Dynamic soil-N-based fertilization approach for optimized N management

The world’s growing population and the resulting rising global food demand have intensified the race to increase agricultural productivity. Over the past century, widespread fertilizer application has enabled a dramatic increase in this productivity. However, inefficient nitrogen (N) fertilizer application methodologies have resulted in excess fertilization that is far beyond the optimal crop demand. Consequently, the N that is not utilized by plants is released into the environment, reducing farmer profits and contributing to unwanted side effects and pollution. Groundwater pollution by nitrate (NO₃⁻), associated with excess fertilization, is a primary reason for drinking water quality reduction¹, causing severe health issues² and substantial economic damages³. The NO₃⁻ -polluted groundwater’s return flow to the surface water can cause serious harm to aquatic environments⁴. However, due to extensive groundwater storage and slow groundwater flow, N flux to streams will continue for several generations⁵. Air pollution from excess fertilization is also a growing concern. It is estimated that the synthetic N fertilizer supply chain contributes to ~2.1% of global CO₂ emissions⁶. Moreover, excess fertilization is connected to N oxide emissions, as N₂O, which is a potent greenhouse gas, and NOₓ (NO + NO₂), promoting harmful ground-level ozone^7,8. Together, these factors underscore the global need for innovative methods that optimize fertilizer application to meet the plant demand in real-time, while achieving maximum yield and minimal losses. Vegetables, as intensively managed crops, pose a significant challenge in N management. This agricultural sector consumes 8.5 million megagrams (Mg or metric tons) of synthetic N annually, accounting for 8% of global fertilizer use, disproportionate to its occupied cropland⁹. High fertilization rates, coupled with characteristically short growing seasons and shallow root systems, lead to low nitrogen use efficiency (NUE) and significant environmental losses¹⁰.

We suggest that persistently low NUE and consequent over-application of N are driven by two main aspects: (1) Agronomic: Plant N uptake rate is typically described by Michaelis-Menten kinetics, where the uptake rate increases with increasing soil N concentration, up to a saturation point^11,12. This necessitates applying N to maintain a high concentration in the root zone rather than simply providing the plant’s required N mass. This, along with the need to preserve high soil moisture through irrigation, leads to the down-leaching of mobile nutrients such as NO₃⁻ and to excess N application above the plant’s actual demand. This surplus is often embedded into standard fertilization protocols that prioritize maximum yield. (2) Economic: The perceived cost of fertilizers is typically lower than the potential economic loss from yield reduction due to nutrient deficiency, which motivates farmers to apply N in additional surplus as a risk-aversion strategy^13,14. Current approaches for improving N management focus on the optimization of the predetermined fertilization protocols¹⁵. Such protocols are predetermined, based on generalized experiments, and are not designed to account for specific field variability. Therefore, such protocols have a limited potential for improving NUE. Field variability stems from: (1) fluctuating crop N demand, which is influenced by external factors such as plant diseases, droughts, and other agronomic stresses¹⁶, rather than by nutrient availability alone; (2) dynamic in-season N demand, affected by crop growth rate and phenological stages, which are mainly influenced by environmental conditions such as temperature and radiation; and (3) diverse soil properties that affect water and nutrient transport and availability¹⁷. Accordingly, predetermined fertilization approaches have inherent limitations in supplying the right amount of N at the right time¹⁸.

Fertilization management during the growing season is often guided by assessing the plant’s nutritional state. This can be performed through destructive tissue analysis, comparing leaf or petiole nutrient content against established N-dilution curves¹⁹, or via non-destructive proximal sensing, such as chlorophyll content meters and spectral indices, which correlate to plant N status and enable high-frequency monitoring^20,21. While plant-based methods are effective at detecting nutrient deficiencies, they are often less reliable for diagnosing soil nutrient excess – the primary driver of environmental leaching as elevated N levels in the soil do not always manifest as increased concentrations in plant tissue²². Furthermore, dynamic fertilization management using plant N status is limited by the inherent physiological time lag between root-zone nutrient availability and the detectable plant response. Adjusting fertilization to actual soil N concentration can directly account for the high field variability and offers the advantage of a direct, rapid indicator of both nutrient excess and shortage.

