Dataset of Comparative Observations for Land Surface Processes over the Semi-Arid Alpine Grassland against Alpine Lakes in the Source Region of the Yellow River

The Yellow River (YR) is the fifth-longest river in the world and the “lifeblood” of northern China. The YR basin is dominated by a semi-arid and semi-humid climate, and the ecology and water resources of the region are facing great uncertainty in the context of global warming (Feng et al., 2016; Wang et al., 2019, 2021). The source region of the Yellow River (SRYR), with an area of 12.2 ×10⁴ km², is the most important runoff-producing area for the YR, which accounts for only 16.4% of the YR basin area, but provides approximately 38% runoff (Zhang et al., 2013). Sometimes called “China's Water Tower”, the SRYR is considered to be an "extremely sensitive region" to climate change (Lu et al., 2018). The air temperature has increased by more than 1.5°C in the past 60 years with a trend of 0.33°C (10 yr)⁻¹ (Li et al., 2021c), 1.2 times the warming rate of the Tibetan Plateau (TP) (Meng et al., 2020). This has especially been the case since the late 1990s, when the warming rate significantly increased to 0.48°C (10 yr)⁻¹ (Li et al., 2021c). However, the increase in the precipitation rate [6.65 mm (10 yr)⁻¹] was only 71% of what was observed in the TP (Meng et al., 2020). The land surface elements of the SRYR, including snow, glacier, permafrost, grassland, wetland, and lakes, are sensitive to climate change and play a role in the critical feedback on regional climate and water resources through interactions between the cryosphere, hydrosphere, and atmosphere.

Lakes are a sentinel of climate change (Adrian et al., 2009), which is a significant local climate driver due to the large thermal inertia and surface area (Clites et al., 2014; Van Cleave et al., 2014). Ngoring Lake and Gyaring Lake in the SRYR, each at an elevation exceeding 4200 m above sea level, comprise the highest large fresh lakes in the TP. The YR runs across these two lakes. They are in the transition region of westerlies and the Asian monsoon climate and are sensitive to changes in atmospheric circulation (Liu et al., 2021). This region belongs to a semi-arid climate, with an annual average air temperature and precipitation of −3.5°C and 326.8 mm, respectively (Li et al., 2021c). The landscapes around the lake are dominated by alpine meadows and sparse degenerated meadows with a small amount of wetland. Three land-surface process observation sites (lake, grassland, and lakeside) have been established successively since 2010 (Li et al., 2015), helping to fill in the gap of long-term in-situ observations of lake-atmosphere interactions in the freshwater and outflow lakes in the TP. The lakes in the SRYR have a few typical characteristics of the TP lakes. At the same time, some lakes in the SRYR are outflow lakes and freshwater lakes, mainly fed by precipitation, which differs from most of the lakes which are fed by glaciers and precipitation in the central and western TP.

After long-term continuous observations, this area has become an important platform that supports the study of lake-atmosphere interactions on the TP (Wen et al., 2015, 2016; Li et al., 2016, 2017, 2018a, b, 2021a; Ao et al., 2018; Lang et al., 2018). Together the flux observation stations at Nam Co Lake (Biermann et al., 2014; Wang et al., 2017, 2020), Qinghai Lake (Li et al., 2016a), and Siling Co Lake (Guo et al., 2019) has provided basic monitoring data for ecological protection and high-quality development of the YR. In this study, the lake-atmosphere interaction observation system and dataset in the Ngoring Lake basin are introduced. These comparative observations better quantify the differences in the energy and water exchanges between the lake-atmosphere and grassland-atmosphere, helping to obtain a deeper understanding of the underlying mechanisms which drive land–atmosphere interactions in the SRYR.

The remainder of this paper is organized as follows. The integrated comparative observations are described in section 2. Section 3 describes the instruments, variables, data processes, and data availability. Section 4 highlights the meteorological characteristics of the lake, grasslands, and lakesides and their differences. Section 5 describes the data integrity. Section 6 summarizes the conclusion and perspectives.

