NitroNet: Smart System to Quantify Nitrous Oxide Emissions FINAL TECHNICAL AND SCIENTIFIC REPORT U.S. DOE ARPA-E Award No. DE-AR0001385 PI NAME: Mark A. Zondlo Professor of Civil and Environmental Engineering Princeton University CEE / EQuad E206A Princeton, NJ 08544 Email: mzondlo@princeton.edu Phone: (609) 258-5037 Fax: (609) 258-2760 Project team members: Vlad Sevostianov, Nathan Li – Dept. of Civil and Environmental Engineering, Princeton University Josh Collins, Paul Guiguizian, Howard Bell – Intelligent Material Solutions Inc. Julie Bushell - Ethos Connected Inc. Paul Zorner, John Wolfe - EcoFluxNet Inc. RECIPIENT ORGANIZATON: Princeton University Princeton, NJ 08544 DUNS NUMBER: 00-248-4665 PROJECT/GRANT PERIOD 5/21/2021 - 8/31/2024 Final Report SUBMISSION DATE: July 16, 2025 1 A. Public Executive Summary Agricultural croplands are the largest anthropogenic source of nitrous oxide (N2O), the third most important greenhouse gas. Emissions are characterized by “hot spots” and “hot moments”, meaning emissions are highly heterogeneous in space and time. Emissions are driven by nitrification and denitrification processes which depend upon complex factors such as soil characteristics (type, compaction, pH, moisture, topography), management practices (fertilizer type and application, tillage, crop type, field history, irrigation), biogeochemistry (organic carbon, microbial composition), and meteorology (precipitation, temperature, and wind). For these reasons, quantifying cropland emissions by either measurements or modeling is extremely challenging with large uncertainties. While the bioeconomy has much promise for replacing conventional fuel sources (e.g. sustainable aviation fuel) with large benefits to rural economies and domestic fuel supplies, assessing the full climatic impact of these changes is poorly known due to the challenges of measuring cropland emissions of N2O. In this project, we developed and deployed a new optical imaging technique to quantify the spatiotemporal patterns of N2O emissions in real-time in the field. In essence, we developed a “cat scan” imaging system of an agricultural field through casting an “optical web” of eye-safe laser beams across a field. The “NitroNet” system is composed of two “towers” where mid-infrared lasers scan across a field. The laser beams are reflected back to the towers through inexpensive mid-infrared retroreflectors that line the perimeter of a field on simple stakes, fence posts, or utility poles. Mid-infrared spectroscopy allows for the highest sensitivity and selectivity for N2O detection compared to other spectral regions because the fundamental (strongest) absorption bands are in this spectral window. Absorption spectroscopy is used to determine path-integrated concentrations. The path-integrated measurements from the reflected laser scans of each reflector results in a complex pattern of overlapping path-integrated concentrations, allowing for a tomographic reconstruction of the concentrations across the field. When coupled to meteorological data, spatially-resolved maps of emissions from the field are obtained at a spatial resolution only limited by the spacing of the retroreflectors on the field’s perimeter. The NitroNet system was deployed in a soybean field in eastern Nebraska in summer 2024, and a controlled release of N2O was conducted within the 1.2 acre field demonstration with a modest number of (16) reflector positions (~ 4 per side on average). NitroNet scanned across the field continuously in an automated manner every 37 minutes with no re-alignment needed after the initial setup. NitroNet located the emission source ±2 m of the actual release location (confined to within 0.26% of the measured area) and measured emissions that were ±15% of the actual release rate. Because NitroNet’s spatial resolution scales only by the number of reflectors and returning light on the detector (demonstrated to at least 400 m), the specific results are equivalent to an 80 acre field (570 x 570 m) with a spatial resolution of 0.20 acre (±16 m). These metrics are much improved over DOE SMARTFARM goals of a spatial resolution of 1 acre (±36 m) in an 80 acre field and 20% accuracy. More retroreflectors aligning the perimeter would allow for even higher spatial resolution, and higher power lasers are now commercially available could extend the distances to 1 km for even larger fields. Hotspot locations of N2O in the field were also confirmed with the development and deployment of a drone-based N2O sensor. Flux detection limits were below to well below the SMARTFARM target of 10 g N2O-N ha-1 day. 2 The technologies developed in this grant include the following: 1) NitroNet system for spatiotemporal quantification of N2O in croplands; 2) inexpensive retroreflectors for mid-infrared sensing; 3) open-path, drone-based N2O sensor for mapping N2O in croplands, other agricultural (feedlot), and industrial (e.g. wastewater) sectors; and 4) center pivot irrigation methods for quantifying N2O emissions. Patent applications, and associated invention disclosures, have been filed on these intellectual properties. Finally, a startup company, EcoFluxNet™ Inc. (www.ecofluxnet.com), has been formed from the NitroNet program to quantify N2O and CH4 emission reductions that farmers implement. EcoFluxNet has exclusive license agreements to the IP from Princeton University from this award. EcoFluxNet Inc. is a Delaware C-Corporation. EcoFluxNet Inc. is also a joint partner in the non-profit venture VeriAg™ (https://veriag.com) with Global Sustainability Developers™ (GSD) (https://www.gsdpartnership.com). VeriAg™ will pay farmers to make sustainable choices through emissions reduction technologies (verified through EcoFluxNet Inc.) and soil health/carbon/water improvements (verified through GSD Inc.). VeriAg will use the VeriFood™ (https://dev.ethosconnected.com/solutions/verifood) platform to trace these sustainable agriculture products from farmers to end users. VeriAg, with traceability and verifiable carbon credits, will