Abstract
Peroxyacetyl nitrate (PAN) is an important photochemical product and affects ozone (O3) formation in the troposphere. Yet, the long-term observation of PAN remains scarce, limiting the full understanding of its impacts on photochemical pollution. Here, we observed PAN from 2018 to 2022 in urban Fuzhou, Southeastern China. We found that, in contrast to upward trend of O3, PAN concentrations shown a significant decreasing trend at an average rate of −0.07 ppb/year. NO2, CO, UVB, and T contributed to the decreasing trend of PAN according to Machine learning analyses, while the effect of O3-represented atmospheric oxidation capacity on PAN was fluctuating from year to year. Chemical box model revealed active PA production and depletion in Fuzhou. Thus, despite the decreasing PAN concentration, PAN chemistry effectively promoted O3 formation by rising ROx levels, leading to increases of 2.18%–58.4% in net O3 production rate in different years. Our results provide valuable insights into the evolution of photochemical pollution in urban environments.
Similar content being viewed by others
Introduction
Peroxyacetyl nitrate (CH3C(O)O2NO2,PAN) is an important secondary product in the process of photochemical pollution1. The precursors of PAN are volatile organic compounds (VOCs) and nitrogen oxides (NOx)2. Specifically, the certain VOCs like acetaldehyde (CH3CHO), acetone (CH3C(O)CH3), methacrolein (MACR), methyl ketone (MVK), and methylglyoxal (MGLY)3 are oxidized by OH, O3 or NO3, to form the peroxyacetyl radicals (CH3C(O)O2·, PA), which then reacts with NO2 to produce PAN4. The dominant loss pathway for PAN is thermal decomposition of PAN, while PAN can also be lost indirectly by reactions of PA radicals with NO, HO2, and NO35,6. Due to its thermal instability, PAN has a relative short atmospheric lifetime and can release NO2 and PA radicals in a warm environment4. Thus, PAN is regarded as a temporary reservoir for NO2 and regulates the spatial distribution of O3 production on a regional scale.
A number of field observations of PAN have been conducted at different types of sites. The PAN concentrations varied widely from ppt to ppb, and the levels of PAN at the urban sites were generally higher than those at background sites (summarized in Supplementary Table 1). The studies of PAN in China mainly focused on the investigation of formation mechanism and its impact on O3 production, quantification of transport effect, as well as source apportionment7,8,9,10,11. The formation of PAN was usually assessed through relative incremental reactivities (RIR) analysis. The results shown that PAN production at sites such as urban Beijing and Xiamen was constrained by VOCs alone7,12, while others such as rural Qingdao and suburban Hong Kong were controlled by both VOCs and NOx13,14,15,16. In urban areas, PAN and O3 are both typical photochemical products, but they respond differently to precursors. PAN was more directly related to PA precursors such as CH3CHO, MGLY, free radicals, and other sources of oxygenated volatile organic compounds (OVOCs)17,18. Previous studies have found that PAN had inhibition or promotion effects on O3. Under conditions of significant NOx deficiency and low atmospheric oxidation capacity, PAN tended to compete with O3 and terminated the radical chain reactions13. Alternatively, PAN promoted O3 production by providing NO2 as well as PA radicals and boosting the ROx (OH, HO2, and RO2) radicals circle12,15.
In the past decade, photochemical pollution in China has been widely concerned. A general trend of decreased O3 was observed in Chinese urban agglomerations, such as the Beijing-Tianjin-Hebei (BTH) region, the Yangtze River Delta (YRD) region and Sichuan Basin, except for the Pearl River Delta (PRD), which showed an increasing trend19,20. And the drivers for the trend of O3 were extensively investigated from chemical, transport and meteorological perspectives21,22,23. However, unlike O3, previous studies on PAN were mostly conducted in specific photochemical smog episodes or four-season observation24,25,26. Long-term observations of PAN are rare even around the world1,27. The only related work in China was conducted in 2015–2019 at a regional background site in Beijing, where they found an almost upward trend of PAN accompanied by the rising O3 level28. Thus, there is still a research gap in long-term PAN observations in Chinese urban areas. Considering that PAN and O3 respond differently to precursors and atmospheric conditions, studies on long-term trend of PAN, in addition to O3, will contribute to a more complete understanding of the photochemical pollution evolution.
Fuzhou is one of the fastest urbanizing regions in southeastern China and photochemical pollution in Fuzhou has been concerned in recent years29,30. In this study, a 5-year observation from 2018 to 2022 was conducted in Fuzhou. To our best knowledge, this is the first multi-year PAN observation in Chinese urban areas. Here, the inter-annual trend of PAN in urban Fuzhou was investigated using the Kolmogorov–Zurbenko Adaptive (KZA) filter method. The key influencing factors of PAN were identified and their impacts on the long-term trend of PAN were further quantified using the explainable machine learning (ML) method. Finally, a chemical box model was used to explore the photochemical mechanism of PA (thus PAN) and more importantly, the impact of PAN chemistry on O3 formation in different years. This study deepens our understanding of PAN chemistry and contributes to our overall insight into of the photochemical pollution evolution in urban environment.
