GEOSCCM was used to simulate atmospheric N2O with geographically resolved surface emissions from soil, ocean, and anthropogenic sources and full stratospheric chemistry with stratospheric N2O destruction due to photolysis and O(1D) oxidation (Nielsen et al., 2017; Liang et al., 2022). GEOSCCM has been evaluated extensively in multi-model assessments and shown to accurately represent the mean atmospheric circulation, the interhemispheric exchange rate, the mean age of air in the tropical and polar stratosphere, and the mean atmospheric lifetime of N2O (Liang et al., 2022, and references therein). For the current study, GEOSCCM was run at 1° × 1° resolution with 72 vertical layers from the surface to 0.01 hPa. In addition to the standard total N2O tracer, four additional N2O tracers were included to track the following items: (1) aged air from the stratosphere (N2OST) and (2) soil, (3) ocean, and (4) anthropogenic sources freshly emitted in the troposphere. Following the approach of Liang et al. (2008), the tropospheric tracers become the stratospheric tracer, N2OST, when they are transported across the tropopause and retain that identity even when N2OST re-enters the troposphere, thereby providing a model estimate of the stratospheric influence on tropospheric N2O.

The full GEOSCCM simulation spanned 1980–2019, but this study focuses on the final 20 years from 2000–2019 for the correlation analysis between model surface N2O anomalies and QBO and PLST. As described in detail in Liang et al. (2022), the GEOSCCM N2O lifetime decreased slightly after 2000 (from 119±2 years in the 1990s down to 116±2 years in the 2010s), and model emissions were optimized to account for the observed change in the atmospheric N2O growth rate. GEOSCCM temperature and QBO do not necessarily correspond to observations since both are internally generated by the GEOS general circulation model, which is free running rather than forced by a reanalysis meteorology. GEOSCCM PLST and QBO were computed in the same way as the observed indices, as described below in Sect. 2.4.1 and 2.4.2, respectively. The GEOSCCM N2O fields were saved as monthly means and were detrended and converted to anomalies by subtracting a deseasonalized fit to the model time series sampled at Mauna Loa (MLO). The N2O time series at MLO is a good proxy for the global N2O trend, and thus its subtraction provides a convenient, single-station approach for calculating anomalies of the N2O mixing ratio for contour plots.

Mean seasonal cycles for NOAA surface N2O observations and GEOSCCM N2O tracers were estimated using a bootstrapping method in which 20 % of the time series was randomly removed and the remaining 80 % was fit to a third-order polynomial plus first four harmonics. These steps were repeated over 500 iterations to estimate the range of uncertainty in the harmonic components of the fit.

Interannual variability in the atmospheric growth rate of N2O in the NOAA surface NH, SH, and global time series was calculated by first removing the seasonal cycle from the monthly mean time series by computing a 12-month running average, 1Xi=(Ci-6+2∑k=i-5i+5Ck+Ci+6)/24, where C is the original monthly mean time series and X is the deseasonalized time series. The slope of the deseasonalized time series then was computed as a central difference, 2Si=12xi+1-xi-12, where S is the centrally differenced slope and the scalar 12 converts S from units of parts per billion per month to parts per billion per year. To account for the increasing growth rate of atmospheric N2O, the absolute slopes S were converted to atmospheric growth rate anomalies by removing an optimal (increasing) linear fit determined by recursive least-squares regression.

Interannual anomalies in the magnitude of the seasonal minimum were calculated by detrending the raw monthly mean N2O data with a third-order polynomial, after which a climatological seasonal cycle was constructed by taking the average of the detrended data for all Januaries, Februaries, etc. This climatological annual cycle was subtracted from the original raw data to produce a deseasonalized (but not detrended) time series. A running 12-month annual mean of this curve was then computed as in Eq. (1) but where C is now the deseasonalized time series rather than the original monthly mean time series. This analysis focused on mid- and high-latitude sites in the NOAA dataset. At stations with gaps in the monthly data, the original third-order polynomial fit was used as a placeholder in the running mean. The running mean was subtracted from the deseasonalized curve to remove the secular trend and other low-frequency variability, thus isolating the residual monthly anomalies.

