivity (NWP) of intercrop systems was calculated as a ratio based on a formula adapted from (35):

where NWP of cereal monocrop (C) or intercrop (S) system is the nutritional water productivity, WP of cereal (C) and intercrop (S) system is defined as the amount of agricultural output produced per unit of water used. Water productivity (25) for monocrop and intercrop systems was calculated as

Publication bias is a challenge in meta-analysis. This bias often occurs when the published studies report larger or more significant effect sizes (e.g., the effect of a treatment). Also, published studies are more likely to have found significant and/or larger yield changes than unpublished studies. Our study acknowledges that if any publication bias was present, it impacted the results of this meta-analysis since we used the resultant yield data from each selected publication. However, due to the limited number of articles identified, we did not subject the data to any statistical correction. Using the R statistical software (version 3.6.0) (39), paired t-tests were used to detect the difference between the sole cereal and intercropped cereal. We also used generalized linear mixed analysis (GLMM) at 95% confidence levels to determine if differences existed between the intercrop system and the corresponding cereal monocrop for Y, WP, NC, and NWP. Descriptive statistics such as means, standard deviations, bubble chart box and whisker plots were used to analyses outputs. Box and whisker plots can show stability and general distribution of the data sets. The bar charts visualized the relationship between two or more variables and helped assess co-dependent variables. On the other hand, box and whisker plots can show stability and general distribution of data.

The intercrop data consisted of nine articles (Supplementary Information 1 and Tables 1, 2) representing five countries from two continents. The locations were China (3 articles), South Africa (2), Kenya (1), Ethiopia (1), and India (1). The cereal component crops included maize (Zea mays L.), pearl millet (Pennisetum glaucum), sorghum (Sorghum bicolor) and wheat (Triticum aestivum). At the same time, the legumes intercropped were dry beans (P. vulgaris), cowpea (Vigna unguiculata), pea (Pisum sativum), groundnut (Arachis hypogea), and chickpea (Cicer arietinum). The various crop species identified highlighted the potential of intercropping in improving crop diversity. Crop diversity is critical in food and nutrition security (40) and dietary diversity (41). Maize intercrop systems were dominant (five out of eight); this is consistent with the global importance of maize as a staple crop (40). Also, water use issues are more prudent in maize production systems due to the crops’ higher requirements than the other cereal crops (42–45). Most intercrop systems (six out of eight) were grown in semi-arid regions (average 550 mm annual rainfall), indicating the potential of intercrop systems to do well under conditions of low water availability. However, the limited number of articles found during the literature search highlights the need for more research focusing on WU and NY in intercrop systems.

Table 1 presents the grain yield (Y) of selected cereal crops and legumes (monocropping and intercropping) reported in the selected studies. The mean values for the cereal monocrops (n = 23 experiments) ranged from 0.9 to 11.0 t ha^–1; the highest and lowest cereal Y was observed from maize and sorghum, respectively. For the intercrop, the mean values for cereal Y ranged from 0.8 to 8.9 t ha^–1; similarly, maize and sorghum exhibited the highest and lowest Y, respectively. The mean Y for legume monocrops ranged from 0.1 to 3.9 t ha^–1, whereas for the intercrop system, it ranged from 0.1 to 2.1 t ha^–1 (Table 1). Groundnut and cowpea exhibited the highest and lowest mean Y for monocrop and intercrop systems, respectively. The differences in species mean Y could be attributed to differences in Y potentials for each crop species (46, 47). Compared to maize, groundnut and soybean, commercial crops with high Y potential, millet, sorghum and cowpea, are regarded as underutilized crop species with limited crop improvement and low yield potential (48).

