To obtain surface concentrations of HCHO and NO2, we conducted in situ measurements at Tokyo Metropolitan University (Tokyo-TMU) during the summer (1–19 July) and fall (17–31 October) of 2022. NO2 was measured using a cavity attenuated phase shift (CAPS) analyzer, which directly detects NO2 by measuring absorption around 450 nm, with a detection limit of less than 0.1 ppb (Choi et al., 2020). Meanwhile, HCHO was obtained using a selected ion flow tube mass spectrometer (SIFT-MS). SIFT-MS utilizes precursor ions such as H3O+, NO+, and O2+ for ionization and detection of target substances (Langford et al., 2023; Roberts et al., 2022). This instrument is recommended as an efficient method for the measurement of HCHO in both indoor and outdoor environments (Zogka et al., 2022).

Surface O3 concentrations were obtained from nearby air monitoring stations using the UV absorption method. These stations are operated by the Atmospheric Environmental Regional Observation System (AEROS) (https://soramame.env.go.jp/, last access: 24 July 2025).

Pandora spectrometer consists of a head sensor, an optical fiber transmission system, and a charge-coupled device (CCD) used as a spectral detector. The data retrieval begins with the raw measurement spectra (L0). The corrected signal (L1) is obtained by applying complex corrections, such as dark correction, latency correction, etc. Spectral fitting (L2Fit) is performed to derive slant column densities relative to a reference spectrum using the differential optical absorption spectroscopy (DOAS) method. Finally, L2 data is produced by converting slant columns into vertical columns utilizing geometrical air mass factors (AMFs) for the direct-sun mode and analytical methods for the sky-scan mode (Rawat et al., 2025). The direct-sun mode measures total NO2 column with high precision (2.7×1014 moleculescm-2) and accuracy (2.7×1015 moleculescm-2). For total HCHO column, a statistical error of 6 % and a systematic error of 26 % have been reported (Spinei et al., 2018). The bias of the sky-scan measurement is approximately -0.02×1016 moleculescm-2 for NO2 and -0.05×1016 moleculescm-2 for HCHO (Tirpitz et al., 2021; Verhoelst et al., 2021). In this study, we explored both the tropospheric column FNR and vertical FNR profile by combining the direct-sun and sky-scan modes of Pandora. In the direct-sun mode, the tropospheric columns of NO2 and HCHO were derived by subtracting the stratospheric contribution. This stratospheric information was derived from the GEOS-Chem model (Sect. 2.3). In addition to column densities in the lower troposphere (up to an altitude of 3 km), the sky-scan measurements provided vertical distributions from the surface up to an altitude of 3 km with several layers of measurements. These profiles are crucial for examining the vertical characteristics of NO2 and HCHO, as well as O3 production in the troposphere.

We used the Pandora data at four JPN stations, Sapporo, Tsukuba-NIES, Tokyo-TMU, and Fukuoka. These stations, listed from north to south, were chosen to investigate FNR across different latitude environments. A brief description of these Pandora stations is provided in Table S1 in the Supplement. These Pandora instruments routinely alternate between direct-sun and sky-scan modes on a standard schedule (Cede et al., 2021a). Data processing was performed using the Blick software, which converts L0 (raw measurement spectra) to L2 products (e.g., vertical column densities, profiles, etc.). For FNR calculations, L2 products were employed. To maximize the available scientific data, we applied a new filtering method adopted from Rawat et al. (2025). The cut-off values were defined as the mean plus three standard deviations of the independence uncertainty for data with a high-quality flag. Data with independence uncertainty exceeding the cut-off value was removed. We also excluded data with a solar zenith angle (SZA)>75° (Mouat et al., 2024). Using this filtering method, the data volume increased significantly by 5 %–30 % for NO2 and 20 %–70 % for HCHO, compared to the filtering method that uses high and medium data quality flags.

To characterize the FNR thresholds, we investigated the response of O3 to emission perturbations using the GEOS-Chem model. GEOS-Chem is a three-dimensional chemical transport model driven by assimilated meteorological observations from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO) (http://www.geos-chem.org, last access: 29 May 2025). To simulate O3, HCHO, and NO2 for the year 2022, we used the high-performance GEOS-Chem (GCHP) model, version 14.4.0 (The International GEOS-Chem User Community, 2024). GCHP is described by Eastham et al. (2018). Improved advection, resolution, performance, and community access are described by Martin et al. (2022). In the current study, we configured four model runs.

