-
The Picarro G2301 gas mole fraction analyzer provides simultaneous and precise measurements of CO2, CH4, and H2O, with a sensitivity of 1 ppb and negligible drift. This instrument is widely used in atmospheric science and air quality applications for quantification of emissions. The G2301 uses WS-CRDS (wavelength-scanning cavity ring-down spectroscopy) to measure the laser ring-down time difference over a path length of up to 20 km with high accuracy (Fang et al., 2012). It meets the WMO and ICOS (Integrated Carbon Observation System) performance requirements for CO2 and CH4 atmospheric monitoring (Laurent, 2015). The G2301 was fully operational, almost without interruption, during the study period. We use 5-min-averaged CH4 data at the surface (90 mMSL) to ensure high accuracy (< 0.7 ppb).
-
We used the Bruker IFS 125HR FTIR to observe CH4 in Xianghe, which is the highest-precision instrument among ground-based remote sensing instruments (± 0.9 ppb; Esler et al., 2000). The maximum optical path difference of the FTIR in Xianghe was 180 cm, with a spectral resolution of 0.0035 cm−1. To obtain more observations of trace gases, we used two observation modes: TCCON and NDACC-IRWG. NDACC-IRWG adopts a CaF2 beam splitter and an InSb detector with a spectral range of 1800–5400 cm−1, while TCCON uses CaF2 and InGaAs with a spectral range of 3900–12000 cm−1. The FTIR works during the day [between 0900 and 1600 local time (LT)]. For each day, there are approximately 40 spectra measured in TCCON observation mode and 5 measured in NDACC observation mode, and these are used to retrieve CH4. Figure 2 shows the typical spectra measured by the two modes. The TCCON and NDACC retrieval strategies, i.e., retrieval windows and interfering species, are shown in Table 1. A specific filter is applied for the InSb spectrato increase the signal-to-noise ratio.
Observation modes TCCON NDACC Algorithm GGG2014 SFIT4 Retrieval windows (cm−1) 5872.0–5988.0
5996.45–6007.55
6007.0–6145.02611.6–2613.35
2613.7–2615.4
2835.55–2835.8
2903.82–2903.925
2941.51–2942.22Interfering species CO2, H2O, N2O H2O, HDO,CO2, NO2 A priori profile TCCON tool (daily) WACCM v4 Products Total column Profile Table 1. TCCON and NDACC CH4 retrieval strategies in Xianghe.
-
The optimal estimation method (Rodgers, 2000) was applied to retrieve gas mole fractions from the FTIR solar spectra. Several software retrieval algorithms are applied for different purposes, such as GGG2014, SFIT, and PROFFIT (Toon et al., 1992; Notholt et al., 1993; Hase et al., 2004).
According to the optimal estimation method, atmospheric radiation transfer can be described by a simple mathematical model, as follows:
where Y is the radiation spectrum, X is the set of unknown atmospheric and surface state quantities, as well as the solar and instrumental parameters that affect the radiation spectrum, b is the set of known atmospheric and surface state parameters, F is the forward-transfer radiation model function, and ε is the error of observation. The inversion process involves taking the observed spectrum (Y) and finding the unknown atmospheric and surface state parameters (X).
To carry out the inversion process, a cost function is defined as:
where the first term on the right-hand side represents the difference between the measured and simulated spectra for a given atmospheric state of X and b,
${\boldsymbol{S}}_ \in ^{}$ is the noise covariance matrix, the second term on the right-hand side is the regularization term, constraining the atmospheric solution state X to the a priori state Xa, and${{\boldsymbol{S}}_{\rm a}}$ is the a priori covariance matrix. Since most physical processes in the atmosphere are nonlinear, the cost function, Eq. (2), is minimized iteratively by the Levenburg-Marquardt (LM) Gauss–Newton method. Convergence gives the retrieved state vector,${\boldsymbol{\hat x}}$ , as follows:where
${{\boldsymbol{x}}_{\rm a}}$ is the a priori state matrix,${{\boldsymbol{x}}_{t}}$ is the true state vector, and A is the average kernel (AVK) function matrix, representing the sensitivity of the inversion parameters to the true state of the atmosphere, given by:In Eq. (4), G is called the contribution function matrix, which indicates the contribution of observed values to inversion values, K is a weight function matrix that expresses the sensitivity of simulated values to input parameters, such as the instrument line functions, and
$\gamma $ is the coefficient in the LM method. -
The GGG2014 algorithm was applied to retrieve the column-averaged dry-air mole fraction of CH4 (
${X_{{\rm{C}}{{\rm{H}}_4}}}$ ) from InGaAs spectra. It also performs profile scaling. Specifically,${X_{{\rm{C}}{{\rm{H}}_4}}}$ is obtained from the ratio between the total column of CH4 (${\rm{V}}{{\rm{C}}_{{\rm{C}}{{\rm{H}}_4}}}$ ) and O2 (${\rm{V}}{{\rm{C}}_{{{\rm{O}}_{\rm{2}}}}}$ ), using the following equation:where 0.2095 is the volume mixing ratio of O2 in dry air (Wunch et al., 2011).