The computation of N2O AGR anomalies from Sect. 2.3 created a set of monthly resolved time series for the SH, NH, and global means. These were plotted against various proxies and indices for stratospheric influences and ENSO. In addition, the high-frequency residuals from Sect. 2.3 at various mid- and high-latitude sites were sorted by month, and selected months were plotted against PLST as described below in Sect. 2.4.1. The PLST proxy involves a single value for each year, such that correlation coefficients and p values were computed based on the number of years N with data. For the ENSO and QBO indices, the correlation statistics were computed based on the reduced, effective N (Neff) number of monthly data points after accounting for the autocorrelation that is introduced by the 12-month running mean used to compute the N2O AGR (see Supplement Sect. S1 for more details.)

Figure 1 shows that the GEOSCCM N2O mean seasonal cycle at surface sites is dominated by stratospheric air depleted in N2O that is transported to the surface, rather than by the influence of surface sources. This dominance holds within the uncertainty of the seasonal cycles, as estimated using a bootstrap method. However, the surface emissions tend to pull the total N2O seasonal minimum about 1 month earlier than the N2OST minimum at most sites. Figure 1 also shows that GEOSCCM captures the mean observed seasonal cycle in N2O relatively well at sites in the SH but overestimates the amplitude of the cycle at sites in the NH, with a ∼ 1–2-month delay in phasing relative to observations.

Figure 2 provides a two-dimensional view, using zonally averaged altitude–month contour plots at middle and high latitudes in both hemispheres, of how the signal of stratospheric air depleted in N2O is transmitted to the surface in GEOSCCM. This N2O-poor air accumulates during winter (starting in ∼ December in the NH and ∼ July in the SH) in the polar stratosphere and descends vertically and crosses into the troposphere in spring (March–April) in the NH and early summer (January–February) in the SH. The SH latitude panels in Fig. 2 are plotted with a 6-month shift to help visualize the later seasonal phasing of the stratospheric influence in the SH relative to the NH. After crossing into the troposphere, the N2O-poor air continues to move downward and also mixes equatorward from approximately January to May at SH middle to high latitudes and April to October at NH middle to high latitudes (Liang et al., 2009, 2022). Due to lags in downward propagation and mixing, the modeled surface minimum in the lower troposphere does not occur until late summer to early autumn in both hemispheres (Figs. 1 and 2). Figure S1 in the Supplement shows a three-dimensional view of this process in a series of 12 monthly altitude–latitude plots.

Figure 3 shows that the NOAA N2O empirical background, when organized as a series of zonally averaged altitude–month contour plots at NH latitudes, has features similar to those simulated by GEOSCCM. Both model and observations show a signal of N2O depletion beginning in early spring in the upper troposphere that propagates down to the surface. In both model and observations, the signal is strongest at high latitudes and weakens substantially moving equatorward. However, the NOAA data suggest a faster, more direct downward propagation of the stratospheric signal, which arrives at the surface in August–September, compared to September–October in GEOSCCM. As a result, the phasing of the GEOSCCM surface minimum is delayed ∼ 1–2 months relative to the NOAA empirical background, consistent with the comparison to NOAA surface monitoring data in Fig. 1.

The positive anomalies in Fig. 3 also differ between model and observations, with surface features centered on July at 40–50° N in GEOSCCM, which uses a summer-dominant soil source (Liang et al., 2022), while the NOAA empirical background shows positive surface anomalies in late winter and spring, likely reflecting North American agricultural sources (Nevison et al., 2018). At 60 and 70° N, the stronger contrast between positive and negative anomalies throughout the atmospheric column in GEOSCCM compared to NOAA reflects the model's larger seasonal cycle, as seen also in Fig. 1.

Figure 4 provides a further perspective on the signal of N2O-poor stratospheric air in the NOAA empirical background. When viewed as a 12-month sequence of altitude–latitude contours in the NH, the signal originates at northern polar latitudes in the upper troposphere in late winter and early spring and descends and mixes equatorward, with a peak surface influence around July–September at middle to high latitudes in the NH that overshadows any surface source signal. By late fall and winter, the signal has dissipated at the surface but is forming again in the upper troposphere.

QCLS N2O data from airborne surveys extend up to 14 km and thus provide a broader perspective with respect to altitude of the stratospheric influence on tropospheric N2O. Of the three available airborne surveys (HIPPO, ORCAS, and ATom), HIPPO provides the most complete N2O time series across all seasons. In Fig. 5, the southbound transects from the five HIPPO deployments are detrended and arranged chronologically as altitude–latitude contour plots over an annual mean cycle. These plots form a sequence with a similar movement of N2O-poor stratospheric air from upper levels down to the surface as seen in GEOSCCM and the NOAA empirical boundary data. This progression is most readily seen in the NH in Fig. 5, in which N2O-poor air in the polar lower stratosphere has crossed the tropopause by March. By June it has descended into the middle troposphere and started moving equatorward, reaching its maximum influence at the surface in August. By November, the stratospheric signal is no longer visible at the surface following tropospheric mixing and dilution. This seasonal progression is also evident in a fuller dataset that also includes the ATom and northbound HIPPO transects collapsed into a climatological cycle (Fig. S2 in the Supplement)

