Continuous in situ CO₂ and CO measurements were conducted from 17 April 2021 to 6 March 2023 (a period of nearly two years) at the Shanghai Tower (SHT, 121.51° E, 31.23° N), an urban GHG monitoring site erected atop the world's second-tallest skyscraper. The sampling inlet was positioned at 632.0 m above ground level (a.g.l.); with the tower base at 5 m above mean sea level (a.m.s.l.), this corresponds to approximately 637 m a.m.s.l. This campaign yielded high-precision, in situ data on CO₂ and CO levels at the UCL top in metropolitan Shanghai, a key carbon emission hotspot in China (Fig. 1a). As shown in Fig. 1b, the site is situated within the densely urbanized Lujiazui Financial and Commercial District, where it is encircled by roads and high-rise buildings and lies adjacent to the Huangpu River, Shanghai's most vital and busiest shipping artery, thereby being significantly affected by human activities.

Considering the high wind speeds (ca. 11.5 m s⁻¹) at the SHT top (637 m a.m.s.l.) and the aerodynamic effects of its 120° spiral facade (Zhao et al., 2011), the measurement point can be reasonably inferred to lie near the urban roughness sublayer (RSL), where turbulence initiates the homogenization of urban flow. For determining background CO₂ and CO level ([CO_2, bk] and [CO_bk]), previous studies have employed tracer analysis related to the 5 % lower value of CO (Wang et al., 2010), or data screening focused on intense vertical mixing around noon, (e.g., 10:00–16:00 local time (LT); Fang et al., 2017, 2022). However, during this campaign, noontime tower-top measurements may be subject to the upward transport of local anthropogenic emissions (see details in Sect. 3.2), as suggested by the PBLH variations (daytime: 728 ± 389 m; nighttime: 290 ± 222 m vs. SHT Top of 637 m; Fig. 2). Thus, to derive regional CO_2, bk and CO_bk amplitudes, we preferentially selected stable nocturnal hours (22:00–05:00 LT) during which the sampling height (SHT top) remained above the urban boundary layer. The procedure for excluding data influenced by local emissions or advected pollution is detailed in Sect. 3.3. The excess CO₂ concentration relative to its background level, expressed as [CO_2, ex], can be further decomposed as follows: 1[CO2]=[CO2,bk]+[CO2,ex]=[CO2,bk]+[CO2,ff]+[CO2,bio-dominated]+TH+TV, where [CO₂] is the measured dry-air CO₂ mole fraction (ppm); [CO_2, bk] represents the background CO₂ level determined by the filtering method (Sect. 3.3); [CO_2, ff] denotes the CO₂ increment attributed to fossil fuel combustion; and [CO_{2, bio-dominated}] is the increment dominated by biological processes (positive for net respiration, negative for net photosynthesis). TH and TV stand for the residual influences of horizontal and vertical transport impacts on the observed CO₂ level, respectively. In practice, the CO_{2, bio-dominated} term may partially encompass effects attributed to TH and TV.

To understand the probable transport pathways of air masses arriving at the SHT site, we analyzed back trajectories using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) dispersion model from the National Oceanic and Atmospheric Administration Air Resources Laboratory (NOAA ARL). The calculations were based on gridded meteorological data from the Global Data Assimilation System (GDAS) provided by the National Centers for Environmental Prediction (NCEP). Specifically, MeteoInfo software (version 3.3.12) combined with NCEP/GDAS reanalysis data (ftp://arlftp.arlhq.noaa.gov/pub/archives/gdas1, last access: 25 October 2025, 1° × 1°; Rousseau et al., 2004) was used to calculate 72 h back trajectories with 1 h intervals. Trajectories for each season were computed with an arrival height of 600 m a.g.l., corresponding closely to the sampling inlet height at the SHT site. The spatial source distributions of CO₂ and CO were then analyzed using Potential Source Contribution Function (PSCF) method. This method identifies potential source areas by calculating the conditional probability that a measured concentration exceeds a predefined threshold (e.g., the seasonal mean; see Table S1 in the Supplement) when an air mass resides over a particular grid cell (Polissar et al., 1999). In the PSCF model, the geographical area covered by the trajectory was divided into grid cells with a spatial resolution of 0.5° × 0.5°. The PSCF value for a given cell (i, j) is defined as: 2PSCFij=mij/nij, where nij represented the number of trajectories endpoints passing through the ij unit, and mij represented the number of trajectories endpoints causing atmospheric CO₂ and CO values to exceed the threshold.

