Eight case studies of legume based cropping systems originally undertaken as independent trials by the authors were selected. These varied substantially in terms of region, farm system, and experimental design (Table 1). The studies included simple designs of short duration (3 years) to test the benefit of legumes on the yields of following crops, to long-term experiments that monitored the performance of complex legume-based rotations. The systems included arable only, pasture only (grazed, silaged), and mixed systems producing a wide range of legume and non-legume grain, silage and forage products. A number of cropping practices were also included in the systems, namely monocultures, mixtures, intercrops, undersown crops, catch crops and green manures. Together, the eight studies included 29 different main-crops, 59 different main-crop/sub-crop combinations (a sub-crop being defined as an undersown crop or one of the components of an intercrop) and 25 different crop sequences (Table 1).

A soil-surface N-balance analysis was performed in which N inputs and outputs were measured or estimated to allow calculation of N surplus (N input − N output) and N use efficiency (NUE, N output / N input; OECD, 2001, 2008) for each crop combination for each year of a crop sequence—defined here as a “crop-year.” The input and output data were measured or estimated at the scale of the experimental plot, independently for each plot replicate (where these were present). An N balance for the crop sequence was calculated for each experimental plot by summing the annual balances across the sequence. This sum was then divided by the length in years of the crop-sequence to produce an annual average crop sequence N balance, thereby accounting for the variation in the duration of the crop sequences.

The influence of legume cropping on N balances was analyzed in terms of N input (N-in), the contributions to this from the application of fertilizer (N-fert) and from BNF (N-fix), N output (N-out), N surplus (N-surp), the difference between N-in and N-out, and N use efficiency (NUE; N-out / N-in; OECD, 2008). The analysis also considered productivity in terms of the variables Biomass Output and Biomass Output Efficiency (BO and BOE, respectively). Together these are referred to as the “N balance variables.” The analysis was conducted in 2 parts, the first addressing the annual N balance variables for crop-year and second the annual average crop-sequence N balances.

The average annual rate of N-fix did not increase monotonically with the proportion of legume crops but peaked at an intermediate proportion of legume crops (Figure 2A). To test the significance of this pattern, a generalized additive model (GAM) was fitted to the data in which average BNF was modeled as a smooth function of the proportion of years of a crop sequence that included legumes, excluding the 100% legume sequences. The smooth term was significant [F_{(4.28, 290)} = 38.7, P < 0.001] indicating a peak in BNF of 50–100 kg ha⁻¹ (Figure 2A).

Though the variation in the average rate of N application between crop sequences was, to a large extent, independent of legume cropping, there was a significant reduction in N-fert at approximately 50% legume fraction [F_{(2.66, 291} = 8.83, P < 0.001; Figure 2B]. The trade-off between N-fert and N-fix, the two major sources of N, acts to minimize the sensitivity of total N-in to the extent of legume cropping (Figure 2C), although N-in was slightly elevated at intermediate levels of legume use [F_{(0.92, 293)} = 10.30, P = 0.002]. The trend in N output (Figure 2D) is similar to that of N-fix and N-in including a peak at intermediate levels of legume cropping [F_{(1.99, 292)} = 29.11, P < 0.001].

The trade-off between N-fix and N-fert is, in part, to be expected as a consequence of the negative relationship between N application and BNF assumed in the calculation of N-fix (Equation 2). However, the increase in legume fraction is also associated with a shift in cropping system. For example, many sequences with a legume fraction above 0.6 were of a similar mixed cropping system dominated by a grazed grass-clover ley. This is a largely vegetative combination, producing higher biomass and N output than other crops, but habitually given N fertilizer even though legumes were present in most years. The various compensations between N-fix and N-fert resulted in a broadly similar biomass output over a wide range of legume fractions (Figure 3), from around 25% legume inclusion.

