The same principle but different approaches
In principle, NUEs are defined and quantified to assess how much of the N input to a system is used effectively by the product and harvested from the system5. In practice, three major approaches developed under different experimental and operational settings can lead to divergent results and possibly differing interpretations, even for the same experimental site and for the same growing season (Fig. 1a, b, c). A conceptual framework that depicts the linkages and differences among the three approaches is visualized in Fig 1d. While all three approaches share harvested N (HNT) and N fertilizer inputs (FN) as part of their calculation formulas, the relationship among NUEs obtained with these approaches can be determined by harvested N in the control plots (HNC), fraction of harvested N derived from in-season-applied N fertilizer (%Ndff), and nonfertilizer N inputs (NFN) measured in NUEdiff, NUE15N, and NUEbala approaches respectively. Mainly due to legacy effects of previous years’ fertilizer additions, points B and C in Fig. 1d (determined by HNC and %Ndff respectively) are usually above reference point A (determined by NFN), suggesting a higher NUEbala than that obtained with the other two approaches. Such differences are prevalent, especially in short-term experiments (e.g., experiments for one or several growing seasons).
To showcase the differences among approaches, we constructed an example for China’s cereal cropping system based on the synthesis of values from the literature (numbers in the brackets of Fig. 1d). We developed this example based on mean values from the literature (mainly obtained by national on-farm surveys and field trials) instead of a single experiment because few studies have tested all three approaches simultaneously, and experiments for a single site may lack representativeness. The results suggest that even for the same cropping system (i.e., Chinese cereal cropping system in this case) with consistently defined N inputs and harvest rates, different NUE approaches will result in different values with the following trend: NUE15N (0.24) < NUEdiff (0.32) < NUEbala (0.52). Consequently, it is critical to understand the causes for the differences among those approaches to better achieve efficient N management.
Mind the gap: differences and connections among the three approaches
The most obvious difference among the three approaches is that NUEdiff and NUE15N assess the efficiency of fertilizer use, while NUEbala assesses the efficiency of N inputs beyond fertilizer (FN + NFN). However, when the bioavailability of NFN (BANF, long-term uptake% of NFN) is the same as that of FN (BAF), then the NUE for FN use based on the balance approach (NUEbala_F, = (HNT – NFN × BANF)/FN) is the same as NUEbala (Supplementary Fig. 1). It is possible that BANF is higher than BAF, but even if all of NFN can be converted to crop product during the same growing season (i.e., BANF = 1.0), the adjusted NUE for fertilizer use based on the N balance approach could still be lower than NUEdiff and NUE15N14. Taking our Chinese cereal cropping system case as an example, NUEbala_F ranges from 0.38 to 0.66, depending on the bioavailability of NFN (Supplementary Fig. 1). However, even the lower boundary of NUEbala_F is still higher than NUE15N (0.24) and NUEdiff (0.32), indicating other important drivers for the differences between NUEbala and the other two approaches.
The major driver for the remaining differences between NUEbala and the other two approaches is the legacy effect of fertilization during previous seasons (Table 1). Fertilizer N inputs not only supply the plant N needs for the season of application but also replenish soil N and support the maintenance of the long-term N supply in the soil10. For example, the meta-analysis of Yan et al.14 shows that only 42% of N fertilizer is recovered by crop products during the same growing season, while 34% is retained in the soil, which could be utilized in future growing seasons or lost to the environment.
With the exception of newly formed croplands, most farms have a history of fertilizer or manure amendments. This commonplace legacy or replenishment effect of fertilizer application indicates that a long-term view of fertilizer efficiency is necessary in most cases for understanding the NUEdiff and NUE15N approaches (Supplementary Fig. 2). However, often due to logistical constraints (e.g., funding and site maintenance), the experiments for determining NUE are usually conducted for one to several growing seasons with different considerations of legacy effects, leading to divergence in the results. With single-season observations, the NUEdiff and NUE15N approaches reflect the fertilizer recovery rate during the current growing season, with no or largely discounted consideration of legacy effects. In contrast, the NUEbala is based on the assumption of mass balance in soil N stocks, and does not exclude the long-term legacy effects of N inputs but may over- or underestimate the actual NUE if significant mining or accumulation of soil N occurs during the year of observation (Table 1). Therefore, the difference between single-season NUE15N and NUEbala (e.g., the difference between 0.52 and 0.24 in the Chinese cereal cropping system case) can be used to estimate the size of the legacy effect of applying 15N fertilizer (Fig. 1d).
