The long tail of nitrate pollution in groundwater challenges governance of global water quality

Abstract

Groundwater nitrate pollution threatens drinking water safety. Legacy nitrogen (N) in the vadose zone creates a “long tail” of leaching that can persist for decades after policy actions. Here we model global nitrate dynamics in groundwater from 1961 to 2100. By 2020, shallow aquifers had accumulated 162 ± 6 Tg N as nitrate, exceeding the WHO drinking water limit across 10% of global land area. This contamination is sustained by a large nitrate reservoir of 4037 ± 214 Tg N in the vadose zone, which will continue to generate new contamination hotspots for decades. Even under an immediate transition to zero-N-surplus, 4% of affected regions are projected to remain above the WHO limit beyond 2100. To guide effective governance, we classify global croplands into four management archetypes and identify tailored strategies that balance water quality, food security, and soil sustainability. These findings redefine the temporal scope of governance, identify priority regions, and provide a science-based roadmap towards safe groundwater.

Groundwater nitrate pollution persists for decades due to a large legacy nitrogen reservoir in the vadose zone. Even with immediate emission cuts, 4% of affected regions will exceed WHO limits beyond 2100, requiring tailored governance strategies.

Introduction

Groundwater is the world’s largest available freshwater resource, accounting for approximately 99% of global freshwater stocks and serving as critical natural capital for both human societies and ecosystems¹. It supplies about one-third of global freshwater withdrawals² and is the primary drinking water source for up to 70% of the global population, making its quality a direct and critical determinant of public health and long-term sustainable development^3,4. Among the diverse contaminants threatening groundwater quality, nitrate (NO₃⁻) is one of the most widespread inorganic pollutants. Epidemiological evidence confirms that drinking water with nitrate-nitrogen (NO₃⁻-N) concentrations exceeding 11.3 mg N L⁻¹ directly causes infant methemoglobinemia (a life-threatening disorder that impairs blood oxygen transport)^5,6. Beyond this acute toxicity, emerging evidence links chronic NO₃⁻ exposure to colorectal cancer and other adverse health outcomes even at lower concentrations^7,8. A nationwide Danish cohort study of 2.7 million people identified elevated cancer risk at concentrations as low as 3.87 mg N L⁻¹ and estimated approximately 6000 attributable cases⁹; follow-up cost-benefit analyses suggest that lowering exposure thresholds could yield net societal benefits despite implementation costs¹⁰. Nevertheless, the World Health Organization (WHO) established its drinking-water guideline of 11.3 mg N L⁻¹ for NO₃⁻-N (equivalent to 50 mg L⁻¹ NO₃⁻) primarily to prevent acute methemoglobinemia, the best-documented health risk at the time of standard-setting¹¹.

Despite this clear safety standard, groundwater NO₃⁻ contamination has intensified dramatically in recent decades (Supplementary Fig. 1a, b). Its persistence, driven by a “long tail” effect of subsurface legacy N (defined as the phenomenon where NO₃⁻ pollution persists for decades or even centuries in groundwater, even after surface N inputs are reduced or ceased), has emerged as a key barrier to effective global water quality governance. Since the mid-20th century, the global adoption of intensive N fertilization practices, driven by the Green Revolution’s goal of boosting crop yields, has consistently exceeded the capacity of crops and soils to retain and recycle reactive N¹². This contributes to a planetary-scale environmental risk: the N component of biogeochemical flows lies in the high-risk zone, with current intentional N fixation for agriculture (~ 165 Tg N y⁻¹) exceeding safe limits by more than twofold¹³. This global N surplus has led to a substantial N accumulation in the vadose zone (the unsaturated layer between the land surface and groundwater table, with a global average depth of ~ 22 m; Supplementary Fig. 2)¹⁴, which acts as a reservoir sustaining NO₃⁻ leaching into groundwater. This contamination long tail has transformed NO₃⁻ pollution from a localized agricultural concern into a transboundary challenge, as it decouples short-term policy actions (e.g., N input limits) from long-term water quality outcomes and complicates cross-regional governance coordination across disparate time scales. Major agricultural regions, including the U.S. Central Plains, Europe’s Rhine River Basin, and densely cultivated regions of South and East Asia, already face this disconnect, with groundwater NO₃⁻ levels failing to improve despite decades of regulatory intervention^15–19. Groundwater quality monitoring has thus been integrated into the UN’s 2030 Agenda for Sustainable Development, with Sustainable Development Goal 6 (SDG 6) aiming to ensure universal access to clean water and sanitation²⁰. Addressing the NO₃⁻ pollution long tail is therefore indispensable for closing the gap between global water quality governance and on-the-ground progress toward SDG 6²¹.

To address escalating NO₃⁻ pollution, numerous countries and international organizations have implemented N management policies since the 1970s (Supplementary Fig. 1c), including the EU Nitrates Directive (regulating agricultural N inputs)²² and the U.S. Clean Water Act (setting nutrient discharge standards)²³. These regulatory frameworks establish stringent limits on agricultural N inputs and losses, yet their effectiveness in improving groundwater quality has been severely undermined by the subsurface legacy N long tail. For instance, isotopic tracer studies in the Paris Basin’s Cretaceous aquifer show that NO₃⁻ transit times through the vadose zone exceed 50 years²⁴, meaning current groundwater NO₃⁻ levels still reflect 1970s fertilization practices, when many of these policies were first introduced. This temporal decoupling is even more pronounced in regions with thick vadose zones, such as the arid southwestern U.S. and China’s Loess Plateau^25,26. This creates a critical governance challenge: even if N inputs are halted immediately, existing NO₃⁻ reserves in the vadose zone will continue to leach into aquifers, causing contamination that exceeds safety thresholds and frustrating global efforts to restore water quality. Furthermore, groundwater NO₃⁻ contamination extends beyond drinking water risks to broader environmental degradation: lateral groundwater discharge to rivers, lakes, wetlands, and coastal zones, particularly from shallow aquifers, drives surface water eutrophication, toxic cyanobacterial blooms, and biodiversity loss^27,28, underscoring the urgency of effective governance. While regional studies have investigated groundwater responses to subsurface N legacies^29–31, global-scale projections of the long tail’s impact on NO₃⁻ dynamics, and its implications for harmonizing water quality governance, remain limited. Moreover, recent research documenting substantial global vadose zone NO₃⁻ accumulations³² has not yet been integrated into global governance frameworks, leaving policies misaligned with the actual timeline of contamination recovery³³.