The observation sites were established by the Zoige Plateau Wetland Ecosystem Research Station (Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences). These include three sites, one located over the lake surface observing lake-atmosphere interaction during an ice-free period (referred to as Lake site (LS), 97°38′59″E, 35°01′28″N), another located over grassland 15 km from the LS for a comparison representing grassland–atmosphere interaction [Grassland site (GS), 97°33′16″E, 34°54′51″N], and the third site located at the lakeside, away from the GS by 2 km (referred to as the Lakeside site (LSS), 97°34′12″E, 34°54′24″N) for an advanced comparison. Before setting up the observation system, we calibrated the instruments and then used these instruments to determine site locations and conduct observations. The following rules were used to select the site as a better representative of the field site, which requires an open and relatively uniform underlying surface accessible by car to transfer the instruments. First, the satellite image and the land-use products were used to estimate land-use types and identify several alternative sites with relative uniformity and uncomplicated topography over a relatively large area. Then, a camera carried by an unmanned aerial vehicle was used to select a representative observation region. Lastly, three sets of instruments, including radiation components, automatic weather stations, and an eddy covariance system, were installed at the three selected locations to assess the representativeness and consistency of the data. Finally, the optimum site location was determined. This procedure was used for the site selection of the GS and LSS (Fig. 1).

Figure 1.  (a) Map of the location of observation sites, (b) the source region of the Yellow River, and (c–e) photos of the three observation sites (c: GS, d: LS, e: LSS). The area range of Fig. 1a is in the box of Fig. 1b.

The lake site was built on a submerged rock (about 1.0 m below the water surface) in northwestern Ngoring Lake during an ice-free period from 2011 to 2013, which was used to measure the flux exchange and radiation budget over the TP fresh lakes. The LS was located 200 m to the northwest lakeshore and 50 m to the southwest island. The water depth within 200 m around LS was approximately 3–5 m (Li et al., 2015). To the southeast of the LS was the central area of the lake. Based on EddyPro flux analysis software (LI-COR, Inc., USA), most of the flux footprint at the LS is on the lake surface (Li et al., 2015).

The grassland site, on a relatively flat grassland west of Ngoring Lake, was established in August 2011. The GS is about 2 km away from the lake, with low hills in every direction except to the east. It is well covered with vegetation, having a height of 5–10 cm. The vegetation coverage varies from 0.55 to 0.75 (Li and Li, 2015), and the leaf area index is approximately 1.2–1.4 m² m⁻² in summer (Li et al., 2021b). The soil at the GS site is predominantly sandy silt loam based on our sampling and the classification criteria of the United States Department of Agriculture (https://www.usda.gov/) (Table 1).

A strong thermal contrast appears near the lake shore due to the difference between the water body and land. Thus, we set up a five-layer boundary layer meteorological tower at the lakeside in late 2010. In 2013, an eddy covariance system was established for the long-term monitoring of turbulent fluxes from the lake surface and land, which was also used to make up for the shortage of flux observations during the lake ice period. For the LSS, all instruments were set up on the land near the lakeshore, except for the eddy covariance system, which was set up at the water edge. The footprint covers the water and grassland surfaces depending on the wind direction. According to the footprint analysis and the location of the lake shoreline, the data from wind directions (WD) that range from (35° < WD < 215°) are considered representative of lake surface observations (Lang et al., 2021); while the ranges of 230° < WD < 360° and 0° ≤ WD < 20° are considered to be land, and the other ranges are regarded as the transition area between the land and lake.

The instrumentation sensors at the three sites are listed in Table 2. For the LS and GS, the instruments were installed in the summer of 2011 to compare lake–atmosphere and land–atmosphere interactions. As the instruments were severely damaged by the lake ice, only the data during the ice-free period was available for the LS. For the LSS, the instruments had been set up since 2013. As the GS and LSS sites were very close together, we used the same precipitation measurements, which were observed using an RG3-M from 2011 to 2012, before replacing them with a T200B in 2013. The RG3-M can only measure rainfall, while the T200B can measure both rainfall and snowfall.

For 10 Hz eddy covariance data, a series of data post-processing corrections are performed using the EddyPro flux analysis software. For example, the ultrasonic virtual temperature is converted to air temperature, and corrections for time lag compensation, spike and trend removal, and coordinate rotation were made for all original 10-Hz data. The Webb-Pearman-Leuning density correction is applied to the water vapor fluxes (Webb et al., 1980). Moreover, the data with automatic gain control values (output from the LI 7500) greater than 70 are removed. The turbulent fluxes and meteorological variables are processed in 30-minute average intervals.

