DayCent is a process-based biogeochemical model that simulates the dynamics of both nitrogen and carbon cycling across various terrestrial ecosystems (Del Grosso et al., 2001). DayCent integrates environmental drivers to predict plant growth, soil organic matter (SOM) decomposition, trace gases and changes in other ecosystem parameters within the soil-plant-atmosphere continuum on a daily timestep. With intermediate complexity and the feasibility to be calibrated to local conditions, DayCent is widely used for evaluating ecosystem responses to land use change, management practices and climate variability (Del Grosso et al., 2005; Gurung et al., 2020, 2021; Laub et al., 2024; McClelland et al., 2025).

In this study, we used the DayCent17centEVI model version to quantify N budgets for two Swiss agricultural ecosystems: (a) croplands and (b) grasslands. The SOM in DayCent is split into three compartmental pools, namely active, slow and passive, with different potential decomposition rates. These SOM pools receive plant materials from above and belowground litter. Simulated N flows follow C flows depending on the C:N ratio which varies from 15 to 3 for the active pool, from 20 to 12 for the slow pool, and from 10 to 7 for the passive pool. For newly formed surface biomass, the C:N ratio is a function of the N content of the decomposed material, with higher C:N ratio for lower N content. The model accounts for both organic and mineral N and the relevant processes, including mineralization–immobilization turnover, nitrification and denitrification. These transformations are regulated by environmental factors such as temperature, soil moisture, soil texture, SOM content and oxygen availability. Nitrogen inputs are modelled through simulated atmospheric deposition and biological N fixation, and through input data to the model that record application of both organic and synthetic fertilizers. Simulated losses are outgassing of N2O, N2, NO and NH3 to the atmosphere and leaching (e.g., nitrate and organic forms). It is crucial to acknowledge that NH3 volatilization in DayCent is not sophisticated, which may lead to underestimation of NH3 emissions and overestimation of nitrate leaching.

DayCent enables explicit parameterization of agricultural management practices. It can dynamically accommodate crop systems (e.g., crop types and rotations), cultivation, irrigation, nutrient inputs (e.g., fertilization) and harvest, with the timing of each management event specified on a daily basis in simulations.

DayCent is driven by weather data and soil data. Three basic meteorological variables include: daily maximum and minimum temperature, and precipitation. We used the 1 km × 1 km resolution weather data from the Federal Office of Meteorology and Climatology (MeteoSwiss, 2025). Inputs of site-specific soil properties such as soil texture (sand, silt, clay), soil pH and SOM were obtained from a recently developed national soil database by the National Competence Center for Soil (“Kompetenzzentrum Boden”) (Stumpf et al., 2024). Soil properties originally available at 30 m × 30 m resolution were aggregated to the 1 km model grid using conservative remapping implemented in the Climate Data Operator (CDO) (Schulzweida, 2023). This approach preserves area-weighted means and is useful in upscaling environmental datasets to coarser modelling resolutions (Hashimoto et al., 2025). Because the model operates at 1 km spatial resolution, harmonization of all input datasets was required to ensure consistency across input variables. While this aggregation inevitably smooths fine-scale heterogeneity in soil properties, the most prominent features of the geospatial distribution of soil variables at the national scale are preserved. Other required soil properties including bulk density, field capacity, wilting point and saturated hydraulic conductivity were determined by the pedotransfer functions embedded in the DayCent utility programme (Saxton et al., 1986; Saxton and Rawls, 2006), and root fractions were used the default values in DayCent.

In addition to meteorological and soil inputs, DayCent also needs data of land use and management practices. Our modelling simulations focused on two major Swiss land use categories: croplands and grasslands. Historical land use and areas of croplands and grasslands were from Federal Statistical Office (“Bundesamt für Statistik”, BFS) (BFS, 2024). Grasslands in Switzerland are categorised into three major types: (1) meadows, (2) pastures and (3) summer pastures. In this study, we focused on permanent meadows, which are managed for grass production for livestock feed. According to the definitions used for agricultural subsidies and the national fertilizer guidelines (Sinaj et al., 2017), meadows are further divided into intensively-managed, less intensively-managed and extensively-managed meadows, depending on the intensity of management practices (fertilization level and mowing events). These categories correspond to categories defined for agricultural subsidies and the national fertilizer guidelines (Sinaj et al., 2017), meaning category-specific data are available. Annual areas of different cropping systems and grasslands are provided for 24 agri-climatic zones. These are defined in Wüst-Galley et al. (2020) and have similar broad climatic conditions and agricultural management; they incorporate geographical (i.e., regions) as well as topographical differences in Switzerland (i.e., valley, hill, mountain and summer pastures).

