Using multi-satellite observations to constrain ammonia emissions and unlock their potential over open water

Abstract

Reducing uncertainty in ammonia ($\:\text{N}{\text{H}}_{3}$) emissions, particularly those over open water, which have largely been unexplored, remains a key challenge. This study refines 2019 $\:\text{N}{\text{H}}_{3}$ emissions over the south-central United States (SCUS) using inverse modeling technique with Cross-track Infrared Sounder (CrIS) data and assesses its impact on inorganic $\:\text{P}{\text{M}}_{2.5}$. We also present a novel assessment of $\:\text{N}{\text{H}}_{3}$ emissions constrained by Infrared Atmospheric Sounding Interferometer (IASI) and CrIS datasets both individually and combined. For the first time, we demonstrate the potential of refining $\:\text{N}{\text{H}}_{3}$ emissions over open water using satellite data, specifically over the northwestern Gulf of Mexico (NWGOM). Annual posterior NH₃ emissions exceeded prior estimates over SCUS by 1.43 GgNa⁻¹ (2.5-fold), raising average concentrations by 2.9 ppb (3.4-fold), particularly in Texas, New Mexico, and Oklahoma, and increasing levels of particulate ammonium (1.26-fold), sulfate (1.01-fold), and nitrate (2-fold). Combined IASI/CrIS outperformed individual datasets when compared with surface measurements. Over NWGOM, average $\:\text{N}{\text{H}}_{3}$ concentrations increased significantly by 1.4 ppb, predominantly driven by biological nitrogen fixation. This study highlights the potential of satellite data to refine $\:\text{N}{\text{H}}_{3}$ emissions over open water and emphasizes the role of multi-satellite datasets and high-resolution regional inverse modeling in improving air quality forecasts and global emission estimates.

Introduction

$\:\text{N}{\text{H}}_{3}$ emissions have negative impacts on air quality and contribute to the formation of inorganic particulate matter smaller than 2.5 microns in aerodynamic diameter ($\:\text{P}{\text{M}}_{2.5}$), leading to various health issues such as cardiovascular disease, asthma, and respiratory problems¹. Data and modeling suggest particulate ammonium ($\:{\text{N}\text{H}}_{4}^{+}$) lies primarily in the accumulation mode (0.1 to 2.5 microns in diameter), largely due to the presence of involatile cations - calcium, magnesium, and/or sodium - in coarse-mode particles that repel ammonium uptake there². Accumulation-mode ammonium affects both health and visibility significantly. $\:\text{N}{\text{H}}_{3}$ emissions also influence air quality and climate change through several mechanisms, including altering radiative forcing by affecting aerosol size distributions, concentrations, and compositions³modifying carbon flux⁴changing the phase of secondary inorganic aerosols⁵enhancing light absorption caused by organic aerosols⁶and affecting heterogeneous ice nucleation⁷. Additionally, $\:\text{N}{\text{H}}_{3}$ plays a vital role in the nitrogen cycle by affecting nitrogen-containing compounds like nitrous oxide ($\:{\text{N}}_{2}\text{O}$) and nitrogen oxide ($\:{\text{N}\text{O}}_{\text{x}}$)⁴. Excessive $\:\text{N}{\text{H}}_{3}$ deposition can harm delicate ecosystems by causing soil acidification, biodiversity loss, and eutrophication⁸.

Despite $\:\text{N}{\text{H}}_{3}$’s significant contribution to air quality, visibility, climate change, and public health, limited measurements have complicated the investigation of its effects⁹. Moreover, the scarcity of relevant observations has resulted in considerable uncertainties in modeling $\:\text{N}{\text{H}}_{3}$ in many regions and formulating regulatory control plans¹⁰. The lack of reliable in-situ information outside of intense field campaigns regarding the spatial and temporal distribution of emissions, emission factors, management practices, and farming plans¹¹ has also contributed to uncertainties in bottom-up $\:\text{N}{\text{H}}_{3}$ emission inventories, especially across North America. Furthermore, Nowak et al.¹² reported larger uncertainties in emission estimates at regional scales. The United States (U.S.) National Emission Inventory (NEI) has historically undercounted non-agricultural $\:\text{N}{\text{H}}_{3}$ sources, notably underestimating on-road vehicle emissions by about a factor of two and underrepresenting biomass burning due to data gaps^13,14. On-road mobile sources and fires has been identified as key missing contributors for future inventories¹⁵. Simplified farm management assumptions in the NEI introduce uncertainties. Manure and fertilizer $\:\text{N}{\text{H}}_{3}$ emissions vary with practices, and limited data on application timing and methods likely lead to underestimation, especially under intensive or sub-optimal conditions¹⁶.

Inverse modeling, which integrates observational data, is essential for refining emission inventories and improving model predictions¹⁷. Remote sensing data, such as $\:\text{N}{\text{H}}_{3}$ concentrations from the Cross-track Infrared Sounder (CrIS) and Infrared Atmospheric Sounding Interferometer (IASI), provide valuable spatial information but face limitations like diurnal sampling and measurement uncertainties¹⁸. In contrast, chemical transport models (CTMs) offer detailed spatiotemporal data yet al.so carry uncertainties due to the numerical approximations of atmospheric processes and emission inventories. By combining datasets with a CTM, inverse modeling effectively reduces these uncertainties, leading to more accurate emission estimates. To refine $\:\text{N}{\text{H}}_{3}$ emissions and examine PM_2.5 impacts, various studies have applied inverse modeling with satellite data. Momeni et al.⁹ used the iterative Finite Difference Mass Balance (iFDMB) method with CrIS data to refine $\:\text{N}{\text{H}}_{3}$ emissions in East Asia, observing improvements in $\:\text{N}{\text{H}}_{3}$ and PM_2.5 concentrations. Li et al.¹⁹ compared iFDMB with four-dimensional variational data assimilation (4D-Var) methods using GEOS-Chem and CrIS data over North America, finding that iFDMB achieved comparable accuracy with 4D-Var at coarse resolutions (2 × 2.5) but showed increased errors at finer resolutions (0.25 × 0.3125) due to spatial smearing. Marais et al.²⁰ estimated the United Kingdom (UK) $\:\text{N}{\text{H}}_{3}$ emissions using GEOS-Chem with CrIS and IASI data, finding CrIS emissions (389 Gg) were 43% higher than IASI emissions (272 Gg). Seasonal variations, including a summer peak in CrIS emissions linked to dairy farming, underscored differences affecting environmental assessments.

While many studies have refined $\:\text{N}{\text{H}}_{3}$ emissions, three key areas of research remain unaddressed: (1) adjusting $\:\text{N}{\text{H}}_{3}$ emissions across the continental U.S. over a year with fine resolution, (2) taking advantage of multiple distinct satellite instruments with various overpass times simultaneously, and (3) refining $\:\text{N}{\text{H}}_{3}$ emissions over open ocean water. This study addresses these gaps using the Python-based Data Assimilation Framework (PyDAF)²¹which facilitates $\:\text{N}{\text{H}}_{3}$ adjustments over extended periods. By leveraging the non-overlapping overpass times of IASI and CrIS, we improve coverage, and by tackling challenges in open-water detection, we enable more effective use of satellite data in these regions. This study applies the iFDMB inverse modeling technique to estimate top-down $\:\text{N}{\text{H}}_{3}$ emissions over the south-central U.S. (SCUS), the northeastern Mexico (NEM), and the northwestern Gulf of Mexico (NWGOM) using IASI, CrIS, and combined satellite data for 2019. After refining the $\:\text{N}{\text{H}}_{3}$ emission inventory, we assess the impacts of IASI, CrIS, and the combination of both on $\:\text{N}{\text{H}}_{3}$ emissions and the resulting impact on the concentrations of multiple chemicals in fine particulate matter. The impact of the updated emissions is further evaluated by comparing the model estimates of particle component concentrations with satellite observations and surface measurements from the Ammonia Monitoring Network (AMoN), the NADP’s National Trends Network (NTN), the U.S. environmental protection agency (EPA) surface measurement stations, and Clean Air Status and Trends Network (CASTNET).

Results

Evaluation

Posterior column density with satellite observations

Figure 1-a presents a scatter plot of $\:\text{N}{\text{H}}_{3}$ column densities, while Table 1-a summarizes model performance metrics for a-priori and a-posteriori $\:\text{N}{\text{H}}_{3}$ columns. Posterior simulations significantly improved accuracy across all periods, with correlations rising from near-zero to 0.75 in January-February-March (JFM) and reaching 0.84, 0.88, and 0.79 in April-May-June (AMJ), July-August-September (JAS), and October-November-December (OND), respectively. Annually, correlation improved from 0.32 to 0.73, and IOA increased from 0.35 to 0.84.

Supplementary Fig. 1, 2, and 3 display annual and quarterly spatial $\:\text{N}{\text{H}}_{3}$, where iFDMB captures CrIS-observed patterns but overestimates in the northwest and underestimates in the southwest. These discrepancies result from the Normalized Mean Difference (NMD)-based adjustments, which overestimate smaller differences to offset larger ones, with hotspots intensifying nearby overestimations.