Sampling and testing of soil extract solution for N concentration is a well-established method in vegetable crops^23,24. In one aspect, soil N concentration can serve as an indicator of the available N in the active root zone. Accordingly, some fertilization guidelines suggest that threshold soil N levels should be determined by the soil’s capacity to supply plant demand over a specific period of time²⁵. In another aspect, plant N uptake has been found to increase linearly with soil N concentration up to a saturation point²⁶. This aligns with the nature of the uptake rate described by Michaelis-Menten kinetics^11,12 and suggests that an optimal soil N concentration may be determined, even where corresponding soil N mass exceeds crop N demand²⁷. However, the applicability of such an approach was historically limited by the low temporal availability of data on soil N concentration that derived from manual sampling and analysis of soil porewater via suction cups. Granados et al.²⁸ demonstrated a “prescriptive-corrective management” (PCM) approach where N application was adjusted based on manual, weekly measurements of SPW NO₃⁻ concentration²⁸. This study confirmed that such corrective adjustments could improve N application efficiency. However, the infrequent, manual measurements limited the feasibility of implementing high-frequency, real-time fertilization adjustment. Conversely, in soilless hydroponic systems, where environments are homogeneous and manageable, real-time measurements and automated closed-loop systems have been successfully applied to control nutrient levels^29,30,31. To the best of our knowledge, no such implementation has been achieved in soil-based cultivation, primarily due to previous limitations in continuous soil N monitoring and autonomous adjustment capabilities to achieve a defined soil N concentration.

Recent advancements in soil N monitoring technologies have overcome this limitation by providing continuous information on actual soil N concentration. These technologies offer accurate soil porewater (SPW) NO₃⁻ measurements at a broad concentration range and across diverse conditions of temperature, pH, and DOC presence^32,33. This may enable the development of new dynamic fertilization approaches that correspond to actual soil N levels. Accordingly, this study presents a novel implementation of this technology, achieving, for the first time, real-time, automated daily adjustment of N application based on continuous, in-situ measurements of soil NO₃⁻ concentrations in soil-grown crops. We propose a dynamic soil-N-based (DSNB) fertilization approach that utilizes real-time information on SPW NO₃⁻ concentration to continuously adjust N application and maintain a defined concentration range in the root zone. This defined range was selected to balance the physiological requirements for high yield with environmental considerations, ensuring sufficient N supply throughout the season while minimizing the leaching risk associated with excessive surpluses. The dynamic management of soil NO₃⁻ within this range allows optimal N uptake by plants under varying soil and crop conditions and addresses key agronomic, environmental, and economic challenges³⁴ (Fig. 1). The primary objective of this study was to evaluate whether the DSNB fertilization approach can effectively maintain soil N within a predefined range and, consequently, reduce nitrogen surplus and environmental losses while maintaining crop yield compared to conventional practices. To test this, we compared DSNB fertilization approach against a conventional, predetermined fertilization approach based on regional agronomic best management guidelines (predetermined fertilization) as a control. We tested it across two representative vegetable crops, bell pepper (Capsicum annuum) and lettuce (Lactuca sativa), at different experimental scales, evaluating soil NO₃⁻ dynamics, NUE, yield, and environmental losses through NO₃⁻ leaching and N oxide emissions, under the two fertilization regimes. Bell peppers were grown in a 1000-m² mesh-net house while lettuce was grown in a 75-L lysimeter system. The net-house setup provided a preliminary examination of the DSNB fertilization approach under commercial growth conditions, where N adjustments were performed manually. In contrast, the lysimeter system enabled high control over soil NO₃⁻ concentrations and implementation of fully automated, daily fertilizer adjustments based on an automated, closed-loop PID control. The net house experiment was conducted in an arid region that suffers from severe groundwater pollution due to irrigation with brackish water applied at a high leaching fraction and an excessively high N concentration. Finally, we quantified the potential for N reduction on a broad scale by analyzing fertilization practices in the U.S. vegetable production sector. By comparing actual farmer-reported application rates³⁵ with established agronomic guidelines as a reference for best management practices¹⁵, we evaluated the ‘farmer-driven excess’ that may be mitigated by improving real-time knowledge of soil NO₃⁻ availability.

Root zone N dynamics and fertilizer application

The soil NO₃⁻ concentrations monitored throughout the growing seasons of both crops demonstrated the ability of the DSNB approach to maintain soil NO₃⁻ within the defined range of 40 ± 10 ppm (Fig. 2). In contrast, under the predetermined fertilization approach, soil NO₃⁻ concentrations were either higher or lower than the defined range, exposing plant roots to either excess N or potential N deficiency. In the bell pepper experiment, soil NO₃⁻ concentrations under the predetermined treatment remained excessively high for much of the season, frequently exceeding 100 ppm. Notably, during certain intervals, the measured concentration was higher than the fertilization input concentration. This suggests that N uptake was significantly lower than water absorption in the shallow root zone, leading to a concentrated effect of the remaining NO₃⁻ before it reached the 20 cm measurement depth. Conversely, the DSNB treatment successfully lowered the initial high concentrations and stabilized them through manual adjustments of N application. In the lettuce experiment, the predetermined fertilization led to high NO₃⁻ concentration during the first month, closely reflecting the input fertilizer concentration and suggesting minimal plant uptake. During the second month, which was characterized by vigorous growth and higher demand, concentrations in the predetermined treatment decreased sharply to near zero, indicating a potential N shortage. The DSNB fertilization approach utilized the PID control algorithm to automatically adjust N application in real-time, maintaining SPW NO₃⁻ concentrations near the defined concentration range even during the high-demand growth phase.