To fill this knowledge gap, we develop a global groundwater NO₃⁻ concentration dataset (1961–2020) that explicitly quantifies the long tail of vadose zone legacy N effects, using the Integrated Model to Assess the Global Environment—Global Nutrient Model (IMAGE–GNM) at a 0.5° × 0.5° spatial resolution. Our objectives are to: (1) project global groundwater NO₃^- concentrations from 2021 to 2100 under four future scenarios (business-as-usual (BAU), 50 % N surplus reduction (NSR50), 75 % N surplus reduction (NSR75), and zero-N surplus (ZNS)) under the SSP126 climate pathway, with a focus on how the long tail modifies pollution recovery timelines; (2) quantify the onset and duration of the groundwater NO₃⁻ contamination long tail using a comprehensive assessment framework, and identify priority regions for governance interventions; and (3) evaluate the feasibility of achieving SDG 6 water quality targets and develop region-specific mitigation strategies that account for the long tail, strengthening the link between local actions and global governance. Although climate change can alter precipitation, seepage, and groundwater recharge, and thus nitrate leaching^34,35, here we focus on the SSP126 climate pathway to establish a low-forcing baseline, enabling a conservative assessment of legacy N impacts under comparatively favorable climatic conditions. Our results provide critical scientific support for aligning global water quality governance with the persistent nature of NO₃⁻ pollution, safeguarding groundwater security, and promoting sustainable socio-economic development.

Results

Our simulations show that deep groundwater NO₃⁻ concentrations worldwide remain below the WHO drinking water safety threshold throughout the study period, with only a tiny proportion of regions approaching this threshold (Supplementary Figs. 3, 4). In contrast, shallow groundwater exhibits far more severe NO₃⁻ contamination, posing greater risks to water supply and public health. We therefore focus the following analyses on shallow groundwater—more vulnerable to anthropogenic impacts and critical for rural drinking water supplies (Supplementary Note). These simulations of shallow-groundwater NO₃⁻ dynamics were validated against observations from 5849 monitoring wells across North America, Europe & Central Asia, South Asia, East Asia & Pacific and other regions, showing good agreement (R² = 0.61–0.79; RMSE = 1.95–6.95 mg N L⁻¹; Supplementary Fig. 5).

Temporal trends in groundwater NO₃⁻ storage in 1961–2020

From 1961 to 2020, modeled global NO₃⁻ accumulation in shallow groundwater increased at an average rate of 1.8 ± 0.1 Tg N y⁻¹, reaching a total stock of 162 ± 6 Tg N by 2020 (Fig. 1a). This global growth showed a clear inflection point around 1990, with the accumulation rate slowing from 2.82 ± 0.04 Tg N y^-1 during 1961–1990 to 0.45 ± 0.06 Tg N y^-1 during 1990–2020, primarily driven by policy-induced N reductions in Europe and North America. This global trend, however, masks striking regional disparities in storage dynamics. Europe and North America exhibited groundwater NO₃⁻ storage peaks in the 1990s, coinciding with the long-term implementation of the EU Nitrates Directive and the U.S. Clean Water Act. After this peak, both regions experienced sustained declines in groundwater NO₃⁻ storage, with an average reduction rate of 0.1 ± 0.02 Tg N y⁻¹ (Fig. 1b). In contrast, East Asia & Pacific (2.9 ± 0.2% y⁻¹) and South Asia (1.5 ± 0.1% y⁻¹) (World Bank regional classification; Supplementary Fig. 6) showed accelerating NO₃⁻ accumulation from 1961 to 2020, with rates approaching or exceeding the global average (1.7 ± 0.1% y^_1).

Fig. 1. Historical NO₃⁻ concentrations and temporal dynamics of shallow groundwater.

a Temporal variations in regional shallow-groundwater NO₃⁻ storage (solid-colored lines) and standard deviation from 1000 Monte Carlo simulations (colored shaded areas). The color gradient in the global panel transitions from blue (pre-1990) to red (post-1990), visually highlighting the inflection point around 1990 when growth rates slowed. b Annual change rates of shallow-groundwater NO₃⁻ storage at global and regional scales. Boxplots show medians (horizontal black lines), averages (black squares), outliers (black dots), interquartile range (25–75th percentiles, box edges), and ranges within 1.5 times interquartile range (whiskers), IQR = interquartile range. c Global spatial distribution of shallow-groundwater NO₃⁻ concentrations in 1961, 1980, 2000, and 2020.

Critically, the shallow-groundwater NO₃⁻ stock (162 ± 6 Tg N in 2020) is dwarfed by the vadose zone reservoir, which stored 4037 ± 214 Tg N in 2020 (Supplementary Figs. 7, 8), more than 25 times larger. This massive legacy N in the vadose zone will sustain groundwater contamination for decades. Sub-Saharan Africa, Latin America & Caribbean, and Middle East & North Africa maintained near-stable NO₃⁻ stocks over the 60-year period (Fig. 1a), reflecting relatively low but persistent N input intensities compared to other regions. Notably, post−1990 net increases in NO₃⁻ storage in developing Asia (24 ± 0.6 Tg N) fully offset net reductions in Western regions (7 ± 0.4 Tg N; Fig. 1a). This offset prevented meaningful global progress in mitigating shallow-groundwater NO₃⁻ contamination, despite decades of environmental regulation in Europe and North America.

Spatiotemporal dynamics of NO₃⁻ exceedance in 1961–2020

The global extent of shallow groundwater with NO₃⁻ concentrations exceeding the WHO safety threshold reflects a legacy of unmanaged contamination risks. Although the total exceedance area peaked in the 1990s (Supplementary Fig. 9), 10 ± 0.2 % of the global land area remained contaminated by 2020. These exceedance zones are concentrated in major agricultural heartlands such as Europe, North America, South Asia, and Sub-Saharan Africa (Fig. 1c), regions with high historical and current N fertilizer application rates.

Regional trends in exceedance areas further highlight the influence of policy and agricultural practices. From 2000 to 2020, Europe and North America reduced their NO₃⁻ exceedance areas at rates of 0.7 ± 0.1% y⁻¹ and 1.4 ± 0.2% y⁻¹, respectively. This reduction aligns with the long-term implementation of N policies from the 1970s and corresponds to documented declines in infant methemoglobinemia incidence in affected regions^36,37. In contrast, our modeling shows that South Asia and East Asia & Pacific have experienced persistent growth in exceedance areas since the 1960s, at rates of 1.3 ± 0.2% y⁻¹ and 3.7 ± 0.4% y⁻¹, respectively (Fig. 1c and Supplementary Fig. 9). This growth directly tracks the expansion of synthetic N fertilizer use in these regions, driven by rising food demand and intensive agricultural practices.