The main questions to be answered by this analysis are whether BNF by legumes compensates for reduced input of mineral fertilizer and whether having a legume in a crop sequence reduces overall productivity. When analyzed by crop-year, the two measured variables, biomass and N fertilizer, were not tightly coupled i.e., large quantities of biomass were produced in the absence of fertilizer. Legumes had compensated by providing the N requirements in treatments that received no or reduced fertilizer. In all five assessments by crop-type (Table 3), the legume generated a higher N-in through a combination of measured N-fert and estimated N-fix, than the non-legume comparator.

The observation that a mixture of legumes and non-legumes increases productivity is consistent with the observations of a large European study that analyzed the productivity of grass swards with differing proportions of clover (Suter et al., 2015). They suggested that competition for N, N transfer, and niche complementarity could explain the synergies associated with such mixtures. In our study the effects of the legumes on productivity (here measured as output) was more complex and depended on how productivity was defined. Non-legumes, such as maize and grass, produced about 1.3-fold the biomass of legumes but about 0.75-fold the N in the output, consistent with the generally higher %N of legumes. This general conclusion was modified in favor of the legume for the one comparison that consisted of all-vegetative crops, namely the forage mix (e.g., clover and grass) and the forage non-legume (e.g., grass), where N-out and biomass were both higher in the grass-legume mixture. The highest biomass and N-out among all comparisons were produced by this vegetative forage (all others having some “grain” component). The factor probably responsible for the higher productivity of these forages was the absence of a reduced photosynthetic capacity that normally occurs during reproductive growth, as a result of the requirement of the grain for N, which usually triggers leaf senescence. There might also have been some niche complementarity between the vegetative legume BNF and the non-legume using applied N, resulting in the forage mix having greater output than the comparator (Lüscher et al., 2014).

The N-balance approach has the advantage that it allows systematic comparison of a wide range of disparate studies that could otherwise not be pooled for analysis, but inconsistencies can enter the analysis since no account is taken of internal cycling of N. Potential carry-over effects between years, e.g., through uptake from N pools in soil, “catch” crops that were purposely plowed and did not contribute directly to output, and persistence of legumes as volunteer weeds can all affect single-year estimates of N-balance variables. However, the examination of the balance over the sequence should negate, or at least reduce, these potential complications. When considered over the sequence, both N-out (Figure 2) and to a lesser extent biomass (Figure 3) showed a broad maximum at legume fraction between 0.3 and 0.7, within which highest values were between 0.4 and 0.6. In contrast, the two components of N-in varied systematically and generally in opposition in their contribution. Together these trends confirm the previous conclusion based on crop-year that BNF compensated for reduced mineral fertilizer in generating biomass.

Between legume fractions of 0.6 and 0.8, the contribution of N-fix decreased. Most sequences in this range were forages that typically included legumes, either as intercrops or in relay, but were also given fertilizer, even in years when the legumes were present, such that N-fix was estimated as a smaller input than it was from grain legumes. These sequences nevertheless, through a combination of N-fert and N-fix, maintained high N-in, and N-out and biomass. Therefore, the presence of legumes did not cause any systematic reduction in total output when assessed over the cropping sequence (Figure 3).

The rates of output and N-fix were measured here in experimental systems, and with rotations, that included those designed for productivity in low-input or organic conditions. Nevertheless, N-balance variables and yield were within the range of values commonly found in high input cropping. For example, the combined inputs of N by BNF and fertilizer, N-in, ranged from 56 to 174 kg ha⁻¹ for the legumes, while mean N-fix ranged from 32 to 115 kg ha⁻¹ (Table 3, back-transformed means). In comparison, the typical mineral N input in conventional high input cropping in Europe is around 100 kg ha⁻¹ for a spring cereal and 200 kg ha⁻¹ for a winter cereal or oilseed (Fertiliser Practice, 2014). The most direct comparator for high input cereals is that of grain legume crops (the first comparison at the top of Table 3). Here, N-in for the grain legume was 108 kg ha⁻¹ and biomass output 2.4 t ha⁻¹; while N-in for the non-legume grain was 85 kg ha⁻¹ and biomass output 4.3 t ha⁻¹. Given that grain yields for high-intensity cereals are typically 4–8 t ha⁻¹, the lower value being for spring crops, (Sylvester-Bradley and Kindred, 2009) the non-legume grains in this study are comparable with general yields, while the legume grains show the typical higher N and lower biomass in yield.

