Ground-Based Remote Sensing of Atmospheric Water Vapor Using High-Resolution FTIR Spectrometry

Abstract

Understanding the distribution of atmospheric water vapor (H₂O) is crucial for global warming studies and climate change mitigation. In this study, we retrieved the ground layer, tropospheric and total columns of H₂O using ground-based high-resolution Fourier transform infrared spectrometry (FTIR). The H₂O total columns are obtained from near-infrared (NIR) and mid-infrared (MIR) spectra, and the ground layer and tropospheric H₂O columns are retrieved from the MIR spectrum. The total columns of H₂O retrieved from NIR and MIR have a good consistency (R = 0.989). Additionally, the ground layer H₂O columns have a similar seasonal variation to total columns and tropospheric columns but have a higher seasonal amplitude. The ground layer H₂O columns are close to the total columns and tropospheric columns in winter; however, in summer, the average difference between the ground layer and total columns and the value between the ground layer and tropospheric columns are large. This is mostly due to temperature variation. The temperature has a linear response to H₂O, and the relationship between surface temperature and ln(XH₂O) values in the ground layer, the entire atmosphere and the troposphere show a significantly positive correlation, and the correlation coefficient R is 0.893, 0.882 and 0.683, respectively. Furthermore, we selected the HYSPLIT model to simulate the back trajectories of air parcels in the four seasons in Hefei and find that the air mass transport has a significant impact on the local H₂O change. These results demonstrate that ground-based high-resolution FTIR technology has high accuracy and precision in observing the vertical distribution and seasonal changes of H₂O in different atmospheres.

1. Introduction

Water vapor (H₂O) in the atmosphere is the most abundant greenhouse gas (GHG) and plays a significant role in atmospheric chemistry and climate change [1]. The most important meteorology is that the characteristics of weather change are closely related to regional transport and seasonal variation of H₂O in the atmosphere. Without H₂O, there would be no clouds, rain and snow as these substances cannot exist without H₂O [2]. Moreover, H₂O is also the most important GHG on earth that people depend on for survival; about 90% of the global natural greenhouse effect comes from H₂O, and keeps the earth warm enough to support life [3,4]. An increase in H₂O will lead to an increase in temperature, which will cause more H₂O to be absorbed into the air. Climate warming and water absorption will increase in a spiral cycle [5,6,7]. The observed climate warming leads to an increase in comprehensive H₂O, which in turn forces further climate change [8,9]. Therefore, assessing and quantifying atmospheric H₂O is critical to constraining weather and climate models.

H₂O mostly exists in the troposphere, only 10% of which is up to 850 hPa and 40% up to 500 hPa [10]. H₂O has been measured by many methods and techniques. Spaceborne infrared instruments have the advantage of large-scale measurements but lack long-term and vertical data [11]. In addition, H₂O measured by satellite is often accompanied by large uncertainty, which must be verified before analysis [12]. The in situ sensors have very high accuracy and precision, but the measurement of H₂O concentration near the ground is easily affected by the landform and the vertical transmission of air masses [13,14]. Remote sensing measurement techniques can avoid the interference of the above factors. High-precision and high-accuracy remote sensing can provide key information on H₂O column concentration [15,16,17].

The ground-based high-resolution Fourier transform infrared spectrometer (FTIR) records the direct sunlight to observe the H₂O, which has high precision and accuracy. Moreover, the ground-based FTIR measurement has long-term stability, so it can be used to verify the satellite H₂O information and model simulations. The total carbon column observing network (TCCON) and the network for the detection of atmospheric composition change (NDACC) use high-resolution FTIR instruments to record solar absorption spectra in the near-infrared (NIR) and mid-infrared (MIR) band, respectively, so as to obtain the total column-averaged dry-air mole fraction (DMF) of H₂O [18]. Recently, many projects have been conducted in different geographical regions to compare different methods of H₂O observation. Makarova et al. (2014) made intercomparison techniques of satellite, airborne and FTIR at Tomsk for observations of H₂O [19]. Yana et al. (2017) compared the simultaneous datasets of H₂O measured by the NDACC measurement, RPG-HATPRO microwave radiometer and Novatel ProPak-V3 global navigation satellite system receiver at the St. Petersburg site [20]. Tu et al. (2020) compared column-averaged dry-air mole fractions of H₂O from the collaborative carbon column observing network (COCCON) with retrievals from the TCCON measurements and the NDACC measurements as references at two northern observation sites [21]. Therefore, long-term records of H₂O can be provided, including time and a certain degree of spatial coverage. However, there are also some challenges in retrieving H₂O based on ground-based FTIR spectroscopy, such as the instability of the retrieved vertical profile, the large error of the retrieval result and the low degrees of freedom for signal (DOFs) of the retrieval result [22,23].