Agronomic performance and environmental nitrogen losses

Adjusting the N application rates directly impacted the crop yield per unit of N applied (NUE), as assessed by the partial factor productivity of applied N (kg yield kg N⁻¹, PFP_N)³⁶. In the bell pepper experiment, DSNB fertilization led to a significant 56% increase in the PFP_N (95% confidence interval (CI): 31% to 81%, p < 0.001), accompanied by a non-significant 9.4% reduction in yield (CI: −8.5–25.9%, p = 0.274). (Fig. 3, Table 1). This PFP_N improvement was attributed mainly to reduced N application, with the intention of lowering the soil NO₃⁻ concentration to the defined concentration range (Fig. 2b). In the lettuce experiment, DSNB fertilization produced a moderate 7% increase in PFP_N (CI: −4–19%, p = 0.190), while achieving a significant 23% increase in total yield (CI: 10–35%, p = 0.002). This was achieved by the precise control of the soil NO₃⁻ concentration, which ensured a sufficient N supply throughout the season and prevented a yield-limiting N deficiency, as occurred in the predetermined fertilization. These results highlight how a predetermined fertilization protocol can lead to suboptimal N management, resulting in either excess N application, as observed in bell pepper, or an inadequate N supply, as seen in lettuce. Accordingly, DSNB fertilization offers a great opportunity for leveraging crop productivity while reducing N losses and the corresponding environmental pollution.

Both experiments demonstrated that DSNB fertilization can significantly reduce NO₃⁻ losses through leaching. We observed a 50% reduction in NO₃⁻ leaching in bell pepper and a 56% reduction in lettuce (CI: −87% to −24.2%, p = 0.006). The reduction in NO₃⁻ leaching in bell pepper was somewhat expected since adjusting the soil NO₃⁻ concentration resulted in an application of 42% less N. However, the significant decrease in NO₃⁻ leaching observed in the lettuce crop was particularly notable, as it occurred even when the total N application rate in the DSNB fertilization treatment was slightly higher than that in the predetermined fertilization treatment. This improvement was attributed to better synchronization of N supply with plant demand, which enabled optimal timing and amounts of N application. Cumulative N application and NO₃⁻ leaching patterns for both experiments are presented in Supplementary Fig. S2. Overall, the results suggest that the precise timing of N application may be more significant in reducing leaching than the simple reduction of the total applied N. In addition to a reduction in NO₃⁻ leaching, eliminating excess soil NO₃⁻ can mitigate N oxide emissions^37,38. In bell pepper, DSNB fertilization led to a 53% reduction in N₂O emissions (CI: −156–49%, p = 0.251) and a 42% reduction in NOₓ (CI: −104% to −14%, p = 0.029) emissions. For lettuce, N₂O emissions were under the limit of detection in either treatment. However, NOₓ emissions were reduced by 36% (CI: −66% to −6%, p = 0.024).

Quantification of N over-application in the U.S. vegetable sector

Analysis of the U.S. vegetable production sector revealed a significant “farmer-driven excess” – the ratio by which actual application rates exceed agronomic fertilization guidelines. For the ten crops analyzed, which account for 83% of the N use in the U.S.’s vegetable production sector, fertilizer applications substantially exceeded recommended rates in seven cases. (Fig. 4). In tomato cultivation, the most significant vegetable crop by cultivated area, the farmer-driven excess reached 37%, equivalent to ~9000 Mg per year of surplus N. Lettuce cultivation showed an even greater farmer-driven excess of 66%, contributing at least 12,000 Mg of surplus N annually. Similarly, celery growers were found to overapply N fertilizer by at least 56%. Overall, the farmer-driven excess across the analyzed U.S. vegetable crops accounts for at least 25% of the total N application, equivalent to approximately 33,000 Mg N per year. While this estimation aggregates diverse regional practices, these results demonstrate a broad potential for N reduction through the implementation of adaptive, real-time NO₃⁻ monitoring and adaptive fertilization strategies.