For Run-1, global anthropogenic emissions were based on the Community Emissions Data System version 2 (CEDSv2) (McDuffie et al., 2020). We used the Regional Emission inventory in ASia version 3.2.1 (REASv3.2.1) (Kurokawa and Ohara, 2020), as the regional anthropogenic emissions override the global anthropogenic emissions for Japan. Biomass burning emissions were taken from the Global Fire Emissions Database version 4 (GFED4) (Van Der Werf et al., 2017). Additionally, dust, sea salt aerosol, soil NOx, lightning NOx, and biogenic VOCs emissions were computed offline (Weng et al., 2020). All emissions were configured at run-time using Harmonized Emissions Component (HEMCO, version 3.9.3) (Lin et al., 2021). Table S2 provides a detailed description of the emission inventories used in the model simulation. For meteorology, we used MERRA-2 (0.5°×0.625°), a global atmospheric reanalysis data product. The full-chem model simulations use chemical mechanism kinetics following JPL/IUPAC recommendations (Bates et al., 2024). Photolysis frequencies for stratospheric and tropospheric chemistry are calculated with Cloud-J v7.7.3 (Prather, 2015). Stratospheric chemistry is represented by a linear chemistry mechanism, the Linoz algorithm (McLinden et al., 2000). The Linoz stratospheric chemistry package is recommended for GEOS-Chem simulations of O3. More details on the chemical mechanisms are available at http://www.geos-chem.org (last access: 29 May 2025). In our study, the simulations were run using a 10 min time step for chemistry and a 5 min time step for transport. Moreover, we applied the grid-stretching capability to focus on the Japan region. The grid-stretching procedure followed Bindle et al. (2021), with an initial cubed-sphere grid of C90, a target latitude of 37°, a target longitude of 137°, and a stretch factor of 4. This procedure yielded an average horizontal resolution of 27.78 km over Japan. The simulation generated the vertical extent from the surface to approximately 80 km with a 72 vertical-layer grid. Surface concentrations were obtained from the first model layer. The tropospheric column and the stratospheric column were separated using tropopause information. Furthermore, this troposphere-stratosphere distribution was used to derive the tropospheric column from the Pandora direct-sun observations.

Run-2 was the same as Run-1 but with all anthropogenic HCHO emissions turned off. Secondary HCHO reflects the VOCs activity through photolysis reactions. Previous studies have highlighted the importance of separating secondary HCHO from anthropogenic HCHO for a more accurate interpretation of the FNR (Hong et al., 2022; Xing et al., 2022; Xue et al., 2022). By comparing Run-1 and Run-2, we determined the contribution of primary HCHO and excluded it from the FNR calculation.

Run-3 and Run-4 were the same as Run-2 but with a 20 % reduction in regional NOx and VOC emissions, respectively. Ozone concentrations result from both in situ photochemical creation and external transport processes (Hong et al., 2022; Qian et al., 2024). A key advantage of the model-based method is that it allows us to exclude external transport processes, leading to a more precise classification of the O3 chemical regime. The external O3 transport influence was eliminated by subtracting Run-3 or Run-4 from Run-2. This step further reflects the response of O3 to changes in NOx and VOCs. Figure S1 in the Supplement shows an example of the surface O3 response to VOC and NOx emission reductions, resulting from GEOS-Chem simulations. Both negative and positive O3 changes were observed in response to NOx emission reduction. In contrast, VOC emission reduction consistently led to decreases in O3 levels. The Greater Tokyo Metropolitan Area was strongly influenced by these emission perturbations.

Regarding the responses, the O3 sensitivity regime was categorized following the method of Jin et al. (2017) and Jung et al. (2022). Instead of the term VOC-limited regime, here we used the term “ROx-limited regime” for more accuracy present for the radical-limited regime. A negative change in O3 owing to NOx emission reduction indicates a ROx-limited regime. A NOx-limited regime is defined when the positive change in O3 owing to VOC emission reduction is smaller than that from NOx emission reduction. The FNR threshold values for ROx-limited and NOx-limited regimes were determined as those corresponding to the 95th percentile of the cumulative probability distribution for each regime.

The O3, HCHO, and NO2 outputs from the GEOS-Chem model simulations were utilized to identify the O3 sensitivity regime (Fig. 7). From the scatter plots, the responses of surface O3 to NOx and VOC emission reductions at Sapporo and Fukuoka were within 5 ppbv, which were less pronounced than those at Tsukuba-NIES and Tokyo-TMU (within 10 ppbv). A positive O3 difference indicates that the emission reduction results in a decrease in surface O3. Conversely, a negative difference means that the emission reduction enhances surface O3 concentrations. In our study, generally, VOC emission reductions led to a decrease in surface O3 concentrations. NOx emission reductions could either decrease O3 through photochemical reaction or increase it due to reduced LNOx processes.

