Estimates of agricultural nitrogen surplus across intensive food-producing regions diverge substantially depending on which dataset researchers choose for fertilizer sales, harvest yields, and animal manure applications. In a study published in Nature Sustainability, researchers led by Baibaswata Bhaduri, Giorgos Kourakos, Kenneth Miller, and Thomas Harter demonstrated that holding the underlying physical model structure and land-use patterns constant while varying input datasets creates wide discrepancies in calculated nitrogen loading beneath crop root zones. Using California’s Central Valley as an empirical test bed, the team evaluated how divergent datasets alter environmental loading calculations, showing that the resulting spread among nitrogen surplus outcomes is substantial enough to alter local pollution hotspot rankings, regulatory mitigation priorities, and screening-level evaluations of economic compliance burdens.
Agricultural nitrogen surplus, defined as the mass residual of total nitrogen inputs remaining after accounting for harvest removal, plant-tissue storage, surface runoff, and atmospheric losses, is a primary driver of groundwater nitrate contamination and related environmental degradation in agricultural basins. In intensive farming regions, calculating this mass residual requires comprehensive regional accounting of all chemical, biological, and organic nitrogen streams entering and leaving cultivated fields. However, researchers frequently encounter multiple conflicting sources of data for identical physical variables, such as fertilizer application rates reported by state regulatory agencies versus producer surveys, or crop yields recorded across county agricultural commissioner summaries versus federal statistics. When these differing records are incorporated into nitrogen budgets without quantifying dataset uncertainty, the resulting figures can misinform groundwater protection strategies.
To evaluate the direct influence of dataset selection on nitrogen assessments, the investigative team applied the CV-Groundwater Nitrogen Loading Model, known as CV-GNLM, across approximately 21 million 50-meter by 50-meter raster cells spanning the agricultural landscape of California’s Central Valley. CV-GNLM features a one-dimensional annual mass-balance module operating under the assumption of decadal steady-state storage, where the change in storage equals nitrogen inputs minus nitrogen outputs, constrained to zero. In this mass-balance formulation, total nitrogen inputs represent the sum of atmospheric deposition, irrigation water nitrate, commercial fertilizer application, dairy manure land-applied to associated cropland, exported dairy manure sold to non-dairy cropland, and biological nitrogen fixation from leguminous crops. Nitrogen outputs in the model balance encompass harvested crop removal, surface runoff, atmospheric losses, perennial plant-tissue storage, and potential below-root-zone nitrate loading.
Within the mass-balance calculations, the researchers parameterized specific physical loss and input pathways using distinct empirical rules and environmental observations. Atmospheric deposition, encompassing wet and dry inputs, was derived from the United States Environmental Protection Agency Community Multiscale Air Quality model at a 4-kilometer resolution and adjusted against empirical measurements. Nitrogen added through irrigation water was calculated from nitrate concentrations observed in public water supply wells. Surface runoff was represented across the entire regional modeling domain as a uniform constant of 14 kilograms of nitrogen per hectare per year due to limited spatial runoff data. Atmospheric losses, which capture denitrification, ammonia volatilization, and nitrous oxide emissions, were calculated as an overall 10 percent fraction of all applied nitrogen inputs, based on published literature values.
The allocation of livestock waste introduced detailed spatial rules within the modeling framework, distinguishing between on-farm manure disposal and commercial manure transfers. For land-applied dairy manure utilized on cropland directly managed by the producing dairy facility, the available nitrogen mass was distributed across the mapped receiving area in proportion to each raster cell’s crop-typical nitrogen requirement. By weighting this allocation according to standard crop-typical fertilizer rates rather than actual applied fertilizer amounts, the authors kept the organic manure distribution mathematically independent of the inorganic fertilizer terms that manure applications subsequently modify. For exported dairy manure sold as dry solids to non-dairy cropland, loading was calculated by dividing the total exported dairy nitrogen reaching a given receiving county by that county’s eligible receiving cropland area.
Biological processes unique to specific crop functional types were accounted for through separate input and storage terms. Biological nitrogen fixation was incorporated solely for leguminous crop classes and set to zero for all non-leguminous vegetation, drawing values from published groundwater-protection nitrogen budgets to represent atmospheric nitrogen captured by plant-microbe symbioses independent of added fertilizers. For perennial vegetation such as vineyards, tree fruits, and nut orchards, the mass-balance formulation incorporated a plant-tissue storage term. This accounting category captured nitrogen permanently accumulated in long-lived woody biomass, separating perennial structural retention from the nitrogen removed annually in harvested agricultural commodities.