The experimental results show that fertilization based on actual soil NO₃⁻ can improve NUE, increase yield, and reduce N environmental impact. The significant advantage of the DSNB fertilization approach is its ability to synchronize N supply with dynamic plant N demand by directly responding to the actual NO₃⁻ availability in the root zone. This capability was illustrated in the lettuce experiment, where a sharp decrease in soil NO₃⁻ concentration during the peak growth phase signaled an increase in plant N demand. In response, the DSNB approach increased N supply significantly to maintain concentrations within the target range, thereby optimizing supply to meet demand and resulting in a 23% yield improvement. This synchronization can also function in the opposite direction – for example, a temporary decrease in crop N demand due to external stress may cause a rise in soil NO₃⁻ concentration, allowing the system to reduce N application and prevent environmental N losses. The DSNB approach can also compensate for fluctuations in soil N availability driven by soil processes such as nutrient transport and biogeochemical reactions. While traditional fertilization guidelines attempt to predict these complex soil dynamics alongside varying N demand, our approach bypasses these challenges by responding directly to the net outcome of all processes. Consequently, the system maintains root-zone NO₃⁻ concentrations regardless of the specific driver. Whether a decrease in concentration stems from increased plant uptake or leaching due to heavy rain, the system identifies the nutrient deficit and automatically compensates by adjusting N application to maintain the target range.

The difficulty of current fertilization methods in synchronizing N application with actual crop demand is illustrated by the significant ‘farmer-driven excess’ identified in the U.S. vegetable production sector. This excess is particularly pronounced in leafy green crops, where shallow root systems and high susceptibility to nutrient deficiency lead to over-application as a precautionary strategy. By detecting real-time soil concentration and corresponding variations that reflect actual crop N demand, the DSNB approach can replace predetermined, surplus-based applications with precise, adaptive management, offering a potential annual reduction of ~33,000 Mg N in the U.S. alone.

This assessment focuses on the U.S. sector mainly due to the data availability, however, the identified ‘farmer-driven excess’ and the demonstrated potential for mitigation are globally relevant. The DSNB approach is designed to be crop, soil, and climate-independent, offering a significant practical advantage as it is more adaptive, simpler to manage, and easily adjustable to diverse agricultural conditions.

To evaluate the practical advantages of the DSNB approach, we conducted a comparative economic and environmental benefit analysis (Table 2). In this assessment, we calculated the net benefit by combining crop revenue, fertilizer expenses, and the estimated environmental value of reducing NO₃⁻ leaching and N₂O emissions. Crop revenue was calculated using farm-gate values of 1000 ± 100 USD ton⁻¹ for lettuce and 1250 ± 200 USD ton⁻¹ for bell pepper³⁹. Fertilizer expenditures were estimated using a unit cost of 5.8 ± 0.5 USD kg⁻¹; reflecting current market prices for technical-grade, nitrate-based fertilizers utilized in greenhouse fertigation systems, calculated as the average price from bi-weekly USDA-AMS production cost reports⁴⁰. Beyond direct input costs, we internalized key environmental externalities: NO₃⁻ leaching was valued at 10 ± 5 USD kg⁻¹ N, based on standard environmental economic valuations for agricultural N management⁴¹, and N₂O emissions were internalized using a social cost of 78 ± 15 USD kg⁻¹ N₂O-N, derived from recent multimodel damage function assessments⁴². Our results demonstrate a potential net economic benefit of 14,514 ± 1445 USD ha⁻¹ for lettuce and 553 ± 156 USD ha⁻¹ for bell pepper. For lettuce, the significant net benefit was primarily driven by increased crop yield. In contrast, the positive yet lower benefit margin for bell pepper was attributed to reduced fertilizer and environmental externality costs, despite a marginal, non-significant yield decrease. These findings underscore that improving crop yield is the primary path for achieving DSNB profitability. However, achieving yields comparable to standard practice remains a practical and viable pathway. Bell pepper management was performed manually, as the DSNB implementation had not yet reached full maturity; if full maturity allows for comparable yield with standard practices, theoretical projections indicate a potential net benefit by cost reduction of approximately 3,500 USD ha⁻¹. While the exact capital and operational costs of the DSNB system depend on future commercialization and industrial scaling, the net benefits calculated in this study can provide a baseline for estimating target system costs and potential economic returns.