To reduce the influence of nij outliers, a weight function Wij was applied to obtain a weighted PSCF (WPSCF): 3Wij=1.00nij>80,0.7020<nij≤80,0.4210<nij≤20,0.05nij≤10. Additionally, to evaluate the influence of local surface wind on species concentrations, we employed the polarPlot function from the OpenAir package in R (Carslaw and Ropkins, 2012; Carslaw, 2019), which uses smoothing techniques to generate bivariate polar plots, depicting pollutant concentration as a continuous surface over ws and wd. The conditional probability function (CPF) was also applied to identify specific wind sectors with a high probability of being associated with elevated concentrations.

Figure 2 displays the time series of CO₂ and CO mole fractions measured at the SHT top, alongside ground-level measurements of O₃, NO₂, SO₂, and CO_gl (from the nearby BSMH station) and simultaneous records of rainfall (at BS station) and PBLH (reanalysis data) over the 2-year campaign from April 2021 to March 2023. Additional meteorological parameters, including surface wind, atmospheric pressure, temperature, and dew point, are presented in Fig. S2. Throughout the campaign, CO₂ mole fractions exhibited significant variability, with the 5 min records changing in the range of 362.1–783.7 ppm (388.3–536.5 ppm after 1 h smoothing) and campaign-averaged values (±95 % CI) of 433.50 ± 0.33 ppm. Meanwhile, tower-based 5 min CO mole fractions varied considerably from 58.7 to more than 4000 ppb (60.0 to approximately 2000 ppb after smoothing), with an overall average of 313.59 ± 4.91 ppb. Focusing on CO₂, a crucial GHG contributor, Table S1 summarizes CO₂ values measured at various observation sites from recent reports and studies. Generally, the CO₂ measurement conducted at the SHT site was significantly higher than that of a comparable elevation (e.g., Akedala: 420.2 ± 1.4 ppm; 563.3 m a.m.s.l.; KRE: 427.05 ± 0.64; 534 m a.m.s.l.) and those recorded at high-altitude stations (including LLN, WLG, MLO, JFJ, ABLECAS, TPB, TOH and DMS: ca. 417–428 ppm; above 1000 m a.m.s.l.), whether based on mountains or tower platforms that are situated in global/regional background settings, and also substantially exceeded the global average value of 416.8 ± 0.20 ppm (in 2021–2022; Table S1). Notably, while elevated, the SHT-based CO₂ value remained appreciably lower than those observed at nearby near-ground sites (e.g., ca. 441.6–446.5 ppm at LAN and HZ site, reported by Chen et al., 2024), where CO₂ levels are strongly influenced by local anthropogenic emissions. Similarly, the CO value at SHT site was also elevated above regional background level (233.4 ± 3.8 ppb at DMS by Lin et al., 2025) in the YRD region. These results suggest a distinct surplus of both CO₂ and CO at the UCL top in this highly-urbanized megacity, likely stemming from strong local/regional influences and underscoring the need for further investigation.

It is noteworthy that Shanghai experienced contrasting meteorological conditions and anthropogenic lockdowns during the campaign. In 2021, it received exceptionally high precipitation (1228.2 mm during April–December), influenced by typhoons (e.g., Typhoon In-Fa in July; Wang et al., 2023; Fig. 2), flood seasons, and plum rains; whereas in 2022, it was recorded significantly lower and unevenly distributed rainfall (870.6 mm, a 38-year low, including Typhoon Muifa in September; Lin et al., 2024), alongside epidemic-related lockdowns (Sect. 3.5). Interannual variability in CO₂ was modest (430.94 ± 0.19 ppm in 2021 vs. 433.89 ± 0.15 ppm in 2022), whereas CO was markedly lower in 2021 (290.23 ± 1.88 ppb vs. 313.77 ± 2.26 ppb). This contrast intensified during typhoons, which brought strong winds, heavy rainfall, a sharp V-shaped pressure drop, and cooling (Fig. S2). Specifically, under Typhoon In-Fa, CO decreased by 121.76 ppb (a reduction of 53.4 %), compared to a decrease of only 4.2 ppm (0.98 %) for CO₂; and during Typhoon Muifa, CO dropped by 246.14 ppb (45.5 %) vs. 2.9 ppm (0.67 %) for CO₂. Concurrently, ground-level reactive species (such as O₃, NO₂, and SO₂; Fig. 2) exhibited scavenging behavior consistent with that of CO. These results underscore that short-term meteorological perturbations exert a much weaker influence on CO₂ than on CO, reflecting their divergent atmospheric lifetimes and chemical reactivity. Beyond meteorological factors, the interannual variability also reflects differences in regional anthropogenic emissions. In 2021, extensive lockdowns across regions upwind of Shanghai – particularly the North China Plain and central China – substantially suppressed anthropogenic mobile source emissions, contributing to a lower background. Given its elevated location, the SHT site was particularly sensitive to this reduction, capturing the long-range transport signal from distant upwind regions more effectively than surface sites. In contrast, by late 2022, following Shanghai's prolonged citywide lockdown (March–June 2022) and subsequent nationwide relaxation of COVID-19 restrictions, anthropogenic emissions had rebounded. These emission differences, together with the distinct meteorological conditions noted above, shaped the observed pronounced interannual CO pattern at the elevated SHT site.