Higher values of N-fix than those recorded here should be feasible. While the agronomy underpinning the trials was advanced by conventional standards, no particular attempt was made to maximize the amount of N-fix. Indeed, the common inclusion of N-fert in many of the sequences would have had the opposite effect. Moreover, values of N-fix estimated here were for current varieties that have been bred largely for conventional high-input cropping systems where there is generally much residual N. The potential for improving BNF within existing species and varieties of grain legume needs to be considered. For example, faba bean has been shown to have a significantly greater effect than other legumes on the yield of subsequent cereal grain crops (Wani et al., 1991; Hauggaard-Nielsen et al., 2009, 2012). Therefore, legume breeding should aim for higher levels of performance directed at low-input environments, cropping-systems focused on energy as well as food and fodder production (Porter et al., 2009; Jensen et al., 2012) and in both mixed cropping or intercropping (Brooker et al., 2014).

Legumes may have a range of benefits other than through BNF and potential as a green manure. Some degree of persistence by living legume plants to the next phase of the cropping-sequence was found in a number of plots and while there may be risks of such carry-over (Driscoll et al., 2014), the production benefits from the legume forages and grains that persist into subsequent crops need to be quantified and exploited. Below-ground phenology will also be important. Faba bean presents extensive tap roots capable of reaching deep into the soil profile, improving soil structure and benefiting the rooting profile of crops which follow them in the rotation (Rochester et al., 1998, 2001). In relation to this, the benefit of grain legumes to the yield of the next crop is greater than the benefit brought by carry-over of N in the soil. Some other factors are influencing yield positively, implying a cascade effect to other soil- and system-processes beyond enhanced soil N status (Danso and Papastylianou, 1992; Ehrmann and Ritz, 2014). This pre-crop benefit of grain legumes can lead to yield increases of up to 1.6 t ha⁻¹ in subsequent cereals (Preissel et al., 2015). This is consistent with our own findings which assessed the pre-crop benefits of legumes at the rotational level, and which also suggest higher non-legume yield potential may be achieved in years after legumes. This reflects the higher N inputs achieved by legume supported rotations.

However, in tackling problems such as reducing N losses from agriculture, legumes should also have a role but as part of broader systems of management. Greenhouse gas emissions, for example, depend on the complex interaction of several different biotic and abiotic variables (Snyder et al., 2014). While legume supported crop systems may limit N and C losses (Drinkwater et al., 1998, maize-soybean), management per se is a major factor in achieving this (Rees et al., 2013), especially the use of cover crops, both legume and non-legume grown to accumulate N in the soil for potential incorporation by later crops and to reduce GHG emissions (Thorup-Kristensen et al., 2003; Li et al., 2015).

A more critical appraisal of the traits targeted by legumes breeders should consider these wider system benefits (Sinclair and Valdez, 2012). Improvements are feasible e.g. soybean demands manipulation by either agronomy or re-breeding to render it as a net “contributor” of N, as opposed to a net “taker” from the system (Alves et al., 2003; Perez de la Vega et al., 2011; Baddeley et al., 2013). Moreover, the purpose of legumes in mixed forage systems, both examined here and in the wider literature, is not simply to contribute BNF or in-field benefits, but to bring wider system benefits beyond the farm level e.g., as a forage with particular characteristics or through a combination of outputs that optimize biomass and N content (Danso and Papastylianou, 1992; Crews and Peoples, 2004; Foley et al., 2011).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer SDSK and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.