The WLG station is sited at the top of Mt. Waliguan (36∘17′ N, 100∘54′ E, 3816 m a.s.l.), located in Qinghai Province of China (Fig. 1a). It is one of the WMO GAW global baseline stations and the only one in the hinterland of the Eurasian continent. Mt. Waliguan is an isolated mountain with an elevation of about 600 m relative to the surrounding landmass; it is surrounded by highland steppes, tundra, deserts, and salt lakes (Fig. 1b). With a low population density of about 6 capitals km-2 and hardly any industry within 30 km, WLG has the advantage of being rather isolated from industry, forest, and population centers. It is relatively dry, windy, and short on precipitation with a typical continental plateau climate (Tang et al., 1995). Xining City (the capital of Qinghai Province, located about 90 km northeast of WLG) and Lanzhou City (the capital of Gansu Province, located about 260 km away to its east) are considered to be the nearest large pollution sources that may have impacts on the WLG site. There are several high mountains (∼4000 m a.s.l.) between Xining and Mt. Waliguan. Total column ozone, surface ozone, solar radiation, precipitation chemistry, greenhouse gases, aerosol optical depth, and aerosol scattering and absorption coefficient together with basic meteorological parameters have become operational measurement items at WLG in succession since the year 1991 (Tang et al., 1995). In addition to routine observations, intensive measurements and model analyses were performed to investigate the regional and global representativeness of WLG and the effects of chemical transformations and physical and transport processes on various atmospheric compositions, e.g., surface ozone, short-lifetime reactive gases (such as NO, NO2, SO2, CO, H2O2, HNO3, HCHO, other carbonyls, and non-methane hydrocarbons – NMHCs), greenhouse gases, persistent organic pollutants, metal and isotopes, and aerosols.

We used the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis data to investigate meteorological conditions over the WLG site, with data for 3–4 km altitude representing the ground level at WLG. As shown in Fig. S1 in the Supplement, temperature is high in summer (around 283 K) and low in winter (around 265 K), pressure is high in summer (around 643 hPa) and low in winter (around 635 hPa), wind speed is low in summer (around 3 m s-1) and high in winter (around 5 m s-1), and wind direction (0 means the wind is blowing from the north) is from the southeast in summer (around 140∘) and from the west in winter (around 260∘). These seasonal variation characteristics are similar to those in earlier years at the station as reported in previous work (Tang et al., 1995). We evaluated the ECMWF reanalysis data for 3–4 km altitude using meteorological data from in situ measurements at the WLG station, and we found that the ECMWF data are in good (temperature, pressure) and reasonable (wind speed and direction) agreement with in situ data (see Fig. S2).

We started the ground-based MAX-DOAS measurement program at WLG on 26 September 2010. An automated and compact (13 cm × 19 cm × 14 cm) mini MAX-DOAS instrument from Hoffmann Messtechnik GmbH in Germany, which had been used at the Gucheng site in the North China Plain (Jin et al., 2016), was moved to and installed at WLG (Fig. 1c). This instrument is designed for the spectral analysis of scattered sunlight and the application of the MAX-DOAS technique (Hönninger et al., 2004). The same type of instrument was used in previous studies, e.g., in Beijing and the surrounding area (Ma et al., 2013; Jin et al., 2016). The entrance optics, fiber-coupled spectrograph, and controlling electronics are hermetically sealed in a metal box of about 3 L volume. A stepper motor, mounted outside the box, can rotate the whole instrument to control the elevation viewing angle, i.e., the angle between the horizontal and the viewing direction, and thus it can scan vertically at different elevation angles. The spectrograph covers a wavelength range of 290–437 nm and its entrance slit has a width of 50 µm. A Sony ILX511 charged-coupled device (CCD) detects the light in 2048 individual pixels. The whole spectrograph is cooled by a Peltier stage to maintain a stable temperature of the optical setup and to guarantee a small dark current signal. The measurement process and spectra data logging are controlled by a laptop using the MiniMax software package developed by Udo Frieß at the Institute of Environmental Physics, the University of Heidelberg in Germany.

