Main

Nitrous oxide (N2O) is a primary and highly potent greenhouse gas (GHG), possessing a global warming potential 273 times that of carbon dioxide (CO2)1. It also has a long lifespan, persisting for ~109 years, in contrast to methane (CH4), another important GHG emitted during wastewater treatment2, which lasts ~12 years1. Not only does N2O contribute to climate change, but it also plays a detrimental role in stratospheric ozone depletion1. A variety of sources are responsible for N2O emissions, including soil management, fuel combustion and industrial processes. Among these sources, wastewater treatment stands out as the most crucial non-agricultural anthropogenic emitter, contributing to ~3–7% of the overall N2O emissions3,4. Over the period from 1980 to 2015, N2O emissions from wastewater treatment exhibited a consistent linear increase at a rate of 0.04 TgN yr−1 (ref. 4). This trend is expected to continue due to the growing adoption of biological nitrogen removal processes to meet more-stringent discharge regulations, as well as the expansion of wastewater infrastructure driven by population growth and increased urbanization.

N2O is generated as a byproduct of biological nitrogen conversion in wastewater treatment. Three main pathways are known to govern the generation of N2O: (1) hydroxylamine oxidation, (2) nitrifier denitrification and (3) incomplete heterotrophic denitrification5,6. N2O production is influenced by many environmental and operational factors, including temperature, dissolved oxygen (DO), pH, ammonium, C/N ratio, aeration rate, perturbations/extremities in loading and feeding strategies6,7,8,9. Consequently, N2O emissions from wastewater treatment exhibit substantial spatial and temporal variations10. Recent studies highlighted high N2O emissions from multiple water resource recovery facilities (WRRFs), with some exceeding 8% of the influent total nitrogen (TN) load11,12. For a typical WRRF, N2O emissions can account for as much as 86% of total direct GHG emissions (or Scope 1 emissions) from a facility based on actual measurements13,14. These findings underscore the critical need to first accurately quantify and then mitigate N2O emissions, particularly in light of the net-zero carbon targets set by many municipalities, cities and countries worldwide.

As electricity grids continue to decarbonize and WRRFs increasingly adopt renewable energy, Scope 2 emission decreases, while Scope 1 process emissions are poised to become the primary source of GHG emissions. Currently, the evaluation of such process-level emissions relies on methodologies developed by the Intergovernmental Panel on Climate Change (IPCC) or national/regional guidelines aligned with the IPCC approach. The IPCC proposed a tier-based methodology for estimating N2O emissions based on the availability and completeness of site-specific data15. However, due to a lack of measured N2O emission factor (EF) observations, the widely adopted EF (often the average value) overlooks substantial variations in treatment processes or site-specific conditions. For instance, the IPCC2006 default EF (0.00032 kg N2O-N kg−1 TN) was derived from a single monitoring campaign16, which was reported to underestimate actual N2O emissions by up to 98% at five Scandinavian WRRFs13. In contrast, the updated IPCC2019 increased the default EF by 50 times to 0.016 kg N2O-N kg−1 TN based on 30 observations15. This change directly elevated WRRFs to the top non-agricultural sources of N2O in the United States. While the updated EF aimed to address the limitations and improve the estimation accuracy, the EF observations curated in IPCC2019 span three orders of magnitude and are a mixture of reactor-level and plant-scale emission measurements. Reactor-level EFs only capture N2O flux from the secondary biological treatment reactors (hereafter referred to as bioreactor), thereby overlooking emissions from up- and downstream reactors (for example, primary or secondary clarifier, sludge treatment). As a result, the use of these default EFs to estimate plantwide, regional or countrywide N2O emissions introduces uncertainty due to limitations in accounting for site-specific variations.

Extensive research has been conducted on N2O emissions, but the existing inventory is limited in its ability to account for the diverse characteristics of N2O emissions. To address this gap, this study created a comprehensive database of direct N2O emissions (in %, kg N2O-N kg−1 TN) throughout the entire municipal wastewater collection and treatment system using natural language processing (NLP) methods. By analysing the updated dataset, we identified potential errors arising from mixing reactor-level and plant-scale EFs. In addition, we examined various factors that influence the quantification of N2O emissions from secondary biological treatment and systematically compared unit-level N2O emissions across all wastewater treatment trains. Finally, we refined a bottom-up methodology to estimate national direct N2O emissions from centralized municipal wastewater treatment (excluding sewer and effluent emission), accounting for variations in unit processes and types of biological treatment process. Through these comprehensive investigations and tool development, we ensured a broader and more-representative dataset and improved the understanding of N2O emissions from the wastewater sector.