The FNRsec threshold values were determined using both surface and column information (Fig. 7, cumulative graphs). Based on surface FNRsec, the ROx-limited conditions were distinguished with thresholds of FNRsec<0.2 for Sapporo, <0.29 for Tsukuba-NIES, <0.28 for Tokyo-TMU, and <0.18 for Fukuoka. The NOx-limited regimes were associated with FNRsec>0.36 for Sapporo, and >0.91, >0.75, and >0.72 for Tsukuba-NIES, Tokyo-TMU, and Fukuoka, respectively. The surface FNRsec thresholds were generally lower than the column FNRsec. The finding of higher column FNR is consistent with previous studies (Jin et al., 2017; Souri et al., 2023a). The column FNRsec thresholds for the ROx-limited regimes and NOx-limited regimes at the Sapporo site were <0.86 and >1.52, respectively. The corresponding column FNRsec threshold ranges for Tsukuba-NIES, Tokyo-TMU, and Fukuoka were 0.74–1.7, 0.62–1.53, and 0.43–1.73, respectively. The transitional regime seemed to occur over a wider range of column FNRsec values at lower latitudes (e.g., Fukuoka) compared to higher latitudes (e.g., Sapporo). The differences between surface and tropospheric column FNRsec could be attributed to the vertical characteristics of HCHO and NO2. NO2 molecules were concentrated near the surface, while HCHO was present in the higher layers. As a result, surface FNRsec did not account for HCHO at higher layers, leading to lower threshold values. The surface and column FNRsec thresholds identified in this study are consistent with those reported by Jin et al. (2017) for East Asia (Table 2).

For the column, the transition range for Sapporo was defined as 0.86–1.52 where the cumulative probability of NOx-limited and ROx-limited condition reached 95 %, corresponding to a 5 % probability of misclassification. Consequently, 15 % of NOx-limited and 15 % of ROx-limited conditions were incorrectly classified as transitional (Fig. 7). Similarly, the probabilities of misclassifying NOx-limited and ROx-limited conditions as transitional were 20 % and 25 %, respectively, for Tsukuba-NIES, 35 % and 25 % for Tokyo-TMU, and 15 % and 60 % for Fukuoka. Notably, the probability of misclassification of ROx-limited condition at Fukuoka was higher than at the other locations. Because we simulated for the year 2022, this uncertainty could be reduced by extending the model simulations to longer time periods.

Table 2 presents the column FNR regime thresholds related to surface O3 sensitivity from previous studies. These threshold values vary depending on the methodology, geographic region, and atmospheric conditions. The FNR thresholds using both primary and secondary HCHO (FNRtotal) are higher than those using only secondary HCHO (FNRsec). The transitional regimes are reported for a column FNRtotal of 2.5–4.0 in Guangzhou, China (Hong et al., 2022), and 1.6–2.6 in United States (Jung et al., 2022), whereas the column FNRsec thresholds identified in our study were lower than 2. Column FNR regime threshold values are useful for examining global O3 production, as both satellite and ground-based remote sensing techniques offer extensive spatial coverage (Inoue et al., 2019; Ryan et al., 2023; Santiago et al., 2021).

The diurnal and seasonal cycles of surface and column FNR are shown in Figs. S9 and S10. Surface FNRsec developed in the early morning and decreased in the afternoon. During the 8 h daytime period (08:00 to 16:00), the average surface FNRsec was 0.99 ± 0.46 in July and 0.72 ± 0.64 in October. Using tropospheric vertical column densities derived from Pandora sky-scan observations, the column FNRsec exhibited a higher value compared to those from surface measurements. Like surface FNRsec, the tropospheric column FNRsec grew during the morning. The highest column FNRsec values were found during the summer months (JJA). In contrast, these values were normally below 1 in winter (DJF).

In this section, we analyze the O3 formation sensitivity focusing on polluted days when the maximum daily 8 h averaged surface O3 (MDA8) exceeded 60 ppbv. We identified 8, 30, 38, and 43 exceedance events in 2022 at Sapporo, Tsukuba-NIES, Tokyo-TMU, and Fukuoka, respectively. During O3 exceedance events, photochemical reactions were typically strong, allowing us to determine the sensitive regime with minimum biases. Because the model simulations did not reproduce surface NO2 and HCHO as well as the column densities, we will focus only on column FNR. Although Pandora direct-sun mode provides highly accurate NO2 measurements, the HCHO retrieval attains non-negligible error budget, which can cause more uncertainty and bias in the FNR. In contrast, Pandora sky-scan mode provides lower tropospheric columns of NO2 and HCHO and yields lower uncertainty in the HCHO data product. Additionally, the sensitivity diagnosis within the boundary layer is better represent tropospheric O3 production processes. Therefore, we discuss instead the column FNRsec derived from Pandora sky-scan observations which is sensitive to the boundary layer. We averaged available sky-scan observations around noon (12:00 ± 02:00), when photolysis rates are highest. Figures 8–11 present column FNRsec during O3 polluted events at Sapporo, Tsukuba-NIES, Tokyo-TMU, and Fukuoka, respectively. The advantage of Pandora observations is the ability to examine daily O3 production in greater detail and investigate specific periods such as exceedance events. Based on the analysis of these events, policy strategies could be proposed to help mitigate O3 pollution.