sell carbon credits at a premium to third parties with proceeds distributed to the farmer, EcoFluxNet and GSD. Recent “greenwashing” lawsuits cast doubt on carbon neutrality pledges of large companies. The market is seeking defensible carbon credits that are traceable and verifiable with rigorous data. Currently, very few farmers will risk the higher costs associated with implementing sustainable practices with only the “hope” that higher commodity prices will cover these costs. In contrast, VeriAg, using the EFN (air) and GSD (soil) measurements, will pay farmers to implement sustainable practices, improve rural economies through new income streams to farmers, enable the bioeconomy, and add robustness to the floundering carbon credit markets. Overall, this is a win for rural America through new injection of capital, the bioeconomy through data-verified practices, and the environment through lower emissions of greenhouse gases. 3 B. Acknowledgements The authors acknowledge DOE ARPA-E for supporting this award including helpful discussions from program managers Dr. David Babson, Dr. Olga Spahn, and Dr. Steven Singer. Their flexibility in dealing with our progress associated with supply chain shortages and other delays (limited lab staffing with social distancing) associated with the immediate post-COVID world was also appreciated. We also appreciate the tech-to-market insights from Dr. Calden Stimpson, who was enthusiastic on the potential for this project and provided key insights into the carbon markets and associated commercialization pathways. Finally, we acknowledge Mr. James McSpiritt and Mr. Lars Wendt of Princeton University for critical assistance with field deployment and project logistics, along with other members of the Zondlo group at Princeton University who have contributed to this project including Dr. Lei Tao, Dr. Hongming Yi, Dr. Da Pan, Ryan Boyd, Yunpo Li, and Yunseo Choi. The authors particularly thank Dr. Lei Tao for helping in the very early stages of the project and associated proposal stage. We appreciate the staff at Intelligent Materials Solutions Inc. and Ethos Connected, Inc., especially Ander Christensen for field testing and deployment. The PI thanks Alison Coakley, Soonoo Aria, and Courtney Kohut at Princeton University for endless patience and assistance with budget management and contractual management. C. Accomplishments and Objectives The goals of NitroNet as proposed were successfully achieved with even better performance than proposed specifications. The key tasks and milestones from the beginning of the project are shown below, along with a summary of actual performance. The key project outcomes and results for NitroNet are described in more detail in Section D. Quarterly project reports have more information on each task but are summarized in the table below. WBS Milestone Title Milestone Description Go/No-Go: Go-No/Go Milestone: Refine tasks and milestones for the work plan (if Refine tasks and applicable) M1.1 milestones (if Actual performance: Not applicable – this was done prior to the applicable) contract start date per negotiation with ARPA-E. Characterize laser current tuning, temperature tuning, and optical power as a function of wavelength. Capable of scanning over an absorption line at laser operation temperatures of 20-40ºC with at least 10 mW optical power. Deliverable 2.1: Select COTS lasers (ICL or QCL) Laser for sUAS and tower setups based upon measured performance metrics. M2.1 characterizations Actual performance: (11/21/21) We selected a Thorlabs continuous wave, 4.57 µm quantum cascade laser with ≥ 100 mW of power for the tower sensors and an interband cascade laser at 4.54 µm with > 40 mW optical power for the drone sensor at room temperature. A preliminary design review will be held for the small unmanned aerial vehicle sensor. Deliverable 2.2: Documentation on the optomechanical, M2.2 sUAS sensor design electronic, and software designs. Design targets are a Herriott cell with an optical pathlength between 10-30 m, power consumption of 10 W (excluding laser), mass 2 kg, and volume of 1.5 L. 4 WBS Milestone Title Milestone Description Actual performance: (completed 2/20/22) The optical cell had 58 passes with a mirror separation of 38 cm and a pathlength of 22.4 m. The mass was 1.6 kg including the electronics and the sample volume of 0.32 L. Power consumption was 8 W excluding the laser and 10-15 W for total power. The system precision in air will be quantified in the laboratory against various retroreflectors as a function of area of the retroflector, type of retroreflector, and distance from the laser using COTS components. Deliverable 2.3: Design of the shape, area, size, coating, and manufacturing specifications of plastic retroreflectors. Retroreflector Go/No-Go Decision Point #1: Ability to achieve a precision of at least 0.1 M2.3 backscatter tests ppbv N2O in the laboratory at roundtrip distances of ~ 20 m with a coated, plastic-molded reflector. Actual performance: (completed 2/20/22) Ambient N2O was measured with the retroreflector with a 17 m path with a signal-to-noise ratio of 3100:1 or 0.1 ppbv. sUAS sensor fabricated and demonstrated in the laboratory to achieve 0.1 ppbv N2O at 1 Hz against a calibrated standard of N2O near atmospheric concentrations. Deliverable 2.4: Allan variance plot showing the precision at timescales up to 600 s. Go/No-Go Decision Point #2: The sUAS sensor will have a short term sUAS sensor M2.4 precision of 0.1 ppbv N2O for a one second integration with a stability at fabricated or below this value for at least 600 seconds. Actual performance: (completed 8/20/22) Sensor was fabricated and tested and showed a precision of 0.1 ppbv at 1 s with stability of 0.07 ppbv for up to 600 s. A final design review will be held for the tower system. Deliverable 2.5: Delivery of a design package including documentation of the optomechancial, electrical, and component schematics of the tower system in a design package. Go/No-go Decision point #3: Design targets will include