Results

High N2O emission variations demand differentiated estimations

A total of 376 N2O EF observations from 119 publications were collected (Fig. 1), augmenting the IPCC2019 database by 12-fold. These monitoring campaigns were conducted at ~200 WRRFs worldwide encompassing a wide range of sizes (from <1 to ~760 million gallons per day (MGD)), nitrogen removal performances (9.4–97.0%; Supplementary Fig. 1) and main nitrogen treatment processes (Supplementary Text 1). Different from the existing method that mixes all emission data together, we categorized the measurements at three different treatment scales, including the biological functional units, the whole plantwide scale and side-stream treatment, and all scales exhibit clear discrepancies in both magnitude and uncertainties of N2O emissions (Fig. 2a). The highest mean (95% confidence interval (CI)) EF was found for side-stream treatment at 2.82% (1.53–4.11%) kg N2O-N kg−1 TN, which is 76% higher than the IPCC2019 EF of 1.6% kg N2O-N kg−1 TN. The mean EFs of plantwide (1.64% (1.20–2.07%) kg N2O-N kg−1 TN) and bioreactor (1.01% (0.73–1.29%) kg N2O-N kg−1 TN) measurements are lower, but they show much wider fluctuations related to temporal and spatial variations, operational variations and environmental conditions. The bioreactor datasets show the highest range and span from 0.00003% to 20.69% kg N2O-N kg−1 TN. Bioreactors are known hotspots for N2O emissions as they are designed for biological nitrification and denitrification reactions. According to plantwide EF observations that reported unit-level emissions, 27 out of 39 observations show that bioreactors account for 81.0–99.8% of plantwide N2O emissions17,18. However, studies also show that other unit processes such as primary and sludge treatment can be major N2O emitters, contributing up to 87% of plantwide N2O emissions19. In such cases, using bioreactor-specific EF to represent the whole plant would underestimate the plantwide emissions.

Fig. 1: Distribution of WRRF N2O emission observations.
figure 1

Data sources include the IPCC2019 database and newly added observations. Nitrogen treatment processes are categorized into 3 types for side-stream measurements and 18 types for plant-scale and bioreactor measurements. Average flow rate of studied WRRFs is categorized as follows: <1 (less than 1 MGD), [1, 10] (equal to or larger than 1 MGD but not exceeding 10 MGD), (10, 100] (larger than 10 MGD but not exceeding 100 MGD) and >100 (larger than 100 MGD). TN removal rate is grouped into: <0.5 (below 50%), [0.5, 0.85] (equal to or larger than 50% but not exceeding 85%), >0.85 (above 85%) and ‘Unspecified’ (TN removal rate not available). The numbers in parentheses represent the percentage of each category in the herein-analysed 376 observations.

Source data

Full size image
Fig. 2: N2O emission factors of the updated dataset.
figure 2

a,b, The variations in magnitude and distribution of N2O EFs associated with measurement scope (a) and treatment processes for side-stream treatment (b). The IPCC2019 and IPCC2006 EF values are shown in horizontal lines as references. Boxplots show 25th, 50th and 75th percentiles, and outlier bounds are based on 1.5× interquartile range (IQR) (with IQR equalling the 75th percentile minus the 25th percentile). Red dots represent the arithmetic mean. Violins represent probability distributions using kernel density estimation on either side. n is the number of monitoring data.

Source data

Full size image

It is worth noting that N2O emissions compiled in the IPCC2019 dataset were mainly dominated by bioreactor-scale measurements (Fig. 1). Interestingly, the IPCC2019 default EF is on par with the mean plantwide EF of the updated dataset despite its limited number of EF observations and its mixing of bioreactor and plantwide measurements. Several literature citation errors were found in the IPCC2019 dataset (Supplementary Text 2). After correcting these errors to align with source literature data, the revised the IPCC2019 default EF to be 1.11% kg N2O-N kg−1 TN (Supplementary Fig. 2). Similar corrections have been reported in a recent peer-reviewed study20. This error may indicate a potential systematic bias in various N2O emissions inventories that use the default EFs. Thus, it is important to address this bias with comprehensive data collection and critical analyses with sufficient scale differentiations.

The WRRFs are increasingly adopting side-stream nitrogen treatment processes to reduce the load of nutrients in reject stream or supernatant liquid before it is returned to the main treatment process21. This raises attention as the mean N2O EF of side-stream treatment is ~2.8- and 1.7-fold of bioreactor and plantwide emissions, respectively (Fig. 2a). Relatively high N2O emissions from side-stream treatment are mainly attributed to the high volumetric nitrogen conversion rates22. Reject stream originating from sludge dewatering or digester supernatant have very high NH4+ concentrations (500–1,500 mg N l−1, compared with 23–69 mg N l−1 common loading in influent23), and the average NH4+ removal rate is calculated at 82.3% on the basis of the data collected in this study. Due to such high concentrations, associated volumetric loading and high conversion rate, the possibility and amount of N2O emissions would be understandably higher than mainstream processes. We also found that side-stream N2O emissions vary significantly depending on the treatment process and reactor configuration. For example, the mean N2O EF for nitrification/denitrification (N/DN) processes is 8.91% kg N2O-N kg−1 TN, equalling to ~3 and 7 times of partial nitritation (PN) and partial nitritation-anammox (PNA) processes, respectively (Fig. 2b). Regarding reactor configurations, N2O EFs of one-stage PNA processes (for example, 0.4–2.0% kg N2O-N kg−1 TN)24,25,26 are lower than those of two-stage PNA processes (for example, 2.3–6.6% kg N2O-N kg−1 TN)22,27 because one-stage PNA limits nitrite accumulation, which is correlated with greater N2O emission22.

