Ground-based vertical profile observations of atmospheric composition on the Tibetan Plateau (2017-2019)

. The Tibet Plateau (TP) plays an essential role in modulating regional and global climate, and its influence on climate is affected also by human-related processes, including changes in atmospheric composition. However, observations of atmospheric composition, especially vertical profile observations, remain sparse and rare on the TP, due to extremely high altitude, topographical heterogeneity and grinding environment. Accordingly, the forcing and feedback of atmospheric 25 composition from rapidly changing surrounding regions to regional environmental and climate change in the TP remains poorly understood. This paper introduces a high time-resolution (~15 min) vertical profile observational dataset of atmospheric composition (aerosol, NO2, HCHO and HONO) on the TP for more than one year (2017-2019) using a passive remote sensing technique. The diurnal pattern, vertical distribution and seasonal variations of these pollutants fibre at the end of an adjustable tube. This unit was controlled by a stepping motor to collect the scattered sunlight at different elevation angles, which was then transmitted via a glass fiber bundle to the spectrometer unit. The stepping motor had a precision of 0.01 o , and the field of view (FOV) of telescope was less than 0.3 o . Telescope scanner unit covered elevation angles from -10 o to 180 o , with angle 125 accuracy < 0.1 o . Moreover, an azimuth motor covering 0 o -180 o was installed in the telescope scanner unit to observe atmospheric composition at different azimuth angles. In this study, we set the elevation angle sequence to 1, 2, 3, 4, 5, 6, 8, 10, 15, 30 and 90 o , and exposure time of each individual spectrum to 1 min. The telescope unit was pointed to an azimuth angle of 53 o .

revisiting (~ 16 days) hinders understanding of processes at smaller scales. Accordingly, the current atmospheric composition monitoring platforms and freely available datasets are inadequate to fully understand the sources and impacts of atmospheric composition on the TP, as the formation, aging and transport processes also occurs above the ground (Hindman 70 et al., 2002;Duo et al., 2018;Huang et al., 2007). If we only use satellite column or ground-level observations to infer transport fluxes of atmospheric trace gases, the relative significance of local and transport contributions could be overvalued/downplayed (Hu et al., 2020;Liu et al., 2021a). In addition, the low air density, land surface heterogeneity, relatively cold temperature and strong solar radiation on the TP cause relatively higher planetary boundary layer (PBL) compared to its surrounding lowlands (Yang et al., 2003;Seidel et al., 2010). It promotes the exchange of atmosphere 75 between PBL and stratosphere, which has important impacts on atmospheric chemistry (Skerlak et al., 2014). The lack of sufficient vertical profile observation data limits the understanding of atmospheric composition on the TP and its impact on the global climate.
Limited field experiments were conducted using balloon and lidar on the TP in recent years (Wu et al., 2016;Dai et al., 2018;Fang et al., 2019;Zhang et al., 2020). The instruments used are costly and very limited atmospheric composition 80 species (mostly O3) can be measured. On the other hand, these observations are not open for sharing or very limited available upon request (i.e. only data during specified periods are provided). The coarse temporal resolution of these data is not sufficient for evaluations of chemical transport modeling and climate modeling (Gao et al., 2020a). Multi-axis differential optical absorption spectroscopy (MAX-DOAS) uses scattered sunlight as the signal source, enabling it to achieve low-cost and continuous measurement of vertical profiles of atmospheric composition. 85 Here we describe and provide access to a high time-resolution dataset of vertical profiles of atmospheric composition over the TP for more than one year. This database will play a critical role in improving satellite retrievals and numerical modeling of atmospheric composition over the TP. The uncertainties of air mass factor (AMF) caused by the high surface albedo and the absence of a priori profile on the TP could lead to large uncertainties in satellite retrievals of trace gases. Previously, we have demonstrated that using AMF calculated with MAX-DOAS measured NO2 vertical profiles could remarkably improve 90 the accuracy of retrievals of NO2 column densities (VCD) (Liu et al., 2016;Xing et al., 2017). On top of that, due to complex terrain and weather conditions, uncertainties in emission inventories, and imperfect model parameterization, chemical transport models are difficult to capture the vertical structures of atmospheric composition on the TP (Yang et al., 2018b), the uncertainties of which could be constrained through assimilation of observations from this dataset. Sect. 2 describes the observation site, MAX-DOAS instrument and retrieval algorithms. The vertical profiles of aerosol, NO2, HCHO and HONO,95 and their diurnal and monthly variations are introduced in Sect. 3. Sect. 4 and Sect. 5 present the availability of this dataset and a summary, respectively.
The spectrometer unit included two AvaSpec-ULS2048L spectrometers (UV: 300-460 nm, visible: 460-630 nm) with a 0.6 130 nm spectral resolution. A cooling unit was used to control the spectrometer temperature. Typical instrumental stray light was < 0.05%, root mean square (RMS) of 1×10 −4 (visible) and 2×10 −4 (UV) for ≈1000 scans around noon. A Peltier element was mounted on the spectrometer housing to cool and heat the spectrometer, and the temperature was stabilized at 20℃ with low fluctuations of < 0.05 ℃. We also equipped a charge-coupled device camera (Sony ILX511 with 2048 pixels) to the spectrometer to convert analog signal to digital signal. Moreover, the dark current and electric offset were recorded at 135 night to correct the observed spectra. To avoid the strong absorption of stratosphere, this study analyzed only spectra collected when solar zenith angle (SZA) was less than 75 o .