A crucial input for DayCent simulations is the amount of C and N applied to soils from both organic and synthetic fertilizers. Organic fertilizers comprise animal manure as well as compost, sewage sludge and digestates as assessed by the Swiss National Greenhouse Gas Inventory (NIR, https://www.bafu.admin.ch/bafu/en/home/topics/climate/state/data/climate-reporting/ghg-inventories/latest.html, last access: 28 October 2024). Animal manure, which represents by far the largest amount of organic fertilizer, is assessed considering livestock numbers, livestock species specific excretion rates for C and N and respective losses from stables and manure storage systems. The allocation of organic fertilizers to different crops or grasslands is carried out in according with the following factors, as described in Wüst-Galley et al. (2020): the tendency of farms to apply manure, slurry or poultry manure to different broad crop groups, from Kupper et al. (2022); the relative fertilizer requirements of different grassland types, across different elevation zones, as indicated in Sinaj et al. (2017). For synthetic fertilizer, we used the fertilizer import data from the NIR. We assumed that all imported fertilizer was applied to the fields in the corresponding year as ammonium nitrate which is the major type of synthetic fertilizer. Other management data such as cultivation, planting, fertilization and harvest were taken from the national “Principles for the fertilisation of agricultural crops in Switzerland” book (Sinaj et al., 2017) and modelling setups (Lee et al., 2020b, a).

During the 1990s in Switzerland, important nationwide agri-environmental policy reforms introduced subsidy schemes in order to mitigate nutrient losses. Meanwhile, there was a significant increase in livestock productivity (i.e., milk yield) during this period and in line with decreased livestock population without lowering production, resulting in decreased N inputs from livestock manure. These led to substantial reductions in synthetic and organic fertilizer application to agricultural soils. The fertilizer input data we used reflect these changes. In addition, the use of cover crops also forms part of the agri-environmental policy that was implemented in the late 1990s, with the aim of reducing soil erosion and nutrient leaching, particularly in winter months. However, this potentially relevant measure was not covered by our study due to insufficient statistical data. Other agricultural practices were also assumed to remain unchanged.

The DayCent model that we used has been calibrated and tested by previous modelling studies (Dos Reis Martins et al., 2022, 2024; Necpalova et al., 2018). We used reported values of parameters controlling plant growth (both C and N yields) and N processes (e.g., nitrification and denitrification) by these studies which evaluated against measurement data from several Swiss long-term experiments for croplands (Emmel et al., 2018; Hüppi et al., 2015; Krauss et al., 2017; Mayer et al., 2015) and Swiss Fluxnet sites for grasslands (Feigenwinter et al., 2023b). In this study, the regional simulations for Switzerland were performed at 1 km × 1 km grids for the time period from 1981 to 2020. In total, there are 14 926 and 36 140 simulated grids for croplands and grasslands, respectively. A complete round of DayCent simulations had two stages: historical spin-up and present baseline. We followed Lee et al. (2020a, b) and assumed five spin-up phases characterised in Swiss agriculture: (1) native forest (between 0 and 1399; until equilibrium), (2) emergence of agriculture (between 1400 and 1750), (3) agricultural revolution (between 1751 and 1850), (4) agricultural intensification (between 1851 and 1950), and (5) modern agriculture (from 1950 to 1980). After the historical runs, DayCent was kept running for the studied time period (i.e., 1981 to 2020).

We analysed total N inputs, N outputs, N surplus, NUE, N losses and soil N balance from the DayCent simulations, and quantified N budgets. The total N inputs (Nin) to agricultural systems (croplands and grasslands) include organic (Norg) and synthetic fertilizers (Nsyn), biological N fixation (NBNF) and atmospheric N deposition (Ndep). As mentioned in Sects. 2.1 and 2.2, BNF (including both non-symbiotic soil N fixation and symbiotic plant N fixation) and atmospheric deposition are simulated by the DayCent model, whereas organic and synthetic fertilizers are model inputs. 1Nin=Norg+Nsyn+NBNF+Ndep

Total N outputs (Nout) include N yields of harvested products (Nyield) and all forms of N losses (Nloss). 2Nout=Nyield+Nloss