Overall, N2O emissions from WRRFs exhibit considerable variabilities and are difficult to be represented by a single EF value. The current guidelines that use a uniform EF methodology are inadequate in capturing the diverse nature of WRRFs and the substantial variations observed in N2O emissions.

Factors influencing N2O emissions from biological treatment

The curated datasets of bioreactor N2O emissions include both individual bioreactor as well as plantwide measurements that reported unit-level emissions (Fig. 3a). A wide range of biological treatment processes were analysed, which show a high variation in N2O emissions (Fig. 3b). On the basis of the magnitude of N2O emissions, we can roughly categorize all treatment processes into two groups: the low-emission processes with mean <1% kg N2O-N kg−1 TN and median <0.5% kg N2O-N kg−1 TN (mint and green bars in Fig. 3b) and the high-emission processes with mean ≥1% kg N2O-N kg−1 TN and median ≥0.5% kg N2O-N kg−1 TN (orange bars in Fig. 3b). All these high-emission processes are designed for biological nutrient removal (BNR), so the findings are understandable. Among all BNR processes, some technologies such as sequencing batch reactor (SBR) and membrane bioreactor (MBR) have 2–4 times more emissions compared with other processes such as anaerobic/anoxic/oxic (A2O) and modified Ludzack–Ettinger (MLE). It is encouraging that some common BNR technologies such as anoxic/oxic (A/O), MLE and A2O showed relatively lower emissions comparable to conventional activated sludge (CAS) secondary treatment, indicating that advanced nutrient removal does not necessarily lead to higher N2O emission. Various reasons caused relatively high EFs for the high-emission processes, including additional anoxic and oxic units28, low chemical oxygen demand (COD)/N ratio resulting in a decreased nitrification/denitrification capacity29,30, combined higher salinity and low DO concentration31 and the increase in nitrogen load32, among others (see Supplementary Text 3 for a detailed discussion).

Fig. 3: The impacts of various factors on bioreactor N2O emissions.
figure 3

a, Composition of bioreactor N2O emissions dataset. b, Bioreactor N2O emissions associated with different biological treatment processes. c–f, Impacts of measurement techniques (c,d) and duration (e,f) for low-emission (c,e) and high-emission (d,f) processes. g, For low-emission processes excluding long-term measurements, scatterplot and exponential decrease for N2O EF versus TN removal rate (adjusted R2 = 0.06; degrees of freedom (d.f.) = 139; P = 0.003). h, For high-emission processes excluding off-gas measurements, scatterplot and exponential increase between N2O EF and influent TN (adjusted R2 = 0.18; d.f. = 43; P = 0.002). Boxplots show 25th, 50th and 75th percentiles, and outlier bounds are based on 1.5× IQR (with IQR equalling the 75th percentile minus the 25th percentile). Red dots represent the arithmetic mean, as indicated by the text above the boxes. n is the number of monitoring data.

Source data

Full size image

We also compared the main monitoring techniques/duration and their uses in emissions monitoring, including flux chamber, tracer gas dispersion method (TDM), off-gas and headspace method (Fig. 3c–f). Given the apparent disparity in magnitude and variation between low- and high-emission processes, we separated the two groups to avoid biases. Overall, the campaign duration moves the mean and interquartile range of N2O emissions remarkably for low-emission processes (Fig. 3e), whereas measurement techniques play a crucial role for high-emission processes (Fig. 3d). The average N2O EFs obtained from off-gas and headspace methods are factors of ~2–3 times those from the other methods. This is probably because off-gas and headspace methods allow for more inclusive quantification of N2O emissions. The off-gas method monitors the off-gases from the entire liquid surface of covered reactor tanks33. The headspace method estimates N2O emissions on the basis of the dissolved concentration using Henry’s Law, which represents the maximum emission potential34. In comparison, the flux chamber technique quantifies the diffusive flux of gases across the air–water surface by a few measurement points, which makes it difficult to cover all sources or capture the spatial and temporal variabilities of fluxes2,35,36,37. TDM measurements require a point release at a chosen site that may not reflect the areal nature and dispersed location of the sources13. Therefore, cross-comparison of simultaneous N2O measurements using different monitoring techniques at the same site is needed to further enhance accuracy and reliability. Besides measurement techniques, previous studies indicate that short-term, intermittent sampling is inadequate to accurately estimate N2O emissions compared with online, long-term monitoring campaigns38,39,40 due to potential under- or over-estimation caused by monitoring timing, emission frequency and representativeness of limited samplings for the full emission profile. The mean N2O EF of short-term, intermittent sampling (0.18% kg N2O-N kg−1 TN) was found to be up to seven times lower than long-term, continuous monitoring (1.41% kg N2O-N kg−1 TN) when the flux chamber technique was applied (Supplementary Fig. 3). This suggests that, where possible, long-term campaigns that capture the diurnal and seasonal variations should be conducted to obtain accurate and representative quantification of N2O emissions.