Scenario-driven trends in groundwater NO₃⁻ storage and exceedance areas

Simulations show that shallow-groundwater NO₃⁻ dynamics are strongly shaped by management choices, yet they exhibit irreversible legacies from past N inputs. Under the BAU scenario (unchanged management practices), shallow-groundwater NO₃⁻ storage remains near 2020 levels through 2100, with a coefficient of variation (CV) of only 2.2 ± 0.1%, indicating minimal interannual fluctuation. The exceedance area is similarly stable, with a CV of 2.1 ± 0.1% (Fig. 2 and Supplementary Fig. 9). This stability confirms that without targeted mitigation, current contamination levels will persist for decades. In contrast, the three N surplus reduction scenarios (NSR50, NSR75, ZNS) all drive declines in NO₃⁻ storage and exceedance areas, with the reduction magnitude scaling with N management intensity (Fig. 2). For instance, the ZNS (the most aggressive scenario) yields the steepest declines in storage and exceedance, while NSR50 produces modest but meaningful reductions.

Fig. 2. Temporal trends in global and regional shallow-groundwater NO₃⁻ storage under different nitrogen (N) management scenarios (1961–2100).

Solid lines represent averages from 1000 Monte Carlo simulations: gray (historical), red (Business-as-usual), blue (50 % N surplus reduction), yellow (75 % N surplus reduction), and purple (zero-N surplus). Colored shaded areas indicate standard deviations from 1000 Monte Carlo simulations.

Regional responses to these scenarios vary significantly, reflecting differences in historical policy implementation and current N input levels. North America and Europe & Central Asia exhibit stable or declining NO₃⁻ storage across all scenarios, with annual change rates ranging from − 2.1 to 0.2% and − 1.8 to 0.4%, respectively. This stability can be attributed to “policy-saturated” landscapes, where decades of regulation (e.g., the EU Nitrates Directive) have curbed excess N inputs, limited further contamination and enabled gradual recovery. The gradual decline in methemoglobinemia cases and groundwater NO₃⁻ contamination incidents in these regions further demonstrates policy effectiveness (Supplementary Fig. 1). In contrast, East Asia & Pacific and South Asia require N surplus reductions of more than 50% (e.g., NSR50, NSR75, or ZNS) to achieve meaningful contamination declines (Fig. 2). This threshold highlights severe current N overloading in these regions, where recent agricultural intensification has created a large backlog of legacy N.

Inevitable contamination onset and legacy N role

Projected timing of new groundwater NO₃⁻ exceedances varies little across scenarios in most regions (Fig. 3a), because most areas already exceeded the WHO standards by 2020 (blue shading; affected areas quantified in the inset bar graphs). This consistency confirms that near-term N surplus reductions cannot prevent imminent water quality violations, even under the aggressive ZNS scenario. This inevitability arises from persistent vadose zone legacy NO₃⁻, which continues to migrate to aquifers via delayed transport processes (e.g., slow pore-water flow, diffusion) despite reduced anthropogenic N loading.

Fig. 3. Simulation of NO₃⁻ concentration exceedance in shallow groundwater.

a Timing of shallow-groundwater NO₃⁻ exceeding the safety limit (11.3 mg N L⁻¹) under four scenarios. Colored bars indicate the number of grids where NO₃⁻ first exceeded the limit across different time periods (error bars represent standard deviations from 1000 Monte Carlo simulations). b Timing of shallow-groundwater NO₃⁻ returning below the safety limit under four scenarios. Solid lines show the proportion of grids in exceedance relative to affected regions, those projected to exceed the WHO safety limit at any point during 1961–2100 (red shaded areas represent standard deviations from 1000 Monte Carlo simulations).

Two lines of evidence validate this transport lag. First, soil NO₃⁻ leaching, vadose zone NO₃⁻ inventories, and groundwater NO₃⁻ concentrations exhibit mismatched timelines (Fig. 2 and Supplementary Figs. 7, 10). For instance, soil N leaching may decline rapidly under mitigation, but vadose zone NO₃⁻ stocks remain high for decades, sustaining the groundwater contamination long tail. This prolonged residence time is primarily from two factors: (1) vadose zone thickness—thick vadose zones inhibit denitrification by limiting deep soil microbial activity, thus extending N retention and the long tail³⁸; (2) environmental and soil conditions—clay-rich soils slow NO₃⁻ transport while potentially increasing its long-term storage³⁰, further prolonging the tail. Second, the global vadose zone stored 4037 ± 214 Tg N in 2020 (Supplementary Figs. 7, 8), with this substantial N reservoir in Central North America, Northern South Asia, Southern East Asia, Europe, and Sub-Saharan Africa (Supplementary Fig. 8). Notably, regions with large vadose zone NO₃⁻ reserves are precisely those projected to experience new groundwater exceedances after 2020 (Fig. 3a), directly linking legacy N to future contamination.

Although N surplus reduction cannot prevent contamination onset, it significantly shortens exceedance duration. Under BAU, 66 ± 2 % of “affected regions” (areas where NO₃⁻ concentrations exceed the WHO limit at any point during 1961–2100) will remain in violation by 2030. Mitigation scenarios reduce this proportion to 32–53%, with more aggressive reductions yielding greater improvements. Notably, even under ZNS, 4 ± 2% of affected regions will remain in exceedance by 2100. These persistent hotspots are concentrated in Central North America, Eastern Europe, and Sub-Saharan Africa (Fig. 3b), where thick vadose zones (average > 30 m) or high historical N inputs have created long-lasting legacy effects. This finding underscores that groundwater remediation in these areas will require multi-generational efforts, extending beyond the 21st century.