In this section NOAA surface N2O AGR anomalies from 1998–2020 are plotted against polar lower stratospheric temperature (PLST) and QBO and ENSO indices, with varying lag times as described in Sect. 2. The analysis focuses on the NOAA global, NH, and SH mean products, with the assumption that a significant correlation between the interannual variability in the N2O AGR and one or more of the indices can be interpreted to support a causal influence on the N2O AGR. However, correlation does not prove causation and cannot distinguish possible confounding effects, such as the influence of ENSO on both interhemispheric transport and surface sources. A similar correlation analysis is performed for the GEOSCCM N2O AGR and the model's internally generated QBO and PLST fields, with the assumption that similarities between modeled and observed correlations may also support a causal influence.

Figure 6, which presents correlations between the SH surface N2O AGR and the QBO and PLST indices, first for NOAA observations and next for GEOSCCM output, shows that the QBO is positively correlated to the NOAA N2O AGR. The optimal correlation (R=0.55, p<0.01) occurs for QBO in the upper stratosphere at 20 hPa with a time shift of about 18 (17–19) months relative to the N2O time series. In contrast, spring PLST is not significantly correlated to the NOAA surface N2O AGR in the SH (Fig. 6b). In Fig. 6c, the correlation between GEOSCCM QBO and the SH N2O AGR is weak (R=0.24, p>0.10) but also positive in sign in the upper stratosphere with an optimal shift in the GEOSCCM QBO of about 19 months, similar to that found for NOAA. In Fig. 6d, GEOSCCM PLST is significantly anticorrelated with the N2O AGR averaged over a range of different 12-month intervals, with the highest correlation (R=-0.61, p=0.01) over the 12-month interval from May–April.

Figure 7, which presents the corresponding correlations for the NH surface N2O AGR, shows that, in contrast to the SH, the NOAA NH N2O AGR is correlated only weakly to the QBO index at all altitudes and with a negative sign. The highest correlation in the NH occurs for 50 hPa QBO (R=-0.21, p>0.1) with a 10–14-month lag (Fig. 7a). However, the NOAA NH N2O AGR is anticorrelated significantly to winter PLST (R=-0.69, p<0.001), with an optimal correlation for the 12-month period from July–June encompassing the January–March PLST average (Fig. 7b). GEOSCCM predicts an anticorrelation (R=-0.47, p<0.05) between the QBO and the NH N2O AGR, which is optimal around 50 hPa with 10–14-month QBO lag, similar to NOAA (Fig. 7c), and also predicts an anticorrelation between PLST and NH N2O AGR (R=-0.55, p=0.02) (Fig. 7d).

Figure 8, which presents correlations between the Niño 3.4 index and the NOAA surface N2O AGR, shows that the two are significantly anticorrelated (R∼-0.4, p<0.05) for both the global and SH N2O AGR, with little or no monthly lag in the index. In the NH, the anticorrelation is weaker (R=-0.26, p>0.10) with an optimal lag of 7 months in the Niño 3.4 index relative to the N2O AGR (Fig. 8).

The N2O monthly anomaly correlation analysis is focused solely on PLST, which has one unique value each year that can be plotted against the corresponding N2O anomaly for any given month. The months selected for this correlation analysis were those surrounding the seasonal N2O minimum, which is the most distinct feature of the seasonal cycle for NOAA sites at remote mid and high latitudes. These months were hypothesized, based on previous work, as most likely to be influenced by the descent of N2O-poor air from the stratosphere (Nevison et al., 2011). In contrast, QBO and ENSO are monthly indices for which it is not straightforward to choose a representative month to correlate to the N2O monthly anomaly, given that the anomaly might result from the cumulative effect over multiple months.