Results and discussion
Overview of measurements
The overall pollutants levels and meteorological conditions in urban Fuzhou from 2018 to 2022 are shown in Fig. 1 and Supplementary Fig. 1. During the study period, mean monthly concentration of NO2, NO, particulate matter (PM2.5), sulfur dioxide (SO2) and carbon monoxide (CO) were 12.2 ± 3.42 ppb, 4.13 ± 1.53 ppb, 24.1 ± 6.79 μg/m3, 4.7 ± 1.2 ppb and 0.6 ± 0.1 ppm, respectively, which were lower than those in Chinese urban agglomeration including the NCP region, PRD region, YRD region and Sichuan Basin31. Compared with a previous observation conducted in Fuzhou from 2014 to 2017, the concentrations of these primary pollutants decreased, largely due to the Chinese government’s effect to mitigate emissions30,32. However, the concentration of O3 in Fuzhou during 2018–2022 was 31.4 ± 7.04 ppb, which was significantly higher than before (28.7 ppb)32. This indicates that photochemical pollution became critical in urban Fuzhou.
Monthly mean values of PAN, PA, atmospheric lifetime (τ) of PAN, pollutants (NO2, NO, CO, O3) and photochemical condition (T, UV) during 2018–2022. Shade represents a standard deviation.
Monthly average concentrations of PAN were from 0.20 ppb to 1.50 ppb, with a mean value of 0.50 ppb in this study, which showed a roughly decreasing trend from 2018 to 2022. The maximum PAN concentrations occurred in January 2019. In this month, we also observed almost the highest concentrations of NO2 and CO likely due to low boundary layer (BLH), which made it easy for pollutants including PAN to accumulate (Supplementary Fig. 1). In addition, low air temperature (T) at that time made it hard for PAN to thermally decompose. In comparison, the PAN concentration level in Fuzhou was lower than those in urban regions such as Beijing (1.00 ± 0.97 ppb), Tianjin (0.73 ppb), Jinan (1.89 ± 1.42 ppb), Zhengzhou (1.96 ± 1.55 ppb) and Shanghai (1.30 ppb), and comparable to the cities in southern China with clean air like Shenzhen (0.54 ppb) and Xiamen (0.55 ppb) (Supplementary Table 1). PA radicals were estimated by PAN, NO2, and reaction coefficients. The monthly average concentration of PA in urban Fuzhou was 0.14 ppt (0.008–0.54 ppt), which was higher than those at background site (0.09 ppt) and rural site (0.07 ppt)33, but lower than that at suburban site in Beijing (0.02–0.88 ppt)34. The lifetime of PAN (τ(PAN)) is the combined result of PAN sources and sinks, and affects the PAN concentration in the atmosphere. In this study, the average τ(PAN) was determined to be 8.35 h with a range of 0.66–46.8 h. This range was relatively small compared with the studies in the PRD region, the NCP region, and Xiamen, a city also in southeastern China8,26,34, suggesting a strong sink of PAN in Fuzhou. During the study period, the mean monthly temperatures (294.9 ± 6.4 K) were so high that they were very conducive for the thermal decomposition of PAN. The above results indicated although Fuzhou is a relatively clean area, critical photochemical pollution and active PAN chemistry also occurred.
Long-term trends of PAN
The original time series of PAN and O3 was decomposed to short-term, seasonal, and long-term components and the results are shown in Fig. 2. As both produced from photochemistry of VOCs and NOx, PAN and O3 generally have a close relationship. We found that the original time series of PAN and O3 did have a positive correlation (R = 0.50, p < 0.001), but this was mainly due to short-term component (R = 0.59, p < 0.001), with a smaller contribution from seasonal component (R = 0.21, p < 0.001). The short-term component of PAN often fluctuated towards high values, sometimes as high as 5 ppb. These PAN episodes were mostly associated with precursor transport, short-term emissions or weather fluctuations28. Apart from this, the diurnal pattern of PAN and O3 showed an obvious daytime single-peak (Supplementary Fig. 2a), suggesting that the short-term variations of PAN were still dominated by photochemistry. As for seasonal changes, both PAN and O3 remained high concentrations in April and October (Supplementary Fig. 2b). But PAN had additional high values in winter, which was probably due to less PAN decomposition as a result of low temperature in winter.
The panel numbers of (a, e), (b, f), (c, g) and (d, h) are the original time series, the short-term variations, the seasonal variations and the long-term trend of PAN and O3, respectively. The average correlation (R) between PAN and O3 in different kinds of series are 0.50, 0.59, 0.21, and −0.83, respectively. Note that the black dash lines show the 95% confidence intervals, the black text is the rate of change with lower value and upper value (ppb/year), and *** indicates the significance level p < 0.001.
To our surprise, there was no positive correlation between the long-term trends of PAN and O3 in this study (R = −0.83, p < 0.001). In detail, after filtering out short-term disturbance and seasonal changes, O3 exhibited a general upward trend with an average rate of 1.06 ppb/year (p < 0.001, Theil–Sen test35). This trend of O3 in Fuzhou was consistent with the pre-2021 trends of O3 along the southeastern coast of China and was similar to the pre-2022 O3 trends observed in the PRD region19,23. In contrast to O3, the trend of PAN in Fuzhou experienced a sequence of initial increase (Jan, 2018–Apr, 2019), subsequent decrease (May, 2019–Dec, 2021), and eventual stabilization (Jan, 2022–Dec, 2022) (see in Fig. 2d). Overall, PAN exhibited a significant downward trend at an average rate of −0.07 ppb/year (p < 0.001, Theil–Sen test).