On average, intercrop systems produced 14% [95% CI 0–4.5 t ha^–1] lower cereal Y than monocrop Y. The most considerable yield differences when comparing intercropping to monocropping were observed from the fox millet–groundnut (−53−20%), maize-dry bean (−35 to −10%), and wheat-chickpea (−32 to −12%) systems. The low Y observed for intercropped cereal and legumes relative to their monocrop systems could be attributed to increased resource competition between the intercropped crops. In multicrop systems, competition between component crops is always present as the crops require the same growth resources, e.g., water, nutrients, and solar radiation (4, 49). The competition for these resources is significantly increased when critical growth stages such as canopy expansion, flowering and grain filling overlap, as could have been the case for the aforementioned systems (50). Then again, where there are lower yield reductions, it could be that the cereal and legume crops are complementary. Chimonyo et al. (4) and Smith et al. (15) observed complementary interactions for the maize-pigeon pea intercrop system, which was attributed to asynchronous phenology and less competition for growth resources during critical growth stages. The low Y in the maize–cowpea system was consistent with low plant populations, nitrogen fertilizer rates, and, more importantly, low water availability (314 mm; Supplementary Information 1, Table 2). To reduce yield gaps in cereal–legume systems, there is a need to exploit crop interactions to manage competition between cereals and legumes. Under low water availability, manipulating agronomic practices such as sowing time, sowing density, and N fertilizer rate can enhance species complementarity and total productivity, optimizing productivity (50).

Water productivity for cereal monocrops ranged from 0.40 to 1.85 kg m^–3, whereas for legume monocrops, it ranged from 0.08 to 0.61 kg m^–3. The differences in WP between the cereal and legume crops can be attributed to differences in physiology (51). Several studies have shown that C₄ plants are more efficient at carbon fixation and have a higher WP than C₃ plants owing to photosynthetic pathways (52–54). This could explain why many of the C₄ cereals showcased in the study had higher WP than C₃ cereals and legumes. The differences in WP between the cereal or legume crops could also be attributed to the differences in Y potential across the different crop species (46, 47). The study results highlighted that intercropping (0.47 to 1.90 kg m^–3) improved WP by 19% [95% CI 0.0–0.44 kg m^–3] compared to monocropping. It could be assumed that improved WP could be due to increased capture and use of unproductive water (water that is not taken up and transpired by the plant and is lost to plant production through deep drainage or evaporation) (55). The wide range for WP was consistent with the observed variation in Y (correlation = 0.81, P < 0.05). Maize–groundnut (0.47 kg m^–3), maize – soybean (0.49 kg m^–3), and sorghum–cowpea (0.37 kg m^–3) intercrop systems showed the lowest WP, while maize–pea (1.90 kg m^–3) and millet–groundnut (1.49 kg m^–3) showed the highest WP (Table 1).

We calculated the nutrients that can be harvested per unit area of land (NY) for selected cereals and legumes (monocropping and intercropping) (Supplementary Information 1). Excluding the results of carbohydrate Y, this study indicated that intercrop systems had higher NY than corresponding cereal monocrops. Intercropping improved fiber, Fe and Ca Y by 10, 15, and 135%, respectively (Figure 2). The most substantial improvements in Ca Y were observed in the maize-soybean intercrop system, with a 658% increase compared to the corresponding maize monocrop (Figure 2). The maize-soybean intercrop system also significantly improved protein (63%) and Fe (152%). The improvements can be attributed to the legumes used within the intercrop systems. For instance, soybean was seen to have a comparative advantage over other legumes as they contain a more significant amount of protein, Ca and Fe (56, 57). In this regard, legumes could be considered a good, cheaper alternative to animal protein and Ca (58) and can improve the nutritional value of starch-based foods despite low bioavailability. When a starch-based food is consumed with legumes, it provides other complementary proteins (59) and enhances Ca absorption (60). Starch-based foods lack lysine and tryptophan found in legumes, and the sculpture-containing amino acids limiting in legumes are found in starch-based foods (61). It could be assumed that the protein quality of a starch-based cereal could be improved when consumed with a legume, thus, contributing to the reduction of protein-energy malnutrition (62–64). It is recommended that the assessment of protein content in foods, such as legumes, should be measured by the sum of individual amino acids (61, 65). However, these data are not always readily available, and it is acceptable to estimate protein content based on total nitrogen content (66). This study showed that intercropping cereals with legumes did not always improve Zn contributions. While legumes contain more Zn than cereal crops, the bioavailability and absorption are affected by the total protein content within the legume and phytic acid found in several cereal crops (67). The consumption of cereal and legumes with Zn-rich vegetables should be encouraged.