Spectral retrieval
This study used the QDOAS software developed by Belgian Institute for Space Aeronomy (BIRA-IASB) (http://uvvis.aeronomie.be/software/QDOAS/, last access: 26 April 2021) to retrieve differential slant column densities (DSCDs) of 140 the oxygen dimer (O4), NO2, HCHO and HONO. A sequential zenith spectrum, estimated from interpolation of two zenith spectra recorded before and after an elevation sequence, was selected as Frauenhofer reference spectrum. The retrieval configuring settings followed Xing et al. (2020Xing et al. ( , 2021, and the configurations used in CINDI intercomparison campaign as well (Roscoe et al., 2010;Kreher et al., 2019;Wang et al., 2020) (details are listed in Table 1). Fig. 3 illustrates the DOAS fits of above four species, and reasonably good fitting with tiny values of RMS can be found. Retrieved data with root mean 145 square (RMS) values larger than 5×10 -4 , 5×10 -4 , 6×10 -4 and 5×10 -3 for O4, NO2, HCHO and HONO, respectively, were filtered out. We calculated also the color index (CI), defined as the ratio of spectral intensities at 330 nm to that at 390 nm, to remove the cloud effects (Wagner et al., 2016). We filtered out data when the CI was less than 10% of the threshold that obtained through fitting a fifth-order polynomial to CI data (Ryan et al., 2018). After these processes, 90.17%, 86.41%, 83.22% and 80.19%, respectively, of the original DSCDs data were marked as qualified. 150

Vertical profile retrieval algorithm
The vertical profiles of aerosol extinction and volume mixing ratios (VMR) of trace gases (i. e., NO2, HCHO and HONO) were retrieved using the optimal estimation method (OEM) based algorithm. The radiative transfer model VLIDORT (Spurr et al., 2006) was used as the forward model. The maximum posteriori state vector x was determined through optimizing the following cost function 2 where ( ) , F x b denotes the measurement vector y (DSCDs measured at different elevation angles) as a function of the state vector x (pollutants profiles) and the true atmospheric meteorological parameters (profiles of temperature and pressure, albedo and aerosol phase function). a x stands for a priori state vector. S  and a S represent the covariance matrices of y and a x , respectively. During the retrieval process, exponential decreasing shape was assumed as the initial a priori vertical 160 profile shape of aerosol and trace gases, and the corresponding WRF-Chem simulated AOD and VCDs were also used as input a priori information. The inversion strategy used a Gauss-Newton (GN) scheme (Wedderburn et al., 1974). The iterative weighting function K was calculated using the Jacobians of DSCDs. The inversion consisted of two steps. Aerosol vertical profile was retrieved first, and then fed into the forward model to retrieve trace gases profile.
In this study, we derived the vertical profiles of aerosol and trace gases at 30 vertical layers, covering from 0.0 to 3.0 km, 165 and a 1.0 km correlation length was chosen. For forward simulations, surface albedo and surface altitude were set as 0.08 and 4.2 km, respectively. In addition, a fixed single scattering albedo (SSA) of 0.85 and an aerosol phase function with an asymmetry parameter of 0.65 were selected, due to the low uncertainties in retrieved aerosols with fixed SSA and asymmetry parameter (Irie et al., 2008). The profiles of aerosol and trace gases were filtered out when the degree of freedom (DFS) was less than 1.0 and retrieved relative error were larger than 100%. 170