The instrument was mounted on a bracket, fixed on the building roof, at an azimuth viewing direction exactly towards the north. After a winter of pilot run, we added heating elements to the outside of the instrument so that the temperature of the spectrograph could be kept at a stable but not very low value, e.g., -10 ∘C in winter. In other seasons, the temperature of the optical setup was set at a higher value, typically 0 ∘C, below the ambient temperature. Dark current spectra were measured using 10 000 ms and one scan and electronic offset spectra with 3 ms and 1000 scans. Measurements of these signal spectra were made generally month by month or whenever the working temperature of the instrument was changed. Over the pilot run period in the years 2010 and 2011, measurements had been made with the same sequence of elevation angles as used at the Gucheng site in the North China Plain (Jin et al., 2016), with no elevation angles lower than 3∘ available. After the beginning of the year 2012, the elevation angles were set to be -1, 0, 1, 2, 3, 5, 10, 20, 30, and 90∘ in a sequence. Unfortunately, during the data analysis it turned out that the elevation calibration was wrong by -4∘ (see Fig. S3). Thus finally only a few elevation angles from the original selection were found to be above the horizon. After the correction by -4∘ the remaining elevation angles are 1, 6, 16, 26, and 86∘. From the comparison between the measured and simulated elevation angle dependencies of the different quantities (see Fig. S3), we estimate the accuracy of the corrected elevation angles to be within ±0.5∘. The exposure time for each elevation angle was about 1 min. The data from 3 years of measurements over the period April 2012 through April 2015 are used for this study.

We investigated the effect of the surface topography in more detail. In addition to the setup with the “true” topography (Fig. 3), we also performed simulations with flat surfaces at sea level or 3700 m altitude, because these options might be used as alternative scenarios for radiative transfer models without the option to consider the true topography. In all three setups the atmospheric properties and the altitude of the detector (3800 m) are kept the same. In Fig. 4 the O4 AMFs for the simulations with flat surfaces are plotted versus the O4 AMFs for the true surface topography. While the results for a flat surface at 3700 m are very similar to those for the true topography, the results for a flat surface at sea level show systematically higher values. With increasing aerosol load these differences even increase. This finding can become very important for the interpretation of measurements in mountainous environments if no radiative transfer simulations considering the true topography are available. However, this finding needs further investigations, especially if an instrument is operated on a more isolated mountain. In such cases, the agreement between simulations with true topography and flat surface at high altitude might be worse. Further studies should also investigate the effects for trace gases with different altitude profiles.

Besides the simulation of the air mass factors, we also compared the results for other simulated quantities: in Fig. 5 the dependence of the simulated radiance (top) O4 AMF (center) and color index (bottom) is shown as a function of the elevation angle (including negative elevation angles). Again, the results for the true topography and the flat surface at 3700 m altitude are very similar, while the results for a flat surface at sea level are systematically different. The results of these comparisons indicate that for our measurements the assumption of a flat surface at the approximate altitude of the surrounding terrain is a very good approximation for the true surface topography. However, for zero and negative elevation angles, the consideration of the true topography might become important. Here it is interesting to note that these differences will probably increase for measurements in other mountainous scenarios, especially for more measurements at isolated mountains. These effects should also be investigated in more detail in future studies.

Cloudy-sky conditions can be identified and classified by different quantities (see, e.g., Gielen et al., 2014; Wagner et al., 2014, 2016). In this study, to minimize the computational effort, we only use the color index (CI) measured in the zenith direction (note that the measurements in the zenith direction were not used for trace gas retrievals because of suspected direct sun impact). We chose the wavelength pair 330 and 390 nm: 1CI=signal330nmsignal390nm. Moreover, in order to minimize the potential effects of instrument degradation, the absolute value of the CI is not used for the cloud classification. Instead, two derived quantities are calculated.

For the low elevation angles the atmospheric visibility and thus the length of the light path depend strongly on the aerosol load. Thus measurements with high aerosol loads have decreased sensitivity to the trace gas absorptions and thus have to be identified and removed from further processing. For that purpose the retrieved O4 absorption is used (Wagner et al., 2004; Lampel et al., 2018). In the following, a threshold for the retrieved O4 dSCD at a 1∘ elevation angle of 1.4×1043 molec.2 cm-5 is used, which corresponds to an O4 dAMF of 1.2 (see Fig. S27). This threshold represents an AOD of about 0.1 at 360 nm.

In Fig. S28 the seasonal variation in the O4 dSCDs at 1∘ elevation is shown. High values are typically found in winter, indicating low AOD. In other seasons smaller O4 dSCDs are found, indicating higher AOD. This seasonal dependence is in good agreement with measurements of the AOD (Che et al., 2011).