Beyond practical field management, these results carry significant implications for the evolution of agricultural policy. Current agricultural regulations for N management, developed to protect water resources, primarily focus on limiting the total amount of N fertilizer applied per area⁴³. However, our research reveals that restricting N application quantity is insufficient, whereas synchronization of N application, which relates to the actual crop demand, is more crucial in preventing environmental N pollution. In addition, quantity-based regulation can lead to yield reduction, negatively affecting both farmers’ income and broader food security. Specifically, groundwater contamination from agricultural N constitutes non-point source pollution^44,45, which makes it challenging to attribute specific pollutants to individual farms and enforce limits on excess application. Soil NO₃⁻ monitoring can be used to initiate a monitoring, reporting, and verification (MRV) framework for nitrate leaching, a framework originally developed for greenhouse gas emissions to measure emissions and mitigation efforts⁴⁶. By utilizing MRV, regulators could transition from restricting N application quantity toward regulating the actual mass of N leached into the environment. The adoption of such framework could start with voluntary pilots and cost-share incentives, eventually proceeding as mandatory in specific areas such as nitrate vulnerable zones⁴⁷. Implementing real-time soil NO₃⁻ monitoring provides a pathway towards more precise and adaptive regulatory approaches, moving beyond the limitations of a gross reduction in N application. Ultimately, real-time soil NO₃⁻ monitoring could provide data that are essential for optimized N management, while empowering farmers to adopt sustainable practices, rather than enforcing restrictive regulations⁴⁸.

While this study represents a pioneering effort in implementing soil monitoring technology for dynamic fertilization, several challenges must be addressed to support widespread adoption. A major challenge in the widespread field-scale applicability of soil-N-based fertilization is related to soil heterogeneity and the representativeness of the soil N measurements. This requires additional research into how multi-point measurements in a field can accurately represent the entire area. Accordingly, the density and spatial distribution of monitoring points are essential for proper representativeness⁴⁹. In addition, this study monitors soil NO₃⁻ as a representative of total soil N availability, assuming that in sandy aerated agricultural soils, other N forms such as ammonium (NH₄⁺) are rapidly converted to NO₃⁻ through nitrification. However, this approach is limited in specific environments, such as heavier or poorly aerated soils where nitrification rates are restricted, and NH₄⁺ may become a dominant N form⁵⁰. In addition, some crop cultivars exhibit physiological preferences for NH₄⁺ uptake, hence monitoring NO₃⁻ may not represent the actual N demand of the plant⁵¹. Consequently, current DSNB applications are optimized for NO₃⁻ dominant systems, to address more complex environments where both NO₃⁻ and NH₄⁺ are dominant future research should integrate in the monitoring system also NH₄⁺ sensing⁵².

Another knowledge gap affecting precise N management is the current broad estimation of the optimal soil N concentration for crop N uptake. Existing research often focuses on comparing N application quantities rather than soil N concentration levels. Our study shows that soil N monitoring can improve our understanding of the relationship between soil N availability and plant uptake, providing the infrastructure for future research to precisely define optimal concentration ranges for various crops and phenological stages. This may be addressed by trials that evaluate specific soil NO₃⁻ concentration ranges rather than traditional N application rates to quantify their direct impact on uptake kinetics, yield, and environmental losses. Another critical pathway for further improving NUE lies in the integration of soil NO₃⁻ monitoring with Variable Rate Application (VRA) technology. While VRA provides the technical ability to differentiate fertilizer delivery across a field, DSNB provides the high-resolution information and decision support required to guide those applications in real time. The integration of these technologies can create a dynamic, adaptive system that addresses spatial field variability while synchronizing nutrient supply with temporal fluctuations in plant demand. Ultimately, addressing field-scale representativeness, defining concentration-based thresholds, and integrating real-time monitoring with VRA systems will enable the optimization of N fertilization and lead to a substantial reduction in agricultural N-related environmental pollution.

Soil NO₃⁻ monitoring system (SNS)

