Similar to the Eq. (3), the δ15N_4a-3s could be calculated as follows: 4δ15N4a-3s=δ15N-NH4+-δ15N-NH3s=1000α-11-f, where δ15N-NH_3s is the nitrogen isotope of NH₃ from various types of sources and other items are the same as that in Eqs. (1) and (2). The equation indicates that the variation in δ15N_4a-3s depends on the change of α, f and δ15N-NH_x. Consequently, the core objective of this model is to iteratively update α, f and δ15N-NH₃ in an open system that accounts for the combined effects of NH3/NH4+ conversion and sedimentation, and ultimately to determine the steady-state δ15N_4a-3s.

Firstly, the parameter α is defined as (δ15N-NH4++1000)/(δ15N-NH3+1000) (Coplen, 2011). This value of α could be empirically determined by fitting experimental data (Urey, 1947; Li et al., 2012) and computational quantum chemistry method (Walters et al., 2019) based on the ambient temperature. The present study used the computational quantum chemistry method (Walters et al., 2019) to calculate the α values as the following method: 51000α-1=12522TK-11.31, where T is the ambient temperature (in Kelvin). The values of f and δ15N-NH_x vary synchronously during the iteration. Next, the continuous variations in atmospheric NH₃ and NH4+, together with their deposition processes, were discretized. Assuming the atmosphere is a well-mixed open system, three processes occur simultaneously at each iteration step: partial conversion of NH₃ to NH4+, deposition loss of NH₃, and deposition loss of NH4+. Accordingly, the mass fractions of atmospheric NH₃ and NH4+, as well as their deposited fractions, can be expressed as follows: 6NH3at=1-G4-D3NH3at-1NH4a+t=G4NH3at-1+1-D4NH4a+t-1NH3dt=D3NH3at-1NH4d+t=D4NH4a+t-1, where G4, D3 and D4 represent the transformation ratio of NH₃ to NH4+, and the deposition ratio of NH₃ and the deposition ratio of NH4+, respectively. All three are dimensionless parameters defined for a single iteration step. The superscript t denotes the iteration index, starting from t=1, and does not represent a specific physical duration. The model is iterated to steady state, and the resulting steady-state δ15N_4a-3s is taken as the model output. [NH_3a] and [NH4a+] are the mass fractions of atmospheric NH₃ and NH4+, whereas [NH_3d] and [NH4d+] are the corresponding deposited fractions in that iteration step. The molar fraction of NH4+ in NH_x at iteration step t can then be calculated as follows: 7ft=NH4a+tNHxt=NH4a+tNH4a+t+NH3at, where [NH_x]^t is the mass fraction of NH_x in the atmosphere at the tth time interval, and the meanings of other items are the same as those in Eq. (6). Furthermore, given that α and f are known, the values of [δ15N-NH4+] and [δ15N-NH₃] at the tth time interval can be expressed as: 8δ15N-NH4a+t=δ15N-NHxt-1+1000α-11-ft-1δ15N-NH3at=δ15N-NHxt-1-1000α-1ft-1, It can be assumed that δ15N values of NH₃ and NH4+ ([δ15N-NH4d+] and [δ15N-NH_3d]) deposited to the surface are equal to δ15N-NH₃ and δ15N-NH4+ in the atmosphere at the tth time interval, respectively, as the following: δ15N-NH4d+t=δ15N-NH4a+t9δ15N-NH3dt=δ15N-NH3at This assumption implies that the deposition process itself does not introduce additional isotopic fractionation; its effect is limited to altering the mass composition of the remaining NH_x in the atmosphere, without further altering the isotopic values of the corresponding components. It should be noted that this assumption is not directly substituted into Eq. (4) to calculate δ15N_4a-3s, but rather indirectly influences δ15N-NH_x and f in Eq. (4) by altering the mass fraction and isotopic composition of the remaining atmospheric NH_x, thereby ultimately affecting δ15N_4a-3s.