The data variables are extracted and saved as excel files (Flux.xlsx, Meteorology.xlsx, Soil.xlsx). To retain the observations in their original form as much as possible, no further post-processing processes are taken, except for replacing outliers with the missing value syntax (NAN). The quality grades of turbulent flux data are marked by Eddypro software, within which class 0 is recommended for fundamental research, and class 1 can be used in continuously running systems or for long-term analysis. The Beijing time is used in all the data files (Universal Time Coordinate + 8). The data have been released and are available for download at National Cryosphere Desert Data Center (http://www.ncdc.ac.cn/portal/topic/detail?id=cb682245-7cff-41f1-9d31-4b19eb659229) and Science Data Bank (https://cstr.cn/31253.11.sciencedb.07048) (Table 3).

Air temperature, surface pressure, relative humidity, and wind speed showed a clear diurnal cycle over the lake and grassland. For the air temperature (Fig. 2a), the peak of the daily cycle over the grassland was clearly larger than that over the lake in August (1.26°C), but the converse was observed in September and October. The temperature peak occurred about an hour later over the lake than over the grassland. At night, the daily minimum temperature over the lake was higher than that over the grassland, especially in October, with a difference of 2.60°C, indicative of the significant influence of the large heat capacity of the water body. There was also a typical daily cycle for air pressure (Fig. 2c), with a minimum in the afternoon and a sub-low at night. Air pressure over the lake was almost 0.2 kPa larger than that over the grassland. Although the lake skin surface is considered saturated with evaporation, the relative humidity of the lake surface remained slightly lower than the GS at night in June and July (Fig. 2e). Our further analysis found that this was mainly due to the high relative humidity of grassland in summer 2012, when the surface soil moisture at GS was significantly larger than that in 2013 and 2011 (figure not shown), and the observed latent heat flux was also significantly higher than that of LS during the same period, especially in the daytime (not shown). The strong latent heat flux increased the water vapor density of the surface air at the GS, which led to higher relative humidity. This phenomenon suggests that evaporation in the TP lakes in summer may not be as strong as expected, which may be related to the low temperature caused by the high altitude. The grassland surface temperature during the daytime is much higher than the lake surface. Once the surface soil water limitation is lifted, evaporation is considerable. Wind speed also had a typical diurnal cycle (Fig. 2g), with maximum values in the afternoons into evenings and minimums at night. Around noon, the wind speeds over the lake and grassland were very close, but from afternoon to night, the wind speed at the LS was considerably larger than that at the GS. These differences at different times were related to the conversion of lake-land breeze circulation and the smaller momentum roughness length of the lake surface.

Figure 2.  Monthly average diurnal cycles (left column) and daily average values (right column) of air temperature, air pressure, relative humidity, and wind speed at the GS and LS sites. The numbers on the x-axis of the left column represent the time in 24-h intervals.

The distance between the LS and GS sites was only 15 km, so the downward shortwave radiation (Rsd) was generally consistent between the sites (Fig. 3a), while the upward shortwave radiation (Rsu) difference reflected the albedo difference between the water body and grassland (Fig. 3c). In the Ngoring Lake basin, the lake surface albedo at noon was only 0.04–0.05 (Li et al., 2015), and the grassland albedo was about in summer 0.17–0.20. Longwave radiation was associated with changes in temperature, which showed an increasing and then decreasing trend spanning from June to October (Figs. 3f, h). After August, the nocturnal downward longwave radiation (Rld) over the lake was significantly larger than that over the grassland. For example, the difference in Rld can be about 32.37 W m^–2 on the night in October (Fig. 3e), when the temperature difference between the two sites was about 2.60°C. The difference in upward longwave radiation (Rlu) between the two sites was even more significant, with the maximum daily value of the grassland being about 95.06 W m^–2 higher than over the lake in August and the lake being about 100 W m^–2 higher than the grassland at night in October (Fig. 3g). Large distinct values of Rlu between the lake and grassland suggests an evident difference of their surface temperatures. Both the Rsu and Rlu showed more violent fluctuations over the grassland than those over the lake. The Rlu at two sites peaked in August, but the decreasing rate at the LS was much slower than that at the GS after August. The daily average value of Rlu at LS was also higher than that at the GS (Fig. 3h), indicating that the lake played the role of heat storage in autumn.

Figure 3.  Monthly average diurnal cycles (left column) and daily average values (right column) of radiation components at the GS and LS sites. On the x-axis of the left column, the numbers represent the time in 24-h intervals (Rsd, Rsu, Rld, and Rlu are downward shortwave radiation, upward shortwave radiation, downward longwave radiation, and upward longwave radiation, respectively).