Posterior evaluation with surface measurements

For posterior evaluation constrained by CrIS, Fig. 1-b compares quarterly mean $\:\text{N}{\text{H}}_{3}$ concentrations from prior and posterior model estimates with AMoN observations. The posterior values closely matched AMoN, improving significantly over the prior, which was notably lower. The best agreement occurred in JAS (2.5 vs. 2.6 µg/m³), while OND showed slight overestimation (2.1 vs. 1.9 µg/m³). Table 1; Fig. 1-c summarize improved $\:\text{N}{\text{H}}_{3}$ estimates, with correlation increasing from 0.50 to 0.66 and reduced MAE, NMSE, and RMSE. Figure 1-b and Supplementary Fig. 4 show enhanced $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition, though underestimations persist compared to NTN. Supplementary Table 1 indicates further reductions in MAE, NMSE, and RMSE, with a slight IOA increase for $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition.

Figure 1-d compares $\:\text{N}{\text{H}}_{3}$ concentrations and $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition across different satellites for April–July. CrIS-posterior estimates closely matched AMoN, outperforming prior and IASI estimates, with the combined IASI/CrIS providing the highest accuracy. $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition improved over prior estimates but remained underestimated compared to NTN, with slight overestimations in June. Figure 1-d compares $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition estimates with NTN for April, June, and July. Posterior estimates improved over prior values but remained below observed levels, except for slight overestimations by CrIS and IASI/CrIS in June.

Supplementary Fig. 5 and Supplementary Table 2 compare CASTNET observations with prior and posterior $\:\text{S}{\text{O}}_{4}^{2-}$ and $\:\text{N}{\text{O}}_{3}^{-}$ concentrations. For $\:\text{S}{\text{O}}_{4}^{2-}$, the posterior adjustment leads to a slight increase in modeled concentrations, but they remain lower than the observed values. The MAE decreases marginally from 0.52 to 0.51 µg/m³, while the NMSE and RMSE remain nearly unchanged, and the IOA stays constant at 0.47, indicating limited improvement. In contrast, for $\:\text{N}{\text{O}}_{3}^{-}$, the posterior corrections result in noticeable enhancements: MAE decreases from 0.38 to 0.34 µg/m³, NMSE drops significantly from 1.59 to 0.81, RMSE declines from 0.45 to 0.40 µg/m³, and IOA increases from 0.68 to 0.75, indicating a much better alignment with observed data.

Supplementary Tables 3 and Supplementary Figs. 6 and 7 compare EPA observations with modeled $\:\text{N}{\text{O}}_{2}$ and $\:\text{S}{\text{O}}_{2}$ concentrations. For $\:\text{N}{\text{O}}_{2}$, the model performs well, capturing the seasonal cycle reasonably but consistently underestimating concentrations. The general good performance of simulated NO₂ concentrations contributes to better simulation of$\:\:\text{N}{\text{O}}_{3}^{-}$ concentrations. For $\:\text{S}{\text{O}}_{2}$, the model captures the general quarterly patterns except during OND, but struggles to accurately reproduce observed SO₂ levels. This limitation affects the simulation of $\:\text{S}{\text{O}}_{4}^{2-}$ concentrations, and part of the underestimation of $\:\text{S}{\text{O}}_{4}^{2-}$ can be attributed to uncertainties in modeled SO₂ concentrations.

Meteorological data

Supplementary Fig. 8 shows monthly-averaged temperature and wind speed from the WRF model’s 2019 outputs, validated against MADIS data from Texas stations. The evaluation focused on U10 (east-west wind at 10 m AGL), V10 (north-south wind at 10 m AGL), and T2 (2 m AGL temperature), with accuracy assessed using MAE, NMSE, RMSE, correlation coefficient (R), and index of agreement (IOA). Supplementary Table 4 indicates a strong match, particularly for V10 and T2 (IOA: 0.86 and 0.96). While Supplementary Fig. 9 highlights some discrepancies, WRF effectively captured spatial and temporal variations across Texas.

Table 1 Evaluation of model performance metrics.

Fig. 1

Comparisons of priori and posterior model estimates of $\:\text{N}{\text{H}}_{3}$ colum densities and concentrations with observations (CrIS, AMoN, and NTN). (a) Scatterplots of quarterly- and annually-averaged comparison of the prior and posterior $\:\text{N}{\text{H}}_{3}$ column densities versus CrIS in 2019. (b) (Top) Quarterly comparison of AMoN versus priori and posterior $\:\text{N}{\text{H}}_{3}$ concentrations using CrIS, (middle) quarterly comparison of NTN versus priori, and posterior $\:\text{N}{\text{H}}_{4}^{+}$ concentrations using CrIS, and (Bottom) quarterly comparison of NTN versus priori, and posterior $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition using CrIS. (c) Scatterplots of prior and posterior $\:\text{N}{\text{H}}_{3}$ concentrations versus AMoN. (d) (Top) Monthly comparison between surface measurements and the posterior constrained by IASI, CrIS, and the combined IASI/CrIS. (Bottom) same as Top, but for wet deposition. The “Post: IASI” is posterior constrained by IASI (green), “Post: CrIS” is posterior constrained by CrIS (red), and “Post: both” is posterior constrained by IASI/CrIS (yellow). The calculations, as well as the plots, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Reduced complexity CMAQ model (RCCM)

In order to reduce computational demands, an RCCM was implemented to model $\:\text{N}{\text{H}}_{3}$ concentrations. $\:\text{N}{\text{H}}_{3}$ concentrations simulated by the RCCM were evaluated against those from the standard CMAQ model. Supplementary Fig. 10 compares $\:\text{N}{\text{H}}_{3}$ simulated by the RCCM with $\:\text{N}{\text{H}}_{3}$ simulated by CMAQ. The results illustrate that the RCCM is in close agreement with CMAQ. Detailed information regarding the RCCM is available from Momeni et al.⁹.

Satellite data

Supplementary Fig. 11 presents the spatial quarterly- and annually-averaged distribution of CrIS $\:\text{N}{\text{H}}_{3}$ column density over the domain in 2019. The figure highlights $\:\text{N}{\text{H}}_{3}$ hotspots, particularly in regions with dense agricultural activities and major urban area. Notable concentrations appeared in Texas, New Mexico, Oklahoma, and southern Kansas.

Top-down Estimation of $\:\mathbf{N}{\mathbf{H}}_{3}$ emissions

Adjusted emissions constrained by cris

Figure 2 and Supplementary Fig. 12 display the annual and quarterly spatial distributions of $\:\text{N}{\text{H}}_{3}$ emissions, comparing a-priori with a-posteriori values constrained by CrIS. Over SCUS, the iFDMB method identified substantial annual emission increases of 1.43 tetragram of nitrogen per year (Tg N a^− 1) compared to prior estimates, representing a 2.5-fold increase. The largest increases were estimated during OND, with emissions reaching a 3.6-fold increase (305 gigagrams of nitrogen per quarter (Gg N q⁻¹)), and during JFM, a 3.3-fold increase (235 Gg N q⁻¹). In contrast, AMJ and JAS showed smaller increases, with a 2.6-fold (434 Gg N q⁻¹) and a 1.7-fold (338 Gg N q⁻¹), respectively. Table 2 shows anthropogenic $\:\text{N}{\text{H}}_{3}$ emission estimates from various studies over the contiguous U.S. (CONUS). Compared to previous research, $\:\text{N}{\text{H}}_{3}$ emissions in CONUS appear to be significantly underestimated. These findings underscore the role of high-resolution regional inverse modeling in improving the accuracy of global emission estimates and highlight the need for adaptive strategies to accurately quantify ammonia emissions, providing essential insights for environmental policy and sustainable management.

Fig. 2

The spatial distribution of a-priori (model results with no satellite data), a-posteriori (results using CrIS satellite data), and the difference between the two, $\:\text{N}{\text{H}}_{3}$ emissions in 2019. (a) The spatial distribution of a-priori (model results with no satellite data, blue), a-posteriori (results using CrIS satellite data, red), and the difference between the two, $\:\text{N}{\text{H}}_{3}$emissions in 2019. (b) Annual estimates of prior and posterior $\:\text{N}{\text{H}}_{3}$ emission based on different regions; SCUS: South-Central United States, TX: Texas, OK: Oklahoma, NM: New Mexico, Other: western Arkansas and Louisiana, and southern Kansas, Colorado, and Missouri, NEM: Northeastern Mexico, NWGOM: Northwestern Gulf of Mexico, Domain: domain of interest in this study. (c) Quarterly and annual estimate of prior and posterior $\:\text{N}{\text{H}}_{3}$ emission in the South-Central United States. Emission estimates are expressed in gigagrams of nitrogen per year (Gg N a⁻¹) and gigagrams of nitrogen per quarter (Gg N q⁻¹). The calculations, as well as the plots and maps, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Table 2 Estimates of anthropogenic $\:\text{N}{\text{H}}_{3}$ emissions (Tg N a^− 1).