Subsequently, using Eq. (10), the overall isotopic composition of atmospheric NH_x at the tth iteration step, [δ15N-NH_x]^t, can be further derived. This quantity serves both as a characterisation of the system's isotopic state at the current step and as an input for the next iteration, continuing to participate in the calculations. Correspondingly, [δ15N-NH_x] in the atmosphere at the tth time interval can be written as: 10δ15N-NHxt=ftδ15N-NH4a+t+1-ftδ15N-NH3at, Thus, Eqs. (6)–(10) together form an iterative framework for an open system that continuously updates the mass fractions of NH₃ and NH4+ and their isotopic compositions. The initial conditions of the model are set as follows: at t=1, [δ15N-NH_x] = [δ15N-NH_3s], i.e. the isotopic composition of atmospheric NH_x at the initial time of the system is equal to that of the source-emitted NH₃. Thereafter, the model is solved iteratively in the following sequence:

Set up source [δ15N-NH_3s];

As described in the previous section, the value of δ15N_4a-3s could be predicted by the developed model if the values of G4, D3, D4, T, and [δ15N-NH_3s] were available. The G4, D3, and D4 had strong spatiotemporal variability, primarily driven by the difference in the content and composition of acidic gases in the atmosphere, as well as the variation of surface roughness, wind speed, and etc (Baek and Aneja, 2004; Schrader and Brummer, 2014). This is a systematic project that cannot be validated experimentally by an article. So, here we constructed six simulation scenarios to assess the extent to which changes in the three parameters affected the simulation results of δ15N_4a-3s as listed in Table 1.

These six simulation scenarios were composed of combining G4 being equal to 0.5 times, 1.0 times and 2.0 times D3, as well as D4 being 0.2 times and 0.5 times D3 respectively. The detailed basis for this proportion setting could be found in the Sect. S1 and Tables S1–S2 in the Supplement. Briefly, the G4 is mainly attributed to the comprehensive neutralization reaction ratio of ammonia (NH₃) with acid gases (sulfuric acid (H₂SO₄), nitric acid (HNO₃) and hydrochloric acid (HCl)) in the atmosphere (Baek and Aneja, 2004). Under normal atmospheric conditions, the relatively abundant acidic gas in the atmosphere could make the G4 faster, and the G4 between NH₃ and H₂SO₄ was faster than that of NH₃ with HNO₃ and HCl, as listed in Sect. S1 of the Supplement (Schrader and Brummer, 2014). To evaluate the impact of varying acid-gas concentrations, we established three G4 levels: 0.5 × D3 (low acid-gas content), 1.0 × D3 (moderate acid-gas content), and 2.0 × D3 (high acid-gas content). Surface roughness is an important parameter affecting the ratio of atmospheric deposition of NH₃ and NH4+, with a general pattern of decreasing deposition ratio from mountainous regions to flat terrain areas (Zhang et al., 2001). Numerous studies had also shown that the D4 was significantly lower than D3 in the atmosphere (Shen et al., 2009). Therefore, we set two levels for D4, one was 0.2 × D3 and the other was 0.5 × D3, representing the deposition differences in flat regions and mountainous areas, respectively.

A total of five temperature levels of -10, 0, 10, 20, 30 °C were set for the above six simulation scenarios. In addition, the [δ15N-NH_3s] was set to 0 ‰ in these simulations.

Sensitivity analysis is a statistical technique used to quantify the extent to which variations in different inputs influence the variability of the outputs. In present study, sensitivity analysis was conducted to examine the impact of individual parameters on the δ15N_4a-3s values. This analysis was achieved by assessing how sensitive the model was to alterations in its input parameters. To do this, the model was run with each parameter individually scaled to 0.9 and 1.1 times its original value (Cao et al., 2007). The evaluation involved computing sensitivity coefficients (SC), which quantify the relative changes in the primary output estimates in response to changes in the input parameters, as outlined below: 11SC=absOUT1.1-OUT0.90.2×OUT1.0, where OUT_1.1, OUT_1.0, and OUT_0.9 are the model output results when the input parameter is 1.1, 1.0, and 0.9 times of its original value, respectively. The sensitivity of various parameters can be directly compared because the SC values obtained through Eq. (11) are dimensionless. The absolute magnitude of SC represents the extent to which input parameters affect the output results. Furthermore, the positive or negative sign of SC reveals the direction of this influence: positive values signify that an increase in input parameters leads to an increase in output results, whereas negative values imply the opposite relationship (Cao et al., 2007; Dong et al., 2010).

Figure 1 shows the variation of δ15N_4a-3s with respect to the f value at temperatures of -10, 0, 10, 20, and 30 °C, respectively, by the six model scenarios. In order to aid in comparisons, Fig. 1 also displays that the change in δ15N_4a-3x obtained using Eq. (4), which can be thought of as the δ15N_4a-3s value calculated using the simplified method. Under different temperature conditions, the six simulation scenarios all obtained similar change characteristics, that is, the δ15N_4a-3s value decreased with the increase of f value.

