Background & Summary

The Qinghai-Tibet Plateau (QTP) is the “third pole” of the Earth1, with a total area of approximately 2.5 × 106 km2 and an average elevation of over 4,000 m above sea level2. The air is thin, and the ecological environment is extremely vulnerable3. Hypoxia is a major environmental characteristic of the QTP4. Altitude sickness caused by hypoxia not only affect the daily life and health of short-term travelers but also have an impact on the life expectancy of permanent residents5,6. In recent years, with the regional development of the QTP and the completion and opening of the Qinghai-Tibet Railway, the permanent population in the region is increasing, and the short-term traveler population is growing rapidly7,8. From 2001 to 2020, the total population of Tibet and Qinghai increased from 7.87 million to 9.57 million, and the annual number of tourists increased from 4.48 million to 68.62 million9,10. Against the backdrop of high-quality regional development, hypoxia is increasingly becoming an extremely important environmental risk factor on the QTP. However, due to the underdeveloped health care system in the region, the increasing population exposure, and the low human adaptive capability to the hypoxic environment, the health risks of hypoxia for the population on the QTP are becoming increasingly serious11. Therefore, to obtain the surface oxygen concentration (i.e., oxygen concentration) data of the QTP and analyze the influencing factors are very important.

The oxygen concentration has been considered nearly constant since 1912, when Benedict proposed the composition of the atmosphere and the volumetric concentration of oxygen in the air12. Machta and Hughes collected observations of oxygen concentration between latitudes 50°N and 60°S, again indicating that the oxygen concentration in dry air was nearly constant at 20.946%13. Some studies have shown that atmospheric pressure and oxygen partial pressure decrease with increasing elevation, but the oxygen concentration does not significantly change at different elevations14,15. Through field research on the QTP and its surrounding areas4,11,16, we have discovered that oxygen concentration is not constant and exhibits spatiotemporal variations. Importantly, elevation is not the primary or sole factor controlling oxygen concentration; other factors such as temperature and vegetation also influence oxygen concentration. The relative contributions of elevation, temperature, and vegetation to oxygen concentration are −39.58%, 35.50%, and 24.92%, respectively17,18.

Considering the unique high-elevation hypoxic environment and the increasingly prominent health risks that are caused by hypoxia on the QTP, we conducted 14 field surveys in this region from 2017 to 2022. Through route-based scientific expeditions and fixed-point measurements, we obtained oxygen concentration data and corresponding geographic environmental data from 807 measurement points. In addition, we measured the fractional vegetation coverage (FVC) of 51 measurement points on the QTP using large surface coverage sampling quadrats (1 km × 1 km). Furthermore, based on the measured oxygen concentration data, we estimated the spatiotemporal patterns of oxygen concentration on the QTP. This work filled the gap in the research of oxygen concentration on the QTP, which provided data support for the analysis of influencing factors and spatiotemporal characteristics of oxygen concentration. This research played an important role in deepening the understanding of the environmental risks that the people and economic systems face in high-elevation areas, including the QTP, under the background of global climate change. This study is significant for ensuring ecological security, improving the health and well-being of residents/tourists and livestock, and promoting stable, prosperous, and high-quality development in high-elevation areas.

Methods

Field measurement methods

The QTP is vast, and its different regions exhibit distinct spatial heterogeneities in terms of the topography, landforms, climate, vegetation, soil, land use, and ecosystems. Considering that a limited number of measurement points are insufficient to represent the oxygen concentration variations in the entire region and would cause significant errors in the overall estimation, large-scale scientific field surveys are limited by factors such as manpower, resources, finances, and time. The layout of our field survey measurements was mainly based on two principles. First, spatially, given the tremendous spatial heterogeneity of the QTP and based on the actual traffic routes on the ground, we aimed to cover the different elevations as much as possible, including topography, landforms, climates, vegetation, soils, land uses, and ecosystem types on the plateau. Second, temporally, we concentrated the measurements during the vegetation growing season, which is typically from late July to early August each year, within a window of no more than one month to ensure the comparability of the data from different years, focusing the summer, while also considering autumn and winter. Based on this, we designed a “7 East‒West and 5 North‒South” scientific field survey route (Fig. 1) to conduct systematic observations of oxygen content and related geographic elements.