In this study, we focused on the FTIR technique for H₂O retrieval and on the data from the Bruker 125 HR spectrometer. We analyzed the FTIR retrievals of NIR and MIR, provide their spectral and column average kernels, and compared the column-averaged dry-air mole fractions of H₂O (XH₂O) in the two spectral regions with the aim of understanding the H₂O vertical distribution in the atmosphere. Focused on the measurement of H₂O in MIR, we calculated its error budget and divided the atmosphere into two independent layers based on the calculated DMF. Afterward, we gave the characteristics of H₂O variation in the atmosphere, inferred the relationship between H₂O and surface temperature, and analyzed the main sources of H₂O.

Section 2 gives an introduction to the Hefei site and the measurement instrument and describes the H₂O retrieval method and the error budget. The H₂O total columns, ground layer H₂O columns and tropospheric H₂O columns are compared in Section 3, and the seasonal variation of H₂O is discussed. Moreover, the main sources of H₂O based on the back trajectories of air masses are clarified in this Section. The conclusions are made in Section 4.

2. Instrument and Retrieval Strategy

2.1. Instrument and Site Description

The observation site (31.540°N, 117.100°E, 29 m) is located in the northwest suburb of Hefei City, Anhui Province, China. The site installed a ground-based remote sensing system in January 2014, consisting of a high-resolution FTIR spectrometer (Bruker IFS 125HR, Waltham, MA, USA) and a solar tracker (A547N) produced by Bruker optics, Germany (Figure 1). The solar tracker is mounted on the roof of the laboratory and protected by a protective cover, with tracking accuracy of 0.1°. A meteorological station (ZENO, Coastal Environmental Systems, Dallas, TX, USA), on the roof of the laboratory that is near the solar tracker, has recorded information including surface temperature, surface pressure, wind speed, wind direction, humidity, rain and snow and leaf wetness data since September 2015 [22,23,24].

As shown in

Table 1. The parametric indices of the instrument and spectral data.

Instrument Parametric	NIR Spectra	MIR Spectra
Spectral range	4000–11,000 cm−1	600–4500 cm−1
Spectral resolution	0.02 cm−1	0.005 cm−1
Optical path difference	45 cm	180 cm
Beam splitter	CaF2	KBr
Detector	InGaAs	InSb/MCT
ILS	HCl	HBr

, we use the FTIR instrument to collect NIR spectra (4000–11,000 cm⁻¹) and MIR spectra (600–4500 cm⁻¹) during clear days. To maximize the signal-to-noise ratio (SNR) of the MIR spectra, the spectra are recorded using optical band-pass filters. FTIR infrared spectra have a continuum SNR of more than 80 over much of the spectral range. The NIR spectra are recorded by the FTIR instrument equipped with the calcium fluoride (CaF₂) beam splitter and indium gallium arsenide (InGaAs) detector, and a nominal spectral resolution of 0.02 cm⁻¹ (optical path difference of 45 cm) [22,23]. The MIR spectra with a spectral resolution up to 0.005 cm⁻¹ (optical path difference of 180 cm) are recorded by liquid-nitrogen-cooled indium antimonide (InSb) and mercury cadmium telluride (MCT) detectors along with a potassium bromide (KBr) beam splitter [22,24]. To ensure the stability of the measurement, the instrument is vacuumized under 50 hPa. The instrument line shape (ILS) is required to diagnose the collimation state of the optical path of the spectrometer, so that the total columns and profile of gases from measurements are retrieved precisely [22,25,26]. The ILS is monitored by low-pressure HCl and HBr cell measurements for near-infrared and mid-infrared range, respectively [22,27,28].

2.2. The Retrieval Strategy of NIR H₂O Column

The GGG2020 software package uses a nonlinear least square spectral fitting algorithm (GFIT) to retrieve the H₂O total column [29]. The daily temperature and pressure profiles are also obtained from the NCEP data. Spectroscopic line parameters are adopted from the HITRAN 2012 database [30,31]. A set of spectral micro-windows for H₂O are chosen to retrieve the concentration by considering the minimum absorption of other interfering species (CH4, CO and CO2). GGG2020 calculates the total column of H₂O, and then the column-averaged DMF of the H₂O is computed as:

$$X_{H_{2}O}=\frac{{column}_{H_{2}O}}{colum{n}_{air}^{dry}}=0.2095\times \frac{colum{n}_{H_{2}O}}{colum{n}_{O_2}}$$

(1)

where $colum{n}_{H_{2}O}$ is the column concentration of H₂O and $colum{n}_{O_2}$ is the column concentration of oxygen in the atmosphere. The column of dry air, in molecules/cm², is calculated from the oxygen (O₂) column divided by 0.2095 [32]. The advantage of dividing by O₂ column abundance is that it reduces the systematic uncertainties from the parameters that affect the retrievals of both species similarly, e.g., such as instrument line shape (ILS) and the solar zenith angle (SZA). Figure 2 depicts a typical spectral fitting (observed at 04:44 UTC 3 February 2016, with an SZA of 48.82°) of H₂O in the spectral windows listed in

Table 2. The parameters for NIR H₂O retrieval.