N surplus is total N inputs minus N yields. 3Nsurplus=Nin-Nyield

The NUE is defined as the harvested crop or grass N (Nyield) divided by total N inputs (Lassaletta et al., 2014; Zhang et al., 2021) 4NUE=NyieldNin×100

For N losses (Nloss), we included gaseous losses (Ngas; NH3, NO, N2O and N2) and leaching (Nleaching; e.g., nitrate and organic N compounds) 5Nloss=Ngas+Nleaching

The soil N balance (ΔsoilN) is calculated by subtracting all N outputs including harvested crop N and N losses from total N input 6ΔsoilN=Nin-Nout=Norg+Nsyn+NBNF+Ndep-Nyield-Nloss

Specifically, N flows expressed as percentages of total N inputs were calculated as: 7%Nflow=NflowNin×100

Similarly, the relative changes in soil N stock were calculated as: 8%ΔsoilN=ΔsoilNNin×100

These results are shown in Figs. 5 and 6.

We used a Monte Carlo ensemble approach (200 iterations) to quantify the uncertainty in simulated N losses, NUE and soil N balance arising from key processes represented in DayCent. Nine model parameters regulating N leaching, nitrification and denitrification were selected for the analysis (details are presented in Table S2 in the Supplement). Parameter ranges were defined using probability distributions based on the default values and calibrated ranges reported in previous studies (Dos Reis Martins et al., 2022, 2024). Because these parameters also influence plant N uptake through their effects on N loss pathways, plant-specific parameters were not evaluated separately.

To reduce the computational burden associated with the large number of simulations, we created a representative subset of grid cells using Conditioned Latin Hypercube Sampling (cLHS). This subset (n=400) preserved the spatial variability of climate, soil properties and fertilization intensity across Switzerland. For each Monte Carlo iteration, DayCent was run for all selected grid cells using a unique set of parameter values. Simulated N losses, NUE and soil N balance were then calculated and compared with the baseline simulation. Uncertainty in national-scale estimates was quantified from the distribution of ensemble results and reported as 95 % confidence intervals (CI).

Our compiled N datasets show that livestock manure and synthetic fertilizers, the dominant N sources for crop and grass production in Switzerland, decreased substantially from 204 to 156 ktNyr-1 between the 1980s and the 2010s. This result reflects policy interventions introduced in the 1990s to reduce agricultural N losses. In Switzerland, this period marked a broader societal and political shift, moving priorities from a narrow focus on maximizing food production to more sustainable and environmentally responsible farming approaches. In 1993, agricultural policy was reframed (Decrem et al., 2007; Herzog et al., 2008; Spiess, 2011), with direct payments (subsidies) introduced within an agri-environmental scheme, replacing the earlier model of guaranteed government purchases (Herzog et al., 2008). Moreover, integrated and organic production systems were promoted by additional incentives, organic farming and other ecological programmes. The constitutional amendment in 1996 further reinforced this direction by formally recognizing the multiple roles of agriculture, including ecological stewardship. Cross-compliance were confirmed in 1998 through the Proof of Ecological Performance (PEP) (Decrem et al., 2007; Herzog et al., 2008; Spiess, 2011), which made direct payments conditional to farms maintaining balanced nutrient budgets. These reforms collectively led to a ∼ 25 % reduction in average synthetic N fertilizer consumption during the 1990s (Herzog et al., 2008), accompanied by a comparable decline in manure application due to decreased livestock numbers. By 2005, 97 % of agricultural land in Switzerland was reported being managed in accordance with PEP standards (Herzog et al., 2008).

DayCent simulations suggest that NUE increased in both croplands and grasslands over the simulated 40 years. Nitrogen yields in both agricultural ecosystems remained stable from 1981 to 2020, despite a decline in N inputs, resulting in notable increases in NUE (Figs. 2–4). The model results are consistent with data reported by the Swiss Farmers' Union, which showed stable yields for major crops between 1991 and 2013 (Figs. S2 and S3 in the Supplement). The most pronounced improvements occurred in the 1990s, coinciding with the implementation of policy measures aimed at controlling agricultural N surplus. Grasslands show higher N yields and NUE compared with croplands, with a steady increasing trend with relatively low inter-annual variability (Fig. 2C).