In Fig. 6 the seasonal variation in the sky conditions is shown. It is derived by applying the cloud and aerosol classification algorithms described above. The statistics is based on the number of observations (at 1, 6, or 16∘ elevation angle) of spectra averaged from 10 original spectra (April 2012–April 2015). Only measurements with more than 800 scans are considered. The basic colors indicate the cloud properties (clear, broken clouds, continuous clouds). The full or light colors indicate observations with low or high aerosol loads, respectively.

While the absolute frequency of clear-sky observations (low and high aerosol load) stays almost constant over the year, the relative fraction changes strongly with the highest probability of clear-sky observations in winter and the lowest probability in summer. The relative fraction of low aerosol cases is largest in winter. The corresponding seasonal frequency plots for the different trace gas analyses (after application of the individual rms filters) are shown in Fig. S29. While the absolute amount of valid data is different for the different analyses, the seasonal frequency is almost the same.

In Fig. 7 (left) time series of daily averages of the individual trace gas dSCDs for a 1∘ elevation angle are shown. On the right side, the corresponding monthly mean values are shown. In Fig. 8 the seasonal cycles for all elevation angles (1, 6, 16∘) are shown for clear sky. In this figure, the systematic uncertainties of the trace gas dSCDs are also indicated by the blue dotted lines. The systematic uncertainties can be regarded as indicators for the lower bounds of the detection limit (which might be reached if a large number of measurements are averaged). For NO2, SO2, and HCHO, approximate mixing ratios derived for the measurements at 1∘ elevation angles are also indicated by the y axes at the right side (see Sect. 4). The results for broken clouds are similar to those for clear sky (see Fig. S30). This is useful information to confirm the results for clear sky, since for broken clouds, the atmospheric light paths are often similar to those for clear sky.

The main findings are as follows. i.

For NO2 and HCHO higher dSCDs are found for lower elevation angles (see also Fig. S20), and this indicates enhanced trace gas concentrations in the lower troposphere (at least in the atmospheric layers between about 4 and 5 km a.s.l.) compared to the upper troposphere.

In this section the lower tropospheric mixing ratios of the NO2, SO2, HCHO, and BrO at WLG are estimated based on the respective dSCDs at 1∘ (see Fig. 8) and the relationships between the dSCDs and the tropospheric mixing ratios (Sect. 4).

Below are our estimates.  

NO2. The dSCDs at 1∘ between 0.2×1015 molec. cm-2 (January) and 3×1015 molec. cm-2 (May) correspond to mixing ratios between about 7 (January) and 100 ppt (May).

Measurements of the very reactive trace gases are very scarce at remote sites like WLG. Table 2 summarizes the mixing ratios of NO2, SO2, HCHO, and BrO at WLG measured by MAX-DOAS presented in this study in comparison to those recorded in other studies using different methods.

The passive sampling method was used to measure the weekly and monthly mean mixing ratios of NO2 and SO2 at WLG (Ma et al., 2002a; Meng et al., 2010; Lin et al., 2013). The monthly mean mixing ratios and standard deviations of NO2 at WLG were 0.022±0.010 ppb in January and 48±17 ppt in July for the year 1996 (Ma et al., 2002a). For the measurement in 1996, the NO2 was sampled with filter packs (SP) with exposure times of typically 3–5 d and then analyzed by ion chromatography (ICG) equipment (Yu et al., 1997). During the measurement experiment in 2008, the samplers were exposed about 10 d in a month. The extracted NO2 samples (Ogawa passive samplers) were analyzed using a spectrophotometer (SPM) while the extracted SO2 samples were analyzed by ICG (Meng et al., 2010). The average mixing ratios of NO2 and SO2 at WLG in 2008 typically varied in the ranges of 0.6±0.4 and 0.7±0.4 ppb, respectively (Meng et al., 2010). Long-term continuous measurements, by filter sampling with exposure times of 3–5 d followed by ICG chemical analysis, showed that the mixing ratio of SO2 varied typically in a range of 0.45±0.14 ppb at WLG from 1997 to 2009 (Lin et al., 2013). The NO2 levels obtained from this study are in accordance with those recorded by Ma et al. (2002a), but an order of magnitude lower than reported by Meng et al. (2010). The time differences between Meng et al. (2010) and the other two studies seem not to be the main cause. The sampling filters were likely to be polluted by local human activities, such as increasing frequency of occasional cars and vans to the station, which could not be excluded from the recording due to their long exposure time. The level of SO2 might also be overestimated by Meng et al. (2010) considering that there had not been a significant increasing trend for SO2 at WLG during 1997–2009 (Lin et al., 2013). It can be deduced that the SO2 level from this study is reasonable with comparison to the results from Lin et al. (2013).