Unlike the chemically inert CO₂, CO has a relatively short atmospheric lifetime (τCO, ∼ 1–3 month; Zheng et al., 2019), governed primarily by OH-initiated oxidation; and its sources are dominated by ground incomplete combustion and, to a lesser extent, the oxidation of CH₄ and biogenic hydrocarbons (Ehhalt and Rohrer, 2009; Seiler, 2010). During this campaign, its diurnal cycles observed at the tower top displayed clear non-uniform distributions, also indicative of near-surface influence (further discussed below). To explore local effects, it is necessary to isolate the regional background signal from the tower-top measurements. Prior to that, vertical CO variations between tower top and base provide a practical perspective for assessing near-surface influences, as discussed in Sect. 3.2.

Figure 3 illustrates diurnal profiles of tower-top CO and CO₂ alongside ground-level primary pollutants (including CO_gl) throughout the entire campaign and across different seasons. Monthly cycles for species/parameters are provided in Fig. S3. Both CO₂ and CO exhibited a relatively consistent bimodal pattern, while in terms of CO, its ground mole fraction (avg. 543.11 ppb) was substantially higher and more variable than that (avg. 313.59 ppb) observed at the tower top. In addition, the initial peak at the tower-top occurred around 10:00–11:00 LT (ca. 443.33 ppb for CO and 438.04 ppm for CO₂), lagging the ground peak (e.g., ca. 584.71 ppb for CO_gl; 08:00–09:00 LT) by about 2 h. The UBL uplifts rapidly in the morning, surpassing the SHT top from ca. 09:00–10:00 LT, and remained above it until around 17:00 LT. Accordingly, the intense vertical turbulence during these hours enhanced an upward mixing of gaseous species such as CO, leading to a smaller concentration difference between the tower base and top (CO vs. CO_gl), especially on summer and autumn afternoons. In the evening, rush-hour traffic emissions produced a second peak in ground-level CO and tower-top CO₂ at around 19:00 LT, comparable to the morning spikes despite slight declines of 2.3 % and 0.2 % (Fig. 3a). In contrast, the secondary peak in tower-top CO occurred earlier (∼ 17:00 LT) and was significantly weaker, showing a -19.2 % decrease compared to the first one. This pronounced attenuation was likely due to limited replenishment of air masses at the elevated measurement site following the onset of atmospheric subsidence and the PBLH falling below the tower height. At night, CO₂ and CO measurements at the SHT site (UCL top; above urban PBLH, ca. 329.5 ± 246.2 m; Fig. S3) were largely free from ground influence, contrasting with the near-surface accumulation of pollutants in the stable nocturnal layer, a pattern consistent with previous hierarchical tower observations that also recorded limited nighttime concentration variability at upper levels (Denning et al., 2008; Winderlich et al., 2014). This vertical decoupling also aligns with tethered-balloon-based observations, which indicated that the impact of surface emissions on atmospheric carbon was mostly restricted to the lower atmosphere (0–1 km), even to below 300 m (Li et al., 2014; Liu et al., 2025). Consequently, measurements taken at or approaching the PBL top can provide a more representative estimate of regional carbon emission characteristics, and reduce the uncertainty in source rates (Kondo et al., 2006; Wang et al., 2006; Han et al., 2009). However, it is worth noting that, in contrast to the diurnal cycles observed at background tall towers (e.g., KRE; Fig. S4), the SHT-based measurements from an elevated urban core site were subject to pronounced daytime urban emissions, whereas the nighttime data remained more representative of regional surface flux influences. In this context, tower-top nocturnal measurements at SHT site were considered suitable proxies for the regional background, as detailed in Sect. 3.3 alongside the specific filtering criteria.