Recognizing the variations in N2O emissions caused by monitoring techniques and duration, we excluded long-term samplings and off-gas measurements from low- and high-emission processes, respectively, to analyse correlations between bioreactor N2O EFs and environmental parameters (for example, TN, DO, temperature). For low-emission processes, there is a notable negative correlation between N2O EFs and TN removal rate (Fig. 3g). This finding is consistent with in situ measurements for 10 full-scale WRRFs in Austria where a linear regression model was built to predict N2O emissions on the basis of TN removal rate41. For high-emission processes, a notable positive correlation exists between N2O EFs and influent NH4+, TN and flow rate (Supplementary Fig. 4). Such positive correlation also corroborates previous findings that peak loads of total Kjeldahl nitrogen (TKN) moments coincide with the highest N2O flux42. Therefore, minimizing TN load peaks by equalizing wastewater flows could be an effective mitigation measure to reduce N2O emissions from WRRFs42,43. Furthermore, statistically significant relationships exist between N2O EF and influent NH4+ (Supplementary Fig. 4) as well as between N2O EF and TN concentration (Fig. 3h). This finding aligns with the principles on which the IPCC EF methodology was originally based, but the low R2 indicates that these concentrations alone are not sufficient to explain variations in N2O emission. A multivariate regression model, which includes other parameters such as TN removal rate, influent flow, DO and temperature, has the potential to accurately predict bioreactor N2O emissions44. Further discussion on the impacts of monitoring campaigns and environmental factors is provided in Supplementary Texts 4 and 5, repectively.

Unit-level N2O emissions and mitigation strategies

Since measurements from bioreactors cannot fully capture emissions from the whole plant, understanding N2O emissions from each treatment unit across a typical train of WRRFs is necessary. To do this, we categorized unit processes into three groups (Fig. 4a): primary (also includes preliminary treatment), bioreactor units (for example, anaerobic and/or aerobic reactors used for secondary biochemical oxygen demand (BOD) removal or advanced nutrient removal, secondary clarifier) and sludge treatment (for example, thickening, dewatering and onsite sludge disposal if applicable). Bioreactor units, including either aerobic only or anoxic plus aerobic units are generally the primary N2O emitters, although the contributing range varies greatly from 13.3%19 to 99.8%18 of plantwide emissions. Specifically, aerobic reactors are found to be the largest source (0.46% (0.35–0.56%) kg N2O-N kg−1 TN), followed by anoxic reactors (0.27% (0.001–0.55%) kg N2O-N kg−1 TN). The higher N2O emissions from aerobic reactors can be attributed to three possible explanations, including increased air-stripping, N2O production due to oxygen-inhibited denitrification and N2O production by nitrification during changes in aeration9,29,43. Mitigation strategies have been tested at full or pilot/lab scales, such as precise control of the DO level, balancing ammonia/TN load and avoiding nitrite accumulation, among others.

Fig. 4: Unit-level N2O emissions at WRRF and sewer networks.
figure 4

a, The magnitude and distribution of N2O EFs associated with different treatment unit processes. b, N2O emissions dataset of sewer networks. c,d, The magnitude and distribution of gaseous N2O flux (c) and dissolved N2O concentration (d) in sewer networks. Boxplots show 25th, 50th and 75th percentiles, and outlier bounds are based on 1.5× IQR (with IQR equalling the 75th percentile minus the 25th percentile). Red dots represent the arithmetic mean. Violins represent probability distributions using kernel density estimation on either side. n is the number of monitoring data.

Source data

Full size image

Besides bioreactors, sludge treatment including sludge buffer tanks, aerobic/anaerobic digestion and dewatering also contributes to N2O emissions, with mean (95% CI) of 0.04% (0.02–0.06%) kg N2O-N kg−1 TN. N2O emissions occur in sludge buffer tanks mainly due to the release of dissolved N2O and denitrification during sludge storage14. For emissions from the dewatering process, lower EFs were found when the plants are equipped with anaerobic digestion (for example, 14.1 ± 5.2 mg m−3 influent) than without anaerobic digestion (for example, 33.3 ± 7.4 mg m−3 influent)45. For WRRFs with onsite biosolids disposal, elevated N2O emissions are expected. The mean N2O EFs of commonly used sludge disposal methods, including incineration46, composting47 and biosolids stockpile35 are 0.76% kg N2O-N kg−1 TN, 1.82 g N2O-N kg−1 dry solids yr−1 and 0.06 g N2O-N kg−1 dry solids yr−1, respectively.

