Atmospheric propane (C₃H₈) column retrievals from ground-based FTIR observations in Xianghe, China

Propane (C₃H₈) is an important trace gas in the atmosphere, as it is a proxy for oil and gas production and has a significant impact on atmospheric chemical reactions related to the hydroxyl radical and tropospheric ozone formation. In this study, solar direct absorption spectra near 2967 cm⁻¹ recorded by a ground-based Fourier transform infrared spectrometer (FTIR) were applied to retrieve C₃H₈ total columns between June 2018 and July 2022 in Xianghe in north China. The systematic and random uncertainties of the C₃H₈ column retrieval are estimated to be 18.4 % and 18.1 %, respectively. The mean and standard deviation of the C₃H₈ columns derived from the FTIR spectra in Xianghe are 1.80 ± 0.81 (1σ) × 10¹⁵ molec. cm⁻². Good correlations are found between C₃H₈ and other non-methane hydrocarbons, such as C₂H₆ (R=0.84) and C₂H₂ (R=0.79), as well as between C₃H₈ and CO (R=0.72). However, the correlation between C₃H₈ and CH₄ is relatively weak (R=0.45). Moreover, the FTIR C₃H₈ measurements in Xianghe are also compared against MkIV measurements at several sites around the world. The new FTIR measurements in Xianghe provide us with insight into C₃H₈ column variations and the underlying processes in north China.

1 Introduction

Methane (CH₄) and non-methane hydrocarbons (NMHCs), such as ethane (C₂H₆), acetylene (C₂H₂), propane (C₃H₈), propene (C₃H₆), and isoprene (C₅H₈), are important trace gases that play significant roles in atmospheric chemical reactions related to hydroxyl radical (OH) abundance and tropospheric ozone (O₃) formation (Sze, 1977; Donahue and Prinn, 1990; Tan et al., 2012; Lelieveld et al., 2015). Human activities contribute greatly to the emissions of CH₄ and NMHCs, especially in urban areas (Bourtsoukidis et al., 2019; Saunois et al., 2020). Atmospheric C₂H₆ and C₃H₈ emissions are dominated by oil and gas sources, and they are co-emitted with CH₄. Therefore, numerous studies used the ratio of C₂H₆ and/or C₃H₈ to CH₄ to understand the CH₄ trend (Kort et al., 2016; Franco et al., 2016; Rigby et al., 2017).

The major sink of C₂H₆ and C₃H₈ is the reaction with OH, and the lifetime of C₃H₈ and C₂H₆ is about 2–4 weeks in summer and 2 months in winter (Jacob, 1999; Xiao et al., 2008). Compared to CH₄ with a lifetime on an order of 10 years (IPCC, 2013), the short-lived gases C₂H₆ and C₃H₈ are not well mixed on the global scale and are therefore more representative of regional pollution, as is carbon monoxide (CO) (Toon et al., 2021).

Atmospheric C₃H₈ concentrations at the surface are observed by National Oceanic and Atmospheric Administration (NOAA) Global Monitoring Laboratory (GML) flask sampling measurements at 12 sites (https://gml.noaa.gov/hats/gases/C3H8.html, last access: 29 October 2024). In addition, the HIAPER Pole-to-Pole Observations (HIPPO), Atmospheric Tomography (ATom), and In-service Aircraft for a Global Observing System (IAGOS) aircraft campaigns provide in situ gas analyzer measurements of C₃H₈ with a wide latitudinal coverage, particularly in the Pacific Ocean, the Atlantic Ocean, Europe, and North America (Wofsy, 2011; Thompson et al., 2022; Li et al., 2022). Toon et al. (2021) demonstrated the use of C₃H₈ absorption lines in the mid-infrared region (Harrison et al., 2010) in solar absorption spectra from MkIV interferometers for retrieving the C₃H₈ total columns or vertical profiles at several locations in Sweden, the USA, and Antarctica. Solar absorption infrared spectra are also being collected by ground-based Fourier transform infrared (FTIR) spectrometers within the Network for the Detection of Atmospheric Composition Change – InfraRed Working Group (NDACC-IRWG) (De Mazière et al., 2018). Currently, there are more than 20 NDACC-IRWG global sites, with a good global latitudinal coverage from 78° S to 80° N (https://www2.acom.ucar.edu/irwg/sites, last access: 29 October 2024). However, to our knowledge, no site has reported C₃H₈ retrievals from spectra observed by a Bruker 125HR spectrometer within the NDACC-IRWG.