In addition, Fig. 3a and b present diurnal comparisons between tower-top 10-min observations and the unified hourly values. Despite the absence of real-time CO₂ vertical gradient measurements in this campaign, its variation can be inferred to be smaller than that of CO (even in daytime), owing to its chemical stability and high background abundance. The reactive species CO was also analyzed in relation to variations in co-located ground-level pollutants. The unimodal diurnal cycle of odd oxygen (O_x = NO₂ + O₃; Fig. 3c), an index of photochemical oxidation capability (Lin et al., 2021), featured elevated concentrations during daytime photochemical hours that potentially intensified CO depletion, particularly during May–August (Fig. S3). Temporal variations in the correlations between CO and SO₂ (a tracer of coal combustion (Li et al., 2006)) and NO₂ (a byproduct of fuel combustion, particularly traffic-related; Allen et al., 2013) implied distinct emission regimes (Fig. 3d). In the warmer summer and autumn months, CO showed a stronger correlation with NO₂, underscoring the influence of traffic-related sources. In contrast, the tighter CO–SO₂ association in cold seasons was a possible reflection of sulfur-containing industrial coal combustion. Despite the limited coal combustion and biomass burning for cooking and heating in urban Shanghai, the elevated wintertime concentrations observed in the study region were likely attributable to the widespread use of these fuels in the upwind North China regions (Liu et al., 2016; Zhang et al., 2022).

The observed CO top-base difference (Fig. S5) indicated upward transport (TV) of an urban plume driven by daytime turbulence, and this effect was potentially reinforced during autumn noontime as high CO levels at the tower top coincided with deeper PBLH, particularly in October (Fig. S3). Moreover, the SHT site appeared to be affected by horizontal transport (TH) within the lower troposphere (up to 700 hPa), especially from the northwest upwind direction where autumn crop residue burning remains active (Wang et al., 2019), despite the long-standing prohibition of open straw burning in Shanghai (Tian et al., 2022). As an elevated site within a coastal metropolis, SHT is subject to pronounced sea–land breeze circulations, the periodic nature of which significantly complicates pollutant transport and dispersion (Huang et al., 2025). Accordingly, we analyzed 24 h back trajectories from three representative time periods (morning and evening rush hours, and the stable nighttime period; Fig. S6), as well as 72 h trajectories and their associated cluster concentrations (Fig. 4) at the SHT site across seasons. Regarding diurnal cycles, CO₂ and CO levels were primarily influenced by air masses from the eastern coastal areas and by urban plumes passing through the regions of Jiangsu, Anhui, and Zhejiang. In contrast, higher loads of CO and CO₂ at the SHT receptor site in winter were consistently linked to 3 d aged plumes from inland North China, indicating a greater influence from distant anthropogenic sources driven by land breeze. The sea breeze effect was most pronounced in autumn, with the vast majority of air masses arriving at the SHT site within 24 h being driven by sea breeze (Fig. S6c), broadly consistent with recent findings by Huang et al. (2025). Notably, summertime patterns were characterized by a stronger influence from sources within the YRD region and by sea breeze extending toward the eastern sea area, accompanied by smaller vertical height variations. Furthermore, these anthropogenic impacts were more pronounced on weekdays, as supported by elevated diurnal concentrations of CO and CO₂ (ca. +5.66 ppb and +1.48 ppm, respectively; Fig. S7) compared to weekends.

The principle of background value filtering is to distinguish the signal of a well-mixed atmospheric state from the noise caused by transient emission or sink events. To reduce filtering bias, we applied a meteorological filter to nocturnal tower-top measurements before further statistical processing, as suggested by Ruckstuhl et al. (2012), to derive regional background values of CO and CO₂ (i.e., CO_bk and CO_2, bk). As noted earlier, the SHT site (637 m a.m.s.l.) is located at UCL top and typically remains above the nighttime PBLH, experiencing strong winds (∼ 11.5 m s⁻¹) that favor a well-mixed condition. Even so, the measurements could still be affected by certain meteorological conditions and plume transport events, as discussed below.

Rainfall influence. The elevation of nocturnal PBLH during rainfall (ca. +110 m; Fig. 5a) indicated enhanced vertical mixing, likely driven by cloud radiative effects that suppress near-surface cooling, along with mechanical turbulence enhancement and potential evaporative cooling aloft (Paul et al., 2018). Measurements further showed a more pronounced decline in tower-top CO concentrations (-16.9 % or -43.0 ppb) than in CO₂ (-0.6 % or -2.6 ppm), as compared to stable rain-free nights (22:00–05:00 LT). This distinct nocturnal response can be explained by reduced anthropogenic activity that disproportionately curtails CO emissions but has little effect on persistent CO₂ respiration, coupled with the rainfall-induced downward mixing of cleaner air, to which the locally concentrated CO was significantly more susceptible than the relatively well-mixed CO₂. Therefore, the analysis of concurrent CO_bk and CO_2, bk (and subsequent parameters) was restricted to non-rainy days, with extreme typhoon events also excluded as previously noted.