system overall cost of <$2000/tower in mass quantities, operating temperatures from - M2.5 Tower sensor design 10 to 40ºC air temperatures, scanning capabilities of 180º, wireless access and control, and an average electrical power consumption of 20 W excluding the laser. Actual performance: (completed 8/1/22) We successfully designed the tower laser system with component costs at $1900 in large quantities (target $2000), power consumption excluding laser less 5 WBS Milestone Title Milestone Description than 20 W (8 W), and electronics operation under a wide range of field temperatures (-40º to +50ºC). Fabrication of the tower sensor system. Deliverable 2.6: Tower system and associated components assembled including 16 retroreflectors. Tower sensor M2.6 Actual performance: (completed 5/1/23) Supply chain delays led to a fabricated gimbal to be delayed, and the tower was not completed until six months after the target date. Two tower systems were tested and deployed in spring 2023 outside at Princeton University. The system will be demonstrated outside for measuring atmospheric N2O over a ≥ 400 m roundtrip distance. Deliverable 3.1: Report on the field performance for ambient conditions and precision metrics and system scanning timescales for full system (amount of time spent per reflector). Go/No-Go Decision Point #4: A precision of at least 100 ppbv-m (mW laser optical power)-1/2 will be obtained for a simple retroreflector and M3.1 Outdoor tower tests tower system at ≥ 400 m integrated pathlength for a 5 minute measurement. Actual performance: (completed 6/1/23) The system measured a 400 m optical pathlength with precision equivalent to 0.1 ppbv in 10 s, or 40 ppbv-m with 10 mW of return signal. This timeframe is better than the performance metric by over a factor of 10 in integration time and a factor of two in precision. A controlled release of N2O will be conducted upwind of a grid containing 10 plastic retroreflectors. The sUAS will also collect data around the release point for additional checks. Deliverable 3.2: Timeseries of N2O before, during and after the release and maps from the sUAS. Actual performance: (completed 7/1/23) A 4x4 grid of reflectors at 10 m M3.2 Controlled release spacing with up to 99 m roundtrip distances, plus two on each tower, was setup with a controlled release of 0.3 g s-1 of N2O. Large enhancements up to 8000 ppbv N2O (path-normalized) were observed at the furthest distances when the tank was on with ambient levels (335 ppbv) when the tank was off. Precision <0.25 ppbv was observed in field conditions. Drone-based flights were conducted on a Tarot T18 drone and showed significant enhancements when transecting the plume. Tomographic reconstruction of 16 (4x4 array) retroreflector controlled release with an estimated emission rate and location and associated M4.1 Modeling errors in fluxes. Deliverable 4.1: Report on the emission rate and associated errors. 6 WBS Milestone Title Milestone Description Go/No-Go Decision Point #5: The location of the leak will be correctly identified to within one grid cell of the tomographic reconstruction and within 40% uncertainty of the actual release rate. Actual performance: (completed 11/20/23) Plume location was located within 8.9 m of the release point on the 4x4 array with a 10 m grid spacing and the error of emission rate was 0.21 g N2O s-1, which was within 40% of the true release rate of 0.3 g N2O s-1. NitroNet will be deployed at a farm for select weeks to test setup, system components and system operation. Deliverable 5.1: Report on the deployment and testing protocol including ease-of-setup, light intensity on detector, alignment of system, accuracy of pointing, and wifi operations. Successful criteria is a full-farm scan image of emissions every 30 minutes. Field test M5.1 demonstrations Actual performance: (completed in June 30, 2024) NitroNet was set up in June 2023 in a soybean field at the University Nebraska at Lincoln. Scanning of 32 reflectors (16 positions) was done automatically, though fine manual alignment had to be done after each movement due to looseness in the gimbal azimuth. Alignment was done with a co-aligned green laser at night. Comparison of sUAS, eddy covariance, and NitroNet fluxes at a field demonstration site. Deliverable 5.2: Temporal resolution of fluxes at the daily scale and associated uncertainties from NitroNet and comparison to sUAS spatial patterns and eddy covariance temporal patterns. Actual performance: (July 31, 2024) A second test at the University of M5.2 Flux comparison Nebraska at Lincoln soybean fields was conducted with an improved gimbal adjustment to avoid the backlash (only going one direction and adding resistance to its one direction with a spring). Scanning the field took 37 minutes with a dwell time of two minutes per reflector at the 16 positions. The system operated overnight and into the next day without any alignment problems. A drone-based flight also showed similar patterns of the controlled release of a tank of N2O. NitroNet will be demonstrated in a field environment at the ARPA-E Phase I site. Deliverable 6.1: Timeseries of path-integrated concentrations during the field testing and precision of 20 ppbv-m N2O will be demonstrated for a period of at least one month after fertilization application. M6.1 ARPA-E Phase I tests Actual performance: (completed August 31, 2024) Per discussions with program manager, field testing continued at the University of Nebraska Lincoln site in July with controlled releases due to the logistics of coordinating with the Phase I site teams and integration and field data of the existing field site. NitroNet was deployed for two weeks in June 7 WBS Milestone Title Milestone Description 2023, two weeks in July 2023, and one week in August 2023 for the various field testing. Because nitrogen was not applied to this field, we used a controlled release to demonstrate the capabilities. Precision was 20 ppbv-m or better for each reflector combination. Flux analyses and emissions for the Phase