Xianghe (39.75° N, 116.96° E) is located in north China, about 50 km east of the megacity Beijing (Yang et al., 2020). According to the Emissions Database for Global Atmospheric Research (EDGAR) v6.0 (Crippa et al., 2020) and the Multi-resolution Emission Inventory for China (MEIC) (Wang et al., 2015; Li et al., 2017), there is a large CH₄ emission source in north China coming from fuel exploitation and oil refineries. Therefore, we expect that the C₂H₆ and C₃H₈ concentrations are relatively high in this region. In June 2018, a Bruker IFS 125HR spectrometer, compliant with the NDACC-IRWG protocol, started recording solar absorption spectra in the mid-infrared spectral range. The spectra have been used to retrieve several atmospheric components, e.g., O₃, CH₄, CO, C₂H₂, C₂H₆, HCN, and H₂CO (Ji et al., 2020; Zhou et al., 2020, 2021, 2023; Vigouroux et al., 2020; Sha et al., 2021). In this study, we investigate the C₃H₈ retrieval from ground-based FTIR spectra in Xianghe and discuss the C₃H₈ column variation in north China, based on these new FTIR measurements.

The remainder of this paper is organized as follows. Section 2 describes the Xianghe FTIR site and C₃H₈ retrieval method, and Sect. 3 presents the C₃H₈ variations and correlations with other species. Moreover, the C₃H₈ measurements in Xianghe are compared to ground-based MkIV measurements in other places. Finally, Sect. 4 draws a conclusion.

2 Method

2.1 Xianghe FTIR spectra measurement

The Xianghe FTIR measurement system started in June 2018 and has been described well in previous studies (Yang et al., 2020; Zhou et al., 2021, 2023). Briefly, the FTIR measurement system contains three parts: a solar tracker system, a weather station, and a Bruker IFS 125HR Fourier transform infrared (FTIR) spectrometer. Shortwave infrared (SWIR) and near-infrared (NIR) spectra (4000–11 000 cm⁻¹) with a spectral resolution of 0.02 cm⁻¹ are recorded with an InGaAs detector, and these spectra are used to derive greenhouse gases' total column abundances as a contribution to the Total Carbon Column Observing Network (TCCON). Mid-infrared (MIR) spectra (1800–4500 cm⁻¹), with a spectral resolution of 0.0035–0.0070 cm⁻¹, are recorded with an InSb detector. To enhance the signal-to-noise ratio (SNR) of the spectra, we add specific optical filters into the light path when recording each MIR spectrum as recommended by NDACC-IRWG (Blumenstock et al., 2021; Zhou et al., 2023). A typical MIR spectrum, with a spectral resolution of 0.0051 cm⁻¹, used for C₃H₈ retrieval is shown in Fig. 1. Note that we only operate the FTIR measurement during the daytime and under clear-sky conditions, as the sun is the light source. In general, we carry out 4 to 10 MIR spectral measurements of this type per day for about 200 d yr⁻¹. Each spectrum takes about 10 min to record. The spectra taken between June 2018 and July 2022 (about 4 years) are used in this study.

Figure 1 A typical MIR spectrum observed in Xianghe on 6 April 2022 with a solar zenith angle of 49.8°. The red and green windows indicate the micro-windows used for the C₃H₈ retrieval and for calculating the noise (Eq. 5), respectively. The inset in the right-hand corner shows a zoomed-in view of the retrieval micro-window.

2.2 Retrieval method

To derive C₃H₈ mole fractions from the observed spectra, we follow the optimal estimation methodology (Rodgers, 2000). The forward model (F) simulates the absorption spectra (y) observed by the FTIR system. It includes modeling of the solar spectra at the top of the atmosphere (TOA), the physics of the radiative transfer from the TOA to the ground-based FTIR, and the FTIR spectrometer instrument line shape function (ILS). Then, the observed spectra (y) can be written as

$$\begin{array}{}\text{(1)}& \mathit{y}=\mathbf{F}(\mathit{x},\mathit{b})+\mathit{ϵ},\end{array}$$

where x is the state vector (retrieved parameters); b is the forward model parameters (not retrieved); and ϵ is the error, including the measurement noise and forward model errors. We wish to find the optimal state (x) that minimizes the cost function (J(x)), given by