If the observation period could be extended from a single season to multiple years or even decades, the assessment outcomes from the three approaches will change due to the diminishing legacy effect: the HNC will likely decrease as soil N stock is continuously depleted without replenishment from additional N inputs, while the cumulative %Ndff will likely increase as more 15N will be recovered since the pulse addition of 15N tracer (i.e., points B and C will likely to move towards point A in Fig. 1d). Consequently, the observed NUEdiff and the cumulative NUE15N will likely increase as observation is conducted for a longer period of time (Fig. 1e). Moreover, the NUEbala may fluctuate due to the year-to-year yield variation, but the moving averages of the NUEbala over several years under consistent management practices tend to be stable, as the average annual soil N stock change tends to be negligible compared to the annual N fertilizer inputs over a long period of time17. Consequently, the assessments from the NUE15N and NUEdiff approaches tend to converge with that from the NUEbala approach for long-term observations (Fig. 1e).
The difference between the NUEdiff and NUE15N assessments could be attributed to the following reasons: 1) In addition to directly providing N for plant uptake, fertilizer inputs may change the in-season soil N supply by promoting root-mediated or soil microbe-mediated soil N turnover or abiotic adsorption-desorption exchange, leading to the differences between NUEdiff and NUE15N13, 14, 18. This effect tends to increase the harvested N in the fertilized plots of the NUEdiff approach but does not increase the recovered fertilizer N in the NUE15N approach. Consequently, it may lead to a higher NUE value for the NUEdiff approach. 2) Given the “law of diminishing returns” between N inputs and N yields, the NUE for FN is inherently lower than the NUE for NFN (slope of line DB < slope of line BO in Fig. 1d). Consequently, NUEdiff, which is based on the different yield responses to two N input levels, could show a lower NUE result than the approaches dealing with only one N input (i.e., NUEbala or NUE15N). Overall, NUEdiff can be higher or lower than NUE15N, depending on which reason dominates, but the difference is generally low in most reports (e.g., < 0.1 in Table 1).
Choosing a proper approach
Overall, when the observation is conducted over a long term and the NFN is negligible, the three approaches for NUE assessment (NUEdiff, NUE15N, and NUEbala) tend to converge and the assessments from these approaches could be compared for informing management practices. However, in practice, these approaches are often implemented for only one or a few growing seasons due to many logistical constraints, and the NFN sources are usually diverse and their quantities are uncertain. Therefore, the comparison of NUE assessments from different approaches may be biased by the inherent differences associated with these approaches, and it is essential that the assessment approach is carefully chosen and consistently implemented among practices or regions to avoid such bias.
Fortunately, a proper approach can be chosen by identifying the goal of the assessment and comparison, as well as the associated logistical requirements, including cost. For example, NUEdiff is a straightforward and cost-effective approach to assess the short-term response of harvested N to N fertilization under different environments or management conditions. The NUEdiff approach is suitable for experimental situations, such as research stations, but it is often difficult to perform in an actual farming operation because it is hard to convince farmers to devote some portion of their land to low yields due to lack of N fertilization. NUE15N is most accurate in tracing the fate and distribution of N fertilizer in the soil-crop system, but it is usually applicable at relatively small space-time scales due to the high cost of 15N-materials and 15N-measurements.
In comparison, NUEbala has great advantages in evaluating the resource and environmental performances of N inputs in crop production, especially when the averaged change in soil N stock is small or even negligible compared to the annual N inputs during the observation period. 1) Given the definition of NUEbala and soil N balance, “1–NUEbala” can be considered the fraction of N lost or subject to lose to the environment11. Applying a similar formula to NUEdiff and NUE15N, N losses to the environment are potentially overestimated due to the legacy effect (description of such overestimation existed widely in published papers12, 13, 19). 2) NUEbala is based on data more readily available from farm to national scales. For example, grain yield and N fertilizer rates are typical values used in farmers’ bookkeeping, and they are commonly collected in national or regional surveys or statistics. Admittedly, the data for some NFN sources (e.g., biological N fixation) are still not widely available, and using default values may introduce uncertainties. Such uncertainties could be important to consider when the NFN is not negligible (e.g., > 5%) compared to the fertilizer N rate.
It should be noted that, despite the approach, short-term observations are limited in reflecting the long-term effect of changing management practices15. First, crop yield and N uptake vary from year to year due to changing weather conditions and consequently affect the NUE assessment. Second, the impact of the new management practices on yield and NUE may be buffered by a soil legacy effect. Agronomic measures combined with fertilizer rate reduction are often reported in the literature as being effective for improving in-season NUEdiff or NUE15N while maintaining crop yields20. However, it must be recognized that some of these measures may only improve the soil environment and promote the release of previously accumulated nutrients in the short term and may eventually deplete soil N supply capacity, as well as yield and NUE, in the long term. Accounting for N balance and NUEbala may help to relieve part of the concerns about soil depletion, but NUEbala assessment tends to be biased when the soil N stocks are not in a steady state (Table 1). Consequently, to assess the impacts of different management practices on NUE, we recommend long-term observations (e.g., over a decade, but at least beyond three growing seasons) when possible.