I tests will be quantified with associated uncertainties. Deliverable 6.2: Final report on the spatial and temporal resolution of fluxes at 1 acre resolution with an uncertainty of 20% to true values and a flux detection limit of at least 10 g N2O-N ha-1 day-1 will be achieved. Actual performance: (completed November 30, 2024 – post award analyses) Algorithms were developed to identify the locations and M7.1 Flux emissions magnitudes of fluxes. As described in Section D in this report, we quantified emissions with a controlled release of 0.3 g N2O s-1 over a 100x50 m plot to within 15% and showed equivalent spatial resolution of 0.2 acres on an 80 acre field. Signal-to-noise of the tomographic reconstruction was better than 3000:1 (measured by the precision of 0.1 ppbv in the field at 1 s integration – integration times were actually two minutes per position so the noise would nominally be better), which for this release rate over the field measurement area equilibrates to <8.6 g N2O ha-1 day-1. M8.1ARPA-E acceptance of a T2M assessment, including: (a) initial product hypothesis/es and differentiating features from incumbent technologies; (b) approach to techno-economic analysis; Technology-to-Market (c) customer discovery and transition plan; (T2M) assessment M8.1 (d) IP strategy; accepted by ARPA-E (e) market readiness assessment Actual performance: (completed 8/20/21) White space analyses and initial markets and products were identified. M8.2 Initial impact sheet Submission of draft Impact Sheet that describes the desired impact the drafted project team would like to have by the end of the project. Actual performance: (completed 8/20/21) Submitted to ARPA-E after discussions with Tech-to-Market SETA of ARPA-E. Establish standard data formats with ARPA-E SMARTFARM performers and other stakeholders M8.3 Dataset design Actual performance: (completed 8/20/21) Due to travel restrictions and lab staffing issues from the pandemic, we submitted a dataset plan to ARPA-E in quarterly review meeting and reports for our own testing. 8 WBS Milestone Title Milestone Description Analysis of IP white space and related literature to estimate domain of freedom to operate (FTO), including: (a) a list of searches performed (e.g. key words and combinations); IP white-space (b) search results; M9.1 analysis submitted for (c) formulation of initial patent strategy ARPA-E acceptance Actual performance: (completed 11/20/21) White space analyses of patent and associated patent protection plans were submitted to ARPA- E. ARPA-E acceptance of an initial T2M plan that assesses the value of the technology based on market impact to determine appropriateness of further work. This report should include findings from outreach to at least 5 potential customers to assess realism of team’s understanding of core product characteristics, as well as: (a) 2-3 product narratives based on the initial product hypotheses and identified alternative / first markets (b) critical performance and design features needed facilitate customer First iteration of T2M acceptance M10.1 plan accepted by (c) integration of technical and T2M findings into a preliminary TEA; ARPA-E (d) supply chain assessment (e.g. component costs and channel requirements); (e) potential follow-on funding arrangements. Actual performance: (completed 11/20/22) It was critical that any technology that quantified emissions, such as NitroNet, would need to partner with a carbon registry to be successful beyond selling niche instruments to researchers / flux communities. Hold first meeting with external advisory board, made up of at least three individuals from organizations that could be potential users of the technology, to receive feedback on the initial T2M plan. Establish Stakeholder Actual performance: (completed 2/20/23) The first stakeholder meeting M11.1 Advisory Board with the External Advisory Board was conducted in June 2023. Representatives of banks that supported green technologies, sensor developers, carbon markets, and agricultural products were included on the board. ARPA-E acceptance of an updated TEA that integrates technical and T2M findings and includes: (a) process diagram of inputs and outputs ARPA-E acceptance of M12.1 (b) input assumptions and source information an updated TEA (c) analysis of best, worst, and expected scenarios (d) sensitivity analysis 9 WBS Milestone Title Milestone Description Actual performance: (completed 11/20/23) Source assumptions and pro forma revenue plans were hypothesized for a startup company on the NitroNet project. ARPA-E acceptance of an updated T2M plan that integrates results of technical investigation, techno-economic analysis, and market analysis to assess impact, value, and plans for further technological development. Specific activities include: (a) refined techno-economic analysis; (b) results of market outreach, including formalization of partnerships with other entities to support next-stage development; Updated T2M plan M13.1 (c) target performance characteristics of final prototype necessary to accepted by ARPA-E spur adoption; (d) revision and/or expansion of IP plan; (e) propose organizational development requirements necessary to succeed in next-stage development activities. Actual performance: (completed 6/30/24) Updated business plan was submitted for EcoFluxNet on the NitroNet technologies. ARPA-E acceptance of a final T2M plan that is built off of T2MQ10 and incorporates ARPA-E feedback and additional technical and T2M findings. Final T2M plan M14.1 Actual performance: (completed post award, 6/30/25) A final pro-forma accepted by ARPA-E for EcoFluxNet Inc. was included with a joint venture for carbon credits and a tracking algorithm for these sustainably-produced commodities measured by NitroNet and other companies. 