To further evaluate NUE, we applied the European Nitrogen Experts Panel (EUNEP) framework (Nitrogen Use Efficiency (NUE) an Indicator for the Utilization of Nitrogen in Food Systems) to assess improvements across four decades (from 1981 to 2020). The EUNEP framework classifies agricultural land into six regimes (Figs. 3, 4 and A1 in Appendix), with the characteristic operating space (COS) representing the optimal agri-environmental performance. COS is defined by: (1) efficient use of nitrogen (NUE between 50 % to 90 %), (2) satisfactory N yield, and (3) controllable N surplus (total N input minus N yield). Other regimes outside the COS correspond to distinct agri-environmental issues (see Fig. 3 caption for details). Our results indicate remarkable progress in both croplands and grasslands: in the 1980s, only 26 % of croplands and 15 % of grasslands fell within COS. By the 2010s, these percentages increased to 56 % and 77 %, respectively (Table 1). Extremely high N inputs to grasslands (> 350 kgNha-1yr-1) were largely abolished after 2011 (Figs. 4D and A1B in Appendix), and COS areas expanded geographically (into the central plateau where intensive crop production takes place) between 1981 and 2020, gradually becoming the dominant regime in both ecosystems (Figs. 3 and 4).

Croplands and grasslands show distinct NUE patterns (Figs. 3, 4 and A1 in Appendix). In grasslands, N yields generally increase with N inputs, whereas in croplands, higher N inputs often lead to larger N surplus and excessive pollution (Fig. A1 in Appendix). Croplands exhibit mixed improvements: although COS areas expanded in the 2010s, areas with insufficient productivity (IP) also became more prevalent (Table 1), primarily due to a shift from the excessive pollution/inefficient use of nitrogen/insufficient productivity (EP/IUN/IP) regime to the IP regime. By contrast, grasslands exhibit more consistent improvement, with higher percentage of COS areas and reduced prevalence of EP, EP/IUN and IP regimes (Table 1).

Risk of soil nitrogen mining (RSNM) is considered negligeable (< 1 %) across Switzerland according to the EUNEP framework (Figs. 3, 4, and A1 in Appendix and Table 1). A distinct cluster of cropland points shows low total N inputs (50–100 kgNha-1yr-1) but very high NUE (> 90 %) (Figs. 3 and A1A in Appendix). In the simulations this pattern reflects land use changes. Throughout the whole simulation period, if land use change took place, we assumed in the model that these areas were covered by grass-clover mixtures during the non-cropland years. This assumption keeps the model running in a consistent way. Fertilization was assumed to be absent for these non-cropland vegetated periods, with N inputs only from BNF and atmospheric deposition. Consequently, N yields are low because of no additional anthropogenic N sources, while NUE is high, reflecting efficient N use under near-natural conditions.

Aggregated N losses (gaseous emissions and leaching) from Switzerland's croplands decreased markedly over the past four decades, from 131 kgNha-1yr-1 in the 1980s to 100 kgNha-1yr-1 in the 2010s (Table 2 and Fig. S4A in the Supplement), with leaching accounting for ∼ 80 % of total losses and for more than 40 % of total N inputs (Fig. 5F). By comparison, N losses from grasslands are roughly half as much as croplands (Table 2 and Fig. S4A in the Supplement), but only decreased slightly, from 58 to 55 kgNha-1yr-1, despite a substantial reduction in fertilizer inputs.

Soil N stock dynamics of croplands and grasslands show contrasting characteristics (Fig. S4B in the Supplement). Widespread soil N depletion is found in croplands (Fig. 5H), resulting from larger N outputs (through harvest and losses) than inputs (as described in Methods). It is estimated that croplands have lost a cumulative 537 kgNha-1 between 1981 and 2020. The most rapid depletion at -23 kgNha-1yr-1occurred between 1995 and 2011. Long-term field monitoring also shows soil N depletion at several arable sites (Fig. S5 in the Supplement). At the same time, nationwide long-term monitoring of soil C stocks reported that topsoil (0–20 cm) total organic carbon (TOC) in Swiss croplands have declined from 62 to 55 tTOCha-1 between 1985–1989 and 2015–2019 (Wollmann et al., 2025). These substantial decreases in soil TOC over time may indirectly provide some evidence for accompanied soil N depletion as pointed out by our modelling results. By contrast, grassland soils showed positive N balance (Fig. 6H) and accumulated 728 kgNha-1 over the same period. The accumulation was the fastest in the 1980s and then gradually slowed down. In the final 5 years of the simulations, the soil N stocks in both ecosystems stabilized, suggesting that national mean soil N pools are approaching an equilibrium.