	Center of Spectral $\mathbf{Window} (\mathbf{c}{\mathbf{m}}^{-1})$	$\mathbf{Width} \left(\mathbf{c}{\mathbf{m}}^{-1}\right)$	Interfering Gas
H2O	4571.75	2.50	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, ${\mathrm{CH}}_4$
	4611.05	2.20	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, ${\mathrm{CH}}_4$, ${\mathrm{N}}_2\mathrm{O}$
	4699.55	4.00	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, ${\mathrm{N}}_2\mathrm{O}$
	6076.90	3.85	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, ${\mathrm{CH}}_4$, HDO
	6125.85	1.45	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, ${\mathrm{CH}}_4$, HDO
	6255.95	3.60	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, HDO
	6301.35	7.90	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, HDO
	6392.45	3.10	${\mathrm{H}}_2\mathrm{O}$, HDO
	6401.15	1.15	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, HDO
	6469.60	3.50	${\mathrm{H}}_2\mathrm{O}$, ${\mathrm{CO}}_2$, HDO

. The root-mean-square error of spectral fitting residuals is 0.16% for H₂O. The average signal–to–noise ratio is 87.6.

Figure 3 gives the total column averaging kernels of H₂O observed at different SZAs. The range of SZAs for NIR spectra is mainly from 20 to 70° at the Hefei site. The sensitivity of height dependence of the retrieved profile to concentration perturbations at different atmospheric levels is represented by the total column averaging kernel [22]. The ideal total column averaging kernels should be 1.0 at each altitude, meaning that the retrieved results are the same as the true states. However, the actual total column averaging kernel of H₂O is not equal to 1.0. If the value is greater than 1.0 for a height, it means that the retrieved results overestimate the contribution of that particular layer to the total column budget, and vice versa [33]. As a result, the near-infrared retrieved H₂O total column underestimates the a priori in the lower troposphere and overestimates it at high altitudes.

2.3. The Retrieval Strategy of MIR H₂O Column

In contrast to the above method, the MIR H₂O column is retrieved by the SFIT4 (version 0.9.4.4) based on the optimal estimation method [34]. MIR spectra containing strong and well-isolated H₂O absorption lines are selected as micro-windows for H₂O retrieval [13,35,36]. Additionally, the spectral signatures of interfering gases (CH₄, N₂O, O₃ and HCl) are also considered to reduce fitting residuals.

Table 3. The parameters for MIR H₂O retrieval.

Species	H2O
Retrieval software	SFIT4 V0.9.4.4
Spectroscopy	HITRAN 2012
Temperature, pressure and H2O profiles	NECP
A priori profiles of retrieved species	WACCM v6
Spectral windows (${\mathrm{cm}}^{-1}$)	2732.28–2732.82
	2818.80–2820.13
	2878.55–2880.65
	2892.83–2893.25
Interfering gases	CH4, N2O, O3, HCl

shows the retrieval parameters and interfering gases for H₂O at the Hefei site. Four retrieval windows of H₂O were fitted at the same time. The NCEP reanalysis data were used for the a priori profiles of atmospheric temperature, pressure and H₂O [37]. The spectroscopic line parameters for H₂O and interfering species are derived from the HITRAN 2012 line list [38]. For MIR retrieval, the $X_{H_{2}O}$ results are different to NIR results, because it is difficult to retrieve ${\mathrm{O}}_2$ total column from MIR spectra. The $X_{H_{2}O}$ column is calculated as:

$$X_{H_{2}O}=\frac{colum{n}_{H_{2}O}}{colum{n}_{dry air}}=\frac{colum{n}_{H_{2}O}}{\frac{p_s}{g{m}_{air}^{dry}}-\frac{colum{n}_{H_{2}O}{m}_{H_{2}O}}{m_{air}^{dry}}}$$

(2)

where $p_s$ is the surface pressure; g is the column–averaged gravitational acceleration; $m_{H_{2}O}$ and $m_{air}^{dry}$ are the molecular masses of H₂O and dry air, respectively; and $colum{n}_{H_{2}O}$ is the total column of H₂O observed from NCEP reanalysis data. The ground surface pressures are recorded by the meteorological station, with an accuracy better than 0.1 hPa. A typical H₂O absorbed spectrum was collected at 03:25 UTC 31 December 2020, with a SZA of 56.22° (Figure 4). The root–mean–square errors are 0.267%, 0.372%, 0.452% and 0.205%, respectively, showing that the spectrum is well–fitted.