Since there is no O2 signal available in the mid-infrared spectrum, and the N2 signal is very weak (Zhou et al., 2018), the SFIT4 algorithm calculates the
${X_{{\rm{C}}{{\rm{H}}_4}}}$ from the dry-air column as follows:where
${\rm{T}}{{\rm{C}}_{{{\rm{H}}_{\rm{2}}}{\rm{O}}}}$ is the total column of H2O, Ps is the surface pressure, g is the column-averaged gravitational acceleration, and${m_{{{\rm{H}}_{\rm{2}}}{\rm{O}}}}$ and$m_{{\rm{air\_dry}}}$ are the molecular masses of H2O and dry air, respectively.Note that the a priori information is different between GGG2014 and SFIT4. For the meteorological variables of temperature, pressure, and water vapor, both algorithms use NCEP six-hourly reanalysis data. However, the a priori profiles of CH4 are obtained by different methods. For GGG2014, the daily profiles are generated by a stand-alone tool based on in-situ and aircraft measurements (Toon and Wunch, 2015). For SFIT4, the profiles are derived from the Whole Atmosphere Community Climate Model (WACCM), version 4 (Zhou et al., 2019).
Figure 3 shows information about the AVK in GGG2014 and SFIT4 for different solar zenith angles (SZAs). The AVK represents the sensitivity of the inversion to the true state of the atmosphere (see section 3.2.1). Ideally, the AVK is an identity matrix, indicating that the inversion is sensitive to the whole atmosphere. However, in reality, the sensitivity to the atmospheric state of the AVK varies by height. For example, for TCCON, the total amount of CH4 is always sensitive to the troposphere (> 0.8), regardless of SZA. However, in the stratosphere, the total amount of CH4 varies from 1.2 to 0.6 as the SZA varies from 10° to 85°.
Figure 3. Column averaging kernel (AVK) of the CH4 retrieval in the (a) GGG2014 and (b) SFIT4 codes, colored with different solar zenith angels (SZAs).
This is mainly because, as the Sun obliquely enters the atmosphere (as SZA increases), pressure broadening in the stratosphere contributes to a narrower linewidth than the saturated central region for gases with saturated absorption lines. Therefore, as the mass of air increases, the line becomes more saturated, so the AVK of the stratosphere is partially reduced, resulting in the total amount of column inversions becoming insensitive to stratospheric information. For the AVK in NDACC, the inversion CH4 column mole fraction is more sensitive to the troposphere and the lower stratosphere.
In addition to the total column, SFIT4 can retrieve the partial column of CH4. The profiling capability is not only important for CH4 source or sink research applications, but is also advantageous when validating column-averaged CH4 obtained from satellites (Sepúlveda et al., 2014). The degree of freedom (DOF) of the NDACC-retrieved profile (from ground to the top of the atmosphere) of CH4 is 2.23±0.18 (
$1\delta $ ), meaning that there are two independent pieces of information in the vertical distribution of CH4. According to the DOF, we divide the vertical distribution of CH4 into two independent parts: one from the ground to 12.2 km (representing the troposphere), and the second from 12.2 to 60 km (representing the stratosphere). The DOF in each part is approximately 1, which means that the signal information in each part is independent (Fig. 4). We calculated the dry-air column-averaged mole fractions of CH4 in the troposphere (${X_{{\rm{C}}{{\rm{H}}_{\rm{4}}}{\rm{,tr}}}}$ ) and stratosphere (${X_{{\rm{C}}{{\rm{H}}_{\rm{4}}}{\rm{,st}}}}$ ) (Zhou et al., 2018), as follows:Figure 4. Total column averaging kernel (black), together with the partial column averaging kernels of two individual layers (CH4: surface–12.2 km, partial column DOFs=1, and 12.2–60 km, partial column DOFs=1) of one typical NDACC retrieval in Xianghe.