Figure 4 shows the sensible (H) and latent heat fluxes (LE) over the lake and grassland. The peak of the diurnal cycle of H over the lake was weaker than that over the grassland. For example, the peak of the grassland was more than five times that of the lake, equivalent to about 137.24 W m^–2 higher. At night, the H of the lake was obviously higher than that of the grassland. In terms of daily average, the H of the lake was higher than that of the grassland only in autumn, from September to before freezing. From summer to autumn, the H of the lake increased continuously, while such an increase over the grassland was not obvious (Fig. 4b). It is worth noting that even in June, the diurnal H of the lake was all positive, different from lakes at high latitudes. In high-latitude lakes, the lake surface temperature is lower than the air temperature, and the H is mainly negative from melting lake ice until autumn (Long et al., 2007; Rouse et al., 2008). The maximum diurnal H of the lake appeared in the morning, and the minimum value appeared in the evening, different from that of the grassland. For the latent heat flux, the diurnal cycle of the LE of the lake was relatively insignificant (Fig. 4c). The LE of grassland in summer was much higher than that in autumn, while the LE of the lake continuously increased from June to September, and decreased slightly in October (Fig. 4d).

Figure 4.  Monthly average diurnal cycles (left column) and daily average values (right column) of sensible and latent heat fluxes at the GS and LS sites.

Compared with the LS site, the observations at the GS and LSS sites have better and longer continuity. In this section, we compare these two sites. Air temperatures at both sites show good consistency, except for the peaks and the minimum in a few months (Fig. 5a). In summer, the peak air temperature at GS was slightly higher than that at the LSS, while in winter, the daily minimum air temperature at GS was slightly lower. For air pressure, differences between the two sites were mainly present in seasons other than winter (Fig. 5c). The relative humidity showed similar phases at the two sites. Still, the minimum value of the daily cycle at the LSS was clearly smaller than that at the GS (Fig. 5e). Typical daily wind speed cycles can be observed in this region. Wind speeds began to increase rapidly before noon, peaked in the afternoon, and dropped rapidly around evening (Fig. 5g). The average wind speed at night varied little from month to month, while the average peak during the daytime could reach nearly 8.80 m s^–1 in February but less than 5.07 m s^–1 in August. The maximum wind speeds in this region generally occur in winter, followed by spring, especially from December to the following March (Figs. 5g-h). The wind speed at the LSS was significantly higher than that at the GS.

Figure 5.  Monthly average diurnal cycles (left column) and daily average values (right column) of surface air temperature, air pressure, relative humidity, and wind speed at the LSS and GS sites.

The radiation components at the GS and LSS sites are shown in Fig. 6. Except for December, the Rsd of the two sites was in good agreement. The radiation sensor at the LSS was on the lake edge, whose observation source area represented the water body and land, and its area proportion changed with the water level. As a result, the albedo varied greatly in different periods. Therefore, the Rsu of the lakeside was generally smaller than that of the grassland (Figs. 6c-d). On the annual scale, the Rld had two peaks occurring in July and September (Fig. 6f), and the results of the two sites were in good agreement. The Rlu of the lakeside was higher than that of grassland at night in all months (Fig. 6g), which is closely related to the relatively high temperature of the lake at night. From December to April, the Rlu at the LSS was clearly larger than that at the GS (Fig. 6g), which was related to the decrease in lake water level in winter, and also due to the majority of the land surface of the radiation observation source being covered with pebbles.

Figure 6.  Monthly average diurnal cycles (left column) and daily average values (right column) of radiation components at the LSS and GS sites.

The soil temperature and moisture at GS are shown in Fig. 7, which depicts larger magnitudes for both in shallow soil layers, and that the soil moisture below 20 cm was small all year round (Fig. 7c). This suggests that soil water was mainly stored in the grass root layer. Soil moisture showed two peaks during the year, which appeared in early July and September (Fig. 7d). It was very similar to the reported precipitation pattern in the SRYR (Li et al., 2021c).

Figure 7.  Monthly average diurnal cycles (left column) and daily average values (right column) of the soil temperature and moisture at the GS.