Fig. 1
figure 1

Oxygen concentration measurement position on the QTP from 2017 to 2022.

Full size image

Following the overall layout of the scientific expedition route mentioned above, we conducted 14 route/point-based field surveys on the QTP from 2017 to 2022 (Table 1) and covering a total distance of over 30,000 kilometers. We obtained geographical environmental data, such as latitude and longitude, elevation, temperature, atmospheric pressure, relative humidity, and oxygen concentration, from 807 measurement points and used 1 km × 1 km field sampling quadrats to measure FVC at 51 measurement points on the QTP.

Table 1 Route/point oxygen concentration measurement from 2017 to 2022.
Full size table

At each measurement point, the latitude, longitude and elevation were recorded by GPS, and three groups of instruments (Table 2) were used to simultaneously measure the atmospheric pressure, temperature, relative humidity, and oxygen concentration approximately 1.5 meters above the ground19. To reduce the measurement error, the average of the data that were measured by the three groups of instruments was used as the final measurement result for each point. Since the oxygen concentration meter measures the instantaneous oxygen concentration in the quasi-stationary state of air, the data were measured in a windless environment as much as possible, and the data were read after the instrument was stabilized to minimize the effect of air flow on the measurement.

Table 2 Instruments used for measurement.
Full size table

It should be noted that due to the differences in the measuring instruments, the oxygen concentration data that were obtained in 2017 needed to be calibrated before comparison with data from other years. Additionally, the three TD400-SH-O2 portable oxygen concentration meters used for field measurements in 2018–2020 were purchased in 2018 and replaced in 2021 due to their life expectancy. Three new TD400-SH-O2 portable oxygen concentration meters were purchased in 2022. Therefore, the oxygen concentration data obtained in 2021 and 2022 also needed to be calibrated before comparison. For this reason, during the summer of 2022, we mutually calibrated the CY-12C and TD400-SH-O2 oxygen concentration meters that were used in 2017, 2018–2020, 2021, and 2022. We converted the oxygen concentration data that were obtained in 2017, 2021, and 2022 to equivalent data that were obtained in 2018–2020, which completed the calibration and unification of the data.

Estimation methods

In the previous study, it can be found that elevation, temperature, and vegetation were the three most important factors that affected the oxygen concentration on the QTP17,18. Among these, elevation has a negative contribution to oxygen concentration, while temperature and LAI have positive contributions. Importantly, this relationship remains stable and does not vary with changes in time or space. Based on the location of the field measurement position from 2017 to 2021, the LAI in the corresponding locations and time periods were extracted20. Combining the measured elevation and temperature data, the relative contributions of the elevation, temperature, and LAI to oxygen concentration can be calculated as −39.58%, 35.50%, and 24.92%, respectively, by the correlation coefficient method and principal component analysis18. Since the data covered an extensive spatial range across the entire plateau and included measurements from both summer and winter seasons, it can be considered that the obtained relative contribution rates are highly relevant and applicable. Therefore, we assume that the effects and relative contributions of the major influencing factors of oxygen concentration on the QTP remain constant, allowing us to estimate the spatiotemporal distribution pattern of oxygen concentration.

Considering the sample size and the stability of the data, the spatiotemporal patterns of oxygen concentration on the QTP were mainly estimated based on 422 groups of data that were obtained from field measurements on the QTP from 2018 to 202017. The estimation process is shown in Fig. 2.

Fig. 2
figure 2

Calculation process of oxygen concentration estimation data on the QTP.

Full size image

The measured elevation, temperature, and extracted LAI are standardized by using the Min-Max scaling method, resulting in normalized variables for elevation (NE), temperature (NT), and LAI (NL). These variables are then used as weights based on their relative contributions to oxygen concentration. Subsequently, a weighted sum is conducted to calculate the estimated intermediate variable (Tmp).

$$Tmp=-0.3958\times {N}_{E}+0.3550\times {N}_{T}+0.2492\times {N}_{L}$$
(1)