To synthesize N dynamics and assess agri-environmental performance of croplands and grasslands in Switzerland, we use a novel analytical framework that jointly evaluates N inputs, NUE, N losses and soil N balance, with the latter two expressed relative to total inputs. Compared with the EUNEP framework, this framework explicitly shows the magnitude of N losses and soil N changes. In both agricultural ecosystems, NUE declines with increasing N inputs, while higher N inputs are associated with larger N losses and shifts in the soil N balance (Fig. 7). At N inputs < 100 kgNha-1yr-1, the two systems operate efficiently: NUE largely exceeds 80 %, N losses remain around 10 %, and soil N pools are minimally disturbed. As N inputs increase (150–200 kgNha-1yr-1), croplands frequently lose more than half of N inputs and experience substantial soil N depletion, with decreases in soil N stocks reaching up to 30 % of total inputs (45–60 kgNha-1yr-1). These N depletions in croplands were not explicitly reflected in the EUNEP framework. At high inputs (> 200 kgNha-1yr-1), outcomes diverge: some sites accumulate N in soils when losses are < 50 %, while other places deplete soil N stocks. Under such high N inputs, the magnitude of N losses affects the soil N balance, with positive soil N balance associated with lower N losses. Croplands receiving > 250 kgNha-1yr-1 N inputs were common in the 1980s but rare in the 2010s, reflecting the decline of input-intensive practices. Grasslands consistently outperform croplands under comparable input levels, with higher NUE, lower losses and predominantly positive soil N balances (Fig. 7B). Grassland soils retained more N in the 1980s than in the 2010s, suggesting diminishing accumulation rates over time due to less fertilizer N inputs (see also Figs. 2, A1 in Appendix, and S4 in the Supplement).

Within the current model framework, our simulation results demonstrate at the national scale that NUE increases with a greater contribution of BNF-derived N and decreases with increasing fertilizer N (Fig. 8A ans B). In DayCent, symbiotic N fixation occurs only when mineral N is insufficient to satisfy plant demand, leading to an inverse relationship between BNF-derived N and fertilizer N inputs. Although the model does not distinguish the source of mineral N during plant uptake or loss processes, the scenario simulations indicate that replacing fertilizer sources by only applying a single type of fertilizer results in modest changes in total N losses (ranging from -7 % to 3 %), whereas reducing the total fertilizer input substantially decreases N losses by ∼ 20 % (as presented in Table S3 in the Supplement). These findings suggest that the amount of N input is a stronger determinant of N losses than the source of N input.

Although lower simulated relative N losses were associated with more positive soil N balances (Fig. 8E), the scenario simulations demonstrate that soil N stock changes are determined by the combined effects of N inputs, harvest removal and N losses (Table S3). A 20 % reduction in fertilizer application decreased total N inputs by 12 %–14 % because increased BNF partially compensated for lower fertilizer inputs (Table S3). Nevertheless, soil N depletion in croplands increased and soil N accumulation in grasslands declined, indicating that reductions in N losses alone are insufficient to maintain soil N stocks if accompanied by reduced N inputs. These findings highlight the need to optimize both N inputs and N retention when improving N management.

Swiss agriculture relies heavily on livestock manure (Table 2), which can supply much of crop and forage N demand. However, the heterogeneous distribution of manure across agricultural landscapes causes mismatches between supply and demand in space and time, and logistical constraints such as storage capacity and weather conditions complicate timely application. Hence, synthetic fertilizers remain a crucial supplementary N source to bridge these gaps, offering readily available N for plant uptake (Figs. 5B and 6B).

In addition to fertilizers, BNF is an important N input. In croplands, our simulations suggest that legumes and grass-clover mixtures in the crop rotation contribute 4 %–15 % of total N inputs (Fig. 5D), while in grasslands BNF accounts for 22 %–33 % (Fig. 6D). In both agroecosystems, we find that relying more on BNF and less on fertilizer tend to achieve higher NUE. Compared to fertilizers and BNF, atmospheric N deposition constitutes a smaller share of N inputs in our simulations (Figs. 5D, 6D and Table 2), and has less significant impacts on agricultural production at the national scale.