PAN and O3 are both photochemical products, but they respond differently to the precursors and atmospheric conditions such as temperature, of which PAN has a simpler relationship with its final precursors PA and NOx. Hence, we also decomposed the long-term trends of PA, NOx and T to find the reason for PAN decrease (Supplementary Fig 3). The fluctuated long-term trend of PA indicates that the PA radicals was not the main cause of PAN decline. Instead, the decreasing NO2 over the study years could reduce PAN production, while other studies have indicated that NO2 reduction would increase O3 in VOC-limited urban areas by reducing the titration effect of O320,36,37. Besides, the ascending ambient temperature could promote the thermal decomposition of PAN, which also decreased PAN concentrations. The key factors influencing on the long-term downward trend of PAN and the chemical processes that might be involved were further discussed below.
Key factors of long-term trend of PAN
The XGBoost model with Shapley Additive explanations (SHAP) was employed to explore the influencing factors of PAN. SHAP values characterized the contribution of each feature to the variations in PAN. Mean absolute SHAP values (|SHAP|) of the factors were compared to determine the importance of those features to PAN. Meanwhile, the SHAP proportion (calculated by \(\frac{|{{{\rm{SHAP}}}}_{i}|}{\mathop{\sum }\nolimits_{1}^{i}|{{{\rm{SHAP}}}}_{i}|}\), i is features) was used to quantify the extent to which a factor explains the variations in PAN, and it was defined as “explanation rate” here.
The results show that O3, NO2, UV, T, surface pressure (SP), and CO were the most important factors of PAN (Fig. 3a). It should be noted that these factors had very different explanatory rates for original time series and long-term trend of PAN (Supplementary Table S2). Compared to original time series, NO2 and CO explained significantly more of long-term trend of PAN from 16.4% to 31.5% and from 5.85% to 11.2%, respectively. CO can represent the intensity of primary emissions of pollutants, such as anthropogenic OVOCs38. Thus, the results suggested an increased importance of PAN precursors for long-term trend in PAN. O3 is a proxy of atmospheric oxidation capacity (AOC) and UV and T are chemical-related meteorological factors. These three factors had comparable explanatory rates, together explaining 34.0% of long-term trend in PAN. In addition, NO can lead to loss of PAN by reacting with PA, but the low explanatory rate of NO (2.0%) implied a minor impact of this consuming pathway on long-term trend of PAN. This is consistent with the research in Hong Kong that the destruction rate driven by NO was small15. From the above analysis, it can be concluded that the key factors affecting the long-term trend of PAN were those related to photochemistry (O3, NO2, UV, T, and CO), while those related to physics (e.g., SP, TP, and wind) were of minor importance.
Feature importance (a) and the main effect of major factors on PAN (b) obtained by the XGBoost-SHAP method.
To further investigate the effect of the key factors on PAN, the relationship between the level of features and their SHAP values was analyzed (Fig. 3 and Supplementary Fig. 4). The SHAP values of NO2, CO and O3 increased with their concentrations, indicating that the precursors and AOC promoted PAN production. There were interactions between NO2, O3 and PAN, as depicted in Supplementary Fig. 5. Specifically, in low-NO2 regimes (about less than 6 ppb), high interaction SHAP values were accompanied by low O3 concentration, while the reversed situation was observed in high-NO2 regimes. The hidden reason is that PAN and O3 competed for the precursor when NO2 was not sufficient in the atmosphere. This phenomenon was also frequently found in low NO2-regime, such as rural and background sites13,17,39. UV and T were chemical-related meteorological factors, but their impacts on PAN were quite different (Fig. 3b). The SHAP value of UV had a positive relationship with UV, indicating that the radiation level promoted the photochemical production of PAN. In contrast, the SHAP value of T decreased as T increased, which means the dominant effect of T on PAN was to promote the thermal decomposition of PAN.
Furthermore, long-term component of SHAP values was separated to investigate the temporal variation of the effect of key factors on PAN (Fig. 4). The SHAP value of NO2 and CO had significant downward trends (Theil–Sen test, p < 0.001), which implies a strong decline effect of precursors on PAN from 2018 to 2022. The positive effect of NO2 to PAN declined probably relating to the sustained decrease in NO2 concentration due to pollution control policies in China. Differently, the effect of CO on PAN decreased clearly in early 2021 when coincided with the notable reduction in anthropogenic emissions due to the pandemic40. As illustrated in Supplementary Fig. 6, the concentration of total VOCs likewise shown a significant decrease in 2021. Regarding the photochemical factors, the SHAP values of T and UV showed a downward trend (p < 0.001), indicating that the negative effect of T on PAN enhanced and the positive effect of UV on PAN diminished. Specifically, the T had an upward trend from 2018 to 2022 (Supplementary Fig. 7), which is conducive to PAN thermal decomposition. Meanwhile, the decrease in inter-annual UV variability likely slowed down the photochemical formation rate of PAN. That is, UV and T contributed to the decline of PAN during the study period from the perspective of influencing source and sink of PAN, respectively. It is noteworthy that the SHAP value of O3 had no significant change trend, and was relatively high in 2020 and 2022. The effect of AOC (represented by O3) on PAN is supposed to involve complex chemical processes, which will be investigated using a chemical box model in the following section.