Figure 5 presents the measured H₂O profiles and the a priori profile during the observation period from 2016 to 2020 at the Hefei site. As shown in Figure 5, the concentrations of H₂O are small and very close to the a priori profile above the tropopause, with concentrations in the ground layer that are higher than those in the troposphere. H₂O mostly exists in the ground layer. The averaging kernels matrix is used to depict the sensitivity of height dependence of the retrieved profile to concentration perturbations at the various atmospheric levels [22,39]. The typical averaging kernel matrix of H₂O is shown in Figure 6. The most sensitive height range for H₂O is mainly located below 10km. H₂O retrievals have a good sensitivity to the whole troposphere and lower stratosphere. The trace of averaging kernel matrix is characterized as DOFs [22]. As is shown in Figure 7, the typical DOFs for H₂O over Hefei are about 2.53. This indicates that the retrieved profile can be separated into two independent partial columns, because the criterion for separating columns into separate layers is that the column DOFs of gases is greater than 1.0 [22,40,41]. We calculate the ground layer H₂O DMF from 0 to 2.95 km with the DOFs of 1.06, and the tropospheric H₂O DMF from 2.95 to 15 km.

2.4. Error Analysis of MIR H₂O Retrieval

The retrieval errors of MIR results contain random and systematic sources, and the error analysis is performed using a posterior error estimation method based on Rodgers, which calculates errors by attributing them to each parameter used in the retrieval [39]. The retrieved H₂O column is in fact an estimate of a state smoothed by the averaging kernel. The smoothing error ($S_s$) is the difference between these two states.

$$S_s=\left(A-I\right)S_a{\left(A-I\right)}^T$$

(3)

$S_a$is a prior covariance matrix, I represents a unit matrix and A is the averaging kernel. The retrieval parameter error is the errors in the retrieval parameters, such as temperature profile, SZA, spectroscopic line parameters and field of view [42]. This error covariance matrix is calculated as:

$$S_b=\left(G_y{K}_b\right)S_b{\left(G_y{K}_b\right)}^T$$

(4)

$S_b$ is the error covariance, $G_y$ is the sensitivity of measurement results to measurement and $K_b$ is the matrix of the weighting function of the parameters of the forward model. As shown in

Table 4. Random uncertainty and systematic uncertainty in the error analysis.

Parameter	Random Uncertainty	Systematic Uncertainty
Temperature profile	3K	3K
Solar zenith angle	0.025°	0.025°
Solar line shift	0.005 ${\mathrm{cm}}^{-1}$	0.005 ${\mathrm{cm}}^{-1}$
Solar line strength	0.1%	0.1%
Field of view	0.01	0.01
Line intensity	-	1%
Line T broadening	-	10%
Line P broadening	-	3%
Spectroscopic parameters	2%	2%

, for the temperature distribution used in a priori, we set the uncertainty as 3K. The systematic uncertainty values of the line intensity, line T broadening and line P broadening solar line are selected as 1%, 10% and 3% [22]. By analyzing the saturated absorption line, we concluded that the solar zenith angle, solar line shift and solar line strength uncertainty are selected as 0.025°, 0.005 ${\mathrm{cm}}^{-1}$ and 0.1%. We also considered the field of view uncertainty about 0.001. In addition, we used uncertainty values of 2% for the spectroscopic parameters of H₂O [15].

The results of the error analysis for a typical H₂O retrieval (03:25 UTC 31 December 2020, with a solar zenith angle of 56.22°) are summarized in

Table 5. Mean random and systematic errors for H₂O retrieval.

Parameter	Random Error/%	Systematic Error/%
Smoothing error	0.026	-
Measurement error	0.05	-
Temperature profile	6.43	1.08
Solar zenith angle	0.049	0.049
Zero level shift	0.72	0.72
Field of view	0.017	0.017
Line intensity	-	0.083
Line T broadening	0.023	0.023
Line P broadening	0.964	0.964
Spectroscopic parameters	0.968	0.968
Subtotal error	6.49	1.55
Total error	6.67

. The total random and systematic errors for H₂O column are about 6.49% and 1.55%, respectively. The very low retrieval error has already been proven in previous studies [43,44]. It is clear that the temperature profile uncertainties are the main random error and system error sources, accounting for 6.43% and 1.08%, respectively. The errors due to uncertainties of smoothing error and measurement error are very small, accounting for 0.026% and 0.05%. The systematic error of H₂O is mainly determined by the uncertainty of the temperature profile.