Figure 9a shows that the mean seasonal cycles in SH surface N2O for NOAA and GEOSCCM total N2O (N2Otot) and stratospheric N2O (N2OST), as illustrated at South Pole (SPO), all have similar autumn seasonal minimum, within the range of uncertainty as estimated using a bootstrap method. Figure 9b shows that PLST from the previous spring is significantly anticorrelated to NOAA SPO N2O monthly anomalies in February, when N2O is descending into its minimum. This correlation is observed in both January and February at SPO and several extratropical southern NOAA sites including Cape Grim, Tasmania, and Palmer Station, Antarctica. The sign of the correlation is such that more negative surface N2O anomalies occur during warm years, in which stronger than average descent of N2O-poor air occurs into the polar lower stratosphere over the austral winter and spring. GEOSCCM simulates similar correlations between PLST and austral summer N2O anomalies at these sites for both N2OST and N2Otot in February (Fig. 9c, d) and March, i.e., the correlations are delayed by about 1 month relative to NOAA surface observations.

Figure 10, which compares February altitude–latitude N2O contour plots from the ORCAS and ATom airborne surveys, offers some qualitative support for the observed surface correlations in Fig. 9. The contour plots show more depleted N2O values in the SH polar upper troposphere during ATom-2, which took place in February 2017 after a relatively warm spring in the Antarctic lower stratosphere (strong BDC) compared to ORCAS, which took place in January–February 2016 after a particularly cold spring (weak BDC). Figure 10b shows ATom-2 data over the full 65° S to 75° N latitude span, putting the stratospheric influence coming from the southern polar stratosphere into broader perspective. Figure 10a extends only from 70 to 20° S because ORCAS was confined to that region. Notably, the positive anomaly in the mid-troposphere observed at 40–60° S during ATom-2, which may be a source plume from the Southern Ocean, tends to contradict the hypothesis that SH tropospheric N2O was lower overall in 2017 than in 2016.

In contrast to the SH, PLST in the NH from the previous winter is not correlated significantly to N2O monthly anomalies at extratropical surface sites for either NOAA or GEOSCCM in any of the months surrounding the NH N2O seasonal minimum, with the exception of Mace Head, Ireland, where a negative correlation is found in July in GEOSCCM (Fig. S3).

The results of the correlation analysis based on NOAA surface N2O data are similar to those found in previous studies based on Advanced Global Atmospheric Gases Experiment (AGAGE) surface N2O data (Prinn et al., 2000; Nevison et al., 2007, 2011). In general, these correlations are weak, in part because the variability in surface N2O is very small compared to its mean tropospheric mixing ratio. Nevertheless, PLST in the NH correlates significantly to NOAA surface N2O AGR anomalies (Fig. 7b), and PLST in the SH correlates significantly to monthly N2O anomalies in February near the time of the seasonal N2O minimum (Fig. 9b). The negative sign of these correlations is easily understood and consistent with more downward transport of N2O-poor air and warming of the polar lower stratosphere in years with stronger BDC, with subsequent cross-tropopause transport that deepens the descent of tropospheric N2O into its seasonal minimum and slows the observed AGR of surface N2O.

The reason for the positive correlation of the QBO index with the SH surface N2O AGR (Fig. 6a, c) is less obvious. A similar correlation in the SH (but not the NH) has been observed in other studies but not fully explained (Ray et al., 2020). In our analysis, the positive correlation between the QBO and the SH N2O AGR is strongest for the QBO at 20 hPa with 18 months lag and then weakens with decreasing QBO altitude down to 100 hPa, with a concurrent decrease in the optimal lag, likely due to the time needed for downward propagation of QBO winds (Fig. S4). At 50 hPa, we find an optimal lag time of 10–12 months (R=0.39), consistent with Ray et al. (2020) (who only presented results for QBO = 50 hPa).

Photochemical destruction of N2O is highest when QBO winds above 30 hPa are in the westerly (positive) phase and lower-altitude QBO winds are in the easterly (negative) phase. This configuration is associated with increased vertical upwelling in the tropical lower stratosphere, which transports more N2O to its peak loss region around 32 km (Strahan et al., 2021; Ruiz et al., 2021). Thus, the magnitude of QBO-associated photochemical destruction per se cannot be the main driver of the stratospheric influence on surface N2O, since one would logically expect a negative correlation (i.e., slower growth in the troposphere due to more stratospheric N2O loss during positive QBO). Ruiz et al. (2021) similarly concluded that surface variability in N2O is not correlated directly to the QBO-driven variability in stratospheric loss but rather by dynamical variations in cross tropopause fluxes of air, which are governed at least in part by the BDC.