Inter-annual trend of PAN (a) and major factor SHAP values (b–f) in UV > 0 condition. Note that the red lines indicate the trend tested by Theil–Sen method, the red dash lines show the 95% confidence intervals, and the black text is the percentage rate of change with lower value and upper value (percentage/year). The * and *** show the trend significance level p < 0.05 and p < 0.001, respectively.
Chemical mechanism of PAN
Since PAN was less susceptible to transport in summer34, we used the observation data during this season to study the chemical mechanism of PAN and subsequently its impact on O3 formation. The sensitivity of PAN production towards its major precursors NO2, CH3CHO, CH3C(O)CH3, MACR, MVK, and MGLY during each summer from 2018 to 2022 was examined and Fig. 5a shows the relative increment reactivity (RIR) values for these precursors. NO2 showed the largest RIR values among the major precursors in each summer, with a daytime average value of 1.53 ± 0.53. The strong positive RIR of NO2 proved that reduced NO2 could cause PAN to decrease, which is consistent with the ML conclusion. Similar to NO2, the precursors OVOCs also had positive RIR values and CH3CHO shown the largest RIR value (1.33 ± 0.56) among the simulated OVOCs followed by CH3C(O)CH3 and MGLY, which is in line with previous studies conducted in Nanjing and Beijing during summer16,41. Positive RIR values for both NO2 and VOCs indicate PAN production in urban Fuzhou was in transition regime and limited by both VOCs and NOx. Moreover, the general increase in RIR values suggested an increased sensitivity of PAN to precursors over time.
a Average RIRs values for major precursors of PAN and b average PnetO3 with and without PAN mechanisms in each summer from 2018 to 2022. The simulation period was daytime from 6:00 to 18:00. The error bars represent a standard deviation. The statistic of significance employed here is the pair-sample t-test. * and *** indicate the significance levels p < 0.05 and p < 0.001, respectively.
The modeled formation and loss pathways of PAN and PA radicals among the years are further depicted in Fig. 6. It is evident that the production and depletion rates of PAN were highly variable from year to year, which was inconsistent with the decreasing trend of PAN from 2018 to 2022. In addition, we found that the active PA radicals budget occurred in 2020 and 2022. This was well consistent with the ML result that the effect of O3-represented AOC on PAN largely elevated in these 2 years. The primary loss pathway for PAN was thermal decomposition, with average rates ranging from 1.34 ± 0.79 ppb/h to 5.05 ± 3.47 ppb/h. The PAN loss rate in urban Fuzhou was higher than that in the suburban environment in Qingdao (2.36 ppb/h)13 and was comparable to that in the urban environment of the PRD region (2.79 ppb/h)14, implying a strong potential for contribution to PA radicals. As shown in Fig. 6b, the specific budget of PA radicals reveals that PAN thermal decomposition accounted for 37.2–74.7% and contributed the most to PA formation in each year. Previous studies have demonstrated that the huge destruction of PA (the average rate here was 5.42 ppb/h) could effectively affect OH, HO2 and RO2 radicals13,42. It can be hypothesized that this is likely to in turn affect the production of O3.
The simulated production and destruction pathways of PAN (a) and the major production and loss pathways of PA radicals (b) in each summer from 2018 to 2022.
Impacts of PAN on O3 formation over 2018–2022
The role of PAN chemistry in O3 formation was also investigated by chemical box model by comparing the scenarios in the presence and absence of PAN mechanisms. In the presence of PAN mechanism, PAN can promote or inhibit the growth of ROx (including OH, HO2 and RO2) radicals by affecting PA radicals, which further influences the production of O3. Supplementary Fig. 8 illustrates the simulated ROx radicals with and without PAN mechanism. During the study period, the average enhancement of OH, HO2 and RO2 concentrations by PAN-accelerated ROx radicals cycling was 18%, 20% and 25%, respectively. In the absence of PAN mechanisms, as illustrated in Supplementary Fig. 9, the oxidation rates of NO by RO2 and HO2 decreased from 11.3 ± 4.17 ppb/h to 8.96 ± 3.64 ppb/h and from 17.9 ± 5.67 ppb/h to 14.1 ± 4.88 ppb/h, respectively, though the average rate of NO2 removal by OH also decreased from 1.71 ± 0.54 ppb/h to 1.24 ± 0.24 ppb/h. These findings indicate that PAN chemistry enhanced the generation pathways of O3 more than loss pathway of O3. Ultimately, the net O3 production rate (PnetO3) was significantly elevated when PAN mechanisms was considered (Fig. 5b). The PAN chemistry promoted PnetO3 in an average increase of 25.8%, varying from 2.18% to 58.4% among the years. The growth PnetO3 promoted by PAN chemistry was also comparable with case studies in other regions of China such as Qingdao (18%), Xiamen (near 10%), and Nanjing (14.16%)12,13,41. In conclusion, although PAN concentrations decreased from 2018 to 2022, above results demonstrated that PAN chemistry promoted O3 production to varying degrees in different years by accelerating the ROx cycling in urban Fuzhou.