At the national scale, the simulations reproduced a progressive decoupling between crop yields and fertilizer N inputs from the 1980s to the 2010s. The simulated increasing NUE over recent decades reflects not only improvements in N management but also the diminishing marginal yield response to additional fertilizer inputs that exceed the agronomic optimum. This pattern is consistent with the well-established nonlinear response of crop yield to N fertilization, which indicates that further increases in fertilizer inputs are unlikely to translate into significant yield gains, suggesting that Switzerland has moved beyond the stage of input-driven intensification. The scenario simulations further support this interpretation: reducing fertilizer application rates by 20 % results in only modest reductions in simulated N harvest (3 %–6 %) while substantially decreasing N losses (∼ 20 %), suggesting considerable opportunity to further improve environmental performance without compromising agricultural production. Our findings place Switzerland among the “type III” countries described in global analyses (Lassaletta et al., 2014) – those capable of maintaining (or increasing) productivity while reducing N inputs.

While NUE improvements in Swiss agriculture are encouraging, the caveats are the persistent N losses (especially in grasslands) and negative soil N balance (in cropland). Although absolute N losses declined with decreasing fertilizer inputs, relative N losses showed only modest reductions in croplands and even increased in grasslands. Our sensitivity analysis suggests that the weaker reduction in relative N losses in grasslands may partly reflect the greater sensitivity of grassland N cycling to increasing temperatures, which offset part of the reduction expected from lower fertilizer inputs. A uniform 2 °C increase in air temperature increased simulated N losses by approximately 6 % in croplands and 16 % in grasslands, indicating that warming enhances soil N turnover and gaseous N emissions in DayCent. Consequently, the reduction in N losses resulting from lower fertilizer inputs may have been partially offset by temperature-driven increases in N cycling, particularly in grasslands. In addition, reducing fertilizer application substantially decreased absolute N losses but also altered BNF and soil N dynamics in DayCent, demonstrating that relative N losses are governed by the interaction among N inputs, plant response and internal soil N cycling rather than by fertilizer inputs alone.

Most national- and regional-scale studies focus on N inputs, outputs or surplus, often neglecting soil N stock dynamics. Current knowledge suggests that N surplus is usually larger than changes in soil N stocks (Zhang et al., 2015) and that only countries with insufficient N inputs have been found to undergo soil mining and soil fertility loss (Lassaletta et al., 2014; Zhang et al., 2021), so regional-scale soil N balance is understudied, especially in places with high N inputs. Our modelling results reveal that soil N depletion can occur in cropland soils despite high N inputs and thus highlights that also in places with high N inputs the soil N balance should be evaluated, because negative soil N balances in croplands represents a waste of valuable nutrient resources and a threat to future soil fertility and productivity.

The spatially explicit simulations showed that higher N losses and negative soil N balances were mainly associated with intensively managed agricultural regions dominated by croplands, where fertilizer inputs and harvest removal were higher (Fig. 5). In contrast, grassland generally exhibited lower N losses and mostly positive soil N balance (Fig. 6), consistent with higher N retention associated with continuous vegetation cover. Although the simulations were conducted at a fine spatial resolution that enables analysis of regional differences across Switzerland, the primary objective of this study was to assess national-scale temporal trends and compare two agricultural systems. A more detailed investigation of regional hotspots of NUE, N losses and soil N changes represents an important direction for future research.

Our simulations, supported by sensitivity analyses, indicate that soil N dynamics are governed by the combined effects of N inputs, harvest removal, climate and internal N cycling, rather than by any single environmental or management factor. Maintaining soil N stocks therefore requires balancing N inputs with harvest removal against N losses. Sensitivity analyses further showed that although reducing fertilizer application substantially decreased N losses, the accompanying reduction in total N inputs slightly accelerated soil N depletion in croplands. Moreover, increasing temperature enhanced simulated N losses, suggesting that climate warming may have partially offset improvements associated with reduced fertilizer use. Our simulations identified nitrate leaching as the dominant N loss pathway (Fig. 5F, Table 2; Fig. S6 in the Supplement), explaining the persistence of N losses despite declining fertilizer inputs. This simulated pattern is consistent with previous studies showing that cropland systems are particularly susceptible to N losses because annual cultivation disturbs soil structure and periods of bare soil increase vulnerability to nitrate leaching (Porwollik et al., 2022; Rupp et al., 2024). Excessive fertilizer application, especially in the form of nitrate, significantly increases leaching risk and thereby degrades groundwater quality (Misselbrook et al., 1996; Shepherd et al., 2001; Vinten et al., 1994), reflected by the widespread (∼ 35 %) exceedance of groundwater nitrate guidelines (> 25 mgL-1) across the Swiss Plateau (Covatti et al., 2025). In addition, wheat as the most widely cultivated crop in Switzerland (and the second most common use of cropland), is known to extract large amounts of soil N (Kraaijvanger and Veldkamp, 2020), suggesting that harvested crops can represent a substantial pathway of N export. Together, these factors can contribute to a negative N balance in croplands. Compared with croplands, grasslands with a more continuous plant cover sustain year-round N uptake. Diverse plant communities in grasslands also improve nutrient retention (De Vries and Bardgett, 2016; Leimer et al., 2016), resulting in higher NUE and lower N losses (Fig. 6, Table 2 and Fig. S7 in the Supplement). Extensive and deep rooting systems can also reduce leaching in grassland (Misselbrook et al., 1996).