Transport impact of upwind, high-load airmass. Pollutant transport also plays a significant role in tower-based measurements, as emphasized earlier. Under the regime of the East Asian monsoon, clean maritime airmasses from the east alternate with polluted air from the west and northwest. These continental inflows were particularly evident in the autumn/winter (Clusters 1 and 2, Fig. 4), leading to a marked increase in CO and CO₂ concentrations. The PSCF-resolved potential source areas (Fig. S8) further identified significant inland extensions during spring, in contrast to a contraction toward the vicinity of the SHT site in other seasons, but it can broadly characterize the overall picture of source influence for the SHT site within a radius of approximately 200 km (with PSCF values > 0.9). In the absence of real-time wind speed and direction data at the tower top, tower-base wind fields were adopted as a proxy, and corresponding distributions of CO and CO₂ across nocturnal (22:00–05:00 LT) and diurnal periods are presented in Figs. 5b and S9, respectively. In particular, a relatively diffuse CO hotspot in the diurnal profile indicated considerable regional transport influence on the SHT receptor site. Under nocturnal conditions, both CO₂ and CO (in terms of averages and top 10 % highs) exhibited clustered hotspot patterns in the W, WNW, and NW sectors at ground wind speeds of ca. 1–4 m s⁻¹, suggesting persistent and mild upwind transport at night. These western sectors represent the upwind areas where the nighttime offshore land breeze may also partially contribute to the carbon footprint observed at the SHT receptor site (Huang et al., 2025; Zhao et al., 2022).

The selection of regional background data involved a sequential filtering procedure (Fig. 6a). Foremost, the NGT_MET-based process selected concurrent nocturnal CO and CO₂ measurements (22:00–next 05:00 LT used, constituting 33.3 % of the total hourly dataset) from the SHT site, followed by the exclusions of rainy days, typhoon-affected periods, and winds from W, WNW, and NW sectors, which collectively accounted for 14.5 % of the total. Subsequently, a comprehensive top-down statistical filter (defined as NGT_MET_SDM and detailed in the following section) was employed to screen out local events (resulting in an additional 5.3 % exclusion). Overall, this multi-step filtering process effectively normalized the initially right-skewed distribution of the measurements. The final background dataset consisted of 1604 hourly records collected over 276 nights during this nearly 2-year campaign, yielding background concentrations of 425.6 ppm for CO_2, bk and 191.1 ppb for CO_bk.

Significant short-term variations (e.g., at minute scales) in species concentrations are commonly indicative of local emission or sink processes (Tsutsumi et al., 2006; Seinfeld and Pandis, 2016; Liu et al., 2019). On this basis, we implemented the SDM approach to analyze hourly standard deviation (SD, derived from 5 min resolution data; Fig. S10a) and examined its distribution across the NGT_MET-screened datasets. As shown in Fig. S10b, a high-variability threshold (±4.4 ppm for CO₂ variation and ±34 ppb for CO variation over 1 h intervals) was adopted for retracing local events, with 5.1 % of records subsequently flagged as anomalous. Suspected outliers were then re-check and removed (additional ∼ 0.2 %) from the NGT_MET-screened data using two statistical criteria: the conventional 1.5 × IQR rule (Q1 - 1.5IQR, Q3 + 1.5IQR; Fig. 6a) applied to the overall dataset, and the ±3σ range relative to the smoothed values (Fig. 2) for hourly time series. Following this screening procedure, the final dataset comprised 248 sliding daily CO_bk and CO_2, bk, each of which satisfied the requirement of at least 3 valid nocturnal data within the 8 h period (i.e., 22:00–next 05:00 LT) and served as a dynamic baseline for calculating subsequent daytime excess concentrations. In cases where no valid background was available for a given day, a biweekly sliding average was applied as a substitute to account for temporal variability. Accordingly, Fig. 6b presents CO and CO₂ levels in semi-monthly intervals during non-rainy periods of the campaign, together with their corresponding background values vs. global (MLO, WLG, and JFJ) and YRD regional (DMS) background measurements.