This study displayed the long-term variation of PAN in urban environment based on 5-year observations from 2018 to 2022. Monthly average concentrations of PAN were from 0.20 ppb to 1.50 ppb. Unlike the upward long-trend of O3, PAN concentration exhibited a significant downward trend at an average rate of −0.07 ppb/year. The trend was related to decreasing NO2 and CO, weakening UV, as well as the elevated ambient temperatures, which explained 31.5%, 11.2%, 10.8% and 10.9% of the long-term trend of PAN, respectively. On the other hand, the effect of O3-represented atmospheric oxidation capacity on PAN did not have a significant trend. The chemical box model further confirmed that the PA generation and consumption rates fluctuated from year to year, with 2020 and 2022 being more active. Moreover, although PAN concentrations shown a decreasing trend over time, PAN chemistry increased PnetO3 by 2.18–58.4% in different years by supplying ROx radicals. Overall, the decreasing trend of PAN is likely to occur in Chinese urban areas in the context of reducing primary pollutants, e.g., NO2 emissions and global warming40,43,44. Nonetheless, PAN can still play an important role in O3 photochemical pollution by influencing free radicals and subsequently O3 production, which should be considered in future urban ozone pollution prevention.
Methods
Sampling site and data description
Observations were conducted at the Environmental Monitoring Center Station of Fujian Province (26°6′29″N, 119°18′6″E) in Fuzhou city of Southeast China (see in Supplementary Fig. 10). This sampling site, situated on the eighth-floor rooftop, is surrounded by roads with heavy traffic, residential apartments, and office buildings, which is a representative of urban environments.
The sampling period of ambient PAN was from January 2018 to December 2022. PAN was measured by a commercial analyzer (Meteorologie Consult GmbH, METCON, Germany) with 10-min resolution. This instrument has three parts including automatic sampling unit, calibration unit and detection unit which is composed of gas chromatograph and an electron capture detector. Monthly multi-point curve calibration and weekly single-point calibration were conducted during the observation, and the specific calibration process was described as previous literatures45,46. In addition, PA radicals were estimated by PAN, NO2, and reaction coefficients and the lifetime (τ) of PAN was calculated by NOx and reaction coefficients. The calculation processes are shown in Supplementary Note 1.
Synchronous measurements of O3, NOx, PM2.5, SO2, CO, T, and relative humidity (RH) were conducted. O3, NOx, PM2.5, SO2, CO, and photolysis rate of NO2 (JNO2) were monitored by Thermo Instrument 49i, 17i, 1405DF, 43i, 48i (Thermo Fisher Scientific, Waltham, MA, USA), and photolysis spectrometer (METCON, GmbH, Germany), respectively. T and RH were measured by an integrated sensor (WXT520, Vaisala, Finland). All these observations were averaged into hourly data. The hourly meteorological parameters including UV, 10m v-component of wind (V10), 10m u-component of wind (U10), surface pressure (SP), total precipitation (TP) and BLH were downloaded from ERA547. In addition, around 106 species of VOCs were observed in every July from 2018 to 2022 with hourly interval by a gas chromatography-mass spectrometer (GC-FID/MS, TH-300B, Wuhan, CN).
Kolmogorov–Zurbenko adaptive filter
Concentrations of PAN and O3 are susceptible to complex meteorological conditions, short-term emissions, or disturbances, which makes it difficult to explore their inter-annual trends solely from time series analysis. In this study, a modified KZA filter method was used to filter short-term disturbance and seasonal changes of observed data to better unveil the long-term trends of PAN and O348,49. And the SHAP values of each factor’s original sequence are also filtered through KZA method to obtain their long-term trends. Long-term trends of KZA filter results were tested by the Theil–Sen estimator. The specific method description about KZA filter and Theil–Sen estimator are shown in Supplementary Notes 2 and 8, respectively.
Machine learning model
This study used ML model to establish the relationship between daytime (UV > 0) PAN and explanatory variables, so as to investigate the influencing factors of PAN. Based on physicochemical meanings to PAN, eleven parameters including O3, NO2, NO, CO, RH, T, UV, U10, V10, SP, and TP were selected as explanatory variables. VOCs were not considered in ML model because the VOCs data during the study period were incomplete. Detailed meanings of all selected variables are discussed in Supplementary Note 3.
As common ML models, both Random Forest (RF) model25,50,51, and Extreme gradient boosting (XGBoost) model52,53,54 were used and compared to determine the better model for analysis. The detailed introduction and setting of the two models are provided in Supplementary Note 4. Models were built using 80% of the data as the training set and 20% of the data as the test set. Hyperparameters were tuned using grid search and 5-fold cross-validation (CV) methods. The established model was examined by R2 value, Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The detailed setting of RF model and XGBoost model are described in Supplementary Note 5. In general, both models shown good fitting performance (Supplementary Fig. 11), but the XGBoost model fitted better, with R2 of 0.79, MAE of 0.16, and RMSE of 0.24. Therefore, the XGBoost model was used as a basic prediction part for further discussion.
SHAP interpreter was used to interpret the XGBoost model. According to coalitional game theory55,56, SHAP values were computed according to Eq. (1) and describe the contribution of each feature to the prediction of an individual sample relative to the base prediction value for all samples52,57,58,59,60:
where \({g(x}_{i})\) is a predicted value for each sample \({x}_{i}\) containing M features, \(\phi {(x)}_{{{\rm{base}}}}\), the base value, represents the predicted value for all samples. \({\phi (x}_{i,j})\) is the SHAP value of the feature \(j\) in sample \({x}_{i}\) showing its impact on the prediction. In this study, the importance of input variables to variation of PAN was quantified by absolute SHAP values. SHAP > 0 means that the feature has a positive impact to PAN concentration relative to the base value. While conversely, it exerts a negative contribution. Additionally, the SHAP method is capable of assessing the interactions among the features. The SHAP interaction value of a and b can be interpreted as the difference between the SHAP values of variable a when variable b is present and absent. Details of the SHAP theory and SHAP values computation are given in Supplementary Note 6.