Over Texas, the iFDMB method reveals substantial annual $\:\text{N}{\text{H}}_{3}$ emission increases of 2.1-fold, particularly in the agriculture-intensive regions of northwestern and southeastern Texas (Supplementary Fig. 13), indicating prior underestimations (299 gigagrams of nitrogen per year (Gg N a⁻¹)). During JFM, AMJ, and OND, total $\:\text{N}{\text{H}}_{3}$ emissions rose by 2.5-fold (138 Gg N q⁻¹), 2.3-fold (195 Gg N q⁻¹), and 2.5-fold (131 Gg N q⁻¹), respectively, with local emissions exceeding 60 megagram of nitrogen per quarter (Mg N q^{- 1}) in northwestern Texas and 20 Mg N q^{- 1} in southeastern Texas. Biomass burning, which occurred predominantly from February to September in southeastern Texas²⁸, may have contributed to these high emissions, warranting further investigation. In JAS, total $\:\text{N}{\text{H}}_{3}$ emissions showed a more moderate rise of 1.6-fold (154 Gg N q⁻¹).

In New Mexico, prior local $\:\text{N}{\text{H}}_{3}$ emissions were mostly below 10 Mg N a⁻¹, totaling 26 Gg N a⁻¹. CrIS adjustments significantly increased annual emissions to 213 Gg N a⁻¹, with local emissions exceeding 150 Mg N a⁻¹ in northern regions, likely linked to dairy farms²⁹agriculture²⁹and biomass burning³⁰. Quarterly, emissions rose by 69 Gg N q⁻¹ (18-fold) in JFM and 53 Gg N q⁻¹ (20-fold) in AMJ, with notable increases in JAS (34 Gg N q⁻¹, 3.9-fold) and OND (9 Gg N q⁻¹, 6.2-fold). In Oklahoma, CrIS adjustments raised prior annual emissions from 91 to 194 Gg N a⁻¹ (2.14-fold), with posterior estimates of 42 Gg N q⁻¹ (2.5-fold) in JFM, 58 Gg N q⁻¹ (2.6-fold) in AMJ, 53 Gg N q⁻¹ (1.5-fold) in JAS, and 37 Gg N q⁻¹ (3.7-fold) in OND. Similar trends were found in western Arkansas, Louisiana, and southern Kansas, Colorado, and Missouri, where prior annual emissions of 164 Gg N a⁻¹ increased to 397 Gg N a⁻¹ (2.4-fold), with local emissions exceeding 150 Mg N q⁻¹. The highest quarterly emissions occurred in AMJ (122 Gg N q⁻¹), while OND showed the greatest increase (4-fold).

Over NEM, we lack access to accurate prior $\:\text{N}{\text{H}}_{3}$ emission data, with many grid cells missing values. As a result, prior $\:\text{N}{\text{H}}_{3}$ emissions were significantly underestimated, missing key sources such as roads, industry, and agriculture. PyDAF addresses these gaps by assigning the second-lowest value to grid cells with missing data, refining $\:\text{N}{\text{H}}_{3}$ emissions and improving iFDMB accuracy. Adjustments identified major road sources and two NH₃ hotspots: the Torreón area, dominated by agricultural activities³¹and the Monterrey Metropolitan Area (MMA), driven by industrial activities^32,33aligning with NH₃ patterns in these regions³¹. However, underestimations persisted in regions lacking major sources, as iFDMB struggled to balance emissions adjustments between low- and high-emission areas⁹. Sitwell et al.²⁷ conducted a study incorporating the Torreón area and MMA into the $\:\text{N}{\text{H}}_{3}$ inversion system over North America using CrIS.

NWGOM’s adjusted emissions constrained by cris

We establish the feasibility of utilizing satellite data over NWGOM based on the satellite’s ability to detect ammonia over this region. Long-term data analysis over the Gulf of Mexico shows that the CrIS satellite achieves an ~ 80% single-pixel detection rate in this region. Over NWGOM, the CrIS instrument, with its higher sensitivity and lower detection limit, showed an average monthly $\:\text{N}{\text{H}}_{3}$ column density of 1.11 × 10¹⁵ molecules cm⁻², with maximum values reaching 2.19 × 10¹⁷ molecules cm⁻². In comparison, IASI yielded mean and maximum values of 8.2 × 10¹⁵ and 1.92 × 10¹⁷ molecules cm⁻², respectively. It should be noted that for CrIS the ~ 20% of data points over NWGOM that are below the detection limit are accounted for using the conservative approach of assigning a non-detect value of zero, which reducing potential high biases when averaging data. However, this approach may lead to underestimation in mean values when background levels exceed zero, but unlike over land there is currently little additional information over water to provide more refined non-detect values. White et al.³⁴ underscore that ignoring non-detects can introduce significant bias, particularly in regions with low $\:\text{N}{\text{H}}_{3}$ concentrations. Including non-detects aligns modeled simulation more closely with surface observations, enhancing the accuracy of $\:\text{N}{\text{H}}_{3}$ concentration estimates. To further assess data quality, we conducted an uncertainty analysis over three distinct regions: a non-elevated background region, an elevated region, and the NWGOM as shown in Supplementary Fig. 14. Results show $\:\text{N}{\text{H}}_{3}$ concentrations in the NWGOM generally range between 0 and 2 ppb, with most values above the detection threshold, indicating valid signal. In the elevated region, occasional outliers exceeding 100 ppb were observed but are unlikely to impact the inversion results. Given that the observed values in the Gulf of Mexico —a region characterized by low $\:\text{N}{\text{H}}_{3}$ levels— are still often (e.g., ~ 80% of the time for CrIS) above the detection limit of the satellites, we are in a position to employ them to refine $\:\text{N}{\text{H}}_{3}$ emissions over NWGOM, even though direct validation is currently unavailable.

Figure 2 and Supplementary Fig. 12 show $\:\text{N}{\text{H}}_{3}$ emissions over NWGOM, highlighting a notable increase across all quarters and annually. Prior estimates indicated low emissions of 0.53 Gg N a⁻¹, with annual local emission rates of 0–10 Mg N q^{- 1}. In contrast, revised estimates suggest annual emissions of 86.3 Gg N a⁻¹, with local rates of 50–100 Mg N q^{- 1} annually. Quarterly local emission rates ranged from 10 to 40 Mg N q^-1 during JFM, AMJ, and OND, corresponding to emissions of 27.5, 29.6, and 23.6 Gg N q⁻¹, respectively, and 8.7 Gg N q⁻¹ during JAS, with local emissions of 5–10 Mg N q^{- 1}.

The highest NH₃ emissions were estimated during AMJ and JFM, respectively, with comparatively smaller increases during JAS. This seasonal variation is likely influenced by the mixed layer depth (MLD), rivers discharge, and water temperature in NWGOM. The seasonal cycle of MLD, driven by variations in wind forcing, surface heat flux, and water density stratification, peaks in December–March and reaches a minimum in July–August^35,36,37. In winter, cooler surface temperatures and stronger winds enhance vertical mixing, bringing nutrient-rich deep water to the surface³⁶which supports phytoplankton growth³⁸ and triggers the spring bloom³⁶. This process, occurring from mid-winter to early spring, facilitates a greater influx of nutrients into the euphotic zone (upper 200 m)³⁶enhancing BNF³⁹ and boosting phytoplankton growth³⁶likely increasing $\:\text{N}{\text{H}}_{3}$ emissions in JFM, particularly in March.

In spring, the combination of MLD, sunlight, the spring bloom in March and April³⁶ and high Mississippi and Atchafalaya Rivers discharge⁴⁰ (March–July, as shown in Supplementary Fig. 15) likely enhances $\:\text{N}{\text{H}}_{3}$ emissions. Additionally, the short lifespans of phytoplankton⁴¹ result in their post-bloom decomposition, likely raising BNF⁴²further increasing $\:\text{N}{\text{H}}_{3}$ emissions. During summer, strong thermal stratification inhibits vertical mixing, limiting nutrient exchange in surface layers⁴³ and reducing BNF and consequently $\:\text{N}{\text{H}}_{3}$ emissions. Furthermore, dead zones form in summer (Supplementary Fig. 16) from nutrient-fueled algal blooms driven by Mississippi River discharge, with decomposing algae depleting oxygen and releasing ammonium in deep layers⁴⁴. Moreover, higher summer temperatures reduce $\:\text{N}{\text{H}}_{3}$ emissions by decreasing $\:\text{N}{\text{H}}_{3}$ solubility, causing more ammonia to remain dissolved as ammonium in water^45,46,47 in JAS. By fall, cooling temperatures and stronger winds break down stratification⁴³deepening the mixed layer and increasing nutrient availability⁴³ and ammonium (from dead zone) from deep water⁴⁴which likely supports increases in BNF and $\:\text{N}{\text{H}}_{3}$ emissions.

Changes in $\:\text{N}{\text{H}}_{3}$ emissions are closely linked to variations in phytoplankton populations, as BNF and $\:\text{N}{\text{H}}_{3}$ emissions are driven by nutrient availability. Chlorophyll concentration, a reliable indicator of phytoplankton growth, is shown in Supplementary Fig. 17, which highlights higher chlorophyll levels in JFM and AMJ compared to JAS. Moreover, shipping activity, particularly along the coastline, has an overall contribution to $\:\text{N}{\text{H}}_{3}$ emissions in NWGOM (Supplementary Fig. 18). The results emphasize the necessity of a comprehensive reassessment of NWGOM emissions and other open water emissions. Overall, this study demonstrates the effectiveness of satellite data, such as CrIS, in constraining $\:\text{N}{\text{H}}_{3}$ emissions over NWGOM, highlighting the potential of satellite-based measurements for inverse modeling and data assimilation in open water regions. Although satellite data effectively refine emissions, validation of these adjusted $\:\text{N}{\text{H}}_{3}$ emissions remains challenging due to the lack of measurements over NWGOM during the study period.