10 D. Project activities D1. Summary The NitroNet optical web system provides both spatial and temporal emissions resolution and is nominally only limited by the spacing and number of the reflectors. The system provides a “gold standard” for measuring emissions of interest, particularly N2O in agricultural settings. Agricultural N2O emissions contribute to 56% of anthropogenic budget of N2O, the third most important anthropogenic greenhouse gas with rapidly increasing emissions and a long atmospheric lifetime of ~ 110 years (Tian et al., 2024). Hot spots and hot moments can be effectively measured by NitroNet unlike any other technique, providing a single platform with which to optimize agriculture practices for lowering emissions and verifying emission compliance. We developed and deployed the mid-infrared, optical tomographic imaging system to quantify the locations and magnitudes of N2O emissions in an agricultural field. A controlled release field test in a 1.2 acre soybean field yielded a location accuracy of ±2 m and an emission rate ±15% of the release rate using 16 reflector positions. By scaling the result to an 80 acre field, this is comparable to 0.2 acre resolution of N2O emissions, significantly better than the program-specific goals of 1 acre resolution at 80 acres at 20% uncertainty in emissions. D2. Overall description The NitroNet system consists of low cost plastic mid-IR retroreflectors setup along the perimeter of the field and two tower based optical subsystems each consisting of a quantum cascade laser at 4.57 µm (QCL), mercury cadmium telluride (MCT) detectors, various mirrors, and associated electronics, among other components. Each subsystem is gimbal mounted can rotate horizontally and vertically. The QCL aims at any one retroreflector briefly before going to the next reflector. By quickly measuring at every single retroreflector, a single “scan” is completed. Both towers do this independently and simultaneously. After setup, scans can be done autonomously and repeatedly. All the data from any single scan are then combined and use a newly-developed algorithm to produce a tomographic image of the field. The data itself— independent of any model—is able to identify a hotspot and measure concentrations over the field. This computed output is then combined with Monte Carlo-based inversion models and meteorological data to estimate to identify hotspot precise locations and hotspot emissions. Furthermore, the integrated flux over the whole field can also determined. Besides the QCL, a visible laser is used to help with the alignment. During initial system deployment and setup, the position of each retroreflector is saved and recorded by a joystick- controlled gimbal, which moves the output beam to different retroreflectors. During measurements, the system automatically returns to these locations while completing scans. The resolution of the “pixels” measured on the tomographic reconstruction is determined by the density of the perimeter retroreflectors. More retroreflectors allow for more individual path-integrated measurements, and consequently when combined together, a finer tomographic reconstruction at higher spatial resolution. Ultimately, the maximum resolution is limited by the diffraction limit of the laser, which sets the maximum density at which retroreflectors can be deployed (~1 m at 500 m). 11 D3. Field Deployment The system was deployed on a 1.2 acre soybean field operated by the University of Nebraska, Lincoln. The soybeans were treated with a combination of biochar and biosolids with rain fed irrigation. 32 retroreflectors were setup on tripods at 16 locations (two at each location, one retroreflector pointing in the direction of tower one, and another pointing towards the direction of tower two) (Figure 1). The reflectors were setup evenly spaced, approximately 14 m apart. The two towers were placed at opposite ends of the field (approximately 100 meters apart). A two- minute averaging time was used at each retroreflector (the amount of time the tower pointed at the retroreflector before moving on to the next one). With the gimbal travel and reset times, this yielded approximately 37 minutes per complete field scan. Fig. 1: Deployment configuration of the NitroNet system showing the 16 retroreflector positions (two units at each position, 32 total) and the two towers in the soybean field (~ 1.2 acre plot). The reflectors were evenly spaced (~14 m apart), and the two towers with rotating gimbals were at the two opposite ends of the field (~100 m apart). For testing purposes, a controlled release experiment was performed. A tank of N2O was placed in the field at a known location at a fixed flow rate (10.6 L min-1 flow rate, ~ 0.3 g N2O s-1). Because the testing was being done in real world conditions in an agricultural field, a large hotspot in a known location had to be artificially created to demonstrate the functionality of the system. Otherwise, any hotspots that appeared could not be distinguished between real emissions features 12 or mathematical artifacts, nor would they be repeatable or verifiable. The flow rate was selected by taking the maximum daily emissions from an 80 acre corn field two months after fertilization and concentrating this to a single hotspot (Pan et al., 2021). Furthermore, a UAV with an open- path N2O point sensor was flown over the field to provide a second point of comparison for measuring concentrations and identifying the hotspot location. Due to rotor wash, the pixel resolution of the UAV based sensor was similar to that of the optical web. The entire system as deployed in the soybean field, including the UAV, is shown in Fig. 2. Fig. 2: The optical web and verification UAV in a soybean field. Low cost plastic retroreflectors were deployed along the perimeter of the 1.2-acre soybean plot, with the two optical tower ~ 100 m apart on opposite ends. An open-path sensor on a UAV was also used for comparison. D4. Tomographic Reconstruction and Inversion Modeling The positions of the retroreflectors and the towers, as well as the path-integrated measurements, were used to compute a tomographic map (Fig. 3). This tomographic map does not require wind information or any meteorological parameters, however, for context, wind direction is