Regarding CO_bk, the filtered mole fractions (±95 % CI: 190.1 ± 4.7 ppb) from tower-top measurements roughly aligned with synchronous values recorded at the sites of DMS, MLG, and JFJ (avg. 227.0, 125.2, and 116.6 ppb, respectively; Fig. 6b); however, this contrasted sharply with surface-based observations, such as the 373–413 ppb range reported at the nearby LAN site by Liu et al. (2019) using multiple filtering approaches. Given the high spatial heterogeneity of CO in the lower atmosphere over non-background regions, the tower-based deployments at the UCL top in this campaign appears to offer a superior vantage point for capturing its reliable regional signals. Focusing on CO_2, bk, the derived value at SHT (425.6 ± 0.3 ppm) showed an enhancement of approximately 6–7 ppm over the global/regional backgrounds (417.8–422.3 ppm at MLO, WLG, JFJ, and DMS; Fig. 6b and Table S1), suggesting significant net source effects within the YRD region, China. The dynamics of atmospheric CO₂ are governed by multiple factors, including biospheric processes (photosynthesis and respiration), PBLH, local anthropogenic activities, etc. (Yang et al., 2021; Chen et al., 2022). Previous studies (Chen et al., 2024; Lin et al., 2025) employed filtered DMS data as a background representative for the YRD; however, these data seem to be still confounded by sources on or advected to the mountain, with a notably enhanced influence from net vegetative CO₂ release during winter (Fig. S11). Specially, nighttime SHT data, measured from within the mixed layer directly above the city center, can be proposed as a better background proxy. In parallel, anthropogenic emission variations were captured at SHT (Fig. 6b), as shown by concurrent declines in CO and CO₂ (including background levels) during the March–June 2022 pandemic lockdown in Shanghai and a subsequent rebound during recovery. The strong coherence between CO₂ and CO (a tracer of incomplete combustion of carbon-containing fuels; Seiler, 2010) underscores the significant role of fuel combustion on atmospheric CO₂ levels. Furthermore, during the recovery period, the total growth rates of CO and CO₂ significantly exceeded their dynamic regional backgrounds (CO_bk and CO_2, bk; Fig. 6b). These growth imbalances can be explained by an increasing concentration excess (e.g., CO_2, ex = CO₂ - CO_2, bk), highlighting pronounced local net source effects. A detailed compositional analysis of these excess signals follows in Sect. 3.4. In addition, a comprehensive case study focusing on Shanghai lockdown and recovery phases is presented in Sect. 3.5.

Specifically, the dynamic excesses of CO and CO₂ (i.e., CO_ex and CO_2, ex) were calculated by subtracting the sliding daily background values (CO_bk and CO_2, bk) from their corresponding real-time measurements. Figure 7 displays the resulting hourly excess values and their inter-correlations across seasons. Since CO is mainly co-emitted with CO₂ from fuel combustion sources, a significant positive correlation between them is therefore expected when combustion processes dominate the observed CO_2, ex. Accordingly, the fossil fuel-derived CO₂ component (CO_{2, ff−derived} in Eq. 1) could be quantified using its co-emission ratio with CO (defined as kCO2/CO), as formulated in Eq. (4). It should be noted that kCO2/CO ratios in this study (Fig. 7a) were determined via the reduced major axis rather than the ordinary least square method to appropriately account for uncertainties in both CO_ex and CO_2, ex, consistent with previous studies (Hirsch and Gilroy, 1984; Wang et al., 2010). 4[CO2,ff]=[CO]-[CObk]×kCO2/CO Figure 7a reveals a clear seasonal pattern in kCO2/CO ratios, with higher values in spring (99.9 ppm ppm⁻¹) and winter (76.5) compared to summer (64.3) and autumn (47.8). As mentioned earlier, a lower kCO2/CO ratio in autumn may be related to crop residue burning as a possible source of autumn CO enhancement (Wang et al., 2019). Moreover, the generally higher kCO2/CO estimates throughout this campaign reflect a lower relative CO ratio (i.e., kCO/CO2, 1.5 ± 0.4 %) than reported across China (typically 1 %–6 %) (Liu et al., 2018; Xia et al., 2020; Che et al., 2022; Li et al., 2022; Wu et al., 2022), which may be due to relatively well-mixed atmospheric CO levels at the UCL top and/or a gradual improvement in regional combustion efficiency over time. Notably, high-efficiency fossil fuel combustion is typically associated with traffic and power plants (Yang et al., 2020; Che et al., 2022; Zhao et al., 2024). Significant reductions in kCO/CO2 have also been observed in specific industrial sectors in the YRD region, such as the steel industry, as suggested by Zhao et al. (2024), and these improvements are largely linked to China's Clean Air Action Plan (2013–2017), after which regional kCO/CO2 has stabilized at low levels. Compared with other urban-integrated observations, the derived kCO/CO2 at SHT is comparable to values in Sacramanto and California (∼ 1 %) (Wu et al., 2022), but remains higher than those in Europe areas such as Netherlands (0.2 %–0.4 %; Super et al., 2017) and France (0.3 %–0.7 %; Ammoura et al., 2016; Wu et al., 2022), indicating potential for further decline. This trend was also noticeable under the pandemic-related lockdown, as detailed in Sect. 3.5. Moreover, the relatively weak correlation between summertime CO_2, ex and CO_ex can be explained by disproportionate enhancement of CO₂ signals from biospheric processes (Fig. S3; Wang et al., 2010; Tohjima et al., 2014) and of CO from in situ oxidation of biogenic hydrocarbons (Bi et al., 2022; Fawole et al., 2022). Despite this, strong vertical mixing during summer helped maintain a good correlation of concurrent CO₂ and CO measurements (R = 0.53–0.80, Fig. S12). In densely populated urban areas, total respiration may substantially contribute to CO_2, ex, forming part of the urban plume. In addition to diel variations in photosynthesis/respiration, atmospheric transport processes (TH and TV; Eq. 1) may also affect observed CO_2, ex–CO_ex correlations. In this study, except for the application of a seasonally constant kCO2/CO, we also estimated summertime CO_2, ff using a time-dependent kCO2/CO expressed as the function of local time (Fig. S13). The results show that CO_2, ff estimates would decrease by 4–5 ppm in the evening hours (17:00–20:00 LT), while the diurnal average remained nearly unchanged (6.3 vs. 6.4 ppm; Fig. 7b). Consequently, a seasonal kCO2/CO (47.8–99.9; corresponding to kCO/CO2 of 0.8 %–2 %) was adopted in the subsequent analysis. Figure 7b further illustrates the diurnal cycles of main CO_2, ex components (CO_2, ff and CO_{2, bio-dominated}) across seasons. In addition, Table S2 summarizes the magnitudes of these resolved excesses relative to filtered background levels.