Based on the cross-validation method, we selected m groups of samples (m = 3, 4, 5, ⋯, 419) each time from the 369 groups of measured data as the training set to construct a linear regression model between the measured oxygen concentration and estimated temporary variable. The remaining (422-m) groups of samples were used as the testing set to validate the model. For each value of m, we performed 50,000 random trainings and used the average slope and intercept after training as the model parameters. We calculated the average and standard deviation of the root mean square error (RMSE) using the testing set. When the RMSE standard deviation was the smallest, the model was the most robust. Finally, we obtained the most robust linear regression model between the measured oxygen concentration (OC) and estimated temporary variable (Tmp), with a R2 of 0.72 and p < 0.001.

$$OC=1.0283\times Tmp+20.2509$$
(2)

The spatiotemporal estimation of oxygen concentration on the QTP was mainly based on DEM data21, monthly temperature data from 2001 to 202022, and 8-day LAI data from 2001 to 202020. The spatial resolution of the data was unified to 1 km × 1 km, and spatial clipping was performed according to the extent of the QTP. In addition, the temperature and LAI data were calculated as annual averages, as well as monthly averages for January and July.

Taking the estimation of the annual average oxygen concentration on the QTP as an example, the elevation, annual average temperature and annual average leaf area index were normalized, the relative contribution of each factor to the oxygen concentration was used as the weight, and the weighted sum was substituted into formula (1) grid by grid to obtain the raster data of the estimated temporary variable. Then, the estimated temporary variable data were substituted into formula (2) grid by grid to obtain the annual average oxygen concentration of the QTP. The estimations of average oxygen concentration in January and July were calculated in the same way. The difference between the July and January average oxygen concentrations was obtained by subtracting the average oxygen concentration in July from the average oxygen concentration in January.

Data Records

The dataset “Surface oxygen concentration on the Qinghai-Tibet Plateau (2017–2022)” is available under National Tibetan Plateau Data Center23. A total of 5 data records are contained in the dataset. Of these,

  1. (1)

    Measured oxygen concentration data on the QTP (from 2017 to 2022) [File “Surface oxygen concentration on the Qinghai-Tibet Plateau (2017–2022).xlsx”];

  2. (2)

    Annual average oxygen concentration distribution data on the QTP [File “Annual average oxygen concentration.tif”];

  3. (3)

    January average oxygen concentration distribution data on the QTP [File “January average oxygen concentration.tif”];

  4. (4)

    July average oxygen concentration distribution data on the QTP [File “July average oxygen concentration.tif”];

  5. (5)

    Distribution data of the difference between the July average and January average [File “Difference between July and January.tif”].

Measured data

The measured oxygen concentration data on the QTP were stored in a matrix with 13 columns and 807 rows. Among them, the 13 columns included the measured latitude, longitude, elevation, temperature, relative humidity, atmospheric pressure, oxygen concentration, FVC, and the geomorphologic and vegetation types were extracted from the Geomorphological Map of China (1:250,000)24 and the Vegetation Map of China (1:1,000,000)25, respectively. 807 rows represent the 807 groups of data we measured on the QTP from 2017 to 2022. As shown in Table 3, some of the measured oxygen concentrations on the QTP are shown with our measured points around Qinghai Lake from August 15th to August 17th, 2018, as an example.

Table 3 Oxygen concentration measured data on the QTP (partial).
Full size table

Estimated data

The estimated oxygen concentration distribution data (Fig. 3) on the QTP include the annual average, July average, January average oxygen concentration, and the difference between the July average and January average. The data were stored in GeoTiff format. The geographic coordinate system of the data was GCS WGS 1984. The spatial resolution of the data was 1 km × 1 km. The data had a spatial extent of 73.50°E to 104.67°E and 26.05°N to 39.91°N, with 2741 rows and 1663 columns, and the value of each pixel represented the oxygen concentration. The unit of the data was %.

Fig. 3
figure 3

Estimated oxygen concentration distribution data on the QTP.