To construct a rigorous scenario envelope for the year 2020, the authors integrated agricultural and regulatory data spanning 2018 through 2022, compiling records from the USDA Census of Agriculture, the USDA Agricultural Resource Management Survey, University of California Davis Cost and Return Studies, the California Department of Food and Agriculture fertilizer tonnage sales reports, and the Irrigated Lands Regulatory Program Irrigation and Nitrogen Management Plan database. These records were matched to the 2019 California Department of Water Resources spatial land-use map. For each agricultural crop class, the datasets reporting the maximum and minimum period mean values for fertilizer application and harvested nitrogen were selected as bounding constraints. Combining these fertilizer and harvest extremes with the presence or absence of perennial tissue storage and three distinct manure allocation levels—full manure, half manure, and zero manure—yielded 24 distinct scenario permutations across the Central Valley.
The resulting 24 permutations revealed extensive divergence in estimated below-root-zone nitrogen surplus, with the most pronounced uncertainties emerging on field crops receiving livestock manure applications. To determine how this dataset-driven divergence affects policy decisions, the researchers examined localized pollution hotspot identifications across geographic townships. Within each fertilizer, harvest, and tissue-storage configuration, agricultural pixels falling into the upper quartile of nitrogen surplus were categorized as hotspots. Townships were then scored by their total hotspot pixel counts and ranked from highest to lowest environmental burden on a normalized scale. Comparing these scores demonstrated marked rank instability, indicating that the choice of input dataset alone can change which geographical townships appear most severely impacted by nitrate leaching.
The study also evaluated the potential regulatory and economic consequences of dataset uncertainty by comparing model outputs against published groundwater-protection targets ranging from 10.4 to 51.6 kilograms of nitrogen per hectare per year. For each crop and scenario, the researchers determined the proportional loading reduction required to meet these targets, converting required load reductions into estimated net-revenue losses using an established convex economic response curve where 25 percent and 50 percent loading reductions correspond to approximately 3.5 percent and 14 percent net-revenue losses, respectively. Because this economic response function was applied uniformly across all crop classes, variations in projected economic burden reflected differences in estimated nitrogen surplus and target exceedance rather than divergent crop-specific economic structures, demonstrating how data choices cascade into regulatory cost projections.
Cross-model consistency was evaluated by comparing the CV-GNLM framework against a 2020 CV-SWAT root-zone simulation across 17 focal crop classes, using matched accounting configurations that aligned biological fixation, perennial tissue storage, and harvest coefficients while excluding explicit dairy sources. The authors emphasized that this exercise served as a structural consistency check rather than an independent validation, confirming comparable crop-level magnitudes and orderings between different models. Furthermore, the authors clarified that the modeled nitrogen loading residual represents potential nitrate flux exiting below the root zone rather than direct, real-time nitrate deliveries to water supply wells, because deep vadose-zone travel times, attenuation mechanisms, storage, and saturated-zone hydrogeology delay and reshape final groundwater responses. Consequently, the researchers conclude that regional, national, and global nitrogen assessments must explicitly propagate agricultural dataset uncertainty to prevent misaligned environmental management decisions.
Subject of Research: Agricultural data source uncertainty in nitrogen surplus modeling
Article Title: Agricultural data source choice as a dominant control on nitrogen surplus uncertainty
Article References: Bhaduri, B., Kourakos, G., Miller, K., & Harter, T. (2026). Agricultural data source choice as a dominant control on nitrogen surplus uncertainty. Nature Sustainability. https://doi.org/10.1038/s41893-026-01992-8
Image Credits: AI Generated
DOI: 10.1038/s41893-026-01992-8
Keywords: agricultural hydrology, nitrogen surplus, Central Valley, groundwater nitrate, CV-GNLM, mass balance, fertilizer data, manure management, vadose zone, water quality regulation, environmental modeling, hotspot prioritization, crop nutrient budgets, data source uncertainty, agricultural sustainability, nonpoint source pollution
Cite Scienmag News
Sloane Callahan. (November 12, 2026). Agricultural Data Source Selection Drives Substantial Uncertainty in Nitrogen Surplus Estimates. Scienmag. https://scienmag.com/agricultural-data-source-selection-drives-substantial-uncertainty-in-nitrogen-surplus-estimates/
Sloane Callahan. "Agricultural Data Source Selection Drives Substantial Uncertainty in Nitrogen Surplus Estimates." Scienmag, 12 November 2026, https://scienmag.com/agricultural-data-source-selection-drives-substantial-uncertainty-in-nitrogen-surplus-estimates/. Accessed 17 November 2026.
Sloane Callahan. "Agricultural Data Source Selection Drives Substantial Uncertainty in Nitrogen Surplus Estimates." Scienmag. November 12, 2026. https://scienmag.com/agricultural-data-source-selection-drives-substantial-uncertainty-in-nitrogen-surplus-estimates/