Chemical box model
A chemical box model was developed by the Framework for 0-D Atmospheric Modeling (F0AM) to investigate the formation of PAN and its impact on O3 in each summer of 2018–2022. The box model was performed primarily based on Master Chemical Mechanism 3.3.1 (MCM v3.3.1 https://mcm.york.ac.uk/MCM/)61. Measured hourly interval data of trace gases (VOCs, PAN, O3, NO, NO2, CO, SO2), meteorological variables (T, RH, P, and photolysis frequencies) and reanalysis data (BLH) were used to constrain the box model. Specific setting of the model is introduced in Supplementary Note 7. The sensitivity of PAN production to its major precursors like NO2, CH3CHO, CH3C(O)CH3, MACR, MVK, and MGLY was diagnosed by the RIR analysis. Note that the four second-generation precursors of PAN were not measured and their concentrations were simulated by constraining the VOCs in the model. Like other studies7,41, RIR is defined as the ratios of the variation in PAN production rates to the variation in its precursors (the variation in this work was set as 20% decrease) and calculated as Eq. (2). Based on the constraints with PAN observation in the model, PAN mechanism and PA budget were investigated. PAN-related reactions herein included the PAN formation via PA + NO2 and the PAN loss via PAN + OH and thermal decomposition. And major reactions relating to PA formation and loss in MCM mechanism are summarized in Supplementary Table 5. As for model performance, the simulated and calculated PA radicals were in good agreement with an R2 of 0.96 (Supplementary Fig. 12). The impact of PAN chemistry to O3 formation was estimated by comparing PnetO3 with or without PAN-related mechanisms in box model simulation. And the PnetO3 was calculated by Eq. (3), including NO oxidized by HO2 and RO2, O3 loss by reacting with OH, HO2, unsaturated VOCs and self-photolysis, NO2 loss by OH and NO3 loss by VOCs:
where x is a specific precursor of PAN as mentioned above.
where \({k}_{i}\) is the corresponding chemical reaction constants, \({j}_{{\rm{O3}}}\) is the photolysis rate of O3.
Data availability
Observed data are available on request from J.C. ([email protected]).
Code availability
Data analysis methods are available on request from J.C. ([email protected]).
References
Penkett, S. A. & Brice, K. A. The spring maximum in photooxidants in the northern-hemisphere troposphere. Nature 319, 655–657 (1986).
Fischer, E. V. et al. Atmospheric peroxyacetyl nitrate (PAN): a global budget and source attribution. Atmos. Chem. Phys. 14, 2679–2698 (2014).
LaFranchi, B. W. et al. Closing the peroxy acetyl nitrate budget: observations of acyl peroxy nitrates (PAN, PPN, and MPAN) during BEARPEX 2007. Atmos. Chem. Phys. 9, 7623–7641 (2009).
Mellouki, A., Wallington, T. J. & Chen, J. Atmospheric chemistry of oxygenated volatile organic compounds: impacts on air quality and climate. Chem. Rev. 115, 3984–4014 (2015).
von Ahsen, S., Willner, H. & Francisco, J. S. Thermal decomposition of peroxy acetyl nitrate CH3C(O)OONO2. J. Chem. Phys. 121, 2048–2057 (2004).
Wolfe, G. M. et al. Missing peroxy radical sources within a summertime ponderosa pine forest. Atmos. Chem. Phys. 14, 4715–4732 (2014).
Xu, W. Y. et al. Aerosol promotes peroxyacetyl nitrate formation during winter in the North China Plain. Environ. Sci. Technol. 55, 3568–3581 (2021).
Hu, B. Y. et al. Characteristics of peroxyacetyl nitrate (PAN) in a coastal city of southeastern China: photochemical mechanism and pollution process. Sci. Total Environ. 719, 137493 (2020).
Zhang, G. et al. Summertime distributions of peroxyacetyl nitrate (PAN) and peroxypropionyl nitrate (PPN) in Beijing: understanding the sources and major sink of PAN. Atmos. Environ. 103, 289–296 (2015).
Sun, M. et al. Seasonal discrepancies in peroxyacetyl nitrate (PAN) and its correlation with ozone and PM2.5: effects of regional transport from circumjacent industrial cities. Sci. Total Environ. 785, 147303 (2021).
Zhang, G. et al. Simultaneous observation of atmospheric peroxyacetyl nitrate and ozone in the megacity of Shanghai, China: regional transport and thermal decomposition. Environ. Pollut. 274, 116570 (2021).
Liu, T. T. et al. Seasonal characteristics of atmospheric peroxyacetyl nitrate (PAN) in a coastal city of Southeast China: explanatory factors and photochemical effects. Atmos. Chem. Phys. 22, 4339–4353 (2022).
Liu, Y. H. et al. Formation of peroxyacetyl nitrate (PAN) and its impact on ozone production in the coastal atmosphere of Qingdao, North China. Sci. Total Environ. 778, 146265 (2021).