Comparison of emissions adjustments constrained by various satellites

Figure 3 and Supplementary Figs. 19 and 20 present monthly $\:\text{N}{\text{H}}_{3}$ emissions for May, June, and July, comparing a-priori estimates with a-posteriori emissions constrained by IASI, CrIS, and combined IASI/CrIS data. Over SCUS, in May, prior emissions were low levels at 50.1 Gg N m^− 1 with local emissions ranging from 0 to 5 Mg N m⁻¹, while adjustments based on IASI suggested modest increases (1.35-fold), with local emissions ranging from 0 to 20 Mg N m^− 1, highlighting significant underestimations compared to AMoN data (Fig. 1). CrIS constraints showed notable increases (2.9-fold), particularly in northern Texas (20–60 Mg N m^− 1), and the combined IASI/CrIS data yielded even higher emissions (3.04-fold), aligning closely with AMoN observations (Fig. 1) and providing the most accurate corrections for this period.

In June, a-priori emissions remained low (65 Gg N m^{- 1}, the rate of 0–20 Mg N m^-1), with IASI adjustments showing slight increases by (1.29 fold). CrIS constraints revealed more pronounced emission rises (2.06-fold), particularly in central and northeastern Texas (20–60 Mg N m^-1), while the combined adjustments provided a balanced correction (2.02-fold) particularly across Texas. CrIS and the combined dataset demonstrated strong agreement with AMoN observations (Fig. 1), offering the most accurate corrections for June.

In July, a-priori emissions continued to be low (57.3 Gg N m^-1), with IASI adjustments introducing modest increases (1.26-fold). CrIS constraints indicated substantial emission increases (1.86-fold), particularly in northern Texas. The combined IASI/CrIS adjustments offered more significant increases (1.95-fold) but still showed considerable underestimations. These findings align with a study by Marais et al.²⁰, which found that bottom-up emissions were 27% lower than those constrained by IASI and 49% lower than those constrained by CrIS, further emphasizing CrIS’s higher emissions adjustments compared to IASI.

Posterior $\:\mathbf{N}{\mathbf{H}}_{3}$ concentrations

For posterior $\:\text{N}{\text{H}}_{3}$ concentrations constrained by CrIS, Fig. 4 and Supplementary Figs. 21 and 22 compare a-priori and a-posteriori $\:\text{N}{\text{H}}_{3}$ concentrations for 2019. Across SCUS, annual average $\:\text{N}{\text{H}}_{3}$ concentrations increased 3.4-fold (3 ppb), with the highest levels in New Mexico. Quarterly, $\:\text{N}{\text{H}}_{3}$ concentrations rose by 3.8-fold (2.7 ppb) in JFM, 4-fold (3.4 ppb) in AMJ, 2.5-fold (2.7 ppb) in JAS, and 3.4-fold (2.9 ppb) in OND. In Texas, NH₃ increased 3-fold (2.6 ppb) annually, peaking in AMJ (3.7-fold, 3 ppb) and OND (3.5-fold, 2.8 ppb). JAS $\:\text{N}{\text{H}}_{3}$ enhancements in northeastern Texas likely result from biomass burning. In Oklahoma, $\:\text{N}{\text{H}}_{3}$ increased 2.4-fold (3.3 ppb), with peaks in AMJ and JAS (3.8 ppb). The largest underestimations occurred in OND and AMJ, increasing 3.4-fold and 3.1-fold, respectively. In New Mexico, prior $\:\text{N}{\text{H}}_{3}$ concentrations were substantially underestimated. The annual average posterior $\:\text{N}{\text{H}}_{3}$ concentration increased significantly by 12.8-fold (3.5 ppb), with the largest underestimations occurring in JFM (24.5-fold, 3.8 ppb) and OND (22.1-fold, 3.9 ppb), primarily due to underestimated emissions from dairy farms²⁹agricultural activities²⁹and prescribed and biomass burning³⁰ during these periods. The highest $\:\text{N}{\text{H}}_{3}$ concentrations were simulated in AMJ and OND (3.9 ppb). High-resolution regional simulations more accurately capture $\:\text{N}{\text{H}}_{3}$ hotspots, such as those around Albuquerque, enhancing emission source representation and spatial precision, which are inadequately depicted in coarser-resolution studies over the U.S. In western Arkansas, western Louisiana, and southern Kansas, $\:\text{N}{\text{H}}_{3}$ concentrations reached 6 ppb, while exceeding 6 ppb in Torreón and peaking at 4 ppb in MMA during AMJ.

For posterior $\:\text{N}{\text{H}}_{3}$ concentrations constrained by various satellites, Fig. 5 and Supplementary Figs. 25 and 26 display monthly averaged $\:\text{N}{\text{H}}_{3}$ concentrations for May, June, and July, comparing a-priori and a-posteriori emissions constrained by IASI, CrIS, and combined IASI/CrIS. In May, a-priori average $\:\text{N}{\text{H}}_{3}$ concentrations over SCUS were 0.7 ppb, while IASI constraints increased them to 1.1 ppb (1.6-fold). CrIS constraints resulted in higher concentrations, reaching 2.9 ppb (4.3-fold) over SCUS, particularly in New Mexico (3.3 ppb, 16.1-fold). The combined IASI/CrIS constraints yielded the highest increases, reaching 3.3 ppb (4.8-fold) over SCUS and 3.7 ppb (17.7-fold) in New Mexico. In June, a-priori $\:\text{N}{\text{H}}_{3}$ concentrations over SCUS remained low, while IASI adjustments increased average concentrations to 1.4 ppb (1.4-fold), and CrIS further raised average concentrations to 3.2 ppb (3.2-fold). The combined IASI/CrIS constraints resulted in 3.5 ppb (3.5-fold), significantly enhancing coverage. In July, a-priori concentrations remained low, with IASI adjustments increasing average NH₃ concentrations to 1.4 ppb (1.4-fold) and CrIS raising average levels to 2.6 ppb (2.6-fold). The combined IASI/CrIS approach provided the broadest coverage, bringing average $\:\text{N}{\text{H}}_{3}$ concentrations to 2.8 ppb (2.8-fold), highlighting CrIS’s substantial impact on NH₃ adjustments.

NWGOM’s concentrations

Annual average $\:\text{N}{\text{H}}_{3}$ concentrations constrained by CrIS (Supplementary Fig. 22) over NWGOM increased by 1.4 ppb (24-fold), with the highest posterior concentrations in AMJ (2 ppb) and the lowest in JAS (0.8 ppb). A long-term analysis of CrIS data (2018–2023) confirmed this pattern (Supplementary Fig. 23). Additionally, $\:\text{N}{\text{H}}_{3}$ concentrations were elevated along the coastline (Fig. 4 and Supplementary Fig. 21), likely influenced by shipping activities and biological processes (BNF), as suggested by inverse modeling. To investigate further, we examined chlorophyll concentrations as an indicator of BNF levels over NWGOM. Higher chlorophyll concentrations are associated with increased $\:\text{N}{\text{H}}_{3}$ levels, as shown in Supplementary Fig. 17, where they peak along the coastline. Additionally, chlorophyll concentrations are elevated in AMJ and decline in JAS, supporting the simulated $\:\text{N}{\text{H}}_{3}$ pattern.

To assess the influence of coastal biological processes, we compared $\:\text{N}{\text{H}}_{3}$ levels over the Gulf of Mexico with other North American coastal regions over the long term (2018–2023) (Supplementary Fig. 24). The Gulf of Mexico especially in NWGOM exhibits higher $\:\text{N}{\text{H}}_{3}$ concentrations than most other ocean regions, with elevated levels near the coastline, suggesting a natural process potentially linked to BNF. Additionally, no evidence supports the identification of individual oil and gas operations in the long-term mean or median, and isolated elevated values likely result from a few outlier points surviving preprocessing. While a general contribution cannot be ruled out, specific hotspots corresponding to platform locations are not apparent. Furthermore, shipping activity generally contributes to $\:\text{N}{\text{H}}_{3}$ levels over NWGOM, particularly elevating concentrations along the coastline (Supplementary Fig. 18). For comparison, Paulot et al.⁴⁸ estimated $\:\text{N}{\text{H}}_{3}$ concentrations over NWGOM at 0.4 ppb, whereas our study found an average of 1.4 ppb in 2019—1 ppb higher than previous estimates.