3. Results and Discussion

3.1. Time Series of XH₂O

The DMF of H₂O retrieved from near-infrared spectra was compared with the values retrieved from mid-infrared spectra. For NIR-spectra-derived products, O₂ total columns are simultaneously retrieved with the target species. The NIR-spectra-derived XH₂O are calculated based on Equation (1). For MIR-spectra-derived products, there are no N₂ or O₂ absorption windows that allow sufficient accuracy for column abundance retrieval. Therefore, the MIR-spectra-derived DMF of H₂O are calculated based on Equation (2). For quantitative comparisons, we used the co-located daily means of FTIR measurements at the site. It should be noted that the FTIR instrument collects direct solar radiation, and the measurements depend on clear skies.

The time series of near-infrared and mid-infrared spectra-derived XH₂O data from January 2016 to December 2020 are depicted in Figure 8. Some data are missing from the ground-based FTIR data due to high SZA conditions and poor weather. The variation trend of XH₂O obtained from near-infrared spectra is similar to that obtained from mid-infrared spectra, both showing obvious seasonal changes over the observation period. There are higher amounts of atmospheric XH₂O in summer than in other seasons, and the seasonal changes in the five years are consistent. For NIR-spectra-derived results, XH₂O peaked each August, and the daily averaged XH₂O reached the peak with the range of 7439.34~8443.83 ppm, while there is only 525.17~907.52 ppm for the minimum XH₂O in each early spring and winter. XH₂O amounts from mid-infrared spectra are slightly higher than the corresponding values from near-infrared spectra, especially in summer. XH₂O from mid-infrared spectra reach up to 7894.22~9665.07 ppm each August 2016 to 2019; meanwhile, it reduced to 526.15~875.60 ppm for the minimum in each early spring and winter. This corresponds to the seasonal variation in temperature, which is the major reason for H₂O variation in the atmosphere [21]. The mean bias, defined as the average difference in XH₂O between simultaneous MIR and NIR data, is approximately 480.52 ppm. The difference is due to using different calculation methods. In addition, the SZA differences also cause the difference of XH₂O from mid-infrared and near-infrared spectra, because both retrievals are dependent on solar zenith angles [45].

As shown in Figure 9, the time series of MIR spectra derived XH₂O are compared with the NIR spectra derived data observed from January 2016 to December 2020. To co-locate the MIR data with the NIR data, the daily averaged data observed on the same day were directly compared, and the datasets that have at least four individual measurements within one day are used for the data average. To eliminate the effect of different a priori profiles and averaging kernels on XH₂O, we used the a priori profile of the MIR retrievals to correct the NIR retrieved data [46]. There is a high correlation between MIR and NIR data (R = 0.989). It is concluded that MIR-spectra-derived XH₂O agrees well with the NIR observations. These time series show that MIR and NIR spectra measurements are able to capture the seasonal variations of atmospheric H₂O. However, MIR and NIR data show a higher correlation coefficient in the winter period than that in the summer, and the slope ratio is closer to one in the winter. This is mainly due to the MIR observations having higher H₂O values than the NIR results in summer.

3.2. Time Series of H₂O in the Ground Layer, the Entire Atmosphere and the Troposphere

The ground layer is the most important channel in the atmospheric process of H₂O transmission. In addition, ground-layer H₂O is key to learning the water cycle, helping us to understand the mixing of air masses, dehydration pathways and free-atmosphere moisture [47]. Ground-layer H₂O concentrations are very variable in the middle latitude, and the concentrations of H₂O on cold and dry days are far lower than those on warm and wet days [48].

The daily averaged time series of ground-layer H₂O columns observed from January 2016 to December 2020 are shown in the lower panel of Figure 10. The time series of H₂O columns in the ground layer is fitted by a lowpass filtering fast Fourier transform (FFT) technique in order to reveal the seasonal variation of H₂O in the ground layer over the Hefei site [22,49]. The FFT curve equation is:

$$f\left(x\right)=Hx+C+{{\displaystyle \sum }}_{k=0}^n{a}_k\cos \left(2\pi kx\right)+b_{k}sin\left(2\pi kx\right)$$

(5)

where $x$ represents the annual score of the date, H represents the long-term growth trend, and $a_k$ and $b_k$ represent the coefficients of variation in seasonal variation. In the FFT fitting, the cut-off frequency with 90-day cycles was used to calculate the seasonal variation [22,50]. XH₂O columns in the ground layer reach up to the maximum of 14496.66–18365.01 ppm in each August, while there is only 1277.42–2026.47 ppm for the minimum XH₂O in each early spring and winter. Tu et al. (2020) obtained XH₂O columns from Kiruna and Sodankyla stations from 2017 to 2020 [21], and the variation trend of XH₂O columns at the two sites is consistent with that observed at the Hefei site. The maximum value appears in summer and the minimum value appears in winter. Within the same measurement time, the XH₂O of the observed station decreased from August to January and reached the lowest value. The reason for the similar H₂O seasonal variation across the three sites is mainly because the Hefei, Kiruna and Sodankyla stations have the same atmospheric cycle model.