The following took the results at 20 °C as an example to further illustrate. Specifically, when f was 0.1, the δ15N_4a-3s values across the six simulated scenarios average around 28.8 ‰. Conversely, when f reached 0.9, the δ15N_4a-3s values decreased to a range of 5.89 ‰–14.0 ‰. Among the six model scenarios, the variation in δ15N_4a-3s values was narrower in regions with lower generation ratios (G4) and deposition ratios (D4) of NH4+ (such as MS1 and MS2), whereas it was broader values in regions with higher ratios (such as MS5 and MS6). The finding suggested that there was a minor fluctuation in δ15N_4a-3s values in areas with less acidic gas and flat landscapes, like flat land surfaces and open oceans, whereas a more significant variation was observed in δ15N_4a-3s values in regions characterized by higher acidic gas levels and rugged terrains, such as mountainous cities with high air pollution load.

The change in δ15N_4a-3s values calculated by the simplified method was larger than those δ15N_4a-3s values yielded by all six model scenarios, and their difference was a gradual increase as f values increase. The difference indicated that when the simple method was used for source apportionment of NH4+ in the atmosphere, the deviation would be larger with the increase of f value. This phenomenon originated from NH₃ exhibiting a higher deposition rate compared to NH4+ (Behera and Sharma, 2011, 2012), which results in a greater proportion of NH_x components with lower δ15N values being removed from the atmosphere. Consequently, the δ15N-NH_x in the atmosphere gradually increased. When using the δ15N_4a-3s values calculated by the simplified method to correct the δ15N of NH4+ for the purpose of apportioning sources of NH4+, overcorrection may occur, leading to an overestimation of the contribution proportion of NH₃ emission sources with relatively negative δ15N values, like non-agricultural sources (e.g., vehicle exhaust and NH₃ slip) (Zong et al., 2023; Feng et al., 2023). The larger δ15N_4a-3s value was in the case of lower ambient temperature, indicating that the difference between δ15N_4a-3s calculated by the model scenarios and the simplified method had more obvious influence on the source apportionment of NH4+ in the northern cold region and the winter period when the temperature was lower (Sun et al., 2021).

The sensitivity coefficients of each input parameters (D3, D4, G4, and T) to the δ15N_4a-3s values at 20 °C were calculated by using Eq. (11) across the six model scenarios (see Fig. 2). The sensitivity coefficient of D3 variation on the simulation results ranged between 2.55 × 10⁻⁵ and 9.32 × 10⁻². The sensitivity coefficient of D4 and G4 variations on the δ15N_4a-3s ranged respectively from 1.77 × 10⁻⁵ to 0.474 and from 3.62 × 10⁻⁴ to 0.494, showing a similar variation signature, which intensified as the value of f increases. Additionally, the range of variation in D4's sensitivity coefficient also widened as the value of f increased. However, G4 did not exhibit this characteristic. The variation range of the sensitivity coefficient of temperature (T) variation on the simulation results was narrow, falling between 0.092 and 0.094, and this influence diminished as the value of f increased. Generally, the sensitivity coefficient exceeding 0.5 is considered that the corresponding input parameter has a sensitive influence on the output result (Cao et al., 2007). It suggested that the influence of each individual input parameter on the simulation output result (δ15N_4a-3s) did not show obvious sensitivity in the six model scenarios.

In fact, the degree to which input parameters affect the simulation results is not only related to the sensitivity coefficient evaluated based on a single input parameter, but also to the synergistic effect of changes in all input parameters (Wang et al., 2023). In this model simulation, the changes in the δ15N_4a-3s and f were the results of the input parameter changes. Taking f as a comprehensive input parameter of D3, D4, G4, and T, the sensitivity of f to output results based on the six simulation scenarios was calculated by formula (11). The variation range of the calculated corresponding sensitivity coefficient with f is also shown in Fig. 2. The variation characteristic of this sensitivity coefficient is similar to that of the input parameter D4, showing that the variation range becomes larger and larger with the increase of f. When the f value was greater than 0.32, the maximum of the sensitivity coefficients obtained by the six simulation scenarios began to be greater than 0.5, that is, the changes in the input parameters (f) were generally considered to have sensitive effects on the output results (δ15N_4a-3s) (Cao et al., 2007). The maximum range of variation of this sensitivity coefficient was from 0.831 to 3.90 when the f value reached 0.9.