A soil NO₃⁻ monitoring system (SNS) was utilized for continuous, real-time measurement of soil NO₃⁻ concentrations. The SNS allows for continuous, real-time monitoring of NO₃⁻ concentrations throughout the soil profile. Detailed descriptions of its technical design have been provided in previous studies^32,53. The SNS operates with a UV light source and a UV–VIS spectrometer, which measures the absorbance spectrum of SPW within optical flow cells. The UV spectral measurements were calibrated for NO₃⁻-N concentration using a set of SPW solutions collected from the experiment-specific soil across a range of DOC and NO₃⁻ concentrations, following the procedure described in Yekutiel et al.³⁴. These optical flow cells are connected to custom-designed suction cups positioned at specified depths across the soil profile. A vacuum pumping system maintains a steady, low-flow rate of SPW (<10 mL h⁻¹) from each suction cup through the optical flow cells to enable spectral analysis and determination of NO₃⁻ concentration. Additionally, the SNS facilitates automated porewater collection for further laboratory chemical analysis. These custom suction cups, featuring a small dead volume and high sampling efficiency, are connected to the optical flow cell via narrow tubing (1.9 mm diameter) to minimize dead volume between the soil monitoring area and flow cell. The system was operated using a Raspberry Pi microcontroller. During the bell-pepper experiment, the SNS system encountered technical problems for part of the season, necessitating manual sampling and analysis using a standard colorimetric method⁵⁴. In the lettuce experiment, manual samples were collected once a week and analyzed using the colorimetric method for validation of the SNS system. All samples collected were filtered immediately through a 0.45 µm filter and stored under refrigeration until laboratory analysis, which was conducted within one week of sampling. The accuracy of the SNS spectral measurements was validated by comparison with the standard colorimetric method⁵⁴ using Deming regression and Bland-Altman analysis (see Supplementary Fig. S1). The Deming regression suggests a good agreement between the two methods (R² = 0.90, RMSE = 5.92 mg L⁻¹, N = 177), following the relationship y = 1.07×–3.36. Complementary Bland-Altman analysis shows a random distribution of residuals and a small mean bias of −1.39 mg L⁻¹ with 95% limits of agreement between −13.30 and 10.52 mg L⁻¹. This analysis confirms that the SNS system provides reliable in-situ measurements with no systematic deviation from established laboratory methods.

DSNB fertilization approach

The DSNB fertilization approach involved modifying N application according to real-time soil NO₃⁻ concentration measurements to maintain soil NO₃⁻ levels within a predefined concentration range. This range was determined to be 40 ± 10 ppm, selected based on the maximum concentration threshold reported in other research and designed to prevent any potential yield reduction^26,34,55. While this range may vary across different crops and growth stages and carries inherent leaching risks depending on environmental conditions, it was utilized here as a stable operational benchmark to examine the system’s applicability and ensure sufficient plant uptake under varying demand. The bell pepper experiment served as a preliminary examination of the DSNB approach, where N application was frequently manually adjusted during the season. In the lettuce experiment, the DSNB approach was executed via a fully automated, closed-loop system. The microcontroller utilized for the SNS measurements also handled data processing, calculated the daily N application rate, and managed fertilizer delivery by controlling the operation time of a fertilizer injection pump. The N application rate was determined daily using a PID-based control algorithm^56,57 calculated as:

Where:

• (uleft(tright)) is the calculated N application rate.

• baseline Is the N application rate applied when the system error (eleft(tright)=0)

• (eleft(tright)) is the current error (C_threshold − C_measured).

• (int eleft(tright){dt}) is the integral term, represents the accumulated past error

• (frac{{de}(t)}{{dt}}) is the derivative term, represents the rate of error change.

• ({K}_{p},{K}_{i},{K}_{d}) are the linear coefficients (weights) for the proportional, integral, and differential components, accordingly.

The control logic comprises a baseline value, applied when the soil concentration matches the 40 ppm threshold, and a corrective part calculated based on the proportional, integral, and derivative terms of the error. The algorithm’s parameters were tuned throughout the growing season to stabilize the NO₃⁻ concentration around the 40 ppm threshold at a depth of 10 cm (Table 3). In both experiments, DSNB fertilization approach was compared with a predetermined fertilization approach that followed standard regional growth protocols.

Experimental setup

Two experiments were designed as parallel-group randomized trials (1:1 allocation ratio) to compare the DSNB fertilization approach with a predetermined fertilization approach (Fig. 5). A controlled lysimeter experiment was conducted at the Sde Boker Campus of Ben-Gurion University of the Negev, Israel (30°51′14.2″N, 34°47′01.1″E). The setup included 24 lysimeters (75 L volume, 50 cm height, 44 cm diameter) packed with sandy soil (93% sand, 3% silt, 4% clay, 1% organic carbon) and planted with three lettuce plant each (Lactuca sativa L., “Romit Noga” variety). The lysimeters were organized into six blocks of four units each, with treatments assigned randomly at the block level. The growing season lasted 62 days from December 2024 to January 2025.