Spatial distribution of inorganic species

Given that changes in $\:\text{N}{\text{H}}_{3}$ emissions can significantly impact the concentrations of inorganic $\:\text{P}{\text{M}}_{2.5}$, it is essential to examine variations in inorganic species such as $\:\text{N}{\text{H}}_{4}^{+}$, $\:\text{S}{\text{O}}_{4}^{2-}$, and $\:\text{N}{\text{O}}_{3}^{-}$. For posterior inorganic $\:\text{P}{\text{M}}_{2.5}$ concentrations constrained by CrIS, Fig. 4, and Supplementary Figs. 22 and 27 show the annual and quarterly $\:\text{N}{\text{H}}_{4}^{+}$ concentrations in 2019, simulated using a-posteriori and a-priori $\:\text{N}{\text{H}}_{3}$ emissions. $\:\text{N}{\text{H}}_{4}^{+}$ levels are expected to vary directly with changes in $\:\text{N}{\text{H}}_{3}$ due to their equilibrium relationship ($\:\text{N}{\text{H}}_{3}\rightleftharpoons\:\text{N}{\text{H}}_{4}^{+}$)⁹. After a-posteriori adjustments, annual $\:\text{N}{\text{H}}_{4}^{+}$ concentrations increased significantly across SCUS (0.39 µg/m³, 1.26-fold), particularly in New Mexico and southeastern Texas, while the NWGOM showed elevated levels (0.19 µg/m³, 2.25-fold), especially near Port Arthur and Galveston. During JFM and OND (Supplementary Fig. 27), the largest rises occurred in Oklahoma and eastern Texas, with notable increases in New Mexico’s San Juan Basin and Santa Fa, NEM (around Torreón and MMA), and along NWGOM coastline. AMJ and JAS showed no significant changes, except over NWGOM. Paulot et al.⁴⁸ estimated $\:\text{N}{\text{H}}_{4}^{+}$ concentrations of 0.25–0.5 µg/m³ over NWGOM in 2006, aligning with our 2019 simulations of 0.2–0.4 µg/m³.

During AMJ and JAS, $\:\text{N}{\text{H}}_{4}^{+}$ concentrations remained low, likely due to the formation of particle ammonium sulfate ($\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$) and ammonium nitrate ($\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$), whereas during JFM and OND, $\:\text{N}{\text{H}}_{4}^{+}$ concentrations were comparatively high. The gas–particle partitioning of total ammonia ($\:\text{N}{\text{H}}_{\text{x}}$ = $\:\text{N}{\text{H}}_{3}$ + particulate $\:\text{N}{\text{H}}_{4}^{+}$) is temperature-dependent^49,50. At low temperatures, ammonia is largely sequestered in aerosols as ammonium ($\:\text{N}{\text{H}}_{4}^{+}$), through the formation of $\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$. In contrast, at higher temperatures, $\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$ becomes thermally unstable, releasing $\:\text{N}{\text{H}}_{3}$ back to the gas phase⁴⁹ (Supplementary Figs. 4). Furthermore, to assess the potential for particle-phase formation, we calculated the relative humidity (RH) and deliquescence relative humidity (DRH) of $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$ and $\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$ across the domain. Supplementary Fig. 28 shows RH and the difference between RH and DRH for $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$ and NH₄NO₃ across quarters. The results indicated that RH decreased and temperature increased during AMJ and JAS compared to JFM and OND, thereby facilitating the particle-phase formation of $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$ and $\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$. This explains the lower $\:\text{N}{\text{H}}_{4}^{+}$ concentrations during AMJ and JAS and the higher concentrations during JFM and OND. Notable hotspots near Port Arthur and Galveston in Texas and the San Juan Basin in New Mexico highlight key regions for understanding air pollution sources.

Fig. 3

Spatial distribution of a-priori and a-posteriori $\:\text{N}{\text{H}}_{3}$ emissions for May (left column), June (middle column), and July (right column) of 2019 when different satellite datasets are used. The first row presents prior concentrations (with no use of satellite data), the second row shows posterior concentrations constrained by IASI, the third row shows results constrained by CrIS, and the fourth row shows results constrained by the combination of IASI/CrIS. The calculations, as well as the maps, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Fig. 4

Spatial distribution of annually-averaged (2019) a-priori and a-posteriori concentrations of $\:\text{N}{\text{H}}_{3}$, $\:\text{N}{\text{H}}_{4}^{+}$, $\:\text{S}{\text{O}}_{4}^{2-}$, and $\:\text{N}{\text{O}}_{3}^{-}$ constrained by the CrIS instrument. The left column presents prior concentrations (with no use of satellite data), the middle column shows posterior concentrations constrained by CrIS, and the right column shows the difference between a-posteriori and a-priori concentrations for each chemical. The calculations, as well as the maps, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Fig. 5

Spatial distribution of a-priori and a-posteriori $\:\text{N}{\text{H}}_{3}$ concentrations constrained for May (left column), June (middle column), and July (right column) for different satellites in 2019. The first row presents prior concentrations, the second row shows posterior concentrations constrained by IASI, the third row shows the same constrained by CrIS, and the fourth row shows the same constrained by the combination of IASI/CrIS. The calculations, as well as the maps, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Figure 4, and Supplementary Figs. 22 and 29 display the annual and quarterly $\:\text{S}{\text{O}}_{4}^{2-}$ concentrations of in 2019. In the $\:\text{N}{\text{H}}_{3}$-$\:\text{H}\text{N}{\text{O}}_{3}$-$\:{\text{H}}_{2}\text{S}{\text{O}}_{4}$-$\:{\text{H}}_{2}\text{O}$ system, is initially neutralized by $\:\text{N}{\text{H}}_{3}$, with any excess $\:\text{N}{\text{H}}_{3}$ reacting with $\:\text{N}{\text{O}}_{3}^{-}$⁴⁹. Annually, a-posteriori $\:\text{S}{\text{O}}_{4}^{2-}$ concentrations increased by 1.01-fold (0.83 µg/m³) across SCUS, particularly near Port Arthur and Galveston in southeastern Texas, likely driven by industrial and shipping emissions⁵¹. Higher $\:\text{N}{\text{H}}_{4}^{+}$ levels in the eastern region amplified SO₄²⁻ formation, as $\:\text{N}{\text{H}}_{4}^{+}$ reacted with $\:\text{S}{\text{O}}_{4}^{2-}$ to form $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$ in particle and aqueous phases. Notable increases were also observed over NWGOM and parts of eastern New Mexico. Quarterly, $\:\text{S}{\text{O}}_{4}^{2-}$ concentrations rose during JFM and OND in Oklahoma, eastern Texas, and western Louisiana, with additional increases in AMJ and JAS near Port Arthur, Galveston, and southeastern Texas.

Figure 4, and Supplementary Figs. 22 and 30 present the a-priori and a-posteriori $\:\text{N}{\text{O}}_{3}^{-}$ concentrations, along with their differences. In the $\:\text{N}{\text{H}}_{3}$-$\:\text{H}\text{N}{\text{O}}_{3}$-$\:{\text{H}}_{2}\text{S}{\text{O}}_{4}$-$\:{\text{H}}_{2}\text{O}$ system, ammonium nitrate ($\:{\text{N}\text{H}}_{4}\text{N}{\text{O}}_{3}$) formation depends on two conditions: in ammonia-poor conditions, $\:{\text{N}\text{H}}_{4}\text{N}{\text{O}}_{3}$ remains low or negligible, while in ammonia-rich conditions, $\:\text{N}{\text{H}}_{3}$ reacts with $\:\text{N}{\text{O}}_{3}^{-}$ after $\:\text{S}{\text{O}}_{4}^{2-}$ has been neutralized⁴⁹. Annual averages showed increased $\:\text{N}{\text{O}}_{3}^{-}$ concentrations by 0.37 µg/m³ (2-fold) across SCUS. Elevated posterior NH₃ likely resulted in an ammonia-rich regime, enabling increased $\:\text{N}{\text{O}}_{3}^{-}$ formation after $\:\text{S}{\text{O}}_{4}^{2-}$ neutralization. Notable increases occurred in the San Juan Basin in New Mexico and along NWGOM coastline. Interpreting $\:\text{N}{\text{O}}_{3}^{-}$ results requires distinguishing between ammonia-rich and ammonia-poor regimes. Figure 6 and Supplementary Fig. 31 display the ammonia-rich and ammonia-poor regimes calculated by the R for a-priori and a-posteriori concentrations over the domain.

Figure 6 shows that the posterior R indicated an ammonia-rich regime dominating much of the domain, aligning with differences between a-priori and a-posteriori $\:\text{N}{\text{O}}_{3}^{-}$ concentrations. In this regime, $\:\text{N}{\text{O}}_{3}^{-}$ levels rise as sufficient NH₃ neutralizes $\:\text{S}{\text{O}}_{4}^{2-}$, with excess $\:\text{N}{\text{H}}_{3}$ driving further $\:\text{N}{\text{O}}_{3}^{-}$ formation. During JFM and OND, $\:\text{N}{\text{O}}_{3}^{-}$ concentrations increased across the domain, particularly in areas with high a-priori $\:\text{N}{\text{O}}_{3}^{-}$ levels, driven by elevated $\:\text{N}{\text{H}}_{3}$ emissions creating an ammonia-rich regime (Supplementary Fig. 31). Notable increases occurred over NWGOM and along the coastline. In contrast, AMJ and JAS showed no significant $\:\text{N}{\text{O}}_{3}^{-}$ increases over Texas, likely due to insufficient posterior NH₃ to fully neutralize $\:\text{S}{\text{O}}_{4}^{2-}$. However, increases near Port Arthur and Galveston were simulated during these periods.