Monthly average diurnal cycles and daily average values of the sensible and latent heat fluxes at the GS and LSS sites are shown in Fig. 8. The peak value of the daily cycle of H at the LSS was significantly greater than that at the GS, especially from January to May, when the maximum difference was about 182.87 W m^–2 (Fig. 8a). The difference in LE between the two sites was relatively small (Fig. 8c). Furthermore, to clarify the influence of flux source region difference on the observed value at the LSS site, Fig. 9 depicts the annual variation of sensible and latent heat fluxes over the lake surface, lakeside land, and grassland over a typical year. Both the lake surface and lakeside land data were from the LSS site, divided according to wind direction (see Section 2.3). The LE of the ice period (early December to early April) was significantly different from that of the ice-free period (mid-late April to November), and most of the LE was less than 50 W m^–2 during the ice period. After the lake ice melted in mid-April, the LE increased rapidly and formed a small peak (>80 W m^–2). From the ice melt to late June, the LE of the lake surface decreased slightly. Beginning in July, the LE increased rapidly, peaking in August and September, before decreasing significantly in October. The annual peak of the H of the lake surface appeared in November (>120 W m^–2) before the lake froze. The temperature difference at the air-lake interface reached the annual maximum value during this period. After the lake froze, H decreased rapidly (<40 W m^–2) and continuously decreased from December to April. In May, the H of the lake surface presented the first small peak of the year and then decreased slightly until the end of June. Since July, H has entered a new round of a rising period, reaching the annual maximum before freezing. It should be noted that the LSS is located on the west side of Ngoring Lake, and the westerly wind prevails in this area during the ice period. At this stage, the data observed from the lake surface was relatively scarce, whose values may be lower than the actual average flux of the lake surface. In addition, the effect of the lake-land breeze resulted in the data from the lake being mostly concentrated in the daytime, which was most obvious in the cold season. However, the H of the lake at night was considerable, which may lead to the underestimation of H in Fig. 9a.

Figure 8.  Monthly average diurnal cycles (left column) and daily average values (right column) of the sensible and latent heat fluxes at the GS and LSS sites.

Figure 9.  Half-hour average sensible (H, orange solid circle) and latent heat fluxes (LE, blue solid circle) (a) over the lake surface at the LSS, (b) over the lakeside land at the LSS, and (c) over the grassland at the GS from 1 November 2012 to 31 October 2014.

In sharp contrast to the lake surface, the H peak of the lakeside land appeared in spring and decreased rapidly after July (Fig. 9b). There was a strong H in winter and spring (December 2012 to May 2013), and the H peak in March 2013 was close to 600 W m^–2. However, the H from November 2013 to February 2014 was significantly lower than that of the previous year, mainly due to more snowfall and low surface temperatures during this period, which was a typical "wet winter", different from the spring of 2013, which had little snow. In clear contrast to the strong H observed in the daytime along the lakeside in winter and spring, the negative H was also strong at night, indicating the presence of a significant temperature inversion in the surface layer. The LE was large from May to September and peaked in July; however, most values were less than 300 W m^–2, considerably less than the H peak.

The patterns of H and LE at the GS were similar to that of the lakeside land, but there were significant differences in their magnitudes (Fig. 9c). The H peak of the GS was about 400 W m^–2 during the daytime and –120 W m^–2 at night, not as strong as that of lakeside land. The large-valued area of H existed on lakeside land, which was related to the special microclimate in this area (Li et al., 2012). The albedo of lakeside land is similar to the grassland, receiving similar solar radiation during the daytime, so their surface soil temperatures were similar. The air temperature at the lakeside was relatively low due to the cooling effect of the lake during the daytime, as well as ice melting in spring, creating a stronger temperature difference at the air-land interface and a larger H. Conversely, a stronger inversion appeared over the lakeside land at night and created stronger negative H.

The ice thickness during the frozen period was also recorded manually (Fig. 10). Lake ice in Ngoring Lake generally begins to appear in mid-December every year and melts in early April. The maximum ice thickness was about 0.7 m, usually occurring in mid-February. A field experiment on lake ice albedo and energy budget was carried out in Ngoring Lake in February 2017, in which we found that the snow-free ice albedo was quite small, and the ice-albedo of the TP lakes was widely overestimated in existing lake models (Lang et al., 2018; Li et al., 2018b). Accurate albedo estimation can significantly improve simulations of lake ice thickness (Li et al., 2021a). Due to a small ice albedo, the lake water warmed rapidly before the melting of lake ice in Ngoring Lake (Wen et al., 2016; Kirillin et al., 2021; Wang et al., 2022). Similar phenomena have recently occurred in other TP lakes (Lazhu et al., 2021). In addition to the low air temperature and strong solar radiation in the TP, unstable stratification existed over the lake surface during almost the entire ice-free period, which was clearly different from that of high-latitude lakes, where the stable stratification prevailed over the lake in summer (Rouse et al., 2005). Beyond surface meteorological observations, radiosonde observations were also carried out over the lake surface, from which we found that the daytime boundary layer height over the lake increased rapidly from less than 500 m in winter to more than 2200 m in summer due to a strong sensible heat flux after the cold air passed through the lake in summer (Li et al., 2017), which was extremely rare in previous studies of lake boundary layers.