shown in Fig. 3. The locations of the retroreflectors (shown as blue squares) and the locations of the two towers (shown as red circles) were input as latitude and longitude coordinates (as recorded by a global positioning system (GPS)). A clear enhancement in measured concentrations was visible near the center on the right side of the field. This enhancement is exactly what would be expected from a large point source with the associated wind direction (i.e. downwind of the tank, shown by the black star). The GPS coordinates of the tank were not used in any way for the tomographic map and were only plotted afterwards for visualization purposes. While the tomographic reconstruction provides inferred concentrations for each pixel on the field (based upon the combined path-integrated measurements), it does not provide a flux rate or exact locations for the emission sources. Background concentrations of N2O were around 342 ppbv at the field site, which is consistent with the derived tomographic reconstruction. 13 Fig. 3: Tomographic reconstruction of the field showing the locations of the retroreflectors (blue squares) and towers (red circles) and path-integrated measurements over the soybean field are combined to develop a tomographic images of the field doing a controlled release test. A clear enhancement is observed (yellow pixel) on the right hand side of the field directly downwind of the controlled release (from a tank of N2O gas, location marked with a black star). New tomographic reconstructions were created after every full field scan. An algorithm consisting of two separate iterative Monte Carlo simulations and an image pattern matching comparison was developed and used to identify hot spot location first and then calculate the emission rate. As with the tomographic reconstruction, GPS coordinates of the retroreflectors, towers, and path-integrated measurements were used as input. In addition, the wind 14 speed and direction during the scan time and a qualitative assessment of atmospheric stability (which can be inferred from the time of day and how cloudy it is) also were used as algorithm inputs. The first round of Monte Carlo simulated the most likely positions of the hotspot, producing simulated path-integrated measurements, and then generated a tomographic reconstruction from the synthetic data. Each reconstruction was compared to the original tomographic reconstruction (Fig. 3) from measured data. This process was repeated many times until a “most-likely” location was found after 1000 iterations. If, based off the tomographic reconstruction, multiple hotspots were predicted, then the Monte Carlo simulation iteratively added multiple hotspot locations from which to produce total emissions. Once the likely location of the hotspot(s) was identified, a second Monte Carlo simulation was conducted where the location from the first simulation was used and different emission rates were assessed. A tomographic reconstruction with each emission rate was compared to the tomographic reconstruction from the measurement. A Gaussian plume model was used to simulate the emission at any particular hotspot. This final inversion model (shown in Fig. 4) was used to identify the hot spot location and the emission (flux) rate at that hot spot. The total simulated fluxes over the field were compared to measurement data for added verification. Running over n=1000 Monte Carlo iterations (for location and emission), the code identified the hotspot location to within 2 meters, the emission rate to within 15%, and 97% of the total N2O integrated flux from the measurement space was accounted for. For reference, an emission rate of 0.3 g N2O s-1 distributed uniformally over the ~ 100x50 m measurement area (3.0 ha) yields an integrated flux of 8640 g N2O ha-1 day-1. Based upon past field deployments and tests, a precision of at least 3000:1 at 1 Hz (0.1 ppbv) is obtained for a 1 s integration time in the field. The tower scanning stopped at each reflector position for two minutes, suggesting the likely precision was even better with some improvement due to averaging time (if at least some of the noise is white in character as expected). Furthermore, because integrated pathlengths are reconstructed from 32 measurements, this is another source of improved detection (as opposed to a single pathlength measurement). In short, it is reasonable to assume the 3000:1 precision in measurement translates directly into a detection limit for the retrieved flux. By using these assumptions, the nominal flux detection limit would be 2.9 g N2O ha-1 day-1 which is better than the SMARTFARM performance metrics of 10 g N2O-N ha-1 day-1. Alternatively, if we had taken the measurement over an 80 acre grid with the same setup, the equivalent flux over this area would have been 131 g N2O ha-1 day-1. Given the similar performance of our measurement at these distances, one would only need a 13:1 signal-to-noise ratio in the tomographic reconstruction to achieve these goals which is well within the capabilities demonstrated. Ultimately, more field studies will have to be done under a wide range of conditions to determine the factors that quantify the detection limit. We suspect time of measurement around a field and measurement precision to be the controlling factors. At long distances (somewhere over 500-100 m), we expect the measurement precision to degrade simply due to less light reflecting back to the detector. Then again, the white noise aspects would be ameliorated at least in part by the dwell time spent on each reflector. It is unclear how sources of noise on a specific pathlengths would translate into the full tomographic reconstruction (e.g. if one reflector was not tilted in the right direction so less light is returned), and weighted reconstruction algorithms would need to be tested. Practically speaking, this may make the case for using more reflectors than necessary in the 15 field (again, in this case, we only used 4 reflector positions each “side” of the 50x100 m test plot). The