Aggregation of activity data on crop management can induce large uncertainties in estimates of regional nitrogen budgets

Agriculture is a major source of anthropogenic N2O, with croplands contributing the majority of emissions. Denmark, with intensive agriculture and strong environmental regulation, seeks large GHG reductions, making accurate, policy-relevant modeling of carbon and nitrogen budgets essential. Process-based models like LandscapeDNDC can evaluate management effects but typically require aggregating detailed field data to regional scales, potentially introducing uncertainty. While prior work has examined aggregation of soil and climate inputs, the impact of aggregating management activity data (crop rotations, fertilization) on regional nitrogen budget estimates is not well quantified. This study asks how different management data aggregation strategies influence simulated yields and NB components at regional scale and quantifies the resulting uncertainties using six well-monitored Danish catchments (LOOP) from 2013–2019.

Previous studies show that aggregating soil and climate inputs can variably affect simulated yields, biomass, water balance, and nitrate leaching, depending on region and agro-climatic context. In water-limited regions, aggregation of precipitation and soil data can cause large deviations in biomass and leaching estimates. Two recent studies highlighted the importance of detailed management activity data: Constantin et al. demonstrated management and resolution effects on yields, evapotranspiration, and drainage for winter wheat and maize in Germany, and Butterbach-Bahl et al. showed activity data critically affects CH4 and N2O emission estimates in Vietnamese rice systems. However, the specific effect of aggregating field-activity data (e.g., crop rotations, fertilization timing and amounts) on regional nitrogen budgets remains insufficiently explored, motivating the present quantitative comparison.

Study area and data: Six Danish catchments from the NOVANA LOOP program (2013–2019) encompassing diverse soils, climates, and agricultural practices. LOOP provides detailed farmer-reported management at field scale (crop type, calendars, synthetic and organic fertilizer type, timing, amounts), as well as catchment-scale NB estimates via a semi-empirical method. Crop fractional cover and field boundaries were obtained from the General Farm Register (2011–2020). Soil inputs (texture, SOC, bulk density) were 30.4 m rasters for 0–100 cm depths, pH at 100 m; saturated hydraulic conductivity estimated via pedotransfer functions. Atmospheric N deposition was from the Danish Eulerian Hemispheric Model. Climate (10×10 km) included daily mean temperature, global radiation, and precipitation from DMI. Model: LandscapeDNDC (LDNDC) simulates plant growth, microclimate, water balance, air chemistry, and soil biogeochemistry. Simulations used 1 m soil profiles and management events (tillage, planting, fertilization, harvest, irrigation). Spin-up: 2001–2010 using repeated 2011–2020 inputs. Simulation period: 2011–2020; analysis for 2013–2019. Irrigation for irrigated crops was applied to avoid water stress: 4 events of 30 mm (total 1200 m³ ha⁻¹ yr⁻¹). Organic fertilizer characteristics followed published Danish slurry data. Aggregation approaches:

• Approach A (detailed): Sub-field resolution at 30.4 m using high-resolution soils and actual field-level management and true crop rotations for 20 major crops. Outputs for 2013–2019 were post-processed to match reported observations.
• Approach B (mono-cropping): Six mono-cropping systems per catchment on the representative dominant soil; management averaged at catchment level by crop and year. Results were area-weighted annually by each crop’s fractional cover to produce catchment-scale estimates.
• Approach C (dominant rotations): 20 representative Danish crop rotations (including cereals, rapeseed, potato, sugar beet, peas, silage maize, grass and perennial grass) on each catchment’s representative soil. Management was defined from catchment-level averages. As rotation data lack temporal alignment, outputs were averaged across 2013–2019 and weighted by long-term average crop fractional cover to yield long-term means (no annual series). Evaluation: Model yields compared to reported LOOP/NOVANA yields (with conversions to ensure comparability). Nitrogen budget components were compared among approaches; Approach A was also compared to NOVANA NB summaries (noting NOVANA pools certain losses and assumes zero annual net balance). Statistical comparisons included r² for long-term means (LTS) and annual series (YS where applicable), RMSE by catchment, and component-wise NB differences.