N2O emissions from primary treatment and secondary clarifier are low due to limited biological nitrogen conversions; instead, they are generally associated with dissolved N2O generated from the upstream reactors, that is, sewer networks and bioreactors. While N2O emissions from sewer networks have long been assumed to be negligible and are excluded from GHG inventories15, recent studies did identify influent pump stations as potential N2O emission hotspots48,49. Dissolved N2O concentration in the influent was observed to be highest compared with other unit processes48, indicating that N2O formation occurs in sewer networks. According to the very limited number of monitoring campaigns conducted in sewer networks or WRRF influents (Fig. 4b), gaseous N2O fluxes were detected in the range of 0.3–1.5 mg m−2 h−1 from sewer networks50 and 1.3–11.2 mg m−2 h−1 from WRRF influents48,49 (Fig. 4c). We should caution that only one study reported gaseous N2O fluxes from sewer networks, so it may not be representative. Given that dissolved N2O concentrations in sewer (6.0–18.8 µg l−1) and WRRF influent (3.9–26.0 µg l−1) are comparable (Fig. 4d), similar levels of gaseous N2O fluxes from sewer networks and WRRF influents should be expected. More monitoring campaigns in sewer networks are necessary for better understanding, as gaps in sewer N2O inventories will restrict the sector’s capacity to benefit from any future N2O emission crediting schemes51. Innovative monitoring approaches and modelling are beneficial, given the complexity and scale of sewer networks.

Nationwide wastewater N2O emissions for the United States

Given that N2O emissions from WRRFs are highly process specific, we propose a tier-based EF approach to capture the diverse nature of WRRFs instead of using uniform EFs. The United States was selected as a case study due to data availability, including data for influent flow rate and installed process units for nationwide WRRFs52. The tier-based EF approach considers the variations in N2O emissions resulting from different unit processes and types of biological treatment implemented by individual WRRFs. Thus, it captures the nuanced nature of N2O emissions and provides better estimations, especially for those WRRFs without the capability of direct monitoring. We employed a Monte Carlo approach to upscale plantwide N2O emissions to a national scale while simultaneously quantifying the relevant uncertainty. This refined approach reveals that the mean (95% CI) direct N2O emissions from domestic WRRFs in the United States is 11.6 (9.2–13.9) million metric tonnes (MMT) CO2-eq (excluding sewer and effluent emissions), with ~64% of total emissions stemming from four main types of bioreactor including CAS, other AS (for example, complex mix, high rate and pure oxygen), step feed and A2O (Fig. 5a). CAS and other AS are two main types of bioreactor employed in the United States, treating ~31% and 14% of nationwide wastewater, respectively (Fig. 5b). Although A2O and step feed only treat ~4% of wastewater, their relatively high EFs result in high emissions. Thus, effective N2O mitigation approaches designed for these processes would be needed to achieve net-zero carbon emission targets in the near-term.

Fig. 5: National N2O emissions analysis for US domestic WRRFs.
figure 5

a, Contribution of each unit process to nationwide N2O emissions. b,c, Total treated wastewater (b) and mean relative difference of estimated emissions (c) between this study (tier-based EF approach) and the US EPA (default EF = 1.5% kg N2O-N kg−1 TN) by WRRF configurations and types of bioreactor. PRI+BIO refers to facilities with primary and biological treatment processes, while PRI+BIO+SLG denotes those with additional sludge treatment. d, Cumulative probability curve of nationwide N2O emissions under different scenarios.

Source data

Full size image

Our estimated mean emission corresponds to ~75% of the US EPA reported mean (95% CI) of 15.4 (8.7–41.1) MMT CO2-eq in 2021, calculated by following the IPCC2019 Tier 2 methodologies3. This is attributed to the relatively high default EF (1.5% kg N2O-N kg−1 TN) applied uniformly across all WRRFs in the US EPA inventory, irrespective of differences in facility configurations and treatment processes. Interestingly, if correcting the errors in the IPCC dataset and adjusting EF to 1.1% kg N2O-N kg−1 TN, the updated EPA inventory would be 11.4 MMT CO2-eq, which is comparable with results obtained in this study. However, such single EF value cannot reflect high variations in N2O emissions from different facilities. To compare disparities in estimated N2O emissions between the approach proposed in this study and the EPA method, we calculated the mean relative differences (RD, equalling to [(this study−EPA)/EPA]) and analysed them across different WRRF configurations and bioreactors (Fig. 5c). For WRRFs that use TF, lagoon, CAS, EA, OD and A/O as main bioreactors, the mean RD values span −90% (that is, ‘PRI + BIO’ facilities using lagoon as the main bioreactor) to −25% (that is, ‘PRI + BIO + SLG’ facilities using contact stabilization as the main bioreactor), indicating that the EPA-estimated N2O emissions from these facilities are 9 to 0.3 times greater than in this study. Conversely, the mean RD values of WRRFs that employ Bardenpho, step feed, SBR and MBR as main bioreactors span 8–114%, meaning that EPA underestimates their N2O emissions by 8–114%. This finding underscores the importance of considering unit process and bioreactor-specific factors in accurately estimating emissions, as a single EF value may not effectively represent the wide range of N2O emissions observed in different facilities across a country.