To compare the characteristics of the H₂O distribution at different altitudes. The XH₂O through the entire atmosphere obtained from MIR spectra and XH₂O in the ground layer are also shown in Figure 10. The differences between the two coincident data are shown in the upper panel. As shown in Figure 10, both the XH₂O in the ground layer and in the entire atmosphere show obvious seasonal variation, that is XH₂O is higher in summer and lower in winter and early spring, and the seasonal changes in the five years are consistent. XH₂O reached its lowest value in the middle of January, then gradually increased and reached its peak at the end of July, then began to decline gradually, and finally reached the lowest values in the middle of January of the next year.

The differences in XH₂O between the ground layer and the entire atmosphere show seasonal variations, with the peak in August, and the minimum in early spring and winter. The mean difference is 4113.64 ± 2834.67 ppm. The maximum difference is within the range of 6602.44 ppm~8699.95 ppm and the minimum difference is in the range of 751.27~1150.87 ppm. The averaged difference is 8246.70 ppm in summer, and only 622.97 ppm in winter.

The correlation analysis of XH₂O values in the ground layer and in the entire atmosphere is also illustrated in Figure 11. The two datasets show good correlation, with an R–value of 0.923. It is clear that the correlation coefficient is relatively higher in winter (December–February), which is due to the small differences in the two XH₂O values in winter. However, the correlation is poor in summer (June–August), because the two XH₂O values differ largely in this season.

We also calculated the daily averaged XH₂O in the troposphere from January 2016 to December 2020. Figure 12 compares the daily averaged XH₂O in the troposphere and ground layer. Similarly, the tropospheric and ground layer H₂O columns show obvious seasonal variation, i.e., XH₂O is higher in summer and lower in winter and early spring, and the seasonal variation in the five years is consistent. Compared with the ground-layer H₂O concentrations, the H₂O concentrations in the troposphere are significantly lower than those in the ground layer, with an average difference of 5292.53 ± 3575.93 ppm. The average difference is 9774.72 ppm in summer, and only 1630.48 ppm in winter. The H₂O columns in the ground layer are higher than those in the troposphere and the entire atmosphere, especially in summer. This is because H₂O is mainly distributed in the ground layer [51], and transpiration occurs in summer, mainly at the surface, so a moist atmosphere near the surface leads to more H₂O in the ground layer [27,43]. The correlation coefficient between XH₂O values in the ground layer and troposphere is 0.841 (see Figure 13), and the correlation is relatively higher in winter and lower in summer. The lower correlation coefficient (0.841) between XH₂O values in the ground layer and troposphere compared to the correlation (0.923) between XH₂O values in the ground layer and the entire atmosphere, shown in Figure 11, indicates that ground layer H₂O contributes more to the entire atmosphere than to the troposphere.

3.3. Relationship with Surface Temperature

The temperature variation has a clear influence on the seasonal variation of atmospheric H₂O [20]. The spatial and temporal distribution of H₂O in the atmosphere has a high correlation with weather conditions [52]. The temperature data were all obtained from the meteorological station in the Hefei site, and the time series of surface temperature from January 2016 to December 2020 is shown in Figure 14. It is found that the seasonal variation in surface temperature is similar to that of XH₂O. In order to analyze the correlation between XH₂O and surface temperature at the Hefei site in detail, we compared the ln(XH₂O) values in the ground layer, entire atmosphere and troposphere with surface temperature from January 2016 to December 2020, respectively. Figure 15a–c illustrate the three correlations. The correlation coefficient R-value between the ground layer ln(XH₂O) and temperature is 0.893, the R–value between the ln(XH₂O) derived from the total column and temperature is 0.882, and the R–value between the tropospheric ln(XH₂O) and temperature is 0.683. This demonstrates that the correlation between H₂O columns in the ground layer and surface temperature is the highest, while the correlation of H₂O columns in the troposphere with surface temperature is the weakest. Therefore, it can be seen that the temperature change is more likely to cause the change in H₂O columns in the ground layer, which explains why the XH₂O amplitude in the ground layer is higher than those in the entire atmosphere and the troposphere [47].