Full size image

From Fig. 3, temporally, the difference between July and January showed that the July average oxygen concentration on the QTP was higher than the January average, the July average oxygen concentration was higher than the annual average, and the January average oxygen concentration was lower than the annual average, which was related to the higher temperature and vegetation oxygen production in July. Spatially, the distribution of the annual, July, and January average oxygen concentrations on the QTP showed that the oxygen concentration first showed an obvious east‒west divergence pattern, i.e., the eastern part was higher than the western part, and the oxygen concentration decreased from east to west. Second, the oxygen concentration showed a pattern of north‒south belt alternation and extended east‒west. Third, the oxygen concentration showed a decreasing pattern from the periphery of the plateau to the hinterland as the elevation rose. The difference in oxygen concentrations between July and January showed a clear spatial pattern of east‒west and north‒south divergence, i.e., high in the northeast, low in the southwest, higher in the north and lower in the south.

Technical Validation

Potential error analysis

This study employed a route/point-based measurement approach using a portable electrochemical oxygen concentration meter. The instruments and their usage periods are detailed in Table 4. Specifically, the instruments used in 2017 were the CY-12C digital oxygen concentration meters, while the instruments used in 2018–2020, 2021, and 2022 were the TD400-SH-O2 portable oxygen concentration meters.

Table 4 Oxygen concentration meter used for measurement.
Full size table

Instrumental error

To mitigate instrumental errors, all instruments were calibrated with oxygen standard gases before each measurement. Three sets of instruments were simultaneously used for measurements, and the average data obtained from these measurements were taken as the results. However, it’s important to note that the instruments used in 2017 were the CY-12C digital oxygen concentration meters, which differ from the TD400-SH-O2 portable oxygen concentration used from 2018 to 2022. Furthermore, there were differences in the sensors of the instruments used in 2018–2020, 2021, and 2022. To address this instrumental error, we conducted field-based comparative measurements of oxygen concentration along the “Mado-Shuanghu-Bange” route using different instruments from July 15th to July 25th, 2022. For each measurement point, four groups of instruments were used simultaneously to take measurements (Table 5). Subsequently, the average of the data obtained from each group of instruments was taken during the data processing.

Table 5 Instrument information used in data calibration.
Full size table

Since oxygen concentration was measured simultaneously, relationships were established by pairwise comparisons of data obtained from different groups of instruments (Fig. 4). Ultimately, the data were revised to adhere to the measurement standards of the same group of instruments, thereby eliminating instrumental errors.

Fig. 4
figure 4

Relationship between the oxygen concentrations measured by different groups of instruments.

Full size image

The oxygen concentrations measured by the instruments in groups B, C and D showed excellent linear relationships with each other (Fig. 4), with R² values of 1.0, and all passed the significance test at the 0.001 level. The results indicated that there is almost no difference in the data measured by the TD400-SH-O2 portable oxygen concentration meters of groups B, C and D.

The data measured by the instruments in groups A and B (Fig. 4b,e), groups A and C (Fig. 4c,i), groups A and D (Fig. 4d,m) also showed good linear relationships with each other, with R² values above 0.7, and all passed the significance test at the 0.001 level. The results indicated that the difference in the data measured by the CY-12C digital oxygen concentration meters of group A and the TD400-SH-O2 portable oxygen concentration meters of groups B, C and D is small.

Considering the largest sample size of data measured in 2018–2020 (Table 1) and the best linear regression relationship between the data measured by instruments in group A and group B (R2 = 0.8). Based on the regression equation in Fig. 4, the oxygen concentration measured in 2017 (Fig. 4e), 2021 (Fig. 4g), and 2022 (Fig. 4h) was substituted into the equation as the independent variable, and finally the data were revised to the same standard to eliminate instrumental errors.

Environmental error

As indicated in Table 4, the CY-12C digital oxygen concentration meter operates in an environmental temperature range of 0 °C to 40 °C with a relative humidity range of 0% to 90%. On the other hand, the TD400-SH-O2 portable oxygen concentration meter operates in an environmental temperature range of −40 °C to 70 °C with a relative humidity range of 0% to 95%. We conducted field measurements during both summer and winter on the QTP, and during these measurements, we made every effort to ensure that the instruments operated in ideal environmental conditions to minimize the impact of the environmental factors such as temperature and water vapor on the measuring. The temperatures during the measurements ranged from −11.58 °C to 35.64 °C, and the relative humidity ranged from 7.9% to 93.2%, all of which met the requirements of the instruments (please note that temperature and relative humidity were not measured during the summer of 2017 due to the lack of instruments).