$$\begin{array}{}\text{(2)}& \begin{array}{rl}\mathit{J}\left(\mathit{x}\right)& =[\mathit{y}-\mathbf{F}(\mathit{x}\left){]}^T{\mathbf{S}}_{\mathit{ϵ}}^{-\mathrm{1}}\right[\mathit{y}-\mathbf{F}\left(\mathit{x}\right)]\\ & +[\mathit{x}-{\mathit{x}}_{\mathrm{a}}{]}^T{\mathbf{S}}_{\mathrm{R}}[\mathit{x}-{\mathit{x}}_{\mathrm{a}}],\end{array}\end{array}$$

where S_ϵ is the measurement error covariance matrix, S_R is the regularization matrix, and x_a is the a priori state vector. The Levenberg–Marquardt (LM) method is used to iteratively solve the above equation:

$$\begin{array}{}\text{(3)}& \begin{array}{rl}{\mathit{x}}_{i+\mathrm{1}}=& {\mathit{x}}_i+{\left[(\mathrm{1}+\mathit{\gamma }){\mathbf{S}}_{\mathrm{R}}+{\mathbf{K}}_i^T{\mathbf{S}}_{\mathit{ϵ}}^{-\mathrm{1}}{\mathbf{K}}_i\right]}^{-\mathrm{1}}\\ & \left\{{\mathbf{K}}_i^T{\mathbf{S}}_{\mathit{ϵ}}^{-\mathrm{1}}\left[\mathit{y}-\mathbf{F}\left({\mathit{x}}_i\right)\right]-{\mathbf{S}}_{\mathrm{R}}\left[{\mathit{x}}_i-{\mathit{x}}_{\mathrm{a}}\right]\right\},\end{array}\end{array}$$

where K is the Jacobian matrix, and γ is a parameter to adjust the regularization of a priori information in each iteration step (Rodgers, 2000). Upon convergence, the final state is called x_r, which can be related to the true state (x_t):

$$\begin{array}{}\text{(4)}& {\mathit{x}}_{\mathrm{r}}={\mathit{x}}_{\mathrm{a}}+\mathbf{A}\left({\mathit{x}}_{\mathrm{t}}-{\mathit{x}}_{\mathrm{a}}\right)+\mathit{\epsilon },\end{array}$$

where A is the averaging kernel matrix, representing the sensitivity of the retrieved parameters to the true parameter, and ε is the retrieval uncertainty propagated from Eq. (1).

2.3 Retrieval strategy

In this study, we use the SFIT4 v1.0 retrieval algorithm (Pougatchev et al., 1995; Hase et al., 2004) to perform the forward model simulation as well as the LM inversion. The well-established SFIT4 code has been used extensively to retrieve total/partial column of atmospheric species in the NDACC-IRWG community (Zhou et al., 2016; De Mazière et al., 2018; Ortega et al., 2019).

Table 1 The retrieval window, interfering species, spectroscopy, and fitting parameters for C₃H₈ in Xianghe.

The key C₃H₈ retrieval parameters used in this study are listed in Table 1. The retrieval window is set to 2964.5–2970.0 cm⁻¹, where we have the strongest C₃H₈ absorption line (Harrison et al., 2010). Apart from C₃H₈, several interfering gases (H₂O, CH₄, O₃, C₂H₆, and HDO) also have absorption lines in this window as shown in Fig. 2. To reduce the impact of uncertainties about the abundances of these species, CH₄, O₃, C₂H₆, and HDO columns are retrieved along with the target gas mole fractions. For these three species, their profile shapes are fixed, and only the scaling factors are retrieved. As H₂O absorption lines are strong (Table 1) and H₂O variability are relatively large in the atmosphere, we perform a profile retrieval for H₂O. The state vector includes CH₄, O₃, C₂H₆, and HDO columns, as well as 47 layers' C₃H₈ and H₂O mole fractions.

Figure 2 The transmittances of main species and solar lines (b), as well as the fitting residual (a) mean (black line) and standard deviation (grey shadow) from all 2783 FTIR C₃H₈ retrievals in Xianghe between June 2018 and July 2022. The mean RMSE is 0.317 %.