3.4. The Impact of Air Mass Transport on H₂O

To understand the impact of air mass transport on H₂O change in Hefei, we used the 500 m height as the starting height of the atmospheric boundary layer in the study area and simulated the 72 h backward trajectories of the air masses using the hybrid single-particle Lagrangian integrated trajectory (HYSPLIT) [53]. HYSPLIT is a comprehensive model system jointly developed by the Air Resources Laboratory of the National Oceanic and Atmospheric Administration (NOAA) of the United States and the Australian Meteorological Administration, which is used to calculate and analyze the source, transport and diffusion trajectories of atmospheric pollutants [54]. The HYSPLIT model has a relatively complete transport, diffusion and sedimentation model that can handle various meteorological input fields, physical processes and different types of pollutant emission sources. It is widely used in the study of atmospheric transport and diffusion of various pollutants in various regions [55].

The path of air mass’s impact on an area is mainly related to the position and movement law of the local pressure system, as well as the characteristics of topography and local circulation. The trajectory calculation principle of this model is based on the assumption that particles in the air move with the wind and are at the position P(t) at time t, and the position $P\left(t+\Delta t\right)$ after a variable time step $\Delta t$ is:

$$P\left(t+\Delta t\right)=P\left(t\right)+1/2 \left[V\left(P,t\right)+V\left({P^}^{\prime },t+\Delta t\right)\right]\Delta t$$

(6)

where V is the wind speed, ${P^}^{\prime }$ is an intermediate conjecture position and the variable time step $\Delta t$ during the calculation process is:

$$P\left(t+\Delta t\right)=P\left(t\right)+V\left(P,t\right)\Delta t$$

(7)

We use Ward’s minimum variance method in the clustering analysis [56]. Since we only focus on the horizontal direction of air mass transportation to Hefei station, and in order to avoid Euler distance from dividing two backward trajectories with different speeds and the same path into two different categories in the calculation process, the angle distance algorithm is used to classify the air-flow trajectories to Hefei station and obtain different types of transportation air-flow [57]. Its calculation method:

$$d_{12}=\frac{1}{n}{{\displaystyle \sum }}_{i=1}^n{\cos }^{-1}\left(\frac{1}{2}\frac{A_i+B_i+C_i}{\sqrt{A_i{B}_i}}\right)$$

(8)

$$A_i={\left(X_1\left(i\right)-X_0\right)}^2+{\left(Y_1\left(i\right)-Y_0\right)}^2$$

(9)

$$B_i={\left(X_2\left(i\right)-X_0\right)}^2+{\left(Y_2\left(i\right)-Y_0\right)}^2$$

(10)

$$C_i={\left(X_2\left(i\right)-X_1\left(i\right)\right)}^2+{\left(Y_2\left(i\right)-Y_1\left(i\right)\right)}^2$$

(11)

where $X_0$ and $Y_0$ is the coordinate of the starting point of the trajectory, and $d_{12}$ is the average angle between two trajectories.

The weather data used are the global data assimilation system GDAS data from 2017 to 2018 provided by the NECP, with a spatial horizontal resolution of a 0.5° × 0.5° longitude and latitude grid [58]. The backward trajectory cluster analysis was conducted in four seasons (Figure 16) based on HYSPLIT: March to May is spring, June to August is summer, September to November is autumn and December to February of the following year is winter.

The spring clusters can be divided into three types of air masses: the first type of air masses, which came from Inner Mongolia and reached Hefei via the Shanxi and Henan provinces, accounts for 24.72%; the second type of air masses, which came from Inner Mongolia and reached Hefei via the Beijing–Tianjin–Hebei region, and the Shandong and Jiangsu provinces, accounts for 37.08%. Therefore, the first 61.8% of airmasses came from dry and cold regions, which leads to low H₂O values in spring. The third type of air mass accounts for 38.20% and came from the border area of the Anhui and Jiangxi provinces and went by the south of Anhui Province.

The summer clusters are divided into three air masses, with the first group accounting for 40.91% of the air masses and arriving in Hefei directly from Henan Province. The second type of air mass accounts for 21.59% and came from the Yellow Sea and entered Hefei via the Jiangsu province; the third type of air mass accounts for 37.50% and came from the South China Sea and entered Hefei via the Guangdong, Fujian and Jiangxi provinces, so the last 59.09% of airmasses came from the wet and warm regions, which leads to high H₂O values in summer.

The autumn clusters are divided into three air masses, the first category with 27.91% was from Mongolia, arriving in Hefei via Inner Mongolia Province, Hebei Province, Shandong Province and Jiangsu Province; the second category with 39.53% was from Inner Mongolia Province, arriving in Hefei via the Beijing–Tianjin–Hebei region, the Yellow Sea and Jiangsu Province. Therefore, the first 61.44% of airmasses came from the dry and cold regions, which leads to low H₂O values in autumn, similar to the pattern in spring. The third type of air mass comes from the border area of Anhui and Jiangxi with a percentage of 32.56%.