Overall, the considerable variability in both CO and CO_bk at the SHT site (mean CO_ex = 98.9 ppb; Table S2), consistent with previous observations (Wang et al., 2010; Liu et al., 2019; Xia et al., 2020), indicates their heightened susceptibility to local/regional, short-term emissions and depletion. It should be noted that our team conducted simultaneous ¹⁴C measurements in atmospheric CO₂ and CO at the site of Hengxiwu (HXW, 119.48° E, 30.60° N; 200 m a.m.s.l.) in Anji, Zhejiang, a representative area of intensive urbanization in the YRD region. Based on these observations, CO_2, ff concentrations were quantified via radiocarbon mass balance (according to Levin et al., 2003 and Vásquez et al., 2022), and the results (Fig. S14) revealed a strong positive correlation (R = 0.98) between CO_2, ff and CO_ex, with an averaged kCO/CO2 of 1.45 ± 0.13 % that was highly consistent with estimates from the SHT site (1.5 ± 0.4 %). These findings further confirm that CO emissions in the YRD region originate predominantly from fossil fuel combustion and high combustion efficiency there, thereby validating the application of Eq. (4) with an empirically constrained kCO2/CO for determining regional CO_2, ff. Focusing on CO_2, ex at SHT site, its averaged concentration was 7.5 ppm, exhibiting a distinct seasonal pattern that peaked in winter (11.4 ppm) and dropped to a minimum in summer and spring (∼ 4.4 ppm). Source decomposition indicated that fossil fuel combustion (CO_2, ff) was its dominant source, contributing 6.4 ppm (∼ 85 %) to the total excess, with the remainder (1.1 ppm, ∼ 15 %) mainly due to biological activities (CO_{2, bio-dominated}). Notably, CO_2, ff consistently served as the major source and was the primary driver of CO₂ diurnal cycles (Fig. 3), while CO_{2, bio-dominated} transitioned to a net summertime sink as intense sunlight enhanced photosynthetic CO₂ uptake (Wang et al., 2010; Tohjima et al., 2014). This biological sink offset the apparent effect of ecosystem respiration, pointing to a potential underestimation of its true role in summer. Moreover, the notable CO_{2, bio-dominated} sink (i.e., strong negative values; Fig. 7b) in summer evenings may stem from the oversimplified constant kCO2/CO and/or dilution by advected sea breeze (Fig. 4) from afternoon to evening, observed at the tower top in the evening. Diurnal analysis during winter revealed a slightly amplified CO_{2, bio-dominated} signal (3.1 ppm, representing 27 % of CO_2, ex) during non-noon hours, attributable to pronounced net respiration, alongside a substantia CO_2, ff contribution (8.3 ppm) that increased notably at night (Fig. 7b). The latter half of 2022 coincided with the post-lockdown economic recovery, a period marked by elevated CO₂ and CO levels (Fig. 6b) that were likely linked to combustion-related emissions. The following section aims to investigate the underlying variables responsible for this observed growth.