Additionally, an experiment with bell pepper (Capsicum annuum) was conducted in a semi-industrial 1000 m² net house, allowing for examination of the approach under field conditions. The experiment was conducted at the Yair Agricultural R&D Center, Central Arava Valley, Israel (30°46’40.1“N, 35°14’21.8“E) from September 2023 to April 2024. This experiment comprised 12 plots, 6 plots per treatment (12.5 m × 1.5 m each) at a density of 3.3 plants m⁻². The plots were randomly distributed throughout the net-house using a randomized design. The sample size (n = 24 for lettuce, n = 12 for bell pepper) was determined by the maximum capacity of the experimental facilities, providing sufficient independent replicates to satisfy the requirements for parametric statistical testing (t-tests) and to account for spatial variability. In both experiments, the soil NO₃⁻ concentration was continuously measured at multiple depths within and below the root zone (10, 20, and 40 cm for lettuce, 20 and 50 cm for pepper), with 3–4 replications for each depth. Soil water content was measured in conjunction with soil NO₃⁻ concentration using TDT sensors (SDI-12 TDT, Acclima, Inc., Meridian, ID, USA). NO₃⁻ leaching was quantified by multiplying the below the root zone NO₃⁻ concentration by the volume of the leachate. In the lettuce experiment, the leachate volume was directly measured by weighing the leachate collected from the lysimeters. Conversely, in the bell pepper experiment, the leachate volume was indirectly estimated over a two-month period by calculating the water mass balance and following the equation:

Where (I) is the daily irrigation, ({{ET}}_{p}) is the daily potential evapotranspiration calculated using the Penman-Monteith equation⁵⁸ based on data from an adjacent meteorological station⁵⁹ and ({K}_{c}) is stage specific crop coefficient value⁶⁰. Changes in soil water storage (∆S) were omitted from water mass balance calculation because continuous volumetric water content measurements remained stable throughout the period, suggesting ∆S is negligible.

In both experiments, irrigation was applied through a drip irrigation system (4 drippers of 0.5 L h⁻¹ for each lysimeter, 1.6 L h⁻¹ for bell pepper, spacing 40 cm along the row). In the lettuce experiment, irrigation water was fresh (EC ~ 0.3 mS cm⁻¹), whereas in the bell pepper experiment, it was brackish (EC ~ 2.5 mS cm⁻¹) and applied with a high leaching fraction of ~50% to prevent salt accumulation. Fertilizers were applied through the drip irrigation system using a fertilizer injector from concentrated fertilizer tanks. In the lettuce experiment, concentrated solutions were prepared using PeaK Mono Potassium Phosphate (ICL) and Micro-Combi (Haifa) for PK and micronutrients, with N supplied as NO₃⁻ via Magnisal™ (Haifa Mag). For the bell pepper experiment, fertilization utilized a liquid NPK 7-3-7 fertilizer (Arava, ICL group) containing 66.6% NO₃⁻ and 33.3% NH₄⁺. In the lettuce experiment, all plants from 11 of the 12 lysimeters of each treatment were harvested at the end of the season to determine total fresh yield. The remaining two lysimeters, one of each treatment, were removed mid-season for destructive analysis (not included in the results). Bell pepper plants were harvested 11 times throughout the season. The harvested fruit was weighed and sorted into export and non-export quality categories to calculate total export yield. For both crops, NUE was calculated as the partial factor productivity (kg yield kg N⁻¹, PFP_N)³⁶ by dividing the fresh yield by the total mass of N applied.

Gaseous fluxes measurements

Soil N oxide fluxes were evaluated in the field in both the bell paper and the lettuce experiments, using manual accumulation (for nitrous oxide; N₂O)^61,62,63 and steady-state (for nitric oxide; NO)⁶⁴ static chambers adapted to our field conditions. Briefly, for N₂O flux estimation, a gastight lid made of opaque polyvinylchloride tubes (PVC), connected to a quantum-cascade laser analyzer (QCL; GLA151-N2OM1; ABB, Quebec City, QC, Canada), was placed on the soil bases made of the PVC (19 cm i.d., 0.5 cm wall thickness, 10 cm height). After lid placement, the total chamber volume was ~3 L. For the flux estimation, the N₂O concentrations (ppbv or nL L⁻¹) in the chamber headspace was measured, and the flux was calculated from the rate of the N₂O concentration increase. For the NO flux estimation the steady-state chamber used the same soil bases and had the same total volume as the accumulation chamber but a different lid, allowing free exchange between the chamber air and the atmosphere while connected to the NO-NO₂-NO_x analyzer (42i TL; ThermoFisher Scientific). Instead of measuring the increasing N₂O concentration in the gas-tight accumulation chamber headspace, the method measures concentration differences between the inlet air and the outlet air coming into or from the chamber’s headspace and then reaching a steady-state. Gas fluxes (µmole N₂O m⁻² s⁻¹) and nmole NO m⁻² s ⁻¹) were calculated from the rate of gas buildup during a 5-min incubation or by the difference between the inlet and outlet gas concentrations⁶³. Limit of detection (LOD) of chamber methods was estimated using literature values from measurements of N₂O and NO fluxes using same instrumentation and was found to be 0.05 nmol N₂O m⁻² s^{−1 65} and 0.03 nmol NO m⁻² s^{−1 64} which are efficiently lower than measured fluxes. In the lettuce experiment, measurements were taken ~bi-weekly in five lysimeters per treatment. For the bell pepper experiment, measurements were taken 11 times throughout the growing season, with four replications per treatment.