Winter can also be divided into three air masses; the first and second types of air masses both come from Mongolia. The first type of air masses accounts for 18.39% and reached Hefei via the Inner Mongolia, Shaanxi, Shanxi and Henan provinces, the second type of air masses accounts for 40.23% and reached Hefei via the Inner Mongolia, Shaanxi, Hebei, Shandong and Jiangsu provinces. The third type of air mass, accounting for 41.38%, came from Shandong Province and entered Hefei via Jiangsu Province. Therefore, almost 100% of airmasses came from the dry and cold regions, which leads to very low H₂O values in winter.

4. Conclusions

The H₂O total columns, ground-layer H₂O columns and tropospheric H₂O columns were retrieved from the solar absorption spectra recorded by the ground-based high-resolution FTIR at the Hefei site. The H₂O total columns were retrieved from NIR and MIR spectra, respectively. The average root-mean-square deviation of fitting residuals of NIR and MIR H₂O spectra are 0.16% and 0.324%, respectively. In the retrieval process of the MIR H₂O column, we found that the typical DOFs for H₂O over Hefei are about 2.53. Therefore, the column is divided into the ground layer and the troposphere. In addition, we conducted an error analysis on the MIR H₂O column and found that the random and systematic errors are about 6.49% and 1.55%, respectively. Among them, the uncertainties of the temperature profile are the main random error and system error sources, accounting for 6.43% and 1.08%, respectively. The errors due to uncertainties of smoothing error and measurement error are very small, accounting for 0.026% and 0.05%. The systematic error of H₂O is mainly determined by the uncertainty of the temperature profile.

The time series of the H₂O total column and tropospheric H₂O column were obtained from January 2016 to December 2020. It is clear that the trend of XH₂O measured in the NIR region is similar to that measured in the MIR range, both showing obvious seasonal variations over the observation period, and having a high correlation between the two datasets (R = 0.989). The amounts of atmospheric XH₂O are higher in summer than those in spring and winter, and the seasonal variations in the five years are consistent.

For ground-layer H₂O columns, the amounts of XH₂O reach up to the maximum of 14496.66~18365.01 ppm in each August, and the daily average XH₂O reached the peak of 8713.48 ppm in August, while it is only 1277.42~2026.47 ppm for the minimum in each early spring and winter. The XH₂O values in the ground layer are obviously higher than those in the entire atmosphere and troposphere, the averaged difference is 4113.64 ppm and 5292.53 ppm, respectively (the standard deviation is 2834.67 ppm and 3575.93 ppm, respectively). This is because H₂O is mainly distributed below the troposphere, and H₂O transpiration mainly occurs near the surface, which is related to the enhanced contribution of the moist atmosphere near the ground to the XH₂O value.

The temperature variations have a strong influence on the seasonal variation of atmospheric H₂O, so the relationship of H₂O and surface temperature was also calculated, and it is found that the seasonal variation in temperature is similar to H₂O variation. The relationship between surface temperature and ln(H₂O) values in the ground layer, entire atmosphere and troposphere all show a significantly positive correlation and the correlation coefficient R-values are 0.893, 0.882 and 0.683, respectively. The change in temperature likely causes the change in H₂O in the troposphere.

Further, we selected the HYSPLIT model to simulate the back trajectories of the air parcels at the Hefei site in different seasons and performed the cluster analysis. Firstly, about 61.8% of airmasses came from the dry and cold regions, which leads to low H₂O values in spring. Secondly, about 59.09% of airmasses came from the wet and warm region, which leads to high H₂O values in summer. Thirdly, about 61.44% of airmasses came from the dry and cold regions, which leads to low H₂O values in autumn, similar to the pattern in spring. Finally, almost 100% of airmasses came from the dry and cold regions, which leads to very low H₂O values in winter.

FTIR is a powerful tool for observing H₂O in the atmosphere, and long-term monitoring of H₂O provides the basis for unravelling the atmospheric water cycle. The combined observations of atmospheric H₂O and H₂O isotopes can reveal the processes of troposphere–stratosphere exchange, cloud processes, rain recycling and evapotranspiration. H₂O isotopic measurements can be used to effectively distinguish the representation of processes controlling the distribution of atmospheric humidity in different models. Our future research will focus on the accurate retrieval of the H₂O stable isotopes HDO, H₂¹⁶O and H₂¹⁸O from the solar absorption spectrum, which together with H₂O and its stable isotopes, can clarify the atmospheric water cycle.