For posterior inorganic $\:\text{P}{\text{M}}_{2.5}$ concentrations constrained by various satellites, Fig. 7, and Supplementary Fig. 26, 32, and 33 compare $\:\text{N}{\text{H}}_{4}^{+}$, $\:\text{S}{\text{O}}_{4}^{2-}$, and $\:\text{N}{\text{O}}_{3}^{-}$ concentrations constrained by IASI, CrIS, and their combination across May, June, and July. In May, $\:\text{N}{\text{H}}_{4}^{+}$ concentrations over SCUS increased to 0.25 µg/m³ (1.09-fold) with IASI, 0.27 µg/m³ (1.2-fold) with CrIS, and 0.28 µg/m³ (1.22-fold) with combined constraints. Similar trends continued in June and July, with combined IASI/CrIS yielding the highest $\:\text{N}{\text{H}}_{4}^{+}$ concentrations (0.26–0.29 µg/m³). Across all months, CrIS consistently produced higher $\:\text{N}{\text{H}}_{4}^{+}$ concentrations than IASI, and the combined approach led to the most extensive spatial coverage of elevated $\:\text{N}{\text{H}}_{4}^{+}$ levels, highlighting its effectiveness in refining $\:\text{N}{\text{H}}_{4}^{+}$ estimates. For $\:\text{S}{\text{O}}_{4}^{2-}$ concentrations in May, the prior average value over SCUS was 0.76 µg/m³. IASI constraints led to a slight increase (1%), while CrIS adjustments resulted in a larger rise (1.9%). In contrast, the combined IASI/CrIS adjustments indicated a decrease (1%). This trend continued in June and July, with consistently high prior $\:\text{S}{\text{O}}_{4}^{2-}$ levels across SCUS. IASI and CrIS adjustments again produced increases, while the combined approach showed reductions. Across all three months, CrIS resulted in higher $\:\text{S}{\text{O}}_{4}^{2-}$ concentration increases, particularly in southeastern Texas and Louisiana. For $\:\text{N}{\text{O}}_{3}^{-}$ concentrations in May, prior levels are low across the domain, generally below 0.4 µg/m³. IASI constraints result in minimal adjustments, while CrIS constraints increase concentrations slightly, particularly in southern Texas, reaching around 0.6 µg/m³. The combined IASI/CrIS data show similar patterns, with localized increases in southern Texas. This trend continues in June and July, with prior $\:\text{N}{\text{O}}_{3}^{-}$ levels remaining low and IASI adjustments making minimal impact. CrIS again provides the most significant increases in southern Texas, reaching 0.6 µg/m³while combined constraints offer similar localized enhancements, especially in southern Texas.

Discussion

This study, for the first time, demonstrated the potential of refining $\:\text{N}{\text{H}}_{3}$ emissions over open water using satellite data, specifically over NWGOM and highlights a novel assessment of $\:\text{N}{\text{H}}_{3}$ emissions constrained by multiple satellite datasets, including IASI and CrIS, individually and combined. The bottom-up ammonia inventory carries high uncertainty because of lack of direct source measurements, emission factors, management practices, and farming plans. The NEI has historically undercounted non-agricultural NH₃ sources, with major gaps in on-road vehicle and biomass burning emissions. Simplified farm management assumptions in the NEI likely underestimate NH₃ emissions from manure and fertilizer due to limited data on application timing and practices. The results showed that inverse modeling constrained by the combined IASI/CrIS outperforms individual datasets when compared to $\:\text{N}{\text{H}}_{3}$ surface measurements. The posterior simulation, incorporating satellite information, consistently outperformed the prior simulation both quarterly and annually, with the combined IASI/CrIS approach yielding the most accurate results. The iFDMB method estimated $\:\text{N}{\text{H}}_{3}$ emissions across SCUS in 2019 to be 2.5-fold higher (1.43 Gg N a⁻¹) than prior estimates, particularly in Texas, Oklahoma, Arkansas, Louisiana, and notably New Mexico. Over Texas, substantial increases in $\:\text{N}{\text{H}}_{3}$ emissions were identified in the agriculture-intensive regions of northwestern and southeastern Texas, while biomass in eastern Texas may have also contributed to elevated emissions, warranting further investigation. In Oklahoma, the increase in $\:\text{N}{\text{H}}_{3}$ emissions indicates an underestimation of prior agricultural emissions. In New Mexico, emission increases in the northern regions are likely linked to dairy farming, agriculture, and prescribed or biomass burning. In NEM, inverse modeling identified major sources, including main roads, Torreón, and MMA, where prior emissions were either underestimated or missing. This study revealed that satellite data, particularly CrIS, has the capability to refine NH₃ emissions through inverse modeling, especially in open-water regions where direct validation is limited. Over NWGOM, posterior $\:\text{N}{\text{H}}_{3}$ concentrations were elevated levels averaging 1.4 ppb (24-fold), particularly along the coastline, predominantly driven by predominantly natural processes linked to BNF, while a general contribution from oil and gas platforms and shipping activities remains possible. The results correspond with seasonal variations in MLD, dead zone formation, Mississippi and Atchafalaya River discharge, and chlorophyll concentrations, serving as reliable indicators of natural processes. Across SCUS, Average $\:\text{N}{\text{H}}_{3}$ concentrations increased by 2.9 ppb (3.4-fold), particularly in New Mexico, Texas, and Oklahoma, with corresponding $\:\text{N}{\text{H}}_{4}^{+}$ increases of 1.26-fold (reaching 0.39 µg/m³) in NWGOM, Texas, and urban centers, influenced by seasonal variations and particle-phase formation under low relative humidity. A-posteriori $\:\text{S}{\text{O}}_{4}^{2-}$ concentrations rose annually by 0.83 µg/m³ (1.01-fold) in SCUS, especially in Oklahoma, Louisiana, Texas, and NWGOM, driven by industrial and shipping emissions, with peaks in JFM and OND. Annual average $\:\text{N}{\text{O}}_{3}^{-}$ concentrations increased by 0.37 µg/m³ (2-fold) across SCUS, particularly in the San Juan Basin, NWGOM, and along the coastline, driven by ammonia-rich conditions following $\:\text{S}{\text{O}}_{4}^{2-}$ neutralization. The ammonium-to-sulfate ratio confirmed widespread ammonia-rich conditions, with $\:\text{N}{\text{O}}_{3}^{-}$ levels peaking in JFM and OND. The substantially revised $\:\text{N}{\text{H}}_{3}$ emissions have important implications for $\:{\text{P}\text{M}}_{2.5}$ control strategies. Targeted reductions in $\:\text{N}{\text{H}}_{3}$, through optimized nutrient management, improved manure handling, feed adjustments, and enhanced wastewater treatment, can significantly lower secondary $\:{\text{P}\text{M}}_{2.5}$ formation. Additional measures such as biofilters, anaerobic digesters, and leak detection and repair (LDAR) programs can further limit emissions. Updating inventories, strengthening regulations, and integrating agricultural practices into air quality planning are essential for effective $\:{\text{P}\text{M}}_{2.5}$ mitigation and advancing public health goals. These findings underscore the role of multi-satellite datasets in improving the accuracy of air quality forecasts and emphasize the importance of high-resolution regional inverse modeling in refining global emission estimates. We suggest that top-down global emission estimates be derived region by region with fine resolution rather than treating the globe as a whole, ensuring greater accuracy and representation of regional variations.

Fig. 6

R parameters for identifying ammonia-rich and ammonia-poor regimes for a-priori (left plot) and a-posteriori (right plot) annual concentrations. The calculations, as well as the maps, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Fig. 7

The spatial distribution for a-priori and a-posteriori $\:\text{N}{\text{H}}_{4}^{+}$, $\:\text{S}{\text{O}}_{4}^{2-}$, and $\:\text{N}{\text{O}}_{3}^{-}$ concentrations constrained by different satellites for May. Spatial distribution of a-priori and a-posteriori concentrations of $\:\mathbf{N}{\mathbf{H}}_{4}^{+}$ (left column), $\:\mathbf{S}{\mathbf{O}}_{4}^{2-}$ (middle column), and $\:\mathbf{N}{\mathbf{O}}_{3}^{-}$ (right column) constrained by different satellites for May 2019. The first row also presents prior concentrations, the second row shows posterior concentrations constrained by IASI, the third row by CrIS, and the fourth row by the combined IASI/CrIS. The calculations, as well as the maps, were produced with a Python (3.9.5) script (https://www.python.org/downloads/release/python-395/).

Methods

Modeling setup

Weather research and forecasting (WRF) model

In this research, we utilized WRF version 4.2 to derive a multitude of parameters crucial for CTM simulations. These encompassed air temperature, specific humidity, surface pressure, the U/V components of wind speed, downward longwave and shortwave radiation flux, and precipitation. The simulations were executed with a spatial resolution of 12 km, specifically targeting the Texas State and NWGOM. To run the WRF model, meteorological inputs were sourced from the North American Mesoscale Forecast System (NAM) reanalysis datasets⁵². The NAM data offered a horizontal resolution of 12 km and provided reanalysis data at a temporal resolution of 6 h. This dataset included essential information such as temperature, wind, moisture, soil, and many other relevant parameters that were crucial for the modeling simulations. The WRF domains had sizes of 150 × 150 for the 12-km domain covering Texas and NWGOM, as depicted in Supplementary Fig. 34. The initial and boundary conditions were generated using NAM reanalysis datasets.