Nitrogen management strategies towards groundwater quality goals

To reconcile groundwater NO₃⁻ mitigation with agricultural sustainability, we developed a spatially adaptive management framework based on three core principles: (1) meeting the WHO groundwater quality target (NO₃⁻-N ≤ 11.3 mg N L⁻¹); (2) maintaining current crop yields to safeguard food security; (3) preserving long-term land productivity—defined by nitrogen use efficiency (NUE) < 90 %, as values exceeding this threshold risk soil nutrient depletion³⁹ (Supplementary Fig. 11, “Methods”). This framework classifies global affected regions into four distinct remediation pathways (Types I–IV) based on management difficulty (Fig. 4a, c), with tailored strategies designed to balance trade-offs between water, food, and land sustainability:

Fig. 4. Global distribution of four regional types and their tailored management pathways.

a, c Global groundwater NO₃⁻ management difficulty for 2030 and 2100 water quality compliance deadlines. Pie chart colors and values denote the proportion of each regional type relative to the total global land area. Extending the target year from 2030 to 2100 is projected to shift most Type III/IV regions in North America, Europe, and East Asia to Type I/II, whereas most Type III/IV areas in Sub-Saharan Africa and South Asia are expected to remain the most challenging for remediation. b, d Proportional distribution of management difficulty zones across major contamination-affected countries. e Tailored management pathways for different regional types. NUE = nitrogen use efficiency; “4 R” = right source, rate, time, place.

Type I (26% of affected regions) presents low management difficulty, requiring no additional action. These regions have historically low N inputs, enabling them to meet 2030 targets with existing practices that already balance yield maintenance, sustainable NUE, and water quality. Type I regions are primarily distributed in low-intensity agricultural areas, such as parts of eastern Europe and northeastern North America.

Type II (27% of affected regions) involves moderate management difficulty, requiring targeted management adjustments. To meet 2030 targets, these regions must simultaneously reduce N input and improve NUE. These regions are concentrated in intensively managed agricultural areas with existing policy foundations, including the U.S. Corn Belt (eastern North America), South Asia’s Gangetic Plain and Western Ghats.

Type III (4% of affected regions) endangers drinking water safety for 2% of the global population and presents high management difficulty (entailing a water quality-yield trade-off). Unavoidable yield reductions are necessary to meet the 2030 target while preserving sustainable NUE. This type is predominantly located in high N saturation regions with limited water availability, such as eastern Latin America, Sub-Saharan Africa, northeastern India (South Asia), and southeastern China (East Asia).

Type IV (43% of affected regions) endangers drinking water safety for 7% of the global population and poses extremely high management difficulty (entailing multi-sector trade-offs). Even with yield compromises, 2030 target compliance is unachievable without undermining sustainable land productivity (e.g., NUE > 90% or soil nutrient mining). This type is widely distributed across major agricultural regions, including central-eastern North America, western Europe, western Sub-Saharan Africa, South Asia’s Central Highlands and Eastern Ghats, southwestern China (East Asia), and parts of Southeast Asia (Viet Nam, the Philippines).

Notably, the world’s largest agricultural producers, including the U.S., India, China, Russia, and Canada, host the largest Type III/IV areas (Fig. 4b, d), highlighting the global trade-off between water quality and food security. Even with the target timeline extended to 2100, 19% of regions remain Type III/IV (Fig. 4c), endangering drinking water safety for 6% of the global population. This underscores that the long tail’s legacy effects will require sustained, enhanced governance to prevent permanent land and water degradation.

Based on these four zones, we developed a differentiated management framework to prioritize actions aligned with local feasibility (Fig. 4e). For regions where marginal adjustments suffice to meet targets, including moderate-difficulty Type II zones and specific high-difficulty Type III zones with excessive N inputs or grain import capacity, the core strategy is to implement “4 R” nutrient stewardship (right source, rate, time, place) combined with urease or nitrification inhibitors^40,41. Compared to conventional fertilization, this integrated approach reduces unnecessary N losses by 30–70%, increases NUE by 10–80%, and maintains or even boosts yields⁴⁰. Complementary measures are recommended for two distinct Type III profiles: (1) in zones with excessive N inputs, reducing fertilizer application rates by 30–60% has no adverse effect on yield (e.g., parts of China’s North China Plain)^42–44; (2) in zones with sustained grain import capacity (e.g., Middle Eastern countries, Western Europe), converting some agricultural land to grassland or wetland reduces agricultural intensity and enhances nutrient sinks, thus mitigating N pollution⁴⁵. For regions where water quality targets conflict with food/land sustainability, specifically the remaining Type III zones and extremely high-difficulty Type IV zones, food security and land productivity are prioritized. The core strategy here is to implement agricultural management measures that achieve modest N input reductions paired with aggressive NUE improvements to minimize yield losses⁴⁶. For instance, in Type IV zones of China, replacing broadcast urea with drip fertigation increases NUE by 34% while reducing leaching, even with small N input cuts⁴⁷. Other strategic approaches include adopting legume/cash crop-cereal rotations (proven to increase annual equivalent yields by 38 % compared to conventional cropping systems in the North China Plain⁴⁸) and importing NUE-enhanced crops⁴⁹ to offset local N reductions. Critical complementary measures include the parallel development of cost-effective groundwater NO₃⁻ removal technologies (e.g., ion exchange, bioremediation)⁵⁰ and vadose zone interception systems (e.g., deep-rooted catch crops, permeable reactive barriers)⁵¹. These technologies act as “safety nets” to reduce existing contamination loads, accelerating restoration where N input reductions alone are insufficient.

Discussion

The multi-decadal lag in groundwater quality responses to surface N reductions is a fundamental, yet often overlooked constraint on environmental policy effectiveness. While regional studies have consistently identified legacy N accumulation as the dominant driver of contemporary groundwater NO₃⁻ contamination^30–32, the global-scale trajectory of NO₃⁻ dynamics, including the timing of exceedance onset and the persistence of the “long tail” effect, remains poorly quantified. This knowledge gap has hindered robust tracking and evaluation of SDG 6 targets, which mandate universal access to safe drinking water. Our study addresses this critical limitation by quantifying these global patterns, providing the first systematic assessment of SDG 6 feasibility through the lens of NO₃⁻ pollution’s long tail effect.