These ranges of sensitivity coefficients were generated by six simulation scenarios. To better understand them, the sensitivity coefficients of f to output results (δ15N_4a-3s) derived from the six simulation scenarios are shown in Fig. S1 of the Supplement. The simulation results of MS1 scenario had the lowest sensitivity to f, while the simulation results of MS6 scenario had the highest sensitivity to f, indicating that large D4 and G4 had a greater impact on the simulation results. The finding also indicated that in some urban areas with more acidic gases, especially in mountainous areas, more attention should be paid to the influence of f changes on the source apportionment of NH4+ in the atmosphere.

As mentioned earlier, δ15N_4a-3s is generally calculated using Eq. (3) when apportioning sources of NH4+ in the atmosphere. The independent variable of this equation included the key influencing parameter f. Based on the form of Eq. (3) and the calculation results of the δ15N_4a-3s values, we constructed a calculation function for the δ15N_4a-3s values through nonlinear regression. The form of this nonlinear function is shown in Eq. (12). The coefficient values of B, C, and D in the equation were iteratively fitted for six simulation scenarios. During the fitting process, each simulation scenario took into account different temperature conditions, including -10, 0, 10, 20, and 30 °C. 12δ15N4a-3s=1000αB-11-fC+D, Table 2 lists the iterative fitting values of the coefficients B, C, and D in Eq. (12) for the six model scenarios (MS1–MS6), as well as the determination coefficient (r2) and sum of squares error (SSE), which were often used to evaluate the fitting effect (Sun et al., 2023; Xu, 2017). Figures S2 and S3 of the Supplement show the comparison plots and scatter plots of the δ15N_4a-3s values calculated by the six fitted equations against the six model scenarios (MS) simulated by the developed model. The coefficients B and C are both greater than 0 and less than 1, and the two coefficients gradually increase from MS1 to MS6, indicating that the Eq. (12) for model scenarios from MS1 to MS6 approaches from nonlinear to linear. The determination coefficients obtained from the regression of the six fitting equations with the simulation results range from 0.967 to 0.998, indicating that these equations can fit the calculated δ15N_4a-3s values quite well (see Fig. S3 of the Supplement). As shown in Fig. S2 of the Supplement, the maximum deviation between the δ15N_4a-3s values obtained from the fitting equation and those calculated by the model occurs in the model scenario of MS1. Under -10 °C condition, the deviation reached its maximum value (1.82 ‰) when the f value was 0.9. This largest deviation was also significantly smaller than the variation range of δ15N values of NH₃, which emitted from various types of sources used in the source apportionment of atmospheric NH4+ (Gu et al., 2022; Zhang et al., 2023; Feng et al., 2022; Li et al., 2023). This indicated that these fitting equations would not significantly increase the uncertainty of the source-resolved assessment of NH4+ in the atmosphere when they were used.

In order to evaluate the fitting equations as mentioned above using monitoring data, we conducted a source apportionment simulation of NH4+ again using the atmospheric δ15N-NH4+ in the Yellow River Delta in the summers of 2013 and 2021. Apart from the difference in the method of calculating δ15N_4a-3s values, the model, input data, etc. were consistent with those reported in our previous study (Zong et al., 2023). Briefly, the Bayesian mixing model (MixSIAR) developed by the R language was used for the source apportionment of NH4+. Four main types of NH₃ emission sources, including fertilizer use (-25.21 ± 9.43 ‰), livestock waste (-16.14 ± 7.98 ‰), vehicle exhaust (+6.62 ± 1.89 ‰), and NH₃ slip (-7.12 ± 7.62 ‰) (Table S3), were considered in the model. In the previous study, we used Eq. (3) (termed as simplified method) to calculate the δ15N_4a-3s values. In this simulation, we used the MS1 fitting equation to calculate the δ15N_4a-3s values. This is because these atmospheric particulate matter samples were collected at the Yellow River Delta Ecological Research Station of Coastal Wetland, Chinese Academy of Sciences (37°45^′ N, 118°59^′ E). The Yellow River Delta is an alluvial plain formed by the Yellow River, featuring a very flat terrain (Li et al., 2022; Zhang et al., 2025). Additionally, there are no obvious emission sources from industrial, transportation, and agricultural activities around the sampling site. Many studies regarded this sampling site as an atmospheric background point in North China (Zong et al., 2015; Sui et al., 2015), which met the characteristics of low pollution and a flat terrain in the MS1 scenario. Moreover, the deviation between the δ15N_4a-3s values calculated by the MS1 method and the simplified method is the largest as shown in Fig. 1, which is more conducive to our evaluation of the degree of difference between the source apportionment results obtained by the two methods.