Sparse matrix operator kernel emissions (SMOKE)

Emissions input for running CMAQ was obtained from National Emissions Inventory (NEI). The Environmental Protection Agency (EPA) provides information on the emission of pollutants in the atmosphere through NEI¹⁶which offers a comprehensive and detailed estimate of air emissions of criteria pollutants, criteria precursors, and hazardous air pollutants from various air emissions sources, including NEI point sources, NEI nonpoint sources, NEI on-road sources, and NEI nonroad sources⁵³. For this study, the NEI modeling platform was utilized, incorporating the NEI emission inventory and SMOKE to spatially and temporally allocate the emission values to modeling grids⁴⁸. Emissions from natural sources were estimated using the Biogenic Emissions Inventory System (BEIS3). As for mobile emissions, they were processed based on the 2017 Motor Vehicle Emission Simulator (MOVES) output within the NEI package. The NEI modeling platform from the year 2017 was employed to produce emissions at a 12 km spatial resolution for Texas State throughout the entire year of 2019. While the NEI does not account for biogenic oceanic emissions, we turned to the Emissions Database for Global Atmospheric Research (EDGAR)²² to quantify these emissions for NWGOM region. It is imperative to note that biogenic oceanic emissions primarily arise from Biological Nitrogen Fixation (BNF). BNF refers to the conversion of dissolved nitrogen ($\:{\text{N}}_{2}$) gas into bioavailable forms such as $\:\text{N}{\text{H}}_{3}$ and $\:\text{N}{\text{H}}_{4}^{+}$ through the action of marine nitrogen-fixing organisms, known as diazotrophs⁵⁴.

CMAQ

CMAQ is a CTM developed and maintained by the US EPA that comprehensively predicts the most important processes affecting the chemistry of the atmosphere^55,56. CMAQ comprehensively simulates the chemistry of the atmosphere by considering processes such as advection, diffusion, and wet and dry deposition, and chemical reactions are represented within this model⁵⁷. By using an extensive database of atmospheric chemical reactions, CMAQ predicts the chemical production and loss of hundreds of pollutants to demonstrate the chemistry of the atmosphere. This model was employed to simulate $\:\text{N}{\text{H}}_{3}$ distributions across 35 vertical layers, extending to approximately 50 hPa, covering both Texas State and NWGOM, and maintaining a 12 km×12 km horizontal resolution. Meteorological inputs, coupled with emissions data, were integrated to conduct detailed atmospheric chemistry simulations, leveraging the capabilities of CMAQ model, version 5.0.2. The domain used for CMAQ is shown in Supplementary Fig. 34.

Inverse modeling setup

Satellite data

To refine $\:\text{N}{\text{H}}_{3}$ emissions for 2019, CrIS satellite was employed. Furthermore, for having a comprehensive evaluation that satellite can have better performance, $\:\text{N}{\text{H}}_{3}$ was constrained by IASI and both IASI and CrIS in Jun, July, and August.

The CrIS sensors are infrared sounders that have been launched onboard the sun-synchronous satellite SNPP (launched in October 2011), NOAA-20 (launched in November 2017), and NOAA-21 (launched in November 2022) missions. Since we are interested in the 2019 period for this study, we utilize data from both CrIS-SNPP and CrIS-NOAA-20 observations. During this period, the SNPP platform had a mean local daytime overpass time of 13:30 and a mean local nighttime overpass time of 01:30. During this time period, CrIS-NOAA-20 orbit had an overpass that was 55 min after SNPP. Both of these CrIS instruments we used in 2019 as CrIS-SNPP had a gap in $\:\text{N}{\text{H}}_{3}$ retrievals between March 25 and August 12. CrIS has an across-track scanning swath width of 2200 km and a nadir spatial resolution of 14 km⁵⁸. Different comparisons between $\:\text{N}{\text{H}}_{3}$ retrievals by the CrIS Fast Physical Retrieval (CFPR)⁵⁹ and ground-based observations have been performed^58,59 showed a reasonable correlation. The single pixel detection limit of the CrIS instrument will vary depending on the atmospheric state; it ranges from 0.25 ppb (surface) or 1.5 × 10¹⁵ molecules cm^-2 (total column) under favourable remotes sensing conditions (at a detection rate of 10%) to 0.5 ppb (3.5 × 10¹⁵ molecules cm^-2) under more typical atmospheric states (at a detection rate of 50%)⁶⁰. This study utilizes CrIS observations from both daytime and nighttime overpasses⁹.

The IASI instruments, housed on the MetOp-A and -B satellites, follow sun-synchronous orbits and execute overpasses at 09:30 and 21:30 local standard time (LST). The observational swath of both IASI instruments spans over 2,000 km, featuring a pixel footprint of 12 km in diameter at nadir viewing angles⁶⁰. This study harnesses both daytime and nighttime observations from IASI^61,62. A comparison of IASI data with ground-based Fourier Transform Infrared (FTIR) observations revealed a correlation coefficient (R) of 0.8 and a slope of 0.73⁶³. As documented by Dammers et al.⁶⁰, a commendable concordance exists between IASI $\:\text{N}{\text{H}}_{3}$ data and ground observations sourced from the Ammonia Monitoring Network (AMoN). The IASI instrument, on a single-pixel basis, has a detection limit for $\:\text{N}{\text{H}}_{3}$ ranging from 0.6 ppb at the surface (or 5.0 × 10¹⁵ molecules cm^-2 for total column) under favorable remote sensing conditions with a 10% detection rate, to 1.2 ppb (8.0 × 10¹⁵ molecules cm^-2) under more typical atmospheric conditions at a 50% detection rate⁶⁰.

Inverse modeling technique: iFDMB

The iFDMB inverse modeling^64,65 was employed to refine the $\:\text{N}{\text{H}}_{3}$ emission inventories as implemented in PyDAF (PyDAF: iFDMB)²¹. In the iFDMB, a-priori concentrations retrieved using a forward model were used to linearize the sensitivity of the column density ($\:{\Omega\:}$) to $\:\text{N}{\text{H}}_{3}$ emissions ($\:E$) at every grid point. Then, top-down emissions ($\:{E}_{t}$) were calculated at each iteration as follows⁶⁶:

$$\:{E}_{t}={E}_{a}\left(1+\frac{1}{\beta\:}\frac{{{\Omega\:}}_{o}-{{\Omega\:}}_{a}}{{{\Omega\:}}_{a}}\right)$$

(1)

where $\:{E}_{a}$ presents a-priori emissions from the previous iteration, $\:{{\Omega\:}}_{o}$ the observed column, $\:{{\Omega\:}}_{a}$ the simulated column, and $\:\beta\:$ the initial sensitivity given as:

$$\:\beta\:=\frac{{\Delta\:}{\Omega\:}/{\Omega\:}}{{\Delta\:}E/E}.$$

(2)

A perturbation of 20% to the a-priori emissions, $\:E$, was applied in each grid to determine the initial sensitivity⁹. The iteration process was repeated until the normalized mean difference (NMD) of new emissions with respect to the emissions calculated from the last iteration was less than 1% or 2%⁹. In this study, NMD of 2% has been employed⁹.

One of the inherent challenges with the iFDMB is handling grid points that lack emission values, even when satellite data indicates a value. To address this, PyDAF: iFDMB filled the grid points with missing emission data using the second-lowest value from the emission dataset (since the lowest value is zero, the next available value is used). This method ensures that all grid points with available satellite data have non-zero values, enhancing the accuracy and overall performance of the iFDMB. By filling in missing data, this approach improves the model’s reliability in refining emissions estimates²¹.

Reduced-Complexity CMAQ model (RCCM) for $\:\mathbf{N}{\mathbf{H}}_{3}$

The iFDMB technique performs multiple model simulations to converge on the final results. To reduce the computational cost, an RCCM was implemented to simulate $\:\text{N}{\text{H}}_{3}$ concentrations⁹. In the RCCM, $\:\text{N}{\text{H}}_{3}$ and $\:\text{N}{\text{H}}_{4}^{+}$ are considered as two tracer pollutants of the model and all of the chemical processes of other species are turned off. The developed RCCM included dry and wet deposition, the transport of $\:\text{N}{\text{H}}_{3}$ and $\:\text{N}{\text{H}}_{4}^{+}$, and $\:\text{N}{\text{H}}_{\text{x}}$ partitioning; the subroutine of ISORROPIA-II in the aerosol module calculates the gas-particle partitioning of $\:\text{N}{\text{H}}_{3}$ and $\:\text{N}{\text{H}}_{4}^{+}$. While running RCCM, the hourly sulfate ($\:\text{S}{\text{O}}_{4}^{2-}$), nitric acid ($\:\text{H}\text{N}{\text{O}}_{3}$), nitrate ($\:\text{N}{\text{O}}_{3}^{-}$), chloride ($\:\text{C}\text{l}$), sodium ($\:\text{N}\text{A}$), hydrochloric acid ($\:\text{H}\text{C}\text{l}$) concentrations are read from archived standard CMAQ simulation offline files. Detailed information regarding the RCCM is available in the study by Momeni et al.⁹.