Methodological error

During measurements, the instruments were handheld to measure oxygen concentration at approximately 1.5 meters above the ground. Measurements of oxygen concentration were conducted in calm or low-wind environments whenever possible. It’s important to note that the fractionation effects can cause significant bias in oxygen concentration measurements when there is inconsistency in temperature, humidity, and pressure inside the instrument. Therefore, measurements were taken after ensuring as much stability and consistency as possible between the instrument’s internal and external environments to minimize errors.

Human error

Prior to each measurement, measurement personnel received training to ensure consistency in measurement methods. During measurements, the same person was responsible for reading the data to reduce errors caused by personnel changes.

Measured oxygen concentration data validation

Considering that there were no previous oxygen concentration data on the QTP, it is generally believed in academia that the oxygen concentration is nearly constant at 20.946%13. Therefore, we conducted a one-sample t test of the 807 oxygen concentration data that were measured on the QTP during 2017–2022, with 20.946% as the comparison value. The statistical results (Table 6) showed that the oxygen concentrations that were measured in the summer and winter seasons of different routes were significantly different from 20.946%. This indicates to some extent that the oxygen concentration of the QTP is not constant and shows significant spatiotemporal differences with changes in geographical environment (elevation, temperature, and vegetation, etc.). At the same time, this also proves the relative accuracy of the measured data.

Table 6 Statistical characteristics of oxygen concentration on the Qinghai-Tibet Plateau.
Full size table

Measured FVC data validation

During the field study on the QTP from 2018–2022, we measured the FVC of 51 measurement points by using 1 km × 1 km field sampling quadrats. Based on the China FVC dataset26, we extracted the FVC remote sensing data that corresponded to the measurement position and the corresponding month and established the relationship between the measured data and the remote sensing data to check the accuracy of the measured FVC. The result showed (Fig. 5) that the correspondence between the measured data and the remote sensing data was good, and the R2 of the linear regression model was 0.88 and passed the significance test of 0.001, which indicated that the measured FVC in the field was accurate and reliable. In addition, it also showed that using remote sensing data to obtain the FVC (or LAI) could be a reliable substitute for measured FVC in the oxygen concentration estimation.

Fig. 5
figure 5

Measured FVC data validation.

Full size image

Uncertainty analysis of oxygen concentration estimation method

This study constructed the estimation model based on the cross-validation method. From the measured oxygen concentration data, a certain number of samples was selected as the training set to build the model each time, thus estimating the spatial distribution of oxygen concentration on the QTP. The standard deviation of oxygen concentration was calculated by statistically analyzing the oxygen concentration of each grid under all model estimations, which measures the uncertainty of the estimation method.

From Fig. 6, the standard deviation of oxygen concentration is on the level of 0.0001%, indicating low uncertainty. Specifically, when estimating the January average oxygen concentration, the maximum standard deviation is 0.00192%, and the minimum is 0.00010%. When estimating the July average oxygen concentration, the maximum standard deviation is 0.00166%, and the minimum is 0.00010%. For the annual average oxygen concentration estimation, the maximum standard deviation is 0.00175%, and the minimum is 0.00010%. In most regions of the QTP, the standard deviation is less than 0.0012%. However, in the southern part of the QTP, the standard deviation is relatively large, ranging from 0.0012% to 0.0016%. This is mainly due to the significant elevation variations and spatial heterogeneity in this region, as well as the limited number of measurement points. Figure 6d shows that, except for areas such as the southern and eastern parts of the QTP and the Qaidam Basin, the uncertainty of the January average oxygen concentration estimation is higher than that in July. This is mainly due to the fact that among the data measured in the field from 2018 to 2020, the amount of data measured in summer was larger, with 369 groups, while the amount of data measured in winter was smaller, with only 53 groups. In summary, the oxygen concentration estimation method used exhibits low uncertainty and provides reasonably accurate estimations on the QTP.

Fig. 6
figure 6

Uncertainty of the oxygen concentration estimation method.