The chosen spectroscopic parameters are crucial in the remote sensing technique. In this study, we have tested several line lists, particularly for H₂O (HDO) and CH₄ (see Table 2), including DLR2016 (Loos et al., 2017), HITRAN2020 (Gordon et al., 2022), and ATM2020 (https://mark4sun.jpl.nasa.gov/pseudo.html, last access: 29 October 2024). The ATM2020 line list is created by Geoff Toon (NASA, JPL) based on HITRAN2020, together with some additional atmospheric and laboratory measurements. For C₃H₈, we use pseudo-linelists (PLLs), which are also created by Geoff Toon based on laboratory cross section measurements by Harrison et al. (2010). For C₂H₆, we use PLLs as well. We tested more than 1000 spectra recorded in 2019 in Xianghe, and we observed that the lowest root-mean-square error (RMSE) of the fitting residual is obtained when the ATM2020 spectral database is used for CH₄ and H₂O. Table 1 lists the spectral datasets finally used for each species in the C₃H₈ retrieval strategy.

Table 2 The fitting RMSE of the retrieval window for all spectra in 2019 from several different line lists.

The a priori profiles for C₃H₈, C₂H₆, and CH₄ are derived from the Whole Atmosphere Community Climate Model (WACCM) version 6. We use the averages of the monthly means between 1980 and 2040 (61 years) as the a priori profiles. Since the variations of temperature and humidity are quite large in the atmosphere, using fixed a priori profiles often results in a bad fitting, especially for the first iteration. To provide a better estimation of temperature and humidity profiles, for each measurement, the H₂O (HDO) and temperature vertical profiles are derived from the closest 6-hourly NCEP reanalysis data (Saha et al., 2014) and linearly interpolated to the measurement time.

According to Eq. (2), the cost function J(x) is composed of the measurement and a priori information, each contracted with a weight matrix S_ϵ and S_R, respectively. In this study, the diagonal of the S_ϵ is calculated as $\mathrm{1}/{\mathrm{SNR}}^{\mathrm{2}}$, and the non-diagonal values are set to 0. The SNR is calculated as

$$\begin{array}{}\text{(5)}& \mathrm{SNR}={\displaystyle \frac{\stackrel{\mathrm{‾}}{I_{\mathrm{r}}}}{{\mathit{\sigma }}_{I_{\mathrm{n}}}}},\end{array}$$

where $\stackrel{\mathrm{‾}}{I_{\mathrm{r}}}$ is the max radiation intensity in the C₃H₈ retrieval window (2964.5–2970.0 cm⁻¹; red window in Fig. 1), and ${\mathit{\sigma }}_{I_{\mathrm{n}}}$ is the standard deviation (SD) of the intensity in the noise window (2250.0–2300.0 cm⁻¹; green window in Fig. 1). The Tikhonov L₁ regularization method (Tikhonov, 1963) is applied to generate the S_R, with

$$\begin{array}{}\text{(6)}& {\displaystyle }{\mathbf{S}}_{\mathrm{R}}=\mathit{\alpha }{\mathbf{L}}_{\mathrm{1}}^T{\mathbf{L}}_{\mathrm{1}},\text{(7)}& {\displaystyle }{L}_{\mathrm{1}}=\left[\begin{array}{cccccc}-\mathrm{1}& \mathrm{1}& \mathrm{0}& \mathrm{\dots }& \mathrm{0}& \mathrm{0}\\ \mathrm{0}& -\mathrm{1}& \mathrm{1}& \mathrm{\dots }& \mathrm{0}& \mathrm{0}\\ \mathrm{⋮}& \mathrm{⋮}& \mathrm{⋮}& \mathrm{\ddots }& \mathrm{⋮}& \mathrm{⋮}\\ \mathrm{0}& \mathrm{0}& \mathrm{0}& \mathrm{\dots }& -\mathrm{1}& \mathrm{1}\end{array}\right].\end{array}$$