Observation operator

To apply the iFDMB inverse modeling and to also compare model estimates to satellite observations, the vertical column of the model needs to be calculated and is known as observation operator. Since the CrIS provides both an averaging kernel and an a-priori $\:\text{N}{\text{H}}_{3}$ profile for every observation, and the IASI does not provide an averaging kernel, the methods underlying the observation operator diverge for each satellite. For both satellites, the retrieved $\:\text{N}{\text{H}}_{3}$ columns density for each observation is assigned to the closet grid point in the model grid.

For the CrIS, the vertical column of the model is calculated by summing a modeled partial column from Herron-Thorpe et al.⁶⁷, which is in molecules cm^-2, as follows:

$$\:VC{D}_{i}=\frac{{c}_{i}}{{10}^{6}}\times\:\frac{{L}_{{T}_{i}}\times\:{P}_{i}\times\:{N}_{A}}{R\times\:{T}_{i}},$$

(3)

where the i index is the level number, $\:c$ the concertation of $\:\text{N}{\text{H}}_{3}$ in ppmv, $\:{{L}_{T}}_{i}$i the model layer thickness, $\:P$ the pressure in pascals, $\:{N}_{A}$ Avogadro’s number, $\:R$ the molar gas constant, and $\:T$ the temperature in Kelvins. By substituting the equation of state and the hydrostatic equation into Eq. (3), the vertical column density was given by:

$$\:VC{D}_{i}=-{c}_{i}\times\:{\Delta\:}{P}_{i}\times\:2.119\times\:{10}^{14}.$$

(4)

For the CrIS, since the averaging kernel is provided, the column density was calculated by using the observational operator, $\:H$, to estimate the model $\:\text{N}{\text{H}}_{3}$ profile:

$$\:H\mathbf{c}={\mathbf{c}}_{a}+\mathbf{A}(\mathbf{M}\mathbf{c}+{\mathbf{c}}_{a})$$

(5)

where $\:\mathbf{c}$ is the model-estimated $\:\text{N}{\text{H}}_{3}$ profile, $\:\mathbf{M}$ a matrix that maps the space of the model to the space of CrIS, $\:\mathbf{A}$ the averaging kernel, and $\:{\mathbf{c}}_{a}$ is the a-priori $\:\text{N}{\text{H}}_{3}$ profile, the total vertical column density of the model is calculated as:

$$\:VCD=\sum\:_{i}VC{D}_{i}$$

(6)

In this study, only valid pixels with quality flag values exceeding 3 are used⁹. For the IASI, as there is no averaging kernel, a direct comparison is made between the IASI-$\:\text{N}{\text{H}}_{3}$ column density and the simulated column density in CMAQ. However, such a direct comparison could be predisposed to biases due to the variable sensitivity of retrieved $\:\text{N}{\text{H}}_{3}$ column densities to $\:\text{N}{\text{H}}_{3}$ concentrations at different altitudes, typically peaking around 700–850 hPa^58,68. While the $\:\text{N}{\text{H}}_{3}$ columns simulated by CMAQ also exhibit the most sensitivity to alterations in $\:\text{N}{\text{H}}_{3}$ concentrations within the 700 to 900 hPa range, the absence of averaging kernels renders us unable to quantify the associated uncertainties in these relations¹. Therefore, to calculate the model column density, the equations of 4 and 6 are used. Noting that in this study, we did not consider the IASI observations with negative values however these are valid observations. Moreover, we excluded pixels with a cloud fraction exceeding 0.3.

$\:\mathbf{N}{\mathbf{H}}_{3}$ emissions over open water

$\:\text{N}{\text{H}}_{3}$ emissions over open water, like NWGOM and oceans, are crucial due to their significant contribution to overall emissions (not just emissions over land). The ocean, contributing over 40% of natural and 15% of global emissions, is the largest natural $\:\text{N}{\text{H}}_{3}$ source⁴⁸. Biological nitrogen fixation (BNF)⁵⁴ and factors like nitrate, salinity⁴⁸dead zone⁶⁹shipping, and oil operations⁷⁰ influence these emissions, which are critical for refining global estimates and improving atmospheric model accuracy.

Posterior evaluation

The posterior evaluation involves a comparative analysis of the updated model simulations against observational datasets from satellite retrievals and surface measurements. To assess the model’s performance, both the posterior and prior estimates are compared with the CrIS, IASI, and a combined IASI/CrIS dataset. For CrIS, the evaluation is conducted throughout the entire year of 2019, while the comparison between IASI, CrIS, and combined IASI/CrIS is limited to the warmer months of May, June, and July when ammonia emissions are higher¹⁰.

The posterior emissions are evaluated by comparing the model simulation from updated emissions with surface measurements. The Ammonia Monitoring Network from the National Atmospheric Deposition Program (NADP) is a program that monitors and measures the atmospheric $\:\text{N}{\text{H}}_{3}$ concentrations and deposition of $\:\text{N}{\text{H}}_{4}^{+}$ in the U.S. The ammonia data are provided as bi-weekly averages, with air samples being collected and analyzed at each AMoN monitoring site every two weeks. To ensure an accurate apples-to-apples comparison, the model’s bi-weekly averages were aligned with the bi-weekly averages provided by AMoN. Ammonia concentrations are originally measured in micrograms per cubic meter (µg/m³), while the units of our results are in parts per billion (ppb). For conversion to ppb, the pressure and temperature at the nearest point to each station within the domain were taken into consideration. For the comparison, 6 active stations were employed over the domain with IDs of AR03, AR09, AR15, OK98, OK99, TX41, and TX43, from which the data is downloadable on the NADP website as shown in Supplementary Fig. 34.

For $\:\text{N}{\text{H}}_{4}^{+}$ deposition, wet deposition values were available from NTN, which measures total weekly wet deposition. For an apple-to-apple wet deposition comparison, the weekly averages of the wet deposition in the model were calculated. For evaluation against the $\:\text{N}{\text{H}}_{4}^{+}$ wet deposition, precipitation bias adjustment (PBA) is also applied⁹. The precipitation bias adjustment scales the simulated deposition by the ratio between measured and simulated precipitation to reduce errors in the precipitation estimates from the meteorological model:

$$\:PBA=\frac{\sum\:{p}_{o}}{\sum\:{p}_{m}}\sum\:W{D}_{m}$$

(7)

which, $\:\sum\:{p}_{o}$ is the monthly total accumulated observed precipitation, $\:\sum\:{p}_{m}$is the monthly total accumulated modeled precipitation, $\:\sum\:W{D}_{m}$ represents the monthly accumulated wet deposition estimate from the model. For the evaluation, 11 stations were used over the domain with IDs of AR03, AR16, AR27, NM08, TX03, TX04, TX10, TX16, TX22, TX43, and TX56, as shown in Supplementary Fig. 34.

For $\:{\text{N}\text{O}}_{2}$ and $\:\text{S}{\text{O}}_{2}$ evaluations, surface concentrations were obtained from EPA monitoring stations, with data available from 60 stations for $\:{\text{N}\text{O}}_{2}$ and 39 stations for $\:\text{S}{\text{O}}_{2}$. For $\:\text{N}{\text{H}}_{4}^{+}$, $\:{\text{S}\text{O}}_{4}^{2-}$, and $\:{\text{N}\text{O}}_{3}^{-}$ evaluations, concentrations were sourced from CASTNET, with data available from five stations. The locations of both EPA and CASTNET stations are shown in Supplementary Fig. 34.

To assess the potential for particle-phase formation, we calculated the relative humidity (RH) and deliquescence relative humidity (DRH) of $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$ and $\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$ across the domain. DRH represents the humidity threshold where a hygroscopic substance transitions from particle to liquid⁴⁹. When RH is below DRH, particles remain particulate phase⁷¹. The DRH of $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$ was calculated using the following Eq. 4⁹:

$$\:DRH\left(T\right)=DRH\left({T}_{0}\right)\text{e}\text{x}\text{p}\left\{\frac{{\Delta\:}{H}_{s}}{R}\right[A\left(\frac{1}{T}-\frac{1}{{T}_{0}}\right)-B\text{l}\text{n}\left(\frac{T}{298}\right)-C\left(T-{T}_{0}\right)\}$$

(8)

where T0 is 298 K, R is the universal gas constant equal to 8.314 J/K/mol. For $\:{\left(\text{N}{\text{H}}_{4}\right)}_{2}\text{S}{\text{O}}_{4}$, $\:\text{D}\text{R}\text{H}\left(298\:K\right)$ at 298 K is 79.9%, ∆Hs at 298 K is 6.56 kJ/mol, and A, B, and C are 0.1149, -4.489 × 10 − 4, and 1.385 × 10 − 6, respectively. Moreover, DRH of $\:\text{N}{\text{H}}_{4}\text{N}{\text{O}}_{3}$ is obtained from⁶⁹:

$$\:DRH=\frac{723.7}{T}+\:1.6954$$

(9)

Interpreting $\:\text{N}{\text{O}}_{3}^{-}$ results requires distinguishing between ammonia-rich and ammonia-poor regimes. This study identifies these regimes using the ammonium-to-sulfate aerosol ratio (R)⁷²:

$$\:R=\frac{\left[N{H}_{4}^{+}\right]}{\left[S{O}_{4}^{2-}\right]}$$

(10)

where [species] means the molar concentration of the species. In the ammonium-sulfate aerosol ratio approach, an R value greater than 2 indicates an ammonia-rich regime, while an R value less than 2 signifies an ammonia-poor regime.