Liu, T. et al. Complexities of peroxyacetyl nitrate photochemistry and its control strategies in contrasting environments in the Pearl River Delta region. Npj Clim. Atmos. Sci. 7, 116 (2024).
Zeng, L. W. et al. Atmospheric fate of peroxyacetyl nitrate in suburban Hong Kong and its impact on local ozone pollution. Environ. Pollut. 252, 1910–1919 (2019).
Xue, L. K. et al. On the use of an explicit chemical mechanism to dissect peroxy acetyl nitrate formation. Environ. Pollut. 195, 39–47 (2014).
Wang, Y. et al. Anthropogenic pollutants induce changes in peroxyacetyl nitrate formation intensity and pathways in a mountainous background atmosphere in Southern China. Environ. Sci. Technol. 57, 6253–6262 (2023).
Sun, M., Cui, J. N., Zhao, X. M. & Zhang, J. B. Impacts of precursors on peroxyacetyl nitrate (PAN) and relative formation of PAN to ozone in a southwestern megacity of China. Atmos. Environ. 231, 117542 (2020).
Wang, Y. T. et al. Sustained emission reductions have restrained the ozone pollution over China. Nat. Geosci. 16, 967 (2023).
Wang, W. J. et al. Ozone pollution mitigation strategy informed by long-term trends of atmospheric oxidation capacity (Nov, 10.1038/s41561-023-01334-9, 2023). Nat. Geosci. 17, 172–172 (2024).
Maji, K. J. & Namdeo, A. Continuous increases of surface ozone and associated premature mortality growth in China during 2015–2019. Environ. Pollut. 269, 116183 (2021).
Wu, K. et al. Drivers of 2013–2020 ozone trends in the Sichuan Basin, China: impacts of meteorology and precursor emission changes. Environ. Pollut. 300, 118914 (2022).
Ji, X. T. et al. Impacts of synoptic patterns and meteorological factors on distribution trends of ozone in Southeast China during. J. Geophys. Res. Atmos. 128, e2022JD037961 (2023).
Lu, X. et al. Rapid increases in warm-season surface ozone and resulting health impact in China since 2013. Environ. Sci. Technol. Lett. 7, 240–247 (2020).
Weng, X., Forster, G. L. & Nowack, P. A machine learning approach to quantify meteorological drivers of ozone pollution in China from 2015 to 2019. Atmos. Chem. Phys. 22, 8385–8402 (2022).
Xia, S. Y., Huang, X. F., Han, H. X., Li, X. & Yu, G. H. Influence of thermal decomposition and regional transport on atmospheric peroxyacetyl nitrate (PAN) observed in a megacity in southern China. Atmos. Res. 272, 106146 (2022).
McFadyen, G. G. & Cape, J. N. Peroxyacetyl nitrate in eastern Scotland. Sci. Total Environ. 337, 213–222 (2005).
Qiu, Y. L. et al. A study of peroxyacetyl nitrate at a rural site in Beijing based on continuous observations from 2015 to 2019 and the WRF-Chem model. Front. Environ. Sci. Eng. 14, 71 (2020).
Zheng, B. et al. Trends in China’s anthropogenic emissions since 2010 as the consequence of clean air actions. Atmos. Chem. Phys. 18, 14095–14111 (2018).
Tong, D. et al. Dynamic projection of anthropogenic emissions in China: methodology and 2015-2050 emission pathways under a range of socio-economic, climate policy, and pollution control scenarios. Atmos. Chem. Phys. 20, 5729–5757 (2020).
Guo, Y. et al. Evaluating the real changes of air quality due to clean air actions using a machine learning technique: results from 12 Chinese mega-cities during 2013-2020. Chemosphere 300, 134608 (2022).
Fu, W. C. et al. The characteristics of ambient air quality in urban forest areas and other urban areas of Fuzhou city, China. Environ. Dev. Sustain. 24, 9500–9518 (2022).
Gong, D. C. et al. Characteristics of peroxyacetyl nitrate (PAN) in the high-elevation background atmosphere of South-Central China: implications for regional photochemical pollution. Atmos. Environ. 254, 118424 (2021).
Wei, W., Zang, J. X., Wang, X. Q. & Cheng, S. Y. Peroxyacetyl nitrate (PAN) in the border of Beijing, Tianjin and Hebei of China: concentration, source apportionment and photochemical pollution assessment. Atmos. Res. 246, 105106 (2020).
Sen, K. P. Estimates of the regression coefficient based on Kendall’s tau. Publ. Am. Stat. Assoc. 63, 1379–1389 (1968).
Li, K. et al. Anthropogenic drivers of 2013-2017 trends in summer surface ozone in China. Proc. Natl Acad. Sci. USA 116, 422–427 (2019).
Li, K. et al. Increases in surface ozone pollution in China from 2013 to 2019: anthropogenic and meteorological influences. Atmos. Chem. Phys. 20, 11423–11433 (2020).
Zhang, X. et al. Spatiotemporal patterns and ozone sensitivity of gaseous carbonyls at eleven urban sites in southeastern China. Sci. Total Environ. 824, 153719 (2022).
Singh, H. B. et al. Relationship between peroxyacetyl nitrate and nitrogen-oxides in the clean troposphere. Nature 318, 347–349 (1985).