To determine the α value in Eq. (6), we apply the method of the degree of freedom for signal (DOF) proposed by Steck (2002). The trace of the averaging kernel matrix (A) is the DOF, indicating the pieces of independent information of the retrieval (Rodgers, 2000). First, we use the optimal estimation method (OEM) to get an estimated DOF. Using the OEM, ${\mathbf{S}}_{\mathrm{R}}={\mathbf{S}}_{\mathrm{a}}^{-\mathrm{1}}$, where S_a is the a priori covariance matrix, which is derived from a covariance matrix on the WACCM monthly means between 1980 and 2040 ($(S_{\mathrm{R}}{)}_{i,i}=(S_{\mathrm{a}}{)}_{i,i}^{-\mathrm{1}}={\mathit{\sigma }}_i^{-\mathrm{2}}$; diagonal values), and the non-diagonal values are set as $(S_{\mathrm{R}}{)}_{i,j}=(S_{\mathrm{a}}{)}_{i,j}^{-\mathrm{1}}=\left({\mathit{\sigma }}_i{\mathit{\sigma }}_j\right)/e^{(d_{i,j}/\mathrm{4})}$, where d_i,j is the vertical distance between layer i and layer j (in km). The DOF derived from the OEM is about 1.1, indicating that there is only column information for the C₃H₈ retrieval. Knowing that, we tune the α value in Eq. (6) to make the DOF derived from the Tikhonov method close to the DOF that is derived from the OEM; this approach results in setting α to 1000.

2.4 Retrieval uncertainty

The retrieval error (ϵ) of the FTIR C₃H₈ column contains three parts as follows:

$$\begin{array}{}\text{(8)}& {\displaystyle }(\mathbf{A}-\mathbf{I})({\mathit{x}}_{\mathrm{t}}-{\mathit{x}}_{\mathrm{a}})\mathrm{smoothing}\mathrm{error}\text{(9)}& {\displaystyle }{\mathbf{G}}_y{\mathbf{K}}_b({\mathit{b}}_{\mathrm{t}}-{\mathit{b}}_{\mathrm{a}})\mathrm{model}\mathrm{parameter}\mathrm{error}\text{(10)}& {\displaystyle }{\mathbf{G}}_y\mathit{ϵ}\mathrm{measurement}\mathrm{error},\end{array}$$

where G_y is the contribution function, and b_t and b are the true and used model inputs, respectively. Table 3 lists the systematic and random uncertainty of each component. The vertical distributions of the systematic and random uncertainties are shown in Fig. 3. For the smoothing error, we separate the contributions into target species (C₃H₈), interfering species (H₂O, HDO, CH₄, C₂H₆), and retrieved parameters (slope, phase, wavenumber shift, instrument line shape, solar intensity and shift). For the model parameter contributions, we calculate the C₃H₈ uncertainty contribution coming from spectroscopy, the solar zenith angle (SZA), the temperature profile, the curvature parameter, and the zero level shift (zshift). Since CH₄ and H₂O have stronger absorptions than C₃H₈, and their absorption lines are not perfectly fitted, the impact from the spectroscopy uncertainty of CH₄ and H₂O is calculated as well.

Table 3 The systematic and random (sys/ran) retrieval uncertainties for the total columns of C₃H₈. The “–” means that the uncertainty is less than 0.1 %. 1σ of the target or interfering species is the SD derived from the WACCM model monthly means between 1980 and 2040. The relative SD in the bottom row is the average of daily SD of C₃H₈ columns on all days with at least three measurements, which is to represent the variability of the retrieval.

Figure 3 The vertical profiles of the systematic (a) and random error (b) of the FTIR C₃H₈ retrieval from each component.

The systematic and random uncertainties of each parameter are also listed in Table 3. It is assumed that 10 % of the a priori profile is used to derive the diagonal values of the systematic covariance matrix $({\mathbf{S}}_{\mathrm{a}}{)}_{i,i}^{\mathrm{sys}}={\mathit{\sigma }}_i^{\mathrm{2}}$, and the off-diagonal values of ${\mathbf{S}}_{\mathrm{a}}^{\mathrm{sys}}$ are calculated as $({\mathbf{S}}_{\mathrm{a}}{)}_{i,j}^{\mathrm{sys}}={\mathit{\sigma }}_i{\mathit{\sigma }}_j$ (von Clarmann, 2014). The covariance matrix derived from the WACCM 61 years' monthly means is set to the random covariance matrix ${\mathbf{S}}_{\mathrm{a}}^{\mathrm{ran}}$. Regarding the model parameter uncertainties in Table 3, the systematic–random S_b matrix is created by the mean and standard deviation of the differences between NCEP and ERA5 in Xianghe. The random deviation is about 2 K, and the systematic deviation is about 1.5 K for the whole vertical range. For the target spectroscopic parameters, the relative uncertainties of C₃H₈ are set to 4 % according to the pseudo-database. For the CH₄ and H₂O spectroscopy parameters, the relative uncertainty of 5 % is derived from the HITRAN2020 dataset (Gordon et al., 2022). Note that the spectroscopy uncertainty in Table 3 is the sum of the uncertainties from the line intensity, pressure-dependent parameter (linePAir), and temperature-dependent parameter (lineTAir). For the uncertainties of background curvature, zero offset (shift), and SZA, we use the default values provided by the SFIT4 algorithm (https://wiki.ucar.edu/display/sfit4/SFIT4+Version+1.0.xx+Release, last access: 29 October 2024), which are recommended by the NDACC-IRWG community.