Despite rigorous data processing and analysis, several uncertainties arose considering the lack of EF measurements for two main popular bioreactors, including other AS and contact stabilization, which treat 15% of national wastewater. In addition, an increasing number of WRRFs have started implementing side-stream treatment to enhance nutrient removal, and existing literature indicates notably higher fugitive emissions from these processes. However, there is very limited data on which WRRFs have installed such processes. To capture these uncertainties and potential ranges of nationwide N2O emissions from the US wastewater sector, we conducted scenario analysis (Fig. 5d). The base case and scenario 1 addressed uncertainties caused by EF-unknown bioreactors, using CAS and step feed to represent emissions from these bioreactors, respectively. Both scenarios assume no side-stream treatment. Scenario 2 captured uncertainties of side-stream treatment and assumed the implementation of side-stream treatment at WRRFs >10 MGD. Scenario 2 shows a higher limit of wastewater N2O emissions as only a limited number of WRRFs in the United States employ side-stream treatment so far. The results show that mean (95% CI) N2O emissions of Scenarios 1 and 2 are 12.7 (9.9–15.4) MMT CO2-eq and 19.7 (12.3–27.0) MMT CO2-eq, respectively. The results indicate that side-stream treatment may result in higher N2O emissions. Therefore, it is critical to minimize such emissions during the design and operation of these units, as fugitive emissions may offset or reverse the carbon benefits of the projected energy savings. In addition, it is necessary to identify EFs for side-stream treatment. Overall, since the tiered emission factors were derived from subsectors/processes across the globe, this approach is applicable to the global scale, despite the absence of a global WRRF database for such estimations.

Discussion

This study is presumably the most comprehensive analysis of N2O emissions from municipal wastewater treatment so far and contributes to more reliable assessments of the environmental impact of the wastewater sector at both local and national levels. The results highlight that current estimations based on uniform EFs are largely oversimplified and misestimated. Therefore, the reported tier-based EF approach offers a more accurate and representative estimation of wastewater N2O emissions. Although the method carried out in this study is based on all currently available in situ measurements reported in the literature, which is 12 times more comprehensive than the IPCC2019 database, we do recognize the limitations in observational data, particularly at the whole-plant level, the sewer networks and the understudied units such as step feed, Bardenpho and sludge treatment. Moreover, cross-comparisons of different monitoring methods and site-specific monitoring campaigns that consider spatial and temporal variations should be carried out where possible to mitigate the uncertainty of the use of EFs for this important and highly variable GHG source. When more cross-comparison measurements become available, it is necessary to quantify the uncertainty of the tier-based EF approach in estimating N2O emissions from the wastewater sector and to outline priorities for improved reporting53.

Accurate accounting of N2O emissions from individual WRRFs is crucial for the development and implementation of effective local mitigation strategies, as N2O may account for more than 80% of process emissions. Currently, GHG inventories in this sector rely on voluntary reporting from facilities, and the compilation of spatiotemporal data is a manual process. More automatic and advanced tools could facilitate the development. For example, AI and machine learning tools have the potential to synthesize data, fill gaps in unreported emission sources, add spatiotemporal information to raw data and automate reporting processes. These technologies can greatly facilitate the development of more comprehensive and up-to-date emissions inventories. In addition, modelling methods such as mechanistic kinetic modelling54, ASM3-based models55 and fuzzy-logic control strategy56 have been utilized to simulate production and emissions at full-scale plants, providing valuable insights and applicable solutions for controlling N2O emissions in real time. Overall, by incorporating site-specific data, developing accurate EFs that account for process variations and utilizing advanced tools, we can achieve an improved understanding and management of N2O emissions.

From a broader perspective, the need to accurately measure and reduce N2O emissions becomes more prevalent as strategies to curb carbon-based pollutants (CO2 and CH4) become more defined and implemented. The less-noticed N2O emissions, mostly from agriculture and waste sectors, would be pivotal in achieving long-term climate objectives. This challenge is reminiscent of the difficulties encountered in the recovery of the stratospheric ozone layer after the implementation of the Montreal Protocol, which banned the production of major ozone-depleting substances, including chlorofluorocarbons57. With the phase-out of most ozone-damaging agents, N2O has now become the primary man-made threat to the stratospheric ozone layer58. Urgent mitigation of N2O emissions is crucial to safeguard tropospheric ozone and fulfil long-term climate objectives.

Methods

Natural language processing method

A combination of literature text mining and manual review was used to collect studies that monitored and reported N2O emissions based on full-scale or pilot-scale measurements. The data collection and preprocessing through literature text mining involved five key steps. First, >330,000 full publication records spanning from 1900 to January 2023 were collected from Web of Science on the basis of general wastewater-related keywords (for example, wastewater, activated sludge, sewer, sewage, anaerobic digestion, nitrification, anammox and so on). Second, the titles, abstracts and keywords of the collected publications were pretreated using NLP methods, including n-grams (one, two, three and four adjacent words), word tokenization, lowercasing, stop-word removal and stemming (see ref. 59 for the detailed NLP method). This step aimed to develop a searchable database for the following steps. Third, a more-refined screening process was conducted to retain literatures that are relevant to N2O, emission and wastewater-related terminologies. Fourth, an additional detailed search was conducted using the abstracts (tokenized to sentences) to exclude laboratory-scale and other domain-focused papers (for example, river, lake, surface water, wetland, soil, aquifer, groundwater). A total of 970 papers that are closely relevant were identified. Finally, a manual review of titles and abstracts was carried out. We identified a total of 119 literatures on WRRF N2O flux measurements, 6 literatures that solely measured unit process emissions (that is, sludge treatment (dewatering, storage, composting) and secondary nitrification), and 10 literatures on sewer N2O production and emission. Details of the text mining methodology are available in Supplementary Text 6.