Data analysis and preprocessing

A statistical analysis was conducted on all quantitative data collected in the experiments. Prior to statistical testing, the raw gas emission flux measurements NO and N₂O were preprocessed to ensure data quality. This included identifying and managing outliers using the Interquartile Range method (1.5 × IQR), which were subsequently replaced with the mean of the non-outlier values within the same date and treatment group to ensure robust flux interpolation. Daily emissions were interpolated using the trapezoidal rule^37,38 to calculate cumulative emissions in kg N ha⁻¹ season⁻¹. Based on these measurements, data were interpolated to calculate the seasonal cumulative flux (kg N ha⁻¹ season⁻¹). The resulting cumulative flux for each replication was utilized for the final statistical analysis. All statistical analyses were performed using Jamovi, an open-source statistical software⁶⁶. The assumptions of normality and homogeneity of variance were assessed using the Shapiro-Wilk test and Levene’s test, respectively. The majority of the data met these assumptions and means were compared using the Student’s independent t-test. The sole exception was NO_x emissions in the Bell pepper experiment, which did not meet the assumption, and treatments were compared using the Mann–Whitney U test. Differences between treatment means were considered statistically significant when the probability value (p) was less than 0.05. The data for NO₃⁻ leaching in the bell pepper experiment was not analyzed statistically as the measurement was based on a composite sample and thus lacked sufficient independent replication. For clarity, treatments’ performance is described in the text as the percentage of change relative to the control group (predetermined fertilization). To provide statistical inference strength for these relative outcomes, we calculated 95% confidence intervals (CI) for the percentage change by scaling the lower and upper bounds of the mean difference CI by the control group mean. To address statistical power and the magnitude of observed effects, Hedges’ g and 95% CI were calculated using the standard approximation formula⁶⁷. Detailed statistical results for each parameter are provided in Supplementary Table S1.

U.S. vegetable crop data analysis

To estimate the potential for reducing N fertilizer application on a broader scale, we utilized the U.S. vegetable production sector as a representative case study due to the availability of a comprehensive dataset on actual farmer N application rates. We compared crop-specific N fertilization recommendations and farmers’ actual application rates and defined “farmer-driven excess” as the ratio of actual application rates to the recommended rates. We assumed that a lack of real-time knowledge on soil NO₃⁻ availability is a major contributor to this farmer-driven excess. The annual mass of N that could potentially be reduced was then estimated by multiplying the excess rate by each crop’s cultivated area. Data for actual application rates were obtained from the 2018 National Agricultural Statistical Service (NASS) survey³⁵. Optimal fertilization guidelines were sourced from the University of California Davis (UCD) and the California Department of Food and Agriculture (CDFA)¹⁵. The NASS Chemical Use Survey of Vegetable Producers is a comprehensive dataset providing detailed information on chemical fertilizer use across different vegetable crops from major producing states. For this study, we extracted the reported applied N mass per unit area for 10 vegetable crops from the 2018 multi-state survey, which accounts for 83% of the annual N use in the U.S. vegetable production sector. The selection of these specific crops was based on their significance in terms of total cultivated area, the availability of recommendation guidelines, and the presence of N application data. In instances where specific multi-state data were unavailable, a weighted average, based on the individual state’s reported planted area, was calculated from available state-level NASS reports. The California Department of Food and Agriculture (CDFA), Fertilizer Research and Education Program (FREP), and the University of California Davis’s (UCD) Department of Land, Air and Water Resource (LAWR) collaboratively develop crop-specific fertilization guidelines. These guidelines, synthesized from peer-reviewed journal articles and research reports, aim to provide nutrient information to optimize crop demand while reducing environmental impact. From these guidelines, we specifically used the recommended N application rates for each crop. To maintain a conservative approach, for recommendations contingent on initial soil N concentration, we adopted the maximum recommended value, assuming no initial soil N was available. Additionally, if recommendations were provided as a range or varied between winter and summer growing seasons, an average value was calculated and used in our analysis.