For decades, agronomy has centred on enhancing crop productivity, but the adverse environmental consequences resulting from elevated N losses have shifted the paradigm towards agronomic sustainability, emphasizing ecological performance alongside yield optimization. The challenge, however, lies in simultaneously improving NUE and reducing N losses without inducing significant disturbances to soil N stocks. We propose that a robust framework for future N management must rest on the three interlinked pillars: NUE, N loss and soil N stock dynamics. NUE serves as an indicator of input efficiency, N losses capture environmental externalities, and soil N stocks indicate long-term system stability and resilience. By considering these dimensions together, policy makers and practitioners can design strategies that secure productivity without undermining ecological integrity.

This study demonstrates the effectiveness of coordinated policy intervention that resulted in improvements in NUE and reductions in losses. A successful case was also found in China, where targeted N management programmes between 2007 and 2017 led to simultaneous gains in agricultural and environmental outcomes (Duan et al., 2024). These examples underscore that system-level change is possible within decades when science, practice and policy are aligned.

Many high-income countries face similar challenges as Switzerland of balancing productivity with environmental goals, while low- and middle-income countries risk soil N depletion if inputs remain insufficient or face environmental penalties due to unsustainable intensification (Falconnier et al., 2023). In this study, we provide a transferable analytical approach that integrates NUE, N losses and soil N changes for evaluating agri-environmental performance at the national scale. Context-specific application of this framework could help identify “win–win” strategies that support food security while benefiting ecosystems functioning and resilience. Looking forward, future policy frameworks should encourage integrated N management that explicitly addresses productivity-pollution-soils nexus, possibly coupled with financial incentives to stimulate adoption of advanced nutrient management technologies.

The Monte Carlo analysis showed that the uncertainty associated with simulated NUE, N losses and soil N stock changes was generally smaller than the observed long-term trends, increasing confidence that the simulated temporal changes are robust. Nonetheless, the quantified N budgets in this study were estimated primarily using a modelling approach so that the results should be interpreted with caution. In addition to the evaluated model parameters, other sources of uncertainty were not quantified, such model inputs (e.g., livestock C and N data, and soil data). It is also worths noting that DayCent uses a simple “tipping bucket” module to represent water movement in soil layers, which can overestimate percolation fluxes (i.e., predict quicker drainage). This influences soil water content and consequently affects water-dependent processes (such as crop N uptake, leaching and N2O production). Moreover, DayCent lacks a sophisticated scheme for NH3 volatilization, leading to substantially underestimated NH3 fluxes. This can result in a larger soil nitrate pool because more ammonium is available for nitrification, subsequently enhancing nitrate leaching. Another uncertainty is related to absence of representing cover crops in the model, which have been found to improve multiple ecosystem functions, including total soil N (Liu et al., 2025) at a global level. For Switzerland, Herzog et al. (2005) estimated that the increased use of cover crops since the late 1990s reduced farm-level nutrient balances by 10 % until 2005. The impact of not including them in this study is that we potentially underestimated the reduction in nitrate leaching and therefore also in NUE, but based on the above-mentioned study, the impact is not high.

Overall, our results and findings are robust and provide valuable insights for agricultural policy implementation and the assessment of agri-environmental performance. This work also demonstrates the feasibility of spatially explicit N flow quantification. Future work on monitoring national N use (e.g., constructing N budgets and calculate N balance) can incorporate biogeochemical modelling as part of the methodology.