For ease of comparing the differences, we summarized both the reported previously results and the findings of this study in Fig. 3. For the sources of atmospheric NH4+ in 2013, the most noticeable difference between the two source apportionment methods was the significant increase in the contribution proportion of fertilizer application (from 32.9 % using the simplified method to 68.8 % using the fitting equation method) and the corresponding NH4+ concentration (from 1.33 µg m⁻³ using the simplified method to 2.78 µg m⁻³ using the fitting method). The contributions of the other three emission sources decreased to varying degrees. When we previously used the simplified method for NH4+ source apportionment, we found that literature reports indicated similar annual NH₃ emissions from fertilizer application and livestock farming in North China, and even in Shandong Province (Zhang et al., 2010). Based on this evidence, we determined that agricultural sources were the main contributors to atmospheric NH4+ in the Yellow River Delta region in 2013, accounting for 67.4 % (Zong et al., 2023). Recent NH₃ emission inventory results for Shandong showed that farmland fertilizer application sources were closer to the Yellow River Delta than livestock farming sources (Zhu et al., 2024), and that farmland emissions in summer are significantly higher than those from livestock farming (Li et al., 2021). This was consistent with the higher contribution proportion of fertilizer application to atmospheric NH4+ concentration obtained using the fitting equation method in this study, with the proportion also increasing correspondingly to 82.3 %. Thus, it could be seen that when considering the impact of atmospheric deposition on δ15N-NH_x in the atmosphere, the contribution of agricultural sources would increase to some extent, and correspondingly, the contribution of non-agricultural sources would decrease.

For the sources of NH4+ in the atmosphere in 2021, the most noticeable difference between the two source apportionment methods remained that the contribution proportion of fertilizer application obtained using the fitting equation method was significantly higher (30.2 % by the fitting method vs. 12.5 % by the simplified method) and also significantly higher for the NH4+ concentration in the atmosphere (1.08 µg m⁻³ by the fitting method vs. 0.45 µg m⁻³ by the simplified method). At the same time, the predicted contribution percentage and concentration of NH₃ emissions from livestock farming by the fitting method was much lower than those by the simple method, 17.8 % and 0.64 µg m⁻³ for the former and 29.3 % and 1.05 µg m⁻³ for the latter, respectively. Thus, due to the offsetting increase and decrease in the contributions of these two types of agricultural sources to atmospheric NH4+ concentration, the differences in the contributions of agricultural and non-agricultural sources obtained using the two methods were not significant (agricultural sources: 41.8 % using the simplified method vs. 48.0 % using the fitting method; non-agricultural sources: 58.2 % using the simplified method vs. 52.0 % using the fitting method).

To evaluate the applicability of the proposed method in a typical urban environment, observational data of nitrogen isotopes in particulate ammonium (NH4+) in PM_2.5 collected in Changsha from December 2019 to October 2020 were incorporated in this study (Li et al., 2024). A Bayesian isotope mixing model was applied to quantify the contributions of ammonia sources. Except for the calculation of δ15N_4a-3s, the model structure, isotopic signatures of the four emission source categories (Table S4), and other relevant parameters were consistent with those reported by Li et al. (2024). Aerosol samples were collected on the rooftop of a building at Central South University of Forestry and Technology (28.1° N, 113.0° E) in Changsha. The sampling site is located in a typical urban functional area, surrounded by high-density residential zones and major traffic roads, and is therefore representative of the urban atmospheric environment of Changsha.

Changsha is located in central China and is characterized by relatively flat terrain within the Xiangjiang River alluvial plain (Zhang et al., 2024; He et al., 2024). The region is subject to strong anthropogenic influences, including vehicular emissions, residential activities, and surrounding agricultural sources, leading to relatively high levels of acidic gases in the atmosphere (Zhai et al., 2014; Ma et al., 2019). According to the classification framework proposed in this study, Changsha is best categorized as the MS5 scenario, representing urban areas with relatively flat terrain and high acidic gas levels. Therefore, the fitted equation developed under the MS5 scenario was applied to calculate δ15N_4a-3s in this simulation, in order to assess its applicability in a typical urban environment.