Based on our uncertainty estimation, the total systematic uncertainty and random uncertainty of the C₃H₈ column are both about 18 %, and the dominating contribution is the uncertainty of the background curvature parameter in the forward model. To represent the variability of the C₃H₈, we select all days with at least three individual measurements each day and calculate the daily SD. The average of all the daily SDs is about 15.3 %, and it is close to our estimated random uncertainty.

3 Results and discussions

3.1 FTIR C₃H₈ retrievals in Xianghe

Figure 4 shows the a priori profile and retrieved profiles of C₃H₈. The vertical profile of C₃H₈ from the WACCM model shows that the C₃H₈ mole fraction is high near the surface and decreases with increasing altitude. Such a vertical shape is expected as the C₃H₈ emissions are at the surface, and its atmospheric lifetime is too short to achieve a well-mixed troposphere. Although we perform a profile retrieval on C₃H₈, we only have about 1 DOF. In addition, the Tikhonov regularization matrix constrains the vertical shape when the DOF is typically close to 1.0. As a result, the retrieved C₃H₈ profiles have a very similar vertical shape to the a priori profile. However, the FTIR measurements show that the a priori column overestimates the C₃H₈ column concentration by about 100 %. The column averaging kernel indicates the sensitivity of the retrieved C₃H₈ column to the C₃H₈ partial column for each height. Figure 4 shows that the retrieved C₃H₈ column has good sensitivity to all the layers and slightly varies with SZA.

Figure 4 The a priori and retrieved C₃H₈ profiles (a) and the column averaging kernel (CAVK) varying with SZA (b).

Figure 5 Panel (a) shows the time series of FTIR a priori C₃H₈ columns (black dots), retrieved columns (grey dots), monthly means (yellow line), monthly SDs (yellow shade), periodic function fitting (red solid line), and the fitted offset (dashed red line). (b) The monthly box plot of the C₃H₈ columns. The bottom and top bars represent the 10 % and 90 % percentiles of the datasets, and the blue crosses are the extremely high values above 90 %.

The time series and seasonal variation of FTIR C₃H₈ column measurements are presented in Fig. 5. To better visualize the seasonal variation, the column measurements are fitted by a periodic function $y\left(t\right)=A_{\mathrm{0}}+\sum _{k=\mathrm{1}}^{\mathrm{3}}\left(A_{\mathrm{2}k-\mathrm{1}}\cos \right(\mathrm{2}k\mathit{\pi }t)+A_{\mathrm{2}k}\sin (\mathrm{2}k\mathit{\pi }t\left)\right)$, where A₀ is the offset, and A₁ to A₆ are the periodic amplitudes, representing the seasonal variation. The obtained mean and SD of C₃H₈ columns in Xianghe are 1.80 ± 0.81 × 10¹⁵ molec. cm⁻². The C₃H₈ columns show a high mean value in July and a low value in October. The difference between the median values in July (maximum) and October (minimum) is 1.2 × 10¹⁵ molec. cm⁻². Although the median values of C₃H₈ columns in June–August are larger than those in October–March, we notice that extremely high C₃H₈ columns often occur in the latter period.

3.2 Correlations with CO, CH₄, C₂H₂, and C₂H₆ in Xianghe

As mentioned above, the infrared spectra observed by the Xianghe FTIR system have also been used to retrieve CO, CH₄, C₂H₂, and C₂H₆ columns using NDACC-IRWG-recommended retrieval strategies (Ji et al., 2020; Zhou et al., 2023), which allows us to investigate the correlation between C₃H₈ and these species. We are particularly interested in the correlation on a regional scale. Therefore, to reduce the impact from the background, we calculate the Δgas (Δgas = gas − monthly median) for all these species. Figure 6 shows the correlation scatter plots between ΔC₃H₈ and ΔCH₄, ΔCO, ΔC₂H₂, and ΔC₂H₆. High correlation coefficients (R) are found between ΔC₃H₈ and ΔC₂H₆ (R=0.84) and between ΔC₃H₈ and ΔC₂H₂ (R=0.79). It indicates that C₂H₂, C₂H₆, and C₃H₈ (NMHCs) are co-emitted in this region. The slope of ΔC₂H₆ and ΔC₃H₈ is 6.03 ± 0.03, which suggests a corresponding mixing ratio of C₂H₆ and C₃H₈ mole fractions during the production in north China. CO, as a pollutant tracer, also has a good correlation with C₃H₈ (R=0.72). According to the MEIC, both CO and NMHCs are emitted from the energy production, industry, residential, and transport sectors.