Full size image

Estimated oxygen concentration data validation

By cross-validation, not only can a robust estimation model be established, but the model can also be validated. From the measured oxygen concentration data, a certain number of samples was selected each time as the training set to build the model, and the remaining samples were used as the test set to validate the model. For each training set number taken, 50,000 random trainings were performed, and the average and standard deviation of RMSE were calculated using the test set. Therefore, the relationship between the number of samples in the training set and the RMSE of the test set could be established.

From Fig. 7, as the number of samples in the training set increased, the corresponding averaged RMSE of the test set showed a downward trend. When the sample number of the training set was greater than 50, the averaged RMSE of the test set basically remained unchanged, slowly decreasing between 0.088–0.096. As the sample number of the training set further increased, the standard deviation of the RMSE of the test was larger due to the fewer samples in the test set. Considering both the average and standard deviation of the test set RMSE, the most robust estimation model of oxygen concentration on the QTP was constructed with a sample number of 76 for the training set (the standard deviation of the test set RMSE was the lowest, 0.0019), and the accuracy of the model was also relatively high (the average of the test set RMSE was 0.0952).

Fig. 7
figure 7

Oxygen concentration estimation model accuracy assessment.

Full size image

Limitations and future work

Due to various factors, our data have certain limitations. (1) Our data were mainly obtained through route/point-based field surveys, and due to limitations in manpower, material resources, financial resources, and time, we could only measure the oxygen concentration on the QTP in a line-by-line manner and could not obtain large-scale and long-term oxygen concentration data. (2) During the measurement, we used portable electrochemical oxygen meters. Despite our efforts to minimize various errors, the precision of the oxygen concentration is limited due to the instrument’s resolution and measurement principles. (3) Due to the limited understanding of the factors that are affecting oxygen concentration, only the most important factors, such as the elevation, temperature and LAI, were considered when estimating the spatiotemporal patterns of oxygen concentration on the QTP, and the influence of factors such as wind and vegetation type were ignored. Additionally, our estimation of oxygen concentration distribution data was based on the assumption that the relative contribution rates of elevation, temperature and LAI to oxygen concentration remain constant. However, in reality, the relative contribution rates of these factors to oxygen concentration can vary in different regions and at different times. Due to the current limitations in data availability, it is challenging for us to calculate the specific relative contribution rates of each influencing factor to oxygen concentration in specific regions and at specific times. (4) We only considered the linear relationships between influencing factors and oxygen concentration. However, the impact of various factors on oxygen concentration may involve nonlinearity and complexity. The effects of these factors may not simply add up linearly, and there could be interactions among them.

To further observe the oxygen concentration on the QTP, we selected 64 meteorological stations in Qinghai, Tibet, Yunnan, Sichuan, Gansu and Xinjiang as observation points and established a network for observing the oxygen concentration on the QTP. This will enable long-term observation of oxygen concentration on the QTP and allow for a deeper analysis of the factors that influence oxygen concentration on the QTP based on the observation stations. In addition, we will further improve the estimation model of oxygen concentration on the QTP, fully consider the tremendous spatial heterogeneity, study the intrinsic relationship of the factors affecting oxygen concentration, and deeply analyze the spatial and temporal characteristics of the contribution of each factor to oxygen concentration, to more accurately estimate the oxygen concentration.

Usage Notes

This dataset provides the measured and estimated oxygen concentration data on the QTP from 2017 to 2022, which is the first dataset of oxygen concentration on the QTP. This dataset fills the gap in the research on oxygen concentrations on the QTP and provides data support for oxygen concentration-related studies. For example, scholars can use these data to study the health risks of hypoxia in the people and livestock on the QTP5. Oxygen concentration is influenced by multiple factors, such as the elevation, temperature, and vegetation etc. A deeper understanding of the natural zones and geographic patterns of the QTP can be obtained based on the spatiotemporal distribution characteristics of oxygen concentration27.

At the same time, it should also be noted that due to the limitations in manpower, resources, finances, and time, the time span of the measured oxygen concentration data on the QTP is only from 2017 to 2022. The data records of oxygen concentration and related geographical environmental data only have 807 measurement points. Further, due to the limited accuracy of the model, the type of estimated data for oxygen concentration on the QTP included the annual average, January average, July average, and the difference between July and January, with a spatial resolution of 1 km × 1 km; the unit of the oxygen concentration data was %.