The results indicated that carbohydrates’ nutritional water productivity (NWP) values were moderately higher under intercropping than monocropping (Figure 3 and Supplementary Information 1, Table 8). The values ranged from 238 to 5,047 g m^–3 for intercrop systems, whereas for monocropping, the values ranged from 258 to 4,888 g m^–3. The increase in carbohydrate NWP and the NWP for all the nutrients under investigation agreed with our hypothesis that intercropping cereals with legumes would improve NY. From a water perspective, the results suggest that improving NWP might require less water to produce comparable yields in intercropping to those produced in cereal monocrops. However, when considering human nutrition and health, an increase in carbohydrate NWP could be detrimental to poor rural households that consume a high-energy diet. As suggested by our results, the legume component’s benefits are that any carbohydrate content reductions resulting from a cereal yield reduction are offset by legume carbohydrate content. In addition, there is a bonus of the addition of other nutrients. When comparing the results of monocropping and intercropping, there was an overall improvement in NWP for protein, fiber, Fe, Zn and Ca by 55, 52, 60,45, and 82%, respectively (Figure 2). Under low water availability and within cereal–legume intercrop systems, the slight improvement in carbohydrate yield can be considered a worthwhile trade-off for an overall improvement in NY. This improved balance between nutrients, especially protein and carbohydrates, can reduce protein-energy malnutrition and obesity–under and over-nutrition, respectively.

The most considerable improvement for NWP was for Ca, which showed an 82% improvement under intercropping relative to cereal monocrops (Figure 2). For Ca-NWP, intercropping (32 to 3,287 mg m^–3) indicated higher values than monocropping (21 to 945 mg m^–3). The improvements in Ca-NWP were associated with the legume species used in the systems. A closer look at maize systems showed that the highest Ca-NWP improvement was when soybean was used, while the least improvements were observed when peas were used. Soybean has high Ca content that averages 35.7 mg 100g^–1 compared to other legumes such as pea (4.9) and groundnut (19.3). As a commercially important food and feed crop, soybean has undergone significant genetic improvements for improved nutrient yield (68). Although soybean is not, particularly drought tolerant, intercropping it with cereals under water-scarce conditions could be viable for improving the overall system Ca NWP.

The estimated percentage contribution of a cropping system (cereal monocrop and intercrop) for a family of four (comprising of a male and female adult, a female adolescent and a toddler) is presented in Figure 4 and Supplementary Information 1, Table 9. An ideal cropping system for optimum nutritional benefits provided more than 100% estimated average requirement (EAR for a family of four for a year). Maize–pea and millet–groundnut systems and their corresponding maize monocrop systems provided more than 100% of the EAR for a family of four for carbohydrates, protein, Fe, and Zn. Our results illustrated that despite intercropping improving Ca yields, the improvements were inadequate to meet the EAR for a family of four. It was interesting to note that maize-pea systems could provide ten four-member families with EAR for carbohydrates for a year. Millet–groundnut systems produced the highest nutrient contribution of Zn. Most intercrop systems reduce the EAR ratio between carbohydrates and other nutrients. For example, the maize–soybean intercrop system had the lowest carbohydrates to Fe ratio (1: 0.7), millet–groundnut had the lowest carbohydrates to Zn ratio (1:1.5), and wheat–groundnut had the lowest carbohydrates to Ca to protein ratio (1:0.5:0.1). The result shows that different crop species combinations result in different contributions to household nutrition. This would suggest that more than one type of legume species or adding other crop species could improve the balance of NC for a household across the nutrients relative to carbohydrates.