Figure 6 The correlation plots between co-located ΔC₃H₈ and ΔCH₄, ΔCO, ΔC₂H₂, and ΔC₂H₆ hourly means in Xianghe between June 2018 and July 2022. The dashed grey line is the linear fit, N is the number of the FTIR measurements, and R is the Pearson correlation coefficient.

The FTIR measurements show that the correlation between ΔC₃H₈ and ΔCH₄ is relatively weak (R=0.45). Note that the variation of the CH₄ column is also affected by the stratospheric partial column (Sepúlveda et al., 2014). The DOF of the FTIR CH₄ retrieval is about 2.5, allowing us to derive the tropospheric and stratospheric CH₄ partial columns separately (Zhou et al., 2018). However, even after eliminating the interference from the stratosphere, the tropospheric CH₄ partial column still has a weak correlation with C₃H₈ (R=0.43). It is probably due to the fact that the CH₄ major emissions in north China are from rice cultivation, waste, and animals instead of oil and gas production (Ji et al., 2020), and the CH₄ measurements include the emissions from much farther away as compared to the C₃H₈ measurements because of its long lifetime (Callewaert et al., 2023).

To further investigate the ratio of ΔC₂H₆ to ΔC₃H₈, the time series of their ratios, together with the monthly correlation coefficients between both time series between June 2018 and June 2022, are illustrated in Fig. 7. The ratio of each month is derived from the linear fitting using all co-located ΔC₂H₆ and ΔC₃H₈ hourly measurements in that month. A relatively low correlation between these two species is found in summer as compared to other three seasons. The mean and SD of the ratios are 5.4 ± 2.1 for the whole period. The ratio is lowest in summer and highest in winter, with seasonal means of 6.6, 3.8, 5.4, and 8.3 in spring, summer, autumn, and winter, respectively.

Figure 7 The time series of the ratio of ΔC₂H₆ to ΔC₃H₈ monthly means and SDs (green, on the left-hand vertical axis scale), together with their monthly correlation coefficients (grey, on the right-hand vertical axis scale) between June 2018 and June 2022.

Figure 8 The C₃H₈ (a) and C₂H₆ (b) columns observed by ground-based Bruker IFS 125HR in Xianghe and MkIV spectrometer at six sites. Note that the C₃H₈ columns observed at MTB and Xianghe are multiplied by 10 to have a better view.

3.3 FTIR measurements in Xianghe against MkIV measurements

Here, the C₃H₈ and C₂H₆ columns derived from the FTIR measurements in Xianghe are compared to the ground-based MkIV C₃H₈ retrievals at six sites in Sweden and the USA (Fig. 8). MkIV data use the GFIT inverse retrieval code to derive the C₃H₈ columns from the MkIV observed spectra between 2964.5 and 2970 cm⁻¹ with a spectral resolution of 0.5 cm⁻¹. The mean uncertainties of the MkIV retrieved C₃H₈ and C₂H₆ column are estimated to be around 8 × 10¹⁵ and 7 × 10¹⁴ molec. cm⁻², respectively (Toon et al., 2021). Note that the C₃H₈ and C₂H₆ retrievals from the MkIV spectrometers at 12 sites have been discussed in Toon et al. (2021), and we only select 6 sites as the measurements are very limited at the other 6 sites. The locations and measurement time coverage of sites used in this study are listed in Table 4.

Table 4 The locations and data time coverages of the MkIV measurements at six sites, together with their mean C₃H₈ and C₂H₆ columns. The bottom row is the Xianghe FTIR measurements in this study.