A retrospective look at Shanghai's epidemic measures from late 2019 to early 2023 indicates that this metropolis underwent a phased evolution, commencing with routine containment, pivoting through emergency lockdowns and targeted management, and culminating in a comprehensive reopening. To account for intraseasonal variability, the following analysis focused on monthly comparisons for two specific intervals: the lockdown period (May–June 2022 vs. 2021) and post-pandemic recovery period (September 2022–January 2023 vs. the preceding year), as shown in Fig. 8. As reported earlier, lockdown measures primarily curbed residential mobility and public traffic, with limited disruption to industrial operations (SHMS, 2022). Thus, the most prominent change during this campaign was a sharp 56.2 ± 78.7 % drop in traffic-related NO₂ during lockdown, followed by post-lockdown declines of 21.4 ± 12.2 % in NO₂ and 9.4 ± 4.1 % in CO. These variations, against a backdrop of modest PBLH increase (11.2 ± 8.4 %), highlight overall air quality improvement and specific success of curbing vehicle-related emissions (Wang et al., 2022). It should be noted that SO₂ (largely tied to industrial activities) increased significantly during the lockdown, likely due in part to a certain increase in industrial emissions, but driven more by reduced OH regeneration from NO_x decline, which weakened its atmospheric loss (Ye et al., 2023; Zhang et al., 2023). Ground-level CO also exhibited variable patterns (e.g., between May and June, 2021–2022), likely more influenced by localized sources such as residential fuel burning and emissions from emergency services during the pandemic peak than by regional transport.

Generally, UCL top medium- to long-lived carbon oxides (CO and CO₂) showed a pronounced divergence from that of ground-level short-lived pollutants such as NO_x and SO₂. The ongoing lockdown of about 3–4 months yielded a 38.8 % decline in tower-top CO and 1.7 % in CO₂, matching concurrent declines in their regional background and local excess concentrations (-75.4 ± 47.2 and -44.8 ± 9.6 ppb for CO; -2.8 ± 3.5 and -4.4 ± 1.1 ppm for CO₂). In contrast, the subsequent recovery saw a sharp rebound in atmospheric carbon loading, reflected by net increases in both CO (150.2 ± 191.6 and 145.3 ± 142.9 ppb) and CO₂ (15.2 ± 16.4 and 2.9 ± 5.7 ppm) background and excess components. These coordinated variations indicate that anthropogenic activities in this densely-populated metropolis (Shanghai) exerted a dominant influence on the regional carbon footprint and induced an additional localized excess/deficit. Moreover, the reversal in carbon load differences between the two scenarios implies a fundamental shift in emission structure. This transition is clearly reflected in the combustion efficiency proxy, kCO2/CO, which increased by ∼ 25 during the control period before a decrease of 36 ± 22 in the relaxation phase (Fig. 8). A widespread enhancement in kCO2/CO in response to anthropogenic emission controls has been also observed in other conditions, such as its near doubling during the 2008 Beijing Olympics (46.4 ± 4.6) relative to the pre-Olympic period (23–29 in September 2005–2007) (Wang et al., 2010). This trend is further supported by long-term interannual increases, exemplified by a rise from 10–18 to ca. 50 based on Δ14C measurements in the Los Angeles basin between 1980 and 2008 (Djuricin et al., 2012), and a climb from 46.5 to about 80–90 over the YRD region from 2012 to 2021 (Wu et al., 2022; Zhao et al., 2024).

UCL-top measurements are representative of the vertically integrated and homogenized urban carbon footprint, whereas ground-level observations are strongly influenced by local, transient sources. The 2022 lockdown in Shanghai precipitated a near-total collapse of its urban and intercity transport systems, showing a dramatic plummet of over 90 % in aviation, rail, and maritime passenger volumes (MOT, 2022), an approximate 20 % decline in road traffic, and a severe operational crisis at the Port of Shanghai with soaring vessel queues and cargo backlogs. The recovery period saw a gradual normalization of port operations, concurrent with a lasting behavioral shift toward private car usage over public transport. Variations in combustion efficiency across fuel types (e.g., gasoline, diesel, and biomass) and traffic states (e.g., moving, congested, or stationary) (Westerdahl et al., 2009; Popa et al., 2014; Wu et al., 2022) are also reflected in the differences of kCO2/CO. The observed elevation in kCO2/CO (Fig. 8) during the lockdown may be explained by the sharp reduction in urban traffic, along with a temporarily increased contribution from low-kCO/CO2 diesel shipping near the port (Zhang et al., 2016; Kim et al., 2020; Wu et al., 2022). In the recovery phase, the kCO2/CO dropped by 36 ± 22 from ca. 90 ± 22, and this shift can be attributed to resumption of gasoline vehicle-related traffic and potentially increasing influence from upwind inefficient combustion sources such as biomass burning Tohjima et al. (2014) in autumn and winter, 2022. Generally, in our year-to-year comparison for the same month (without accounting for biospheric influences here), localized CO₂ excess or deficit could be primarily attributed to variations in emissions related to fossil fuels (CO_2, ff).