inexpensive costs of the reflectors and ease of setup would allow for some “noisy” or blocked signals (e.g. bird droppings on reflectors, winds moving them, etc.). Fig. 4: Simulated tomographic reconstruction with inverse modeling. After running Monte Carlo simulations, a plume location (indicated by the balloon like shape on the field) was found whose emission profile associated tomographic map matched the measurement data. A secondary Monte Carlo simulation was then run to find a best emission rate for the emission profile. The plume location was 2 meters from the true known source location, indicated by the black star. 16 D5. Outlook Currently available measuring tools are insufficient for identifying hot spots and hot moments. The NitroNet platform produced tomographic maps of the agricultural field. These tomographic maps provide hotspot detection, as directly observed through concentration enhancements, through long range open-path mid-IR laser spectroscopy. The optical web system is a paradigm shift for the measurement community by providing continuous measurements which can autonomously detect different local emission sources. The new approach provides the temporal benefits of approaches such as eddy covariance, while addressing micrometeorological and geographic constraints, and the spatial benefits of localized measurement approaches such as soil chambers and UAVs, all while covering the whole field simultaneously as opposed to a small region. While the full system has been tested on acre-sized plot here with 100-meter path lengths, the laser measurements and retroreflectors have also been successfully tested at 400 path length differences in prior milestone work in this award. In past work, QCLs have been used with corner cube arrays at kilometer scale distances (Michel et al., 2015). Indeed, as shown in Fig. 5, the inexpensive (~ $10), 4 mm thick retroreflectors developed in this project were compared to commercial-off-the-shelf (COTS) corner cube array reflectors ($1000s). Our retroreflectors showed up to 60% higher reflectivities and a broader acceptance angle than COTS corner cube reflectors (likely due to less beam shift, no gaps between corner cubes as is the case in conventional devices). Therefore, the NitroNet system is capable of measuring over 100s of acres and spatial resolution is really limited to the number and mounting of reflectors on fence posts, stakes, or utility poles. 17 a b Fig. 5a: Retroreflector reflectivities showing the effects of different angles of incidence with different coatings applied to them. At sharp angles there is a clear loss of signal, where individual sub-cells of the retroreflector shadow other cells, blocking the return beam (meaning the retroreflector is functioning more like a mirror at sharp incidence angles). Different retroreflector samples of the same type were averaged together. Fig. 5b: Retroreflector reflectivities showing the effects of different rotation angles with a fixed 0° incidence angle. In principle, the rotation angle should have no impact on reflectivity, but manufacturing defects mean some rotations are slightly worse, though this may also be an artifact of the measurement method. Generally, the developed retroreflectors can be considered rotation invariant as effective surface area during the measurement does not significantly change during the rotation. Retroreflector alignment currently is done manually during setup, which is time-consuming. Algorithms can be implemented without the use of any new hardware for autonomous alignment in the future. These could be run for micro-adjustments during measurement to also ensure that if 18 any reflector moves due to be struck or shifting from wind, etc., alignment could again be rapidly restored. Furthermore, any motor backlash from the gimbal could be more easily accounted for. Implementation of the optical web approach at scale will provide insights into how farming processes can be optimized and constrain models. New, unknown biological processes can be discovered to further refine our understanding of nitrogen cycling. Even without deploying the optical web at every single farm, the optical web can determine constraints and corrects to existing models and more common measurement approaches or models. The retroreflectors used here are broadband and can work at many wavelengths. By changing the wavelength of the laser, the optical web can be readily adapted to other gases of interest, such as ammonia or methane, and deployed in locations like wastewater treatment plants and industrial facilities. Finally, we note that the retroreflectors are robust in that they survived several months in field conditions (wind, rain, hail, dust, insects) without apparent loss of signal. We also tested several retroreflectors on a stratospheric NASA research aircraft at airspeeds of mach 0.8, altitudes of 17 km, temperatures to -70ºC, and impaction by hydrometeors and dust at aircraft speeds. No apparent loss in reflectivity was also observed after 40 flight hours, showing the robustness of the new retroreflectors. In summary, the entire NitroNet system – from laser scanning to inexpensive reflectors in the mid-infrared – is ready for further commercialization efforts. Ongoing efforts beyond SMARTFARM include additional deployments in corn, soybean, and lettuce fields and improving optomechanical aspects for ease of field use. One could envision NitroNet as simply a rotating mid-IR laser beam that picks up “reflections” whenever the backscattered signal reaches some threshold where no real alignment or setup is required and reflectors are placed randomly throughout a field. An algorithm could “remember” the altitude and azimuth of consistently high signals and automatically scan those positions as needed. Such improvements would lower labor costs in deployment and drive the cost structure for integration down even more. EcoFluxNet Inc. is actively pursuing such improvements in the next generations of NitroNet. E. Project Outputs E1. Journal articles Four