WRRF dataset

Most of the studied WRRFs employed advanced BNR processes such as MLE, A2O, oxidation ditch (OD), SBR and MBR. These BNR processes exhibited average TN removal rates ranging from 53% to 86% (Supplementary Fig. 1). We also separately grouped WRRFs permitted to meet secondary limits (nominally 30 day average of BOD5 and total suspended solids not exceeding 30 mg l−1 each), including lagoon, CAS and tricking filter (TF). These processes, in contrast, had mean TN removal rates that are within the range of 38–51%. Furthermore, we included side-stream treatment for the first time, such as PN, PN/PNA and N/DN, which are emerging processes designed for high concentrations of COD and nitrogen. Their mean nitrogen removal rates vary from 63% to 77%. Detailed descriptions of each treatment process are provided in Supplementary Text 1.

Sewer network dataset

So far, there are only 26 measurements of sewer N2O emissions. Sewer N2O emissions were measured at various locations, including manholes and pumping stations along sewer collection systems, as well as from the WRRF influent. Studies that focused on WRRFs but provided quantification of the emissions at the influent were also included, given that WRRF influents represent the sewage condition at the end of sewer networks. Out of the total of 26 measurements, only 4 observations reported both gaseous and dissolved N2O concentrations at WRRF influent pump stations. Twelve observations solely reported dissolved N2O in wastewater/sewage, while 8 observations only reported gaseous N2O emissions. The remaining 2 observations studied city-wide N2O emissions from sewage networks.

Statistical analysis

Previous studies reported N2O EFs in different units, such as g N2O-N kg−1 TN load, g N2O kg−1 TN load, g N2O kg−1 TN removed, g N2O-N kg−1 NH4+ removed and g N2O person−1 yr−1. We normalized the unit of N2O EFs to N2O-N as a percentage (by mass) of influent (plant) TN load to be consistent with the IPCC methodologies. It is worth noting that the unit of N2O EFs for side-stream treatment is N2O-N as a percentage (by mass) of influent (side-stream reactor) TN load. Corresponding assumptions were made to convert the unit if TN concentration is not available. For studies providing TKN, we assumed TN = TKN considering that the differences between TN and TKN were found to be negligible before biological treatment10. For studies solely presenting inorganic N data (sum of ammonia, nitrite and nitrate), we assumed TN = 1.69 × (NH4+ + NO2− + NO3−), given that 59% of TN in domestic wastewater is formed by inorganic N on average23. For sewer studies, the dissolved N2O concentration in µg N2O l−1 and gaseous N2O flux in mg N2O m−2 h−1 were widely reported. Thus, the two units were used to analyse the relative contribution of sewer networks to N2O emissions in this study. Due to high spatial and temporal variations in N2O emissions2,35,36,37, the monitored N2O emissions can vary significantly depending on when, where and how the campaign was carried out. To unveil how these factors affect N2O emissions, we also collected data in terms of campaign duration (from a few hours to years) and mode (continuous or intermittent). For intermediate-term (a few months without capturing the whole spectrum of temperature changes) and long-term samplings (lasting at least 1 yr), temporal variations in N2O emissions were captured by collecting average monthly or seasonal data when this information is available in either continuous or intermittent monitoring campaigns. Otherwise, the widely reported average values were collected. For studies reporting diurnal variations, we only collected average values to avoid the collected N2O EF observations being dominated by a few studies. Mean N2O emissions were compared, and 95% CIs were used to estimate the uncertainty of the mean value. The 95% CI was calculated using the population standard deviation (σ) and sample size (n), assuming Student’s t distribution and α = 0.05.