Since the “Green Revolution”, groundwater NO₃⁻ contamination has followed distinct continental trajectories, shaped by historical differences in agricultural intensification and policy implementation. Europe and North America (early adopters of agricultural modernization) initiated synthetic fertilizer application decades earlier than Asia^52,53. However, these nations have experienced declining N inputs and shrinking contaminated areas due to three key factors that have shortened their contamination long tail. First, comprehensive environmental policies have established regulatory frameworks for N inputs and emissions, a critical external driver of NO₃⁻ pollution reduction^22,23,46. Second, the ratio of N fertilizer consumption to gross domestic product has declined in Western nations, reflecting an agricultural economic shift: N fertilizer primarily supports raw commodity production, while downstream processing and services now contribute more to economic wealth. In response, farmers have reduced N fertilizer use, with further cuts expected as subsidies are eliminated, sustainable farming practices are promoted, and the environmental costs of fertilizer use are gradually internalized into prices⁵². Third, dietary transitions in Europe, combined with land retirement policies (e.g., “set-aside” programs⁵⁴), have driven land-use shifts and reduced agricultural intensification, shortening legacy N effects. In contrast, East and South Asia (rapidly intensifying agricultural regions) have experienced accelerating groundwater NO₃⁻ pollution, driven by post−1980s population growth, agricultural expansion, and high N fertilizer use to meet growing food demand^55,56. These regions face an extended long tail of contamination, as thick vadose zones and ongoing N accumulation amplify legacy effects. Critically, our results show that despite these continental historical differences, subsurface N reserve long tails will sustain groundwater contamination for decades globally, even in policy-mature regions such as Europe. This challenge is particularly acute for Sub-Saharan Africa, where expanding fertilizer subsidies⁵⁷ and increased legume cultivation⁵⁸ coincide with precarious hydrogeological conditions: shallow vadose zones (average < 20 m)¹⁴ and rapid NO₃⁻ transport velocities (> 5 m y⁻¹)⁵⁹ create “acute contamination hotspots”. This global heterogeneity highlights the infeasibility of uniform restoration timelines and one-size-fits-all governance frameworks, demanding an urgent shift to hydro-geologically tailored management that accounts for local subsurface properties and long tail duration.

Our study quantifies a critical temporal disconnect: changes in agricultural N management take decades to translate into groundwater quality improvements. While regulatory frameworks are essential for groundwater protection, sustainable control of agricultural non-point source pollution requires meaningful farmer participation, a gap that persists even in policy-mature regions such as Europe and North America. Farmer skepticism toward water-related legislation (e.g., the U.S. Clean Water Act, EU Nitrates Directive) undermines implementation^60–63, driven by three key barriers. First, the invisibility of delayed pollution: agricultural non-point source pollution is transboundary, and its impacts are deferred by the long tail effect, making compliance benefits far less tangible than short-term productivity gains. Second, short-term economic priorities: producers often prioritize near-term profitability over long-term environmental benefits that may not materialize for decades^60,64. Third, institutional deficiencies: inadequate financial incentives, fragmented monitoring networks, and weak enforcement mechanisms limit compliance, compounded by governance frameworks that fail to account for the long tail’s temporal scope⁶⁵.

To address this gap, our adaptive management framework incorporates yield maintenance as a non-negotiable constraint, aligning policy goals with farmer priorities while accounting for the long tail effect. For future global water quality governance to succeed, three integrated strategies are essential. First, demonstrate visible, incremental progress: scale pilot projects that achieve N reductions without yield losses, and implement long-term monitoring programs to track delayed groundwater quality improvements, building farmer trust in the long-term value of compliance. Second, implement tiered, time-bound incentives: adopt gradually increasing fertilizer taxation to penalize excess N application (allowing farmer adaptation), paired with targeted subsidies for water-saving and N-efficient technologies (e.g., drip irrigation, precision fertilization tools) that deliver immediate economic benefits (reduced input costs and labor). Long-term rewards (e.g., certification, market access) should be linked to management actions rather than measured groundwater quality improvements, balancing economic and environmental goals and addressing the long tail’s temporal lag with patient policy design. Third, strengthen transboundary governance and compliance systems: establish cross-agency partnerships (e.g., between agriculture and environment ministries) to standardize monitoring protocols, share long tail dynamic data across regions, and enforce regulations consistently. These measures enable precision N management that is both environmentally effective (reducing legacy N loads and shortening the long tail) and economically viable for farmers, closing the gap between policy intent and governance outcomes.

We acknowledge three key limitations of this study, which also identify critical avenues for future research. First, limited global aquifer depth data led us to use standardized depth thresholds (5 m vs 5–55 m below the water table)⁶⁶, a convention validated across hydrogeological settings^41,67–70. This assumption inevitably oversimplifies heterogeneous geological environments, particularly karst regions, and our NO₃⁻ transport rate calculation in the vadose zone does not capture unique hydrological processes (e.g., preferential flow through fractures) in such systems. Nevertheless, this simplification is reasonable given the trade-off between spatial resolution and computational feasibility in global-scale modeling. Future work leveraging high-resolution aquifer and karst-specific hydrological datasets will refine this distinction and reduce uncertainty. Additionally, our 0.5° spatial resolution ensures global coverage but smooths fine-scale heterogeneity in intensively cultivated regions. Using the HYDE dataset, we allocate N inputs to land use types according to their fractional area within each grid cell; however, field-to-field management variations remain unresolved. Higher-resolution global datasets for all input variables will enable future studies to better capture local-scale legacy N accumulation patterns.

Second, our model assumes static vadose zone thickness and groundwater table position, a necessity due to the absence of globally consistent groundwater table dynamic data. While denitrification is incorporated as a depth-dependent, first-order decay process, dynamically coupling this to water table fluctuations would enhance model realism. This static assumption may underestimate remediation periods in regions experiencing aquifer compaction (e.g., China’s North China Plain) from groundwater overdraft or pronounced seasonal water table variations. However, because our analysis targets multi-decadal legacy N dynamics, the dominant signal is governed by cumulative historical N surplus rather than interannual hydroclimatic variability. Incorporating such dynamics into future models will underscore the urgency of governance intervention, as compaction slows NO₃⁻ transport and extends the contamination long tail, delaying groundwater recovery.

Third, our future scenario design focused exclusively on N loading dynamics under the SSP126 climate pathway, excluding potential variables such as population growth, technological advancements, and evolving irrigation practices (e.g., reclaimed water use⁷¹) from projections. While integrating full socioeconomic and technological uncertainties would yield more comprehensive results, such complexity exceeds current computational and data constraints for global-scale modeling. Our N surplus reduction scenarios (BAU, NSR50, NSR75, ZNS) serve as “what-if” baselines for assessing legacy N effects under contrasting management intensities, consistent with previous global assessments⁷². This approach does not capture likely regional divergence in future N trajectories, whereby some regions may increase N inputs (and surplus) while others pursue efficiency gains or stricter regulation. However, forecasting region-specific pathways entails substantial socioeconomic, technological, and policy uncertainties beyond the scope of this study. Future studies should therefore develop region-specific N management scenarios to assess how such divergence affects the spatial heterogeneity of legacy N impacts.