where
${\rm{P}}{{\rm{C}}_{{\rm{C}}{{\rm{H}}_{\rm{4}}}{\rm{,tr}}}}$ ,${\rm{P}}{{\rm{C}}_{{{\rm{H}}_{\rm{2}}}{\rm{O,tr}}}}$ ,${\rm{PC}}_{{\rm{air\_dry,tr}}}$ , and${\rm{PC}}_{{\rm{air\_wet,tr}}}$ are the partial columns of CH4, H2O, dry air, and wet air in the troposphere, and${\rm{P}}{{\rm{C}}_{{\rm{C}}{{\rm{H}}_{\rm{4}}}{\rm{,st}}}}$ and${\rm{PC}}_{{\rm{air\_dry,st}}}$ are the partial columns of CH4 and dry air in the stratosphere, respectively.Combined in-situ and FTIR measurements provided information about CH4 in three vertical layers: near the ground, in the troposphere, in the stratosphere, and in the total atmospheric column. This enabled us to analyze the temporal and vertical distribution of CH4 in source emission areas.
-
TCCON is a well-developed worldwide observation network. Details of the sources of known uncertainty are described in Wunch et al. (2011). For
$X_{{\rm CH}_4} $ , the largest sources of error are observer-sun Doppler stretch, shear misalignment, continuum curvature, a priori profiles, and angular misalignment. The total$X_{{\rm CH}_4} $ error, i.e., the sum of each individual uncertainty, is below 0.5% until the SZA is above ~85°. Yang et al. (2019) proved that the FTIR at the Xianghe site meets the TCCON requirements, and the instrument is in good working condition. Thus, it is reasonable to believe that the$X_{{\rm CH}_4} $ uncertainty budget of this site when using the TCCON observation mode is consistent with other TCCON sites.For the NDACC observation mode, according to Rodgers (2000), the difference between the true and retrieved state of the atmosphere can be written as follows:
where I is the identity matrix, εb is the error in forward model parameters, and εy is the measurement error. According to Eq. (9), the total error is divided into three error sources: smoothing error, forward model parameter error, and measurement error (Senten et al. 2008). Table 2 shows the values of those errors for a regular observation. In summary, the systematic and random errors of the NDACC total column are about 3.3% and 1.7%, respectively.
Smoothing error Measurement error Forward model parameter error Total random error Interference species error Temperature (systematic) Line intensity error (CH4) 0.117[%] 0.104[%] 0.014[%] 1.426[%] 2.938[%] 1.737[%] Table 2. CH4 uncertainty budget in NDACC observation mode.
-
CarbonTracker-CH4 is an assimilation system for atmospheric CH4 developed by the National Oceanic and Atmospheric Administration (NOAA). Using observations from surface stations and towers, CarbonTracker-CH4 assimilates global atmospheric CH4 into TM5 (Transport Model 5; Liu et al., 2016). We used the current release, CT2010, to analyze the atmospheric CH4 mole fractions in Xianghe at the surface. In CarbonTracker-CH4, CH4 emissions into the atmosphere are estimated separately for natural and anthropogenic sources. Natural sources include oceans, wetlands, soil, and insects and wild animals. The anthropogenic sources include emissions from fires, coal, oil and gas production, animals, rice cultivation and waste. We used daily averages from 2010 to determine the contributions from different anthropogenic sources in Xianghe.
Observation modes | TCCON | NDACC |
Algorithm | GGG2014 | SFIT4 |
Retrieval windows (cm−1) | 5872.0–5988.0 5996.45–6007.55 6007.0–6145.0 | 2611.6–2613.35 2613.7–2615.4 2835.55–2835.8 2903.82–2903.925 2941.51–2942.22 |
Interfering species | CO2, H2O, N2O | H2O, HDO,CO2, NO2 |
A priori profile | TCCON tool (daily) | WACCM v4 |
Products | Total column | Profile |