The daytime NO and NO2 mixing ratios at WLG in the summer of 2006 were 71±45 ppt and 0.28±0.13 ppb, respectively, as measured by chemiluminescence (CLS) analyzer coupled with a photolytic converter (PhC) (Xue et al., 2011). Also measured by CLS, the daytime average NO mixing ratios at WLG were 0.072±0.079 ppb in the late spring and 0.047±0.032 ppb in the summer of 2003 (Wang et al., 2006; Xue et al., 2013). Assuming the same NO2/NO ratio (i.e., 3.9) as in the summer of 2006, the NO2 mixing ratios at WLG would be 0.28±0.31 ppb in the late spring and 0.18±0.12 ppb in the summer of 2003. The NO2 levels obtained from the CLS method are much higher than those from MAX-DOAS presented in this study. This can be explained by the fact that in situ measurements at WLG are strongly influenced during the daytime by the underlying boundary layer, where NO2 levels are high due to soil emissions and other occasional pollution sources. In contrast, the upwelling air masses have a relatively small influence on the MAX-DOAS measurements, which represent a long optical path.

The ambient concentrations of carbonyl compounds were measured at WLG in 2005, by using the silica gel cartridge–high-performance liquid chromatography (SGC–HPLC) method. The means and standard deviations of HCHO mixing ratios were 4.16±1.89 ppb in August–September and 1.48±0.42 ppb in December (Mu et al., 2007). The HCHO mixing ratios obtained by the SGC–HPLC method are nearly an order of magnitude higher than those by MAX-DOAS presented in this study. Mu et al. (2007) reported that the variability of HCHO concentrations was significant (by a factor of >2) if the air samples were collected at different places at the site. HCHO forms through oxidation reactions associated with methane and various NMVOCs, most of which (e.g., isoprene) have high spatial variability in abundance due to short lifetimes and thus can influence local HCHO concentrations significantly. Therefore, isoprene and other very active NMVOCs emitted from natural vegetation can accumulate near the surface and result in high levels of HCHO by oxidation. While in situ sampling methods as used by Mu et al. (2007) measured HCHO on the ground level within the boundary layer, MAX-DOAS measured the HCHO mixing ratios at a large scale, partly including the free troposphere. The model sensitivity experiment shows that the HCHO mixing ratio is larger in the boundary layer than in the free troposphere over WLG (see Sect. 6.4).

Also listed in Table 2 are the levels of NO2, HCHO, and BrO measured by MAX-DOAS in other remote areas at middle and subtropical latitudes. Note that to our knowledge, no results for SO2 were reported in those studies. Our purpose here is to show what low levels of these gases, BrO in particular, had been detected by MAX-DOAS in these remote regions.

Gomez et al. (2014) retrieved the gas concentrations at a very long horizontal path of over 60 km from high-mountain MAX-DOAS measurements at the Izaña Atmospheric Observatory (28∘18′ N, 16∘29′ W, 2373 m a.s.l.) on the Canary Islands using the modified geometrical approach (MGA). The NO2 mixing ratios were within the range of 20–40 ppt in the summer of 2011 (Gomez et al., 2014). Gil-Ojeda et al. (2015) applied the same technique to a longer data set obtained at the same station and showed that the free-tropospheric NO2 mixing ratios varied in the range of 20–45 ppt between 2011 and 2013, with the lowest values in winter and highest values in summer. MAX-DOAS measurements were also performed at Zugspitze, Germany (47.5∘ N, 11∘ E, 2650 m a.s.l.), during February–July 2003 and Pico Espejo, Venezuela (8.5∘ N, 71∘ W, 4765 m a.s.l.), from March 2004 to February 2009. Based on the MGA method, the monthly mean mixing ratios of free-tropospheric NO2 were estimated to be in the range of 60–100 ppt at Zugspitze and 8.5–15.5 ppt at Pico Espejo. The values for HCHO were in the range of 500–950 ppt at Zugspitze and 255–385 ppt at Pico Espejo (Schreier et al., 2016).