Despite these limitations, our results provide a robust, fit-for-purpose baseline: the model identifies the fundamental governance challenges posed by disrupted biogeochemical N flows and offers actionable insights for interventions aimed at returning N cycling toward the Earth’s safe operating space as defined by planetary boundaries. Building on this framework, subsequent studies can incorporate additional climate pathways (e.g., SSP245, SSP585), demographic transitions, and emerging technologies (e.g., precision N management tools, advanced remediation technologies) to further enhance predictive precision.

Methods

Calculation of global vadose zone NO₃⁻ storage

Global vadose zone NO₃⁻ storage was quantified through integrated modeling of annual soil NO₃⁻ leaching rates and vadose zone transport timescales. The calculation framework was initiated with the determination of soil N surplus (N_surplus) via mass-balance principles^30,73:

$$N_{surplus}=N_{fix}+N_{dep}+N_{\min }+N_{org}-N_{uptake}-N_{vol}$$ 1

where N_fix, N_dep, N_min, and N_org represent N inputs from biological N fixation, atmospheric N deposition, mineral fertilizer application, and organic fertilizer application, respectively; N_uptake and N_vol represent N loss via crop uptake and ammonia volatilization, respectively (Supplementary Fig. 12). This study focuses primarily on the major anthropogenic and natural sources, while other potential sources or natural processes that could lead to elevated NO₃⁻ (e.g., geogenic origin and bacterial fixation) are not considered.

We categorized global land systems into five functional types using the History Database of the Global Environment (HYDE v3.2)⁷⁴: upland crop lands, wetland rice systems, intensive grasslands, extensive grasslands, and other ecosystems. The data sources of N_min and N_org and the estimated methods for N_fix, N_dep, N_uptake, and N_vol can be found in the supplementary material (Supplementary Method 1). These data represent the most comprehensive dataset currently available, although inherent uncertainties remain.

We assumed that all reduced N compounds are nitrified to NO₃⁻, following Ascott et al.³². Leaching, surface runoff and denitrification can occur if N surplus is positive (Supplementary Fig. 12). In this case:

$$N_{\mathrm{leach}}=f_{leach}\times \left(N_{surplus}-N_{sro}\right)$$ 2

where N_leach represents NO₃⁻ leaching at the base of the soil zone; N_sro represents the NO₃⁻ runoff loss term; f_leach is the leaching coefficient, which was calculated by the denitrification coefficient (f_den) and land use type coefficient (f_landuse):

$$f_{leach}=\left(1-f_{den}\right)\times f_{landuse}$$ 3

where f_landuse was set to 1 for arable land areas and 0.36 for grassland and natural land^32,75, and f_den was estimated by climate factors (f_climate), soil texture (f_text), soil drainage conditions (f_drain), and soil organic carbon (f_SOC)⁶⁶:

$$f_{den}=MIN\left[\left(f_{climate}+f_{text}+f_{drain}+f_{SOC}\right),1\right]$$ 4

where f_climate was estimated based on soil water content, soil’s total available water capacity for the top 1 m, and mean annual temperature⁶⁶; f_text, f_drain, and f_SOC were obtained from Van Drecht et al.⁷³.

$$N_{sro}=f_{cal}\times f_{red}\times f_{qsro}\times \left(N_{surplus}+N_{uptake}\right)$$ 5

where f_cal, f_red, and f_qsro represent calibration coefficients, land use-specific coefficients, and hydrological runoff coefficients, respectively, with values set following Beusen et al.⁶⁶.

Vadose zone NO₃⁻ storage (1961–2020) was calculated using the cumulative approach proposed by Ascott et al.³². For a given year t, NO₃⁻ storage in the vadose zone (N_vz, Tg N) for a grid cell with vadose zone travel time T_vz (years) and time-variant NO₃⁻ leaching input N_leach (kg N) is:

$$N_{vz}\left(t\right)=\frac{{\sum }_{i=t-T_{vz}}^t{N}_{leach}}{{10}^9}$$ 6

where T_vz is the travel time (y) of NO₃⁻ through the vadose zone, which was obtained by dividing the vadose zone thickness¹⁴ by NO₃⁻ transport velocity in the vadose zone⁵⁹. Vadose zone thickness was derived from a global water table map¹⁴ based on a published global groundwater model (forced by modern climate, terrain, and sea level); NO₃⁻ transport velocity was obtained from Yang et al.⁵⁹, who developed the first open-source dataset using a NO₃⁻ Time Bomb model incorporating lithological, permeability, porosity, and groundwater recharge. Global N surplus (1950–1960) was estimated from the annual growth rate of global N surplus (1961–1971), with 1960 global NO₃⁻ storage calculated as the initial vadose zone NO₃⁻ storage.

Calculation of groundwater NO₃⁻ concentration

Groundwater NO₃⁻ storage was calculated using the IMAGE–GNM model, which couples the IMAGE model with the PCRaster Global Water Balance (PCR-GLOBWB) global hydrological model to simulate N flows and processes in terrestrial landscapes^76–78. IMAGE-GNM incorporates a two-compartment groundwater system (shallow and deep), following the methodologies of Van Drecht et al.⁷³ and de Wit⁷⁹. The shallow groundwater system comprises the top 5 m of the saturated zone, with rapid NO₃⁻ transport to local water courses⁷³ (Supplementary Fig. 12). The deep system consists of unconsolidated aquifers extending 50 m below the shallow system, with slow NO₃⁻ transport via deep groundwater flow to larger streams/rivers (baseflow)^{41,67–70,73} (Supplementary Fig. 12). This depth distinction is justified, as 50 m depth for the deep system covers the common thickness range of unconsolidated aquifers, and represents a reasonable and conservative global estimate^68,73,80,81. Karst regions and other exceptions exist, but potential errors are acceptable given the global study scope. Consolidated aquifers were excluded due to inherent hydraulic conductivity constraints that prevent extensive NO₃⁻ migration and diffusion; while local NO₃⁻ contamination may occur⁸², these properties prevent widespread global pollution, consistent with the study’s objective of identifying large-scale NO₃⁻transport patterns (rather than simulating micro-processes in localized geological settings). The model operates with an annual time step and a spatial resolution of 0.5° × 0.5°⁶⁶.