Wang, Q. & Su, M. A preliminary assessment of the impact of COVID-19 on environment? A case study of China. Sci. Total Environ. 728, 138915 (2020).
Xu, T. et al. Investigation on the budget of peroxyacetyl nitrate (PAN) in the Yangtze River Delta: unravelling local photochemistry and regional impact. Sci. Total Environ. 917, 170373 (2024).
Liu, T. T. et al. Atmospheric oxidation capacity and ozone pollution mechanism in a coastal city of southeastern China: analysis of a typical photochemical episode by an observation-based model. Atmos. Chem. Phys. 22, 2173–2190 (2022).
Zhao, Y. et al. Decline in bulk deposition of air pollutants in China lags behind reductions in emissions. Nat. Geosci. 15, 190–+ (2022).
Zheng, H. et al. Achievements and challenges in improving air quality in China: analysis of the long-term trends from 2014 to 2022. Environ. Int. 183, 108361 (2024).
Zhang, H. L., Xu, X. B., Lin, W. L. & Wang, Y. Wintertime peroxyacetyl nitrate (PAN) in the megacity Beijing: role of photochemical and meteorological processes. J. Environ. Sci. 26, 83–96 (2014).
Volz-Thomas, A., Xueref, I. & Schmitt, R. An automatic gas chromatograph and calibration system for ambient measurements of PAN and PPN. Environ. Sci. Pollut. Res. 9, 72–76 (2002).
Hersbach, H. et al. The ERA5 global reanalysis. Q. J. R. Meteorol. Soc. 146, 1999–2049 (2020).
Rao, S. T. & Zurbenko, I. G. Detecting and tracking changes in ozone air-quality. J. Air Waste Manag. Assoc. 44, 1089–1092 (1994).
Ma, Z. Q. et al. Significant increase of surface ozone at a rural site, north of eastern China. Atmos. Chem. Phys. 16, 3969–3977 (2016).
Grange, S. K., Carslaw, D. C., Lewis, A. C., Boleti, E. & Hueglin, C. Random forest meteorological normalisation models for Swiss PM10 trend analysis. Atmos. Chem. Phys. 18, 6223–6239 (2018).
Yang, C. et al. Machine learning reveals the parameters affecting the gaseous sulfuric acid distribution in a coastal city: model construction and interpretation. Environ. Sci. Technol. Lett. 10, 1045–1051 (2023).
Peng, X. et al. Critical role of secondary organic aerosol in urban atmospheric visibility improvement identified by machine learning. Environ. Sci. Technol. Lett. 10, 976–982 (2023).
Gui, K. et al. Construction of a virtual PM2.5 observation network in China based on high-density surface meteorological observations using the Extreme Gradient Boosting model. Environ. Int. 141, 105801 (2020).
Kononikhin, A. S. et al. Prognosis of Alzheimer’s disease using quantitative mass spectrometry of human blood plasma proteins and machine learning. Int. J. Mol. Sci. 23, 7907 (2022).
Lundberg, S. M. & Lee, S. I. A unified approach to interpreting model predictions. In 31st Annual Conference on Neural Information Processing Systems (NIPS 2017).
Shapley, L. S. 17. A value for n-person games. In Contributions to the Theory of Games, Princeton University Press, 2, 307−318 (1953).
Yang, C. et al. New insights on the formation of nucleation mode particles in a coastal city based on a machine learning approach. Environ. Sci. Technol. 58, 1187–1198 (2023).
Yang, J. et al. Identifying driving factors of soil heavy metal at the mining area scale: methods and practice. Chemosphere 350, 140936 (2024).
Hou, L. L. et al. Revealing drivers of haze pollution by explainable machine learning. Environ. Sci. Technol. Lett. 9, 112–119 (2022).
Palansooriya, K. N. et al. Prediction of soil heavy metal immobilization by biochar using machine learning. Environ. Sci. Technol. 56, 4187–4198 (2022).
Wolfe, G. M., Marvin, M. R., Roberts, S. J., Travis, K. R. & Liao, J. The framework for 0-D atmospheric modeling (F0AM) v3.1. Geosci. Model Dev. 9, 3309–3319 (2016).
Acknowledgements
This work was funded by the National Natural Science Foundation of China (U22A20578, 42305102 and 42277091), the Science and Technology Department of Fujian Province (2022L3025), the National Key Research and Development Program (2022YFC3700304), STS Plan Supporting Project of the Chinese Academy of Sciences in Fujian Province (2023T3013), Fujian Provincial Environmental Protection Science & Technology Plan Projects (2023R004), and Xiamen Atmospheric Environment Observation and Research Station of Fujian Province.
Author information
Authors and Affiliations
Contributions
Z.L. contributed to the methodology, data curation, software, analysis and writing of the original draft. L.X. and J.C. contributed to the conceptualization, investigation, data curation, reviewing and editing the text, supervision, and funding acquisition. C.Y., G.C., X.J., L.L., K.Z., Y.H., M.L., X.F., B.H., and F.Z. provided useful advice and revised the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Lin, Z., Xu, L., Yang, C. et al. Trends of peroxyacetyl nitrate and its impact on ozone over 2018–2022 in urban atmosphere. npj Clim Atmos Sci 7, 192 (2024). https://doi.org/10.1038/s41612-024-00746-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41612-024-00746-7