publications have been drafted from the SMARTFARM NitroNet project, and a brief synopsis of each manuscript, as follows: Sevostianov, V.I., J. Collins, P. Guiguizian, L. Tao, N.P. Li, and M.A. Zondlo, “NitroNet: An optical web for spatiotemporal agricultural greenhouse gas emissions”, to be submitted to Science Advances, expected September 2025. This manuscript will describe the field measurements of the NitroNet system to determine the spatiotemporal patterns of emissions including the controlled release experiment. Also included will be the tomographic reconstruction and analyses to convert tomographic images to emissions. Sevostianov, V.I., J. Collins, Y. Choi, J. DiGangi, G. Diskin, J.A. Miech, L. Tao, N.P. Li, and M.A. Zondlo, “Thin mid-infrared reflectors”, to be submitted to Nature Photonics, August 2025. This manuscript describes the thin, inexpensive retroreflectors designed as part of this project. These have applicability like street signs and simple reflectors in the visible but now for the mid- 19 infrared spectral region. These retroreflectors will open the door for new mid-infrared chemical sensing technologies. Sevostianov, V.I., L. Tao, P. Guiguizian, J. Collins, and M.A. Zondlo, “Open-path nitrous oxide sensor for UAV-based field measurements”, to be submitted to Envi. Sci. Technol., October 2025. This manuscript describes the UAV-based nitrous oxide sensor including field measurements and performance attributes. Sevostianov, V.I., L. Tao, P. Guiguizian, J. Collins, and M.A. Zondlo, “Long-path, scanning nitrous oxide sensor for agricultural applications”, to be submitted to Optics Express, November 2025. This manuscript will describe the optomechanical and spectroscopic design attributes of the NitroNet tower system. It is possible that a separate manuscript may be needed on the Monte Carlo inverse Gaussian methods to deduce source strength and source location, though at this time we intend this to be folded into the overall NitroNet publication in Science Advances. E2. Invention disclosures and patent applications, licensed technologies US Patent App. 63/631,897, Zondlo, M.A., J. Collins, P. Zorner, P. Guiguizian, A. Michaels, and V. Sevostianov, "Systems and methods for determining analyte concentration in air over an agricultural field", April 9, 2024. US Patent App. 63/631,904, Zondlo, M.A., V. Sevostianov, N. Li, J. Collins, L. Tao, P. Guiguizian, "Systems and methods for spatiotemporal analyte measurements over a field using an optical web", April 9, 2024. US Patent App. 63/631,902, Zondlo, M.A., V. Sevostianov, J. Collins, P. Guiguizian, H. Yi, L. Tao, D. Pan, D. Moore, J. McSpiritt, N. Li, "Open-path optical sensors for analyte measurements", April 9, 2024. US Patent App. 63/523,690, Sevostianov, V., M.A. Zondlo, N.P. Li, J. Collins and L. Tao, "Mid- IR thin low-cost retroreflector", June 28, 2023. All four patent applications have been exclusively licensed to EcoFluxNet Inc. by Princeton University. E3. Website EcoFluxNet Inc.: www.ecofluxnet.com (Delaware C-Corpation startup derived from NitroNet/SMARTFARM award) VeriAg, 501c non-profit: www.veriag.com (non-profit, joint venture formed by EcoFluxNet Inc. and Global Sustainability Developers, Inc. to sell traceable and verifiable carbon credits with revenues split between farmers and the two partner companies for collecting verifiable data to support the credits) E4. Networks or collaborations fostered The PI and EcoFluxNet Inc. team has been invited for talks and panels at the Alternative Fuels and Chemicals Coalition annual workshops the past three years. This trade group is an 20 excellent means to introduce high-value carbon credits including the bioenergy sector, particularly sustainable aviation fuel and ethanol/biodiesel markets. E5. Other products such as data Large datasets used in publications will be archived on a permanent repository with a digital object identifier, per policies of individual publications. Princeton University’s Data Commons is a permanent data repository available to the public for scientific datasets. Because the data archived cannot be amended or changed once uploaded, we will wait until final acceptance of associated papers before archiving any datasets with the publications. F. Follow-on Funding Source Funding Committed or Received Grantham Foundation $330,000 for NitroNet to inform DNDC models Princeton University IP Accelerator Fund $100,000 for center pivot NitroNet Private investors, EcoFluxNet Inc. startup* $150,000 *EcoFluxNet started selling SAFE notes in its initial fundraising that started June 2025 References Michel, A.P.M., D.J. Miller, K. Sun, L. Tao, L. Stanton, and M.A. Zondlo (2016), “Open path, quantum cascade laser-based sensor for path-integrated, in-situ methane measurements”, J. Atmospheric and Oceanic Technology, 33, 2373-2384, doi:10.1175/JTECH-D-16-0024.1 Pan, D., I. Gelfand, L. Tao, M. Abraha, K. Sun, X. Guo, J. Chen, G.P. Robertson, and M.A. Zondlo (2021), "A new open-path eddy covariance method for nitrous oxide and other trace gases that minimizes temperature corrections", Global Change Biology, 28, 1446-1457, doi:10.1111/gcb.15986. Tian, H., Pan, N., Thompson, R. L., Canadell, J. G., Suntharalingam, P., Regnier, P., Davidson, E. A., Prather, M., Ciais, P., Muntean, M., Pan, S., Winiwarter, W., Zaehle, S., Zhou, F., Jackson, R. B., Bange, H. W., Berthet, S., Bian, Z., Bianchi, D., Bouwman, A. F., Buitenhuis, E. T., Dutton, G., Hu, M., Ito, A., Jain, A. K., Jeltsch-Thömmes, A., Joos, F., Kou-Giesbrecht, S., Krummel, P. B., Lan, X., Landolfi, A., Lauerwald, R., Li, Y., Lu, C., Maavara, T., Manizza, M., Millet, D. B., Mühle, J., Patra, P. K., Peters, G. P., Qin, X., Raymond, P., Resplandy, L., Rosentreter, J. A., Shi, H., Sun, Q., Tonina, D., Tubiello, F. N., van der Werf, G. R., Vuichard, N., Wang, J., Wells, K. C., Western, L. M., Wilson, C., Yang, J., Yao, Y., You, Y., and Zhu, Q. (2024): Global nitrous oxide budget (1980–2020), Earth Syst. Sci. Data, 16, 2543–2604, https://doi.org/10.5194/essd-16- 2543-2024, 2024. 21