We assumed that for a given year t (y), NO₃⁻ entering the vadose zone for a grid cell (C_N(t), kg N), experiences a lag time for T_vz years before reaching the groundwater. Specifically:

$$C_N\left(\mathrm{t}\right)=N_{leach\left(t-T_{vz}\right)}$$ 7

It was also assumed that groundwater flows in homogeneous and isotropic aquifers⁷⁰, and the mean time of residence T_r (y) of each groundwater body is calculated as follows:

$$T_r=\frac{1000\times D\times P}{Q_{inflow}}$$ 8

where P is soil effective porosity (m³ m⁻³); Q_inflow represents the inflow rate of groundwater recharge (mm y⁻¹); D represents aquifer thickness (m). We assumed shallow groundwater partially recharges deep groundwater (Supplementary Fig. 12), with a groundwater fraction coefficient (0–1) obtained from Beusen et al.⁶⁶. Mean travel time was equated to vertical flow travel time in the shallow system (uniform vertical flow distribution assumed)⁷³; lateral flow travel times to the fluvial systems vary considerably, with lateral flow travel time at depth z (T_age(z), y) calculated per Meinardi⁸³:

$$T_{age}\left(\mathrm{z}\right)=-T_r\times \ln \left(1-\frac{z}{D}\right)$$ 9

where z is aquifer depth (m), with the top of the aquifer at depth z = 0 and the bottom of the aquifer at depth z = D.

Denitrification occurs during NO₃⁻ transport in shallow groundwater (Supplementary Fig. 12), with N loss via denitrification for a grid cell at time t and depth z (C_N-den(t, z), kg N) calculated through a first-order decay process:

$$C_{N-den}\left(t,z\right)=C_0\times \left(1-e^{-k\times T_{age}\left(z\right)}\right)$$ 10

where k is the first-order decay rate constant (y⁻¹), calculated as ln(2)/DT₅₀. DT₅₀ (y) is the NO₃⁻ half-life in groundwater, dependent on aquifer lithology⁷¹; C₀ is the initial NO₃⁻ concentration entering shallow groundwater, equivalent to $C_N\left(t-T_{age}\left(\mathrm{z}\right)\right)$.

NO₃⁻ outflow from shallow/deep aquifers to surface water for a grid cell at time t and depth z (C_N-out(t, z), kg N) was calculated as⁸⁰:

$$C_{N-out}\left(t,z\right)=C_0\times e^{-k\times T_{age}\left(z\right)}\times f_{out}$$ 11

where f_out is the NO₃⁻ outflow coefficient (y⁻¹), equal to 1/T_r. NO₃⁻ storage for a grid cell (S_N, kg N) at time t and aquifer depth z is:

$$S_N\left(\mathrm{t},\mathrm{z}\right)=C_0\times e^{-k\times T_{age}\left(z\right)}\times \left(1-\frac{1}{T_r}\right)$$ 12

Total groundwater NO₃⁻ storage (kg NO₃⁻-N) in shallow/deep aquifers per grid cell was obtained by integrating S_N over aquifer depth (0 to D m). Groundwater NO₃⁻ concentration (mg N L⁻¹) was then calculated by dividing total NO₃⁻ storage by grid area, aquifer thickness, and porosity. Detailed methods are provided in Van Drecht et al.⁷³, Beusen et al.⁶⁶, and Liu et al.⁸⁰.

All model parameters were derived from previous studies^{66–68,84,85}; model performance has been comprehensively evaluated in existing research, including global/regional validation of simulated N/phosphorus storage in groundwater/surface water and model sensitivity analysis^66,80,86,87. Our key enhancement to this validated framework is the incorporation of N residence time in the vadose zone to capture the “long tail” mechanism. This addition does not modify the core model structure or introduce new parameters.

To predict vadose zone/groundwater NO₃⁻ storage (2021–2100) and assess the feasibility of achieving targeted NO₃⁻ concentration reductions, four future management scenarios were modeled: business-as-usual (BAU), 50% N surplus reduction (NSR50), 75% N surplus reduction (NSR75), and zero-N-surplus (ZNS). The BAU scenario assumes 2020 cropping patterns and fertilizer application rates remain unchanged until 2100. For the other scenarios, instantaneous N surplus reductions (50, 75, and 100%) were imposed; these reductions can be achieved via multiple management practices, land use changes or fertilizer application adjustments.

Estimates of future onset and mitigation timings of global shallow groundwater NO₃⁻ exceedance

A modified Sequential T-test Analysis of Regime Shifts (STARS)⁸⁸ was used to analyze groundwater NO₃⁻ concentrations across all scenarios (1961–2100), identifying the timing of NO₃⁻ exceedance of the WHO safety limit (11.3 mg N L⁻¹) and subsequent recovery. This data-driven method identifies time series breakpoints without prior assumptions about regime transition timing⁸⁹, a key advantage for this study. Algorithmic improvements to STARS are provided in the Supplementary Method 2.

A parameter sensitivity analysis based on spatiotemporal cross-validation was conducted to avoid subjectivity in parameter selection. Coefficients of variation (CV) for detection rates across different parameter combinations (window length l = 3, 5, 7, 10; significance level p = 0.01, 0.05) were calculated for nine (3 × 3) spatiotemporal scenarios, then converted to robustness scores (0–1 interval) via exponential transformation (exp(-CV)). p = 0.05 and l = 5 were ultimately selected as analytical parameters. This approach draws on CV applications in sensitivity analysis⁹⁰ and integrates spatiotemporal cross-validation frameworks⁹¹, providing an objective quantitative standard for parameter selection.

N management framework for groundwater NO₃⁻ pollution control

A spatially differentiated N management framework for groundwater NO₃⁻ pollution control was developed by identifying the minimum N surplus reduction required to resolve groundwater NO₃⁻ contamination by target years, combined with an NUE indicator framework. Global affected regions were categorized into four types, with tailored management strategies for each (Supplementary Fig. 11).

Three basic management principles were established: (1) achieve the WHO water quality targets; (2) maintain sustainable land productivity; (3) preserve 2020 (baseline) crop yields. To meet water quality targets, immediate N surplus reduction is required to ensure shallow groundwater NO₃⁻ concentrations return below the safety limit by the target year. The minimum required N surplus reduction ratio (α) was thus determined from the four scenario projections, where α = 0, 0.5, 0.75, and 1 indicate minimum reductions of 0, 50, 75, and 100%, respectively.

The allowable N surplus at the target year (S_target) to meet water quality targets is constrained by:

$$S_{target}\le \left(1-\alpha \right)\times S_{baseline}$$ 13

where S_baseline is the 2020 baseline N surplus per grid cell (kg N), calculated as baseline N input ($I_{baseline}$) minus baseline N output ($O_{baseline}$).