Figure 8 shows that the C₂H₆ column is the largest in Xianghe, apart from several extremely high values at JPL-B and FTS. The seasonal variations of C₂H₆ columns are similar at these sites, especially for JPL-B, MTB and Xianghe, with a high value in northern spring and a low value in northern autumn. Note that it is hard to derive the seasonal variation of C₂H₆ columns at ESN, FAI, TMF, and FTS because measurements were carried out in several months. The mean and SD of C₂H₆ columns at JPL-B are 1.96 ± 0.52 × 10¹⁶ molec. cm⁻², which is about 25 % less than at Xianghe ((Xianghe − JPL) $/$ Xianghe × 100 %). Keep in mind that the C₃H₈ columns at MTB and Xianghe have been multiplied by 10 in Fig. 8. The C₃H₈ column at Xianghe is quite low as compared to other sites, which is only larger than that at MTB (mountain site) but much less than that at the mid-latitude sites. The mean and SD of C₃H₈ columns at JPL-B are 2.14 ± 1.33 × 10¹⁶ molec. cm⁻², which is about 12 times larger than that in Xianghe. The seasonal variations of C₃H₈ columns are similar at JPL-B and Xianghe sites too, with a high value in northern summer and a low value in northern winter. The good correlations (R>0.6) between C₃H₈ and C₂H₆ columns at JPL-B and FTS have been demonstrated in Toon et al. (2021), which is similar to what we observe at the Xianghe site. However, the ratios of ΔC₂H₆ to ΔC₃H₈ at JPL-B and FTS are 0.16 ± 0.10 and 0.78 ± 0.10, respectively, which are much less than the ratio observed at the Xianghe site of 6.03 ± 0.03. It indicates that the emission of C₃H₈ is much larger in the Los Angeles basin, California, than in north China.

4 Conclusions

The Xianghe FTIR 125HR system measures the solar absorption spectra following the NDACC-IRWG guidance. For the first time, the FTIR MIR spectra in Xianghe are used for the C₃H₈ column retrieval, using the well-established SFIT4 code, between June 2018 and July 2022. In this study, the retrieval strategy, retrieval uncertainty, and retrieval information are presented and discussed. Due to the wide and weak absorption of C₃H₈, we only derive the C₃H₈ column instead of its vertical profile. The systematic and random uncertainties of the C₃H₈ retrieved column are estimated to be 18.4 % and 18.1 %, respectively. In the C₃H₈ retrieval window, CH₄ and H₂O absorption lines are not perfectly fitted, indicating there is still room left to improving the line lists of these two species.

The mean and SD of the C₃H₈ column derived from the FTIR measurements in Xianghe are 1.80 ± 0.81 × 10¹⁵ molec. cm⁻². A month-to-month variation is observed with a high value in July and a low value in October. The difference between the median values in July (maximum) and October (minimum) is 1.2 × 10¹⁵ molec. cm⁻². As C₃H₈ is co-emitted with CH₄, CO, C₂H₂, and C₂H₆ during oil and gas production, we calculate the correlation between ΔC₃H₈ and these species in Xianghe. Good correlations are found between C₃H₈ and C₂H₆, between C₃H₈ and C₂H₂, and between C₃H₈ and CO. However, the correlation between C₃H₈ and CH₄ is relatively weak, which is probably due to CH₄ emission in north China being dominated by rice, cultivation, and waste, instead of oil and gas production and fossil fuel combustion. By comparing the C₃H₈ and C₂H₆ columns in Xianghe with six other sites around the world, provided by the ground-based MkIV spectrometers, we find that the C₂H₆ column in Xianghe is the largest. However, the C₃H₈ column in Xianghe is only larger than the columns observed at the mountain sites and polar sites, and it is much less than the C₃H₈ columns observed at mid-latitude sites in the USA. Currently, the reported uncertainty of MkIV C₃H₈ measurements is relatively large at about 8 × 10¹⁵ molec. cm⁻², which is much larger than that of the mean FTIR C₃H₈ measurements in Xianghe. Further investigation is needed to understand the large difference between FTIR and MkIV C₃H₈ measurements.

In summary, we successfully retrieve C₃H₈ columns from the FTIR MIR spectra in Xianghe, which provides us with a new dataset to understand the variation of C₃H₈ in north China. The retrieval strategy of C₃H₈ in this study should work at other Bruker 125HR FTIR sites as well, especially those close to a city or oil and gas field, e.g., Paris, Toronto, and Boulder. Nevertheless, efforts are still needed within the NDACC-IRWG community to generate a global harmonized FTIR C₃H₈ column dataset.