US national synthesis of N2O emissions

Direct N2O emissions from municipal WRRFs at national scale were estimated using the United States as a case study, considering data availability such as for influent flow, detailed wastewater and solid treatment train, and geographical location, among others. Emissions from sewer networks and effluents were excluded due to the lack of sufficient data. Indirect N2O emissions generated by electricity consumption, chemical usage, offsite biosolids disposal (for example, land application, landfill, incineration) and other sources occurring in a WRRF’s value chain (for example, construction activities, transportation, employee commuting) were not considered, as they are beyond the scope of this study. We employed the Monte Carlo approach to upscale plant-level N2O emissions to a national scale and to estimate uncertainties. For a series of Clean Watersheds Needs Survey (CWNS) datasets published in 2004, 2008 and 2012, the 2004 CWNS reported installed unit processes for all plants60. Starting from the 2008 survey, the United States transitioned to reporting installed unit processes only for those plants that upgraded their treatment trains. Therefore, we merged the three CWNS datasets to capture complete unit process data and identified a total of 14,788 WRRFs with unit process information (Supplementary Table 1). Given that treatment train details (that is, primary, secondary and sludge treatment) are missing for WRRFs obtained from the 2004 CWNS dataset, we identified their treatment trains on the basis of unit processes (Supplementary Table 2). Then, we further assigned types of bioreactor (Supplementary Table 3) and types of sludge treatment (Supplementary Table 4) for the whole dataset. Detailed processes are provided in Supplementary Text 7.

Specifically, we grouped unit processes into three categories (that is, primary, biological treatment units and sludge treatment) with consideration of 13 different types of bioreactor. Different onsite sludge disposal methods such as incineration, composting and air-dried biosolids stockpile were also considered. We made this categorization considering the understanding that N2O emissions from wastewater treatment are specific to each unit process and bioreactor. National wastewater N2O emissions were calculated as the sum of the product of N2O EF and influent TN load at each WRRF (equation 1). Given that the data for N2O EFs were positively skewed (Supplementary Table 5), three common right-skewed distributions, including Weibull, gamma and lognormal, were tested to identify which distribution fits the data best (R package ‘fitdistrplus’). For each simulation, we generated 10,000 random values from the corresponding best-fit distributions centred around mean and deviations. Detailed processes for determining which distribution best fits the data for each process unit are provided in Supplementary Text 8.

$${E}_{\rm{N}_{2}{\rm{O}}}=\mathop{\sum }\limits_{i=1}^{i=14,788}\left({\rm{TN}}\times \frac{{Q}_{i}}{Q}\times \sum _{j}{{\rm{EF}}}_{j}\right)\times \frac{44}{28}\times {\rm{GWP}}\times {10}^{-9}$$
(1)

where \({E}_{\rm{N}_{2}{\rm{O}}}\) is the nationwide annual N2O emissions from all domestic WRRFs (MMT CO2-eq yr−1); i indicates the studied WRRFs in the United States; TN is the annual nitrogen load entering a centralized wastewater system, calculated using equations (7–35) in the US EPA report (equation 2)3 (kg N yr−1). Qi and Q are the annual wastewater volumes treated in the ith facility and nationwide WRRFs (m3 yr−1); EFj is the N2O emission factor for unit process j (kg N2O-N kg−1 TN); GWP is the 100-yr global warming potential of N2O (273 kg CO2-eq kg−1 N2O) according to the Sixth Assessment Report of the IPCC1.

$${\rm{TN}}={\rm{{US}}_{{pop}}}\times p\times {\rm{Protein}}\times {F}_{{\rm{NPR}}}\times {N}_{{\rm{HH}}}\times {F}_{\rm{{NON}-{CON}}}\times {F}_{\rm{{IND}-{COM}}}$$
(2)

where USpop is the US population in 2021 equalling to 331,893,745 based on the US Census Bureau61; p is percent of wastewater collected for centralized treatment (83%); Protein is the consumed protein per capita (34.68 kg capita−1 yr−1), calculated by multiplying protein supply per capital per day (118.79 g capita−1 d−1 (ref. 62)), protein consumed as a fraction of protein supply (0.8 according to Table 6.10A in IPCC201915) and 365 d yr−1; FNPR is the fraction of nitrogen in protein (0.16 kg N kg−1 protein); NHH is the correction factor for additional nitrogen from household products (1.17); FNON-CON is the factor for non-consumed protein added to wastewater (1.13); FIND-COM is the factor for industrial and commercial co-discharged protein into wastewater treatment (1.25). The uncertainty range for each parameter is provided in Supplementary Table 6.

For WRRFs with onsite sludge composting and biosolids stockpile, N2O emissions from onsite sludge disposal were calculated (equation 3) and added to the national estimation. For WRRFs with onsite incineration, equation (1) was used to calculate emissions from sludge incineration.

$${E}_{{\rm{N}}_{2}{\rm{O}},{\rm{S}}}={\rm{{{EF}}_{S}}}\times {Q}_{i}\times {R}_{\rm{S}}\times 365\times \frac{44}{28}\times {\rm{GWP}}\times {10}^{-12}$$
(3)

where \({E}_{{\rm{N}}_{2}{\rm{O}},{\rm{S}}}\) is the annual N2O emissions from WRRFs with onsite composting and biosolids stockpile (MMT CO2-eq yr−1); EFS is the N2O EF for sludge composting (1.82 ± 1.32 g N2O-N kg−1 dry solid yr−1) or biosolids stockpile (0.06 ± 0.04 g N2O-N kg−1 dry solid yr−1); RS is the biosolids generation rate (0.24 kg dry solids m−3 of treated wastewater). The uncertainty range for each parameter is provided in Supplementary Table 6.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.