According to the EU N Expert Panel³⁹, an NUE (ratio of N uptake to N input) above 0.9 risks soil nutrient mining, compromising sustainable land productivity. Target year N input ($I_{target}$) must therefore satisfy:

$$I_{target}\ge \frac{N_{uptake-target}}{0.9}$$ 14

where $N_{uptake-target}$ is N removal via crop harvests in the target year (kg N). To maintain baseline yields ($N_{uptake-baseline}$), we set $N_{uptake-target}=N_{uptake-baseline}$. Thus, I_target should satisfy:

$$I_{target}\ge \frac{N_{uptake-baseline}}{0.9}=I_{target-\min }$$ 15

where $I_{target-\min }$ is the minimum allowable N input in the target year.

Meanwhile, the $I_{target}$ must also satisfy:

$$I_{target}=O_{target}+S_{target}\le O_{target}+\left(1-\alpha \right)\times \left(I_{baseline}-O_{baseline}\right)$$ 16

where $O_{target}$ is the target year N output (kg N). Since ammonia volatilization losses are negligible compared to crop harvest N removal, we assumed N output = N uptake. Thus, $O_{target}=O_{baseline}$, and I_target is further bounded by:

$$I_{target}\le O_{baseline}+\left(1-\alpha \right)\times \left(I_{baseline}-O_{baseline}\right)=I_{target-\max }$$ 17

where $I_{target-\max }$ is the desired maximum N input in the target year.

Global regions were categorized into four types based on α, $I_{target-\min }$, and $I_{target-\max }$: Type I (α = 0): Low management difficulty. Water quality targets are achievable with current practices, maintaining sustainable land productivity and baseline yields; Type II (α = 0.5 or 0.75; $I_{target-\min }\le I_{target-\max }$): Moderate management difficulty. Targets are achievable via N input reduction and NUE improvement, adhering to all three management principles; Type III (α = 0.5 or 0.75; $I_{target-\min }>I_{target-\max }$): High management difficulty. Water quality targets cannot be achieved while maintaining baseline yields, an unavoidable water quality-yield trade-off; Type IV (α ≥ 1): Extremely high management difficulty. Targets require NUE > 100%, which undermines sustainable land productivity (soil nutrient mining), entailing multi-sector trade-offs.

Model validation and uncertainty analysis

Model validation was conducted via two approaches: direct comparison with observational data, and cross-reference with prior global estimates, confirming the model’s ability to capture spatial/temporal patterns of groundwater NO₃⁻ dynamics. First, simulated groundwater NO₃⁻ concentrations were compared with monitoring data from the U.S⁹², EU⁹³, and India^94,95, plus observed values compiled from 264 peer-reviewed publications/ datasets (Supplementary Reference List). Across 5,849 global data pairs (North America, Europe & Central Asia, South Asia, East Asia & Pacific, other regions), simulations showed good agreement with observations (R² = 0.61–0.79; RMSE = 1.95–6.95 mg N L⁻¹; Supplementary Fig. 5). Second, simulated vadose zone/groundwater NO₃^- storage was cross-referenced with published global/regional estimates: vadose zone NO₃⁻ storage in 2000 (3405 ± 188 Tg N) falls within the range of prior assessments (Ascott et al³².: 605–1814 Tg N; Almaraz et al.⁹⁶: ~ 15,200 Tg N for deep vadose zones); groundwater NO₃⁻ storage in 2020 (185 ± 7 Tg N) is slightly lower than Beusen et al.⁶⁷ (245 Tg N), a discrepancy attributed to their globally uniform 1 m vadose zone thickness assumption, which underestimates NO₃⁻ retention times in thick vadose zones, leading to overestimated groundwater NO₃⁻ inputs. These cross-comparisons confirm model consistency with previous studies.

Groundwater NO₃⁻ storage estimation and the timing of NO₃⁻ contamination/exceedance recovery are primarily affected by N leaching uncertainty. Monte-Carlo simulations (1000 runs) were thus conducted to quantify uncertainty, using a normal distribution with original N leaching as the mean and a CV of 40 %. Standard deviations of NO₃⁻ concentrations in the vadose zone/groundwater (Supplementary Figs. 13, 14), and the timing of NO₃⁻ exceedance/recovery (Supplementary Fig. 15) were calculated. Shannon entropy analysis⁹⁷ was also conducted to assess uncertainty in NO₃⁻ exceedance/recovery events (Supplementary Fig. 16), where higher entropy values indicate greater event occurrence uncertainty⁹⁸. Results show that regions with high uncertainty in NO₃⁻ exceedance/recovery account for 1.4% of global land area and 14.7–24.8% of potentially contaminated areas, respectively. Over 98 % of global land area has a standard deviation of < 20 years for NO₃⁻ exceedance timing; 82–99% of contaminated areas have a standard deviation of < 20 years for recovery timing. Collectively, these findings confirm low prediction uncertainty, with the primary variability source being N leaching.

Author contributions

W.Z., X.J. and X.L. conceptualized and designed the study. W.Z., X.J. and L.N. performed the analysis and created illustrations. W.Z. and X.J. wrote the manuscript with input from all authors. W.H., T.Y., T.T., A.B. Y.X., X.W., Y.L., M.S., and X.L. reviewed and revised the paper. W.Z., X.J., L.N., W.H., T.Y., T.T., A.B., Y.X., X.W., Y.L., M.S. and X.L. contributed to the discussion and interpretation of the results.

Peer review

Peer review information

Nature Communications thanks Brent Clotheir and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China (42225707, U2444217, 42430706) and the National Key Research and Development Program of China (2025YFD1500204).

Data availability

Global datasets used in this study were obtained from open-access sources (see below). All data necessary for replication are provided via a Figshare dataset (10.6084/m9.figshare.30723884). Mineral/organic fertilizer application rates: 10.11888/Terre.tpdc.300446; Atmospheric N deposition: 10.1594/PANGAEA.942069; Agricultural yield data: https://www.fao.org/faostat/en/#data; Land use data: 10.17026/dans-25g-gez3; Regional classification: World Bank country/lending groups (https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups); Vadose zone nitrate transport velocity: 10.6084/m9.figshare.14782959.v2; Groundwater table depth: https://glowasis.deltares.nl/thredds/catalog/opendap/opendap/Equilibrium_Water_Table/catalog.html; Population density: https://www.earthdata.nasa.gov/data/catalog/sedac-ciesin-sedac-gpwv4-popdens-r11-4.11?utm_source=chatgpt.com.

Code availability

A complete model description is provided in the Methods section. Python modeling code is available at 10.5194/gmd-8-4045-2015. Additional custom code is available from the corresponding authors upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75014-8.