At Sapporo, the O3 concentrations increased during spring (Fig. 8). All eight events with MDA8 O3 exceeding 60 ppbv occurred in April and May 2022. Pandora observations were unavailable on 21 and 22 May 2022. Given the transition range determined at Sapporo (0.86<FNRsec<1.52), the column FNRsec exhibited O3 sensitivity in transitional and NOx-limited regime. Therefore, reducing NOx emission would be the optimal strategy for mitigating O3 levels. The vertical sensitivity indicated that the ROx-limited regime was typically confined to the near surface layer during events classified as transitional using column FNRsec. Concurrently, the NOx-limited regime frequently formed aloft, whereas the transitional regime formed in the mid-levels. Figure S11 illustrates the corresponding HCHO and NO2 profiles for these exceedance events. Higher concentration of NO2 on 21 and 22 April, concentrated below 0.5 km, led to a ROx-limited regime within this layer. For events categorized as NOx-limited (6–7 May) according to the column FNRsec, the NOx-limited regime dominated at higher altitudes and expanded downward to the surface, while the ROx-limited regime retreated. Even during periods of high MDA8, the vertical profiles of both NO2 and HCHO on 23 April and 6 May were low, suggesting that O3 could be influenced by regional or vertical transport. This influence could be recognized using Pandora HCHO and NO2 observations; however, further studies are needed to distinguish these events from local photochemical O3 production.

Figure 9 is similar to Fig. 8, but for Tsukuba-NIES. Compare with Sapporo, more O3 exceedance events were found at Tsukuba-NIES, with 30 events in 2022. These events primarily took place during spring and summer. Like the results at Sapporo, the column FNRsec exhibited transitional and NOx-limited conditions during spring events, whereas O3 chemistry was almost entirely NOx-limited during summer events, likely due to substantial enhancement in HCHO production (Fig. S11). Regarding vertical sensitivity, we determined a lower probability of ROx-limited regime during the spring events. During summer events, O3 formation sensitivity throughout the lower troposphere was consistently NOx-limited, consistent with the column FNRsec classification. Thus, emission policies focusing on continued reductions in NOx would effectively improve tropospheric O3 pollution during summer.

The exceedance events of O3 spanned a broad period from spring through fall at both Tokyo-TMU and Fukuoka (Figs. 10 and 11). The column FNRsec were higher during summer leading to a dominant classification in the NOx-limited regime. Transitional conditions were observed more frequently during spring and fall. At Tokyo-TMU, only one exceedance event (7 April) was found to be sensitive to ROx radicals. The locally elevated NOx emissions on that event shifted the chemical environment downward to lower FNRsec values (Fig. S12). Vertical sensitivity profiles showed a ROx-limited regime within the lower 1 km on 7–8 April. For Fukuoka, the transitional condition occurred more frequently during the exceedance events, emphasizing the need for simultaneous control strategies targeting both NOx and VOCs in this region.

The use of FNR indicator with satellite and ground-based observations can be readily implemented. However, FNR has inherent limitations in inferring the complex interactions among O3 production, its precursors, photolysis rates, and water vapor (Jin et al., 2025; Schroeder et al., 2017; Souri et al., 2026; Yang et al., 2026). For example, during the early morning and late afternoon, weak light intensity can lead to large biases in the interpretation of FNR because O3 production is less sensitive to its precursors under low photolysis conditions. In this study, we minimized the potential biases in predicting O3 chemistry by using Pandora data under conditions of strong photolysis rates. This may underutilize the rich hourly observations provided by Pandora and geostationary satellites. Jin et al. (2025) developed an approach to link the FNR with O3 chemistry using a steady-state O3-NOx-VOC model combined with GOES-CF simulations, and applied it to TEMPO observations. Additionally, Souri et al. (2026) optimized the OMI and TROPOMI data to investigate the daily mapping of net O3 production using a deep neural network approach. A similar approach could also be applied to Pandora data to derive accurate hourly net O3 production estimates.