The dynamics of the QBO, its interaction with the BDC and ultimate influence on surface N2O are complex. However, the positive correlation of QBO, peaking at 20 hPa, with the SH N2O AGR, could be explained in the context of Strahan et al. (2015), as described in detail in Sect. S2. Briefly, the QBO has an associated meridional circulation, which transports N2O-poor air poleward from the region of peak photochemical destruction in the tropics between about 30 and 10 hPa in altitude. Paradoxically, this meridional circulation transports less N2O-poor air toward the poles during the phase when the QBO is positive above 30 hPa (i.e., when photochemical destruction is highest). The N2O-poor air subsequently is trapped in the Antarctic polar vortex and undergoes BDC-driven diabatic descent in isolation from mixing with lower latitudes, arriving largely intact in the lower stratosphere and eventually at Earth's surface, with a long lag time consistent with the 18-month lag found in Fig. 6a.

In the NH, the planetary wave activity that drives the BDC is stronger due to the more variable topography and stronger land-sea contrasts. Consequently, the BDC-driven descent into the winter pole is more strongly seasonal and the NH polar vortex is less isolated (Holton et al., 1995; Scaife and James, 2000; Butchart, 2014; Kidston et al., 2015), such that any signal associated with the QBO meridional circulation does not transport intact to lower altitudes (Strahan et al., 2015). This may explain why, for both NOAA surface stations and GEOSCCM, the NH N2O AGR is more strongly correlated to PLST (a proxy for the BDC) than it is to the QBO, consistent with Ruiz et al. (2021).

In contrast to the NH, the NOAA SH surface N2O AGR does not correlate to PLST. This result is somewhat puzzling given the significant correlation between PLST and NOAA N2O February anomalies at SH high-latitude surface sites (Fig. 9). It appears that the impact of the stratosphere in austral summer as tropospheric N2O descends into its seasonal minimum is not sufficient to influence the N2O AGR across the whole SH over the entire year. The SH N2O AGR results may reflect the strong preservation of the QBO signal that is ultimately transported into the troposphere, as discussed above, combined with the relatively weak BDC in the SH and/or the interference of ENSO-driven signals discussed below.

The NOAA surface station N2O AGR correlation with the ENSO index is significant in the SH (R=-0.42, p<0.05, 0–2-month phase shift) but weaker in the NH (R=-0.26, p>0.10, 7-month optimal phase shift) (Fig. 10). The correlation in the SH could in part reflect meteorological shifts in the tropical low-level convergence pattern during positive (El Niño) conditions. For atmospheric gases with a positive north–south latitudinal gradient, these shifts result in a lessened influence of winds from the NH on the tropical SH and an increased influence of southeasterly winds. The NOAA Samoa site at 14° S, which strongly influences the cosine-latitude-weighted SH average, is known to be affected by these kinds of wind shifts (Prinn et al., 1992; Nevison et al., 2007). The fact that the N2O AGR correlation with ENSO is considerably weaker in the NH than in the SH suggests a limited impact of ENSO on NH N2O and supports the hypothesis that reduced north-to-south transport during positive ENSO contributes to the correlation observed in the SH.

The negative correlation between N2O AGR and ENSO also may reflect a true reduction in the biogeochemical N2O source during the positive ENSO phase, for example, due to drought over tropical land or due to reduced upwelling in the tropical ocean (Ishijima et al., 2009; Thompson et al., 2013). The most well-documented biogeochemical response of N2O to ENSO events occurs in the eastern tropical South Pacific (ETSP), a well-known oxygen minimum zone (OMZ) and hot spot of oceanic N2O emissions (Areivalo-Martinez, 2015; Ji et al., 2019). El Niño conditions decrease upwelling in the ETSP, thereby reducing the surface productivity, deepening the oxycline, contracting the OMZ, and decreasing the N2O sea-to-air flux (Ji et al., 2019; Babbin et al., 2015).

However, less than a quarter of the total N2O budget likely comes from oceanic emissions, of which the ETSP is only one component (Yang et al., 2020; Canadell et al., 2021). This raises questions about whether a reduced ETSP source (or a strengthened source during La Niña periods) has enough leverage to control the overall N2O AGR. Ruiz et al. (2021) removed the stratospheric influence from surface N2O data to infer a source of ∼ 1 Tg N (about 5 % of the total annual N2O source) associated with the 2010 La Niña event, which could have come from tropical land or ocean, or some combination of both. Similarly, Kort et al. (2011) found evidence of strong episodic pulses of ∼ 1 Tg N from tropical regions, based on maxima in QCLS N2O data measured in the middle and upper troposphere during aircraft campaigns in 2009. These pulses were not tied specifically to an ENSO event but rather more generally demonstrated the strength of the tropical N2O source.