The simulation results indicate (Fig. 4) that, compared with the traditional “simplified method”, the use of the fitting equation developed in this study reveals that non-fossil fuel-related emissions dominate in the Changsha region, with their contribution increasing from 55.5 % (agricultural emissions: 28.3 ± 13.4 %, biomass combustion: 27.2 ± 14.3 %) to 59.2 % (agricultural emissions: 30.8 ± 21.1 %, biomass combustion: 28.4 ± 22.3 %). Meanwhile, the contribution from fossil fuel-related sources decreased significantly, falling from 44.5 % (coal combustion: 25.5 ± 14.1 %, vehicle emissions: 19.0 ± 10.5 %) to 40.8 % (coal combustion: 32.7 ± 24.5 %, vehicle emissions: 8.1 ± 8.1 %). It is worth noting that the optimised source contribution structure is closer to the NH4+ emission inventory for Changsha (Li et al., 2025).

From a seasonal perspective, both methods indicated pronounced seasonal variations in atmospheric NH4+ sources in Changsha, generally characterized by an overwhelming dominance of non-fossil sources in summer and a substantial increase in fossil-fuel-related contributions in autumn and winter. In summer, agricultural emissions remained the largest source throughout. After applying the fitted equation, the contribution of agricultural emissions increased from 29.0 % to 57.0 %, whereas that of biomass burning decreased from 26.8 % to 28.5 %. The contributions from coal combustion and vehicle emissions decreased to only 9.7 % and 4.8 %, respectively, and the total contribution of non-fossil sources further increased from 55.8 % to 85.5 %. The optimized source apportionment was more consistent with emission inventory studies conducted in the Changsha–Zhuzhou–Xiangtan region and across China, which have shown that agricultural NH₃ emissions peak in summer and constitute the dominant component of regional NH₃ emissions during this season (Xu et al., 2020; Zhao et al., 2023).

Beginning in autumn, fossil-fuel-related sources gradually became dominant, and the total contributions estimated by the two methods were relatively close (simplified approach: 57.2 %; fitted-equation method: 53.1 %). This result is also more consistent with local observations in Changsha. Xiao et al. (2020) reported that atmospheric NH4+ in Changsha already exhibited a clear combustion-source dominance in autumn 2017, with fossil-fuel-related sources contributing 51.8 ± 14.9 %.

Winter exhibited the strongest fossil-fuel signature, although the two methods showed marked differences in the inferred structure within fossil-related sources. The fitted equation method identified coal combustion as the overwhelmingly dominant source in winter, with a contribution of 55.9 %, substantially higher than the 24.4 % estimated by the simplified approach. At the same time, the contribution of vehicle emissions was sharply reduced from 34.4 % to 13.5 %. According to the fitted equation method, the contributions of non-fossil and fossil sources in winter were 30.6 % and 69.4 %, respectively, compared with 41.2 % and 58.8 % derived from the simplified approach, further highlighting the dominant role of fossil-fuel emissions in winter. This interpretation is consistent with previous studies. Xu et al. (2020) and Pan et al. (2016) pointed out that winter heating increases coal consumption, leading to substantial increases in NH4+ and NO3- concentrations, and that during severe winter haze episodes, the NH₃ precursor of urban NH4+ is mainly controlled by fossil-fuel emissions and related industrial processes, while the influence of agricultural sources is weakened.

In the MS1 case, the difference in source contributions in 2013 was significantly greater than that in 2021, which was mainly related to the change in the atmospheric f value (Zong et al., 2023). Specifically, the f value decreased from 71.9 % in 2013 to 41.6 % in 2021. As shown in Fig. 1, with decreasing f values, the discrepancy between the simplified approach and the fitted-equation method gradually became smaller, thereby reducing the difference in source apportionment results obtained by the two methods. In contrast, the f value (Li et al., 2024) in MS5 was 74.6 %, which was close to that of MS1 in 2013. Combined with Fig. 1, this suggests that the variation in the difference between the two methods was closely related to G4 and D4: when G4 and D4 were at relatively low levels, the difference between the two methods was comparatively small; however, in regions with higher NH4+ formation ratios (such as MS5 and MS6), the discrepancy between the two methods became more pronounced.

Overall, the use of experimentally validated model parameters (G4, D3, and D4) can further refine the simulation results of source apportionment. Two independent cases, namely the background site MS1 and the urban site MS5, consistently demonstrated that, after accounting for atmospheric deposition, the fitted-equation method can more reasonably constrain the δ15N_4a-3s value, thereby yielding NH4+ source apportionment results that are more consistent with actual regional emission characteristics. This indicates that the established fitted equation has good applicability under different environmental gradients and can significantly improve the reliability of NH4+ isotope-based source apportionment, thus providing a more robust tool for identifying regional ammonia emission sources and formulating emission reduction strategies.