Figure 10.  Records of manually measured ice thickness during two frozen periods.

The data integrity of the different variables in each year is shown in Fig. 11, with the value of 100 indicating complete continuous data (no missing data). For most variables, the integrity is more than 80% in all years, while the latent heat flux, precipitation, and soil heat flux data at 5 cm have gaps. The information in the figure facilitates data selection when analyzing lake–atmosphere interactions, driving land surface models, and evaluating model results.

Figure 11.  The annual data integrity of different variables. The value of 100 indicates complete continuous data, with no missing data. (WS: wind speed; WD: wind direction; Ta: air temperature, RH: relative humidity, Pres: air pressure, Rsd: downward shortwave radiation; Rsu: upward shortwave radiation; Rld: downward longwave radiation; Rlu: upward longwave radiation; SHF: soil heat flux; LE: latent heat flux; H: sensible heat flux).

To quantify the difference between the lake and the grassland, we calculate the statistical indicators of the observed variables in different periods. Table 4 shows the averaged values and the 5% and 95% percentiles for the variables at the GS and LS sites. The comparisons of air temperature show that the lake is generally a heat island from June to September, with the exception of July. This is further confirmed by the 55 and 95% percentile averages. The average air temperature from June to September is 7.69°C over the lake and 7.60°C over the grassland. This contrast can be reflected in the net radiation and surface heat fluxes. Considering the contrast between the lakeside and the grassland (Table 5), the GS and LSS show smaller differences compared to the GS and LS in terms of air temperature, air pressure, relative humidity, and radiation components, except for the upward longwave radiation. The average latent heat fluxes of lake and grassland are 36.42 W m^–2 and 41.09 W m^–2, respectively (Table 5). The average sensible heat flux of the grassland is 29.05 W m^–2, markedly less than the 51.98 W m^–2 observed over the lake, which significantly affects synoptic weather over these regions (Li et al., 2017).

Due to a lack of field observations, the lake–atmosphere interaction and the difference between it and the land–atmosphere interaction in the TP remain inadequately investigated despite the presence of thousands of lakes of different kinds. In addition, the applicability of the lake model to the TP has not been fully evaluated (Lazhu et al., 2016; Wen et al., 2016; Lang et al., 2021; Li et al., 2021a). Some special hydrothermal characteristics of TP lakes (e.g., rapid warming before ice melting) have not been discovered and explored until recently (Kirillin et al., 2021; Lazhu et al., 2021; Li et al., 2021a; Wang et al., 2022). The parameterizations in existing lake models mainly depend on lake observations in high-latitude lakes, and little consideration is given to the characteristics of TP lakes. Therefore, this observation-based dataset can help promote the scientific understanding of the lake–atmosphere interaction over the TP, to assess the biases and gaps between the model simulations and reality, and to facilitate the improvement of lake models and land surface models in high-altitude regions. Based on this, we hope more characteristics related to lake–atmosphere interactions on the TP will be explored, and more suitable lake models will be established and improved. Such efforts will allow for the lake models to be coupled to land surface process models and even climate models, providing support for the full study of the water cycle and climate change on the TP.

Acknowledgements. This study was supported by the National Natural Science Foundations of China (Grant Nos. 41930759, 41822501, 42075089, 41975014), the 2nd Scientific Expedition to the Qinghai-Tibet Plateau (2019QZKK0102), The Science and Technology Research Plan of Gansu Province (20JR10RA070), the Chinese Academy of Youth Innovation and Promotion, CAS (Y201874), the Youth Innovation Promotion Association CAS (QCH2019004), iLEAPs (Integrated Land Ecosystem-Atmosphere Processes Study-iLEAPS). Thanks to all the scientists, engineers, students, and other people who participated in these field observations, instrument maintenance and data processing.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.