Aerosol
As indicated in Fig. 5, maximum AOD over the TP occurred in August (1.19), almost doubled the minimum that happened in April (0.57). However, the high levels of AOD in summer might have been overestimated due to poor integrity of AOD data in summer ( Fig. 4: 12.91% in July and 22.58% in August). Relatively enhanced levels of AOD started from September and persisted till February. Such an enrichment was likely to be associated with both the increased anthropogenic emissions 185 due to tourism and transported pollution plume from South Asia. The ground surface of CAS (QOMS) station is exposed continually throughout the year, rarely covered by ice, snow or vegetation. Valley and mountain winds in this region began to strengthen from October, and strong winds persisted throughout the whole winter to blow soil and dust particles into the atmosphere (Xu et al., 2015;Liu et al., 2017;Chen et al., 2018).  Fig. 6(a) also indicates that strong aerosol extinction coefficients occurred occasionally in the middle layer, associated with long-range transport of particles Chen et al., 2018). The total and seasonal averaged vertical profiles of aerosol extinction are displayed in Fig. 7. All 195 averaged profiles exhibit exponential decreasing shape with maximum values occurred near the surface. Aerosol extinction coefficient in the lower layer varied with seasons, with maximum in autumn (1.55 km -1 ) and minimum in spring (1.00 km -1 ).
The ratios of aerosol extinction coefficients in the middle layer to those in the lower boundary layer were 44.96%, 41.11%, 44.35%, 40.01% and 52.31% for the total-averaged, spring-averaged, summer-averaged, autumn-averaged and winteraveraged aerosol profiles, respectively. These numbers decreased to 24.81%, 22.72%, 25.87%, 21.29% and 30.02% for the 200 upper layer. Relatively higher levels of aerosol above ground during winter is associated with strong mountain-valley breeze, which can blow dust on the ground surface into high layers . Aerosol pollution is also more severe in winter in South Asia , which can be transported as an important source of aerosols over the TP Chen et al., 2018;Zhang et al., 2018). Fig. 8 (a-e) illustrates the diurnal variations of aerosol vertical profiles for different seasons. Lower aerosol extinction mainly 205 occurred before 10:00, and it gradually increased and spread to higher altitudes with the rise of planetary boundary layer (PBL) height after 10:00. Moreover, aerosol extinction showed bi-peak patterns in all four seasons. One peak appeared between 10:30-12:30, and the other one occurred after 14:00. This bi-peak pattern was in line with previous investigation by Liu et al. (2017), dominated by the effects of local aerosol emissions, local dust geomorphology and mountain valley breeze . We observed also that the diffusion height of aerosol exhibited maximum in summer and minimum in 210 winter, mainly due to differences in PBL height driven by temperature in four seasons. in cold periods also caused the relatively high NO2 levels.

Nitrogen dioxide (NO2)
respectively. These numbers dropped to 42.86%, 47.60%, 46.27%, 40.77% and 41.14% for the upper layer. The vertical gradient of HCHO in autumn was relatively larger than that in other three seasons, which was partially associated with stronger surface HCHO sources in autumn.
On the diurnal variations of vertical profiles of HCHO, lower concentrations were observed before 10:00, but the abundance 255 gradually increased with sunrise and elevation of temperature (Fig. 8 (k-o)). HCHO peaked within 10:00-16:00 in spring and winter, while peak values of HCHO also appeared after 16:00 due to prolonged daytime in summer and autumn. The maximum and minimum diffusion heights of HCHO appeared in summer and winter, respectively, assoc iated with PBL height.

Nitrous acid (HONO) 260
The photolysis of nitrous acid (HONO) is a significant source of hydroxyl (OH) radical. HONO mainly originates from primary emissions from vehicles, ships, biomass burning and soil, the photolysis of nitrate particles (  in spring (0.56 ppb) (Fig. 11). Different from aerosol, NO2 and HCHO, HONO concentrations in the middle and upper layers were extremely low. The proportions of HCHO HONO concentrations in the middle layer to those in the lower boundary layer were 8.86%, 13.68%, 7.82%, 9.05% and 10.44% for the total-averaged, spring-averaged, summer-averaged, autumnaveraged and winter-averaged aerosol profiles, respectively. These numbers decreased to 3.14%, 5.30%, 3.52%, 2.72% and 3.06% for the upper layer. On the diurnal variations of vertical profiles of HONO, we found that HONO concentrations were, 275 consistently in four seasons, lower before 10:00 and peaked around 10:00-14:00. Such asThis pattern in the monitoring site was different from that in other low altitude cities (Wang et al., 2013;Lee et al., 2016;Xing et al., 2021).