• Yield simulation accuracy: Approach A had the highest agreement with observations (LTS r² = 0.93; YS r² = 0.77, nYS = 175, nLTS = 27). Approach B: LTS r² = 0.92; YS r² = 0.72. Approach C: LTS r² = 0.77 (no annual series). All correlations were significant (p < 0.05).
• Catchment/crop yield performance with Approach A: Best catchment YS-r = 0.77 (LOOP1). Best crop YS-r = 0.78 (winter barley). Lowest crop YS-r = 0.41 (potato). Deviations may stem from limited site data for parameterization, cultivar differences not fully captured in calibration, potential overestimation in reported yields, and conversion factor uncertainties.
• Approach A NB components (means across LOOPs): Gaseous N losses totaled 24.3 kg N ha⁻¹ yr⁻¹, comprising N2 16.6 (3.6–23.9; 68.3%), NH3 5.6 (4.6–7.7; 23.1%), N2O 1.5 (0.9–2.9; 6.3%), NO 0.6 (0.3–1.0; 2.3%). NO3 leaching averaged 14.7 (6.9–27.9) kg N ha⁻¹ yr⁻¹. Total N in harvested products (grain, straw, cut grass) averaged 132.8 (115.9–159.1) kg N ha⁻¹ yr⁻¹.
• Aggregation effects on gaseous N losses relative to Approach A: Approach B overestimated by 31.4% (+7.6 kg N ha⁻¹ yr⁻¹). Approach C overestimated by 17.6% (+4.3 kg N ha⁻¹ yr⁻¹).
• Aggregation effects on NO3 leaching: Approach C strongly overestimated leaching by 204.9% (+30.2 kg N ha⁻¹ yr⁻¹; 44.9 vs 14.7 kg N ha⁻¹ yr⁻¹). Approach B yielded similar average leaching but with a different temporal/spatial pattern.
• Yields/biomass across approaches: Approaches B and C produced about 5.7% higher N in harvested biomass than Approach A (~+7.5 kg N ha⁻¹ yr⁻¹ on average), but overall yields were comparable among approaches.
• Comparison with NOVANA (Approach A): Simulated total N input averaged 186 vs NOVANA 204 kg N ha⁻¹ yr⁻¹, partly due to inclusion of marginal lands (extensive grassland) in simulations and differences in N deposition datasets. Simulated organic fertilizer inputs were lower than reported, while simulated N harvest was higher by ~14 kg N ha⁻¹ yr⁻¹, leading to higher simulated N-use efficiency (71% vs 54%). NOVANA does not separate gaseous vs aquatic N losses and assumes zero annual net NB, whereas LDNDC accounts for soil and biomass N pool changes.
• Spatial patterns (Approach A): Marked sub-field variability in NB components at 30.4 m resolution; ~30% of area showed negative net NB (N depletion), and ~43% had N-use efficiency between 60–80%. LOOP6 had highest N losses; LOOP2 had highest N-use efficiency. Annual soil C change averaged +261.7 kg C ha⁻¹ yr⁻¹ (150.0–449.3), with some years of loss.
• Temporal and catchment variability: Highest gaseous fluxes in LOOP6/7; peak in 2018 for LOOP7 (41.7 kg N ha⁻¹ yr⁻¹). Highest NO3 leaching in LOOP2 (mean 27.9; max 35.6 kg N ha⁻¹ yr⁻¹ in 2016). Highest biomass N in LOOP7 (avg 159.1) and LOOP1 (144.1) kg N ha⁻¹ yr⁻¹.

Aggregating management activity data introduces sizable and systematic biases in simulated nitrogen budget components at regional scale, even when yield simulations remain reasonably accurate. The mono-cropping approach (B) effectively captures anomalies but systematically overestimates gaseous N fluxes and underestimates other components because it cannot represent soil C and N sequestration dynamics and realistic rotations. The rotation-based approach (C) tends to substantially overestimate nitrate leaching and moderately overestimate gaseous losses, reflecting the loss of temporal specificity in management and rotation timing. These discrepancies are particularly consequential in regions like Denmark where sandy soils make NO3 leaching a key environmental pressure. The findings emphasize that for applications such as regional NB estimation, GHG inventories, and policy assessment, detailed management representation (Approach A) markedly improves fidelity of NB components and spatial targeting of mitigation (hotspots, field-to-field variability). Where computational or data constraints necessitate aggregation, modelers must weigh trade-offs and quantify uncertainties introduced by the chosen approach.

The study quantitatively demonstrates that simplifying the representation of crop management when upscaling to regional modeling can induce large, component-specific uncertainties in nitrogen budgets. While yields may be simulated reasonably under aggregated approaches, gaseous N losses and nitrate leaching can be substantially biased, with mono-cropping (B) and dominant-rotation (C) schemes overestimating total gaseous losses by ~31% and ~18%, and C overestimating nitrate leaching by ~205%. Detailed, field-level management (A) best reproduces observed yields and provides more reliable NB component estimates and spatial patterns needed for targeted mitigation. These results caution that differing management representations can compromise the accuracy and comparability of national/international nutrient budgets. Future work should enhance integration of high-resolution, time-specific management data (e.g., from administrative registers and remote sensing), improve soil and climate input fidelity, and expand calibration/validation across crops and cultivars to reduce model structural and parameter uncertainties.

• Reported yields and fertilizer/manure inputs (NOVANA) carry uncertainties due to farmer reporting and standardized N content assumptions; conversions (fresh-to-dry weight, C-to-dry weight) add error.
• Crop parameterization was based on limited site datasets and may not capture cultivar-specific differences, affecting yield accuracy for some crops (e.g., potato).
• Irrigation timing/amounts were assumed (triggered events), potentially contributing to discrepancies in NO3 leaching.
• Field-measured NO3 leaching upscaling (empirical models) and model structural differences contribute to mismatches; Approach A underestimates measured leaching at LOOP catchments.
• High-resolution soil maps and derived properties (e.g., Ks via PTFs) have known uncertainties affecting simulations.
• Approach B uses a representative soil and averaged management, and cannot represent soil C and N sequestration dynamics.
• Approach C lacks annual temporal alignment for rotations and management, precluding year-specific comparisons and contributing to leaching overestimation.
• Differences in atmospheric N deposition inputs between model and NOVANA affect NB comparisons.