BrO was detected directly in the midlatitude marine environment for the first time, in the region north of the Canary Islands (around 35∘ N, 13∘ W), when a research cruise including ship MAX-DOAS (SMAX-DOAS) measurements was carried out over the Atlantic Ocean in October 2000 (Leser et al., 2003). Typical levels of BrO in the marine lower troposphere were <1 to 3.6 ppt in most cases, and upper limits of NO2 for clean air in the Atlantic west of Africa were observed to be 24 to 100 ppt (Leser et al., 2003). BrO was measured by long-path MAX-DOAS (LP-MAX-DOAS) at the Mace Head observatory on the west coast of Ireland (53∘20′ N, 9∘54′ W) in August 2002. The BrO mixing ratios were shown to vary from below the detection limit (0.8 ppt) at night to a maximum daytime concentration of 6.5 ppt, indicating that significant bromine activation occurs over the open ocean (Saiz-Lopez et al., 2004). Tropospheric BrO profiles over the remote Atlantic were obtained by aircraft- and ship-based MAX-DOAS measurements by Volkamer et al. (2015). In the marine boundary layer (MBL) mixing ratios close to zero were obtained, which increase with altitude in the free troposphere. In the middle troposphere mixing ratios up to 1 ppt were found. Fitzenberger et al. (2000) derived BrO profiles from balloon-borne measurements in Kiruna (Sweden) with mixing ratios up to about 1 ppt in the free troposphere.

It can be seen that the levels of NO2 at WLG derived from this study are comparable to those measured by MAX-DOAS in other remote areas. However, the upper limit for the free-tropospheric mixing ratios of BrO at WLG (about 0.3 ppt) derived from this study is lower than those observed under the MBL conditions at the same latitude band.

We performed chemical box model simulations to evaluate the effects of the variability and trend in NO2 levels on the photochemistry and oxidation capacity of the background atmosphere over WLG. The model, named NCAR's Master Mechanism (MM), was originally developed by Madronich and Calvert (1989) and applied to estimate the chemical budget for ozone at WLG by Ma et al. (2002a). For this study, we simulated the net ozone production and OH in summer for the free-tropospheric conditions, which were characterized by air masses with relatively high ozone and low water vapor and by excluding very-short-lifetime non-methane hydrocarbons (NMHCs) like isoprene, as described by Ma et al. (2002a). All simulations were constrained by the same diurnal cycles of temperature, water vapor, and ozone and fixed NMHC values measured at WLG, as reported in the work of Ma et al. (2002a). Three different NOx levels were considered and fixed to constrain the simulations, with scenarios described below.  

M02. The daily average of NO2 mixing ratio is 48 ppt, corresponding to the result derived by the SP method from Ma et al. (2002a).

As mentioned above, the free-tropospheric conditions, which generally have less ozone production with respect to the planetary boundary conditions at WLG as characterized by Ma et al. (2002a), are taken into account. We also made a sensitivity test, for which the short-lived NMHC conditions from Ma et al. (2002a) were added to the M20 conditions (denoted as M20-BL). For the scenario with the added NMHC species, high intermediate oxidation products were predicted for the planetary boundary condition (M20-BL) compared to the free-tropospheric condition (M20). For example, the daily average HCHO mixing ratio was simulated to be 1.9 ppb in M20-BL and 0.4 ppb in M20. Compared to the M20 condition, the instantaneous net ozone production and destruction are more in balance over the course of the day under the M20-BL condition, with a daily average PO3 value of 0.3 ppb d-1, indicating that ozone is slightly net produced.

We also performed an additional model simulation (denoted as M20-BrO, not shown in the figure) to investigate the effects of bromine chemistry on PO3 and OH in the free troposphere over WLG, by including the gas-phase bromine chemical mechanism provided by the MM. All input variable values were the same as in M20, except that the simulation was constrained with a daily maximum BrO mixing ratio of 0.3 ppt (occurring around noon) for M20-BrO. The result shows a daily net ozone destruction of -1.0 ppb d-1, slightly lower than that for M20. Compared to M20, the OH concentrations in the daily cycle decrease in M20-BrO, with the peak COH value ranging from 7.9×106 molec cm-3 for M20 to 7.4×106 molec cm-3 for M20-BrO. The perturbation of BrO on the net ozone production estimated here (i.e., about -0.2 ppb d-1) may be considered negligible considering that the variability of other environmental conditions, such as the NMHCs and water vapor concentrations, can have larger effects on the simulated PO3 value (Ma et al., 2002a).