Validations against satellite retrievals
In this study, OMI NO2 and HCHO products were used to validate the described dataset. For better comparison, MAX-DOAS data were averaged over the OMI overpass time (from 06:30 to 08:30 UTC) in the region that covers the CAS 280 (QOMS) station. OMI data were spatially averaged over the 15 km grid cells around the MAX-DOAS site, as the spatial resolution of OMI is 15×15 km 2 . The VCDs of NO2 and HCHO were derived by vertical integrations of their vertical profiles. The linear association between MAX-DOAS and OMI measurements are shown in Fig. 12. We found reasonably good correlations between MAX-DOAS and OMI observations, with Pearson correlation coefficients (R) of 0.68 and 0.60 (slope of 1.21 and 1.12, offset of 1.29×10 15 and 2.06×10 15 molec. cm −2 ) for NO2 and HCHO, respectively. OMI NO2 and 285 HCHO VCDs were systematically lower than MAX-DOAS NO2 and HCHO VCDs, which is different from previous comparisons applied in lower-altitude areas (Liu et al., 2016;Xing et al., 2017). This might be caused by large-area averaging effect of OMI and introduced uncertainties in OMI products by cloud cover and high surface albedo.
We observed also that the correlations between MAX-DOAS and OMI observations exhibited significant differences across seasons. The highest correlation appeared in autumn, followed by winter (Table S1). High correlation in these two seasons 290 was associated with increased tourism vehicle emissions, regional transport of NO2, and enhanced photochemical formation of HCHO. Failed linear regression analysis mainly appeared in spring and summer, due to high cloud coverage from March to September in the TP (Chen et al., 2006).

Summary
In situ observations, especially profile observations, are scarce but essential in polar regions, including the TP, the so-called "Third Pole". The sources of atmospheric composition over the TP, and how the species interact with regional and global climate remain unclear, mainly owing to lack of observations. Large uncertainties also remain in chemical transport and 300 climate modeling of cold-region processes, without observational constraints. The presented dataset offers high timeresolution vertical profile observations of aerosol, NO2, HCHO and HONO for more than one year on the TP. Distinct features were found on the diurnal variations, vertical distributions, and seasonality of these atmospheric composition species. Data products were also validated with satellite retrievals.
The limitations of this data are described as follows. (1) MAX-DOAS in ultraviolet and visible spectral ranges are typically 305 affected by photon-shot noise, and the retrieval errors usually increase under heavy haze or cloudy conditions. The data with relative retrieval errors larger than 50% were filtered in this study; (2) only the daytime vertical profiles of aerosol, NO2, HCHO and HONO were retrieved since MAX-DOAS relies on scattered sunlight. The spectral collected when solar zenith angle (SZA) are larger than 75 o were filtered to avoid the strong absorption of stratosphere (Xing et al., 2017); (3) the vertical resolution of 100 m is the highest resolution at present, which still needs to be improved with the development of 310 hardware and algorithms in the future.
Formatted: Normal, Left, Automatically adjust right indent when grid is defined, Line spacing: single, Adjust space between Latin and Asian text, Adjust space between Asian text and numbers, Snap to grid This dataset provides better temporal coverage and fills the gaps in the vertical direction. It has potentials to improve both satellite retrievals and reduce uncertainties in chemical transport modeling over the TP, particularly in the vertical direction.
We expect this dataset can be used to understand the sources and dynamic evolution of air pollutants over the TP, to assess the impacts of chemical forcers on climate system at multiple scales, and to facilitate the development and improvement of 315 models in cold regions. The potential applications of this dataset include: (1) Performances of chemical transport models are commonly poor when applied over the Tibet plateau due to complex topography and meteorology, etc. This dataset can be used to reduce the uncertainties of these models, especially in the vertical direction (Liu et al., 2021a); (2) this dataset can assist in the source apportionments of atmospheric composition at different altitudes over the Tibet Plateau; (3) observed atmospheric composition over the TP is valuable inputs for box models to understand atmospheric oxidation capacity at 320 different altitudes on the Tibet Plateau.