(1) Land use

Land use is an important manifestation of human activities, and its changes reflect the magnitude of human activities (Otto et al., 2016; Zhang et al., 2020). The land use data on the QTP were obtained from the Resource and Environment Science and Data Center (available at http://www.resdc.cn/), which covers the years 1995, 2005, and 2015. The dataset includes six land use classes and 25 subclasses. Based on the HAI, recent studies have reclassified the initial land use types and then assigned scores to each new type (Table 1) (Zhao et al., 2015; Li et al., 2018). A maximum score of 10 was assigned to built-up land, which means that this type of land use was the most important area for human activities, while lower scores (8, 8, and 7) were allocated to rural settlements, reservoirs/ponds, and croplands, respectively. The traditional nomadic lifestyle of Tibetan herdsmen enables dense grasslands to have the best grazing potential, followed by moderate grasslands (Fang et al., 2011a; Li et al., 2018). Therefore, scores of 2 and 1 were assigned to these grasslands, respectively. A minimum score of 0 was assigned to all other land cover types, which had the lowest intensity of human activities.

(2) Roads and railways

As some of the most prolific linear infrastructures, roads/railways are important drivers of habitat conversion by humans (Trombulak and Frissell, 2000), especially on the QTP. The data for 2005 and 2015 on the QTP were vectorized from road traffic maps and combined with OpenStreetMap (OSM, http://download.geofabrik.de/). Since the data for 1995 were not available, we used the global roads open access dataset (Center for International Earth Science Information Network Columbia University, and Information Technology Outreach Services University of Georgia, 2013) from 2000 instead. Detailed information can be found in Table S1. The data for the above three periods all included expressways; national-, provincial-, county-, and rural-level highways; and railways. Since the thermal disturbance of frozen ground threatens the surrounding infrastructure within a certain distance, this study added a buffer analysis to the road importance assignment. The further away from the road, the less harmful and less important the impact. The buffer range was ultimately set as 7 km in this study, instead of the 10 km used in previous studies (Li et al., 2018), and the intensity score diminished gradually to either side of the road (Table 2). A detailed explanation is provided in the Discussion section.

(3) Population density

Population density is a direct reflection of the importance of a region (Ellis and Ramankutty, 2008). The 1 km population density datasets of the QTP for 1995, 2005, and 2015 were obtained from the Resource and Environment Science and Data Center (Table S1). The population density was established based on the land use type, nighttime light, and the population distribution weight of residential density (Xu, 2017). Compared with the cities on the east coast, the QTP was extremely sparsely populated (Fang, 2013). To reasonably reflect the distribution of population density in this study, we modified the formula proposed by Venter et al. (2016a). For locations with more than 100 people per square kilometer, we assigned an intensity score of 10. For more sparsely populated areas, we logarithmically scaled the intensity score using the following formula: I n t e n s i t y   s c o r e   = 2.214 × l o g ( p o p u l a t i o n   d e n s i t y + 1 ) (1)

(4) Grazing density

The QTP is a major pastoral area of China (Li et al., 2018). Although long-term grazing activities affect the grassland vegetation biomass, they can also stimulate plant growth and improve soil fertility through excrement (Lin et al., 2010; Zhao et al., 2019). Therefore, grazing activities in the QTP were considered a key indicator in this study. The grazing density data were acquired from the United Nations Food and Agricultural Organization (UN FAO, https://data.apps.fao.org/). We overlaid the cattle and sheep densities to represent the grazing density. Since only the grazing data from 2010 were obtained, a trend extrapolation method (Duan and Luo, 2020) was used to obtain the grazing density data for 1995, 2005, and 2015. First, we calculated the change rates of cattle and sheep production from 1995 to 2015 based on the statistical yearbooks of the Qinghai Province and the Tibet Autonomous Region. Then, we used the cattle and sheep density layer from 2010 to calculate the cattle and sheep density data in these two areas for 1995, 2005, and 2015. Considering the impact of different herds on the ecological environment in this calculation, the grazing intensity of one cattle was regarded as equivalent to that of five sheep (Ouyang, et al., 2016). Finally, we converted the cattle data into the equivalent intensity of sheep and overlaid them together. Referring to the conversion method of population density, we also logarithmically scaled the intensity score using the following formula: I n t e n s i t y   s c o r e   = 2.453 × l o g ( g r a z i n g   d e n s i t y + 1 ) (2)

(5) Thawing hazard regions of frozen ground

Frozen ground degradation poses a huge threat to human activities on the QTP. However, the value of the HAI varies with different thawing hazard levels of the frozen ground. The data of the thawing hazard regions of the frozen ground were derived from our previous research results (Supplementary Figure S1; Ni et al., 2021). It should be noted that the seasonally frozen ground only experiences a periodic seasonal thaw settlement that has a lower hazard. However, in addition to seasonal settlement, permafrost also experiences a long-term, absolute ground settlement (due to the thawing of ground ice, Supplementary Figure S2) that has a relatively large hazard. Therefore, we classified seasonally frozen ground as a minimal hazard region. We divided the permafrost regions into low-, medium-, and high-hazard areas (the low hazard area in this study included the stable area) according to the hazard level of the thawing settlement.

(1) Normalization method

To eliminate the impact of the dimensions and magnitudes of each factor (various physical properties and conditions) and to effectively overlay the four types of data, we normalized the intensity score of the human activity in each type to 0–1. The following standardized processing formula was performed: x i   ′ = ( x i − x i m i n ) / ( x i m a x − x i m i n ) (3) where i is the factor number, x i and x i   ' are the initial intensity assignment and actual calculated value, respectively, and x i m i n and x i m a x are the minimum and maximum values of the factors in the study area, respectively. The higher the index value is, the greater the intensity.

(2) The analytic hierarchy process

The analytic hierarchy process (AHP) is a multiobjective, multicriteria decision-making approach that is used to analyze complex problems (Saaty and Kearns, 1985; Niu et al., 2011). This model requires the creation of a reciprocal pairwise comparison matrix used to determine weighted coefficients for the computation of HAI scores. Generally, the pairwise comparison is subjective, and the quality of the results is dependent on the experts' judgment. Hence, a consistency ratio was proposed to evaluate the accuracy of the results (Saaty and Kearns, 1985; Niu et al., 2015). In a successful expert judgement, the consistency ratio should be <0.1. In our AHP model, the consistency ratio was 0.04, indicating an acceptable assessment. The weight coefficients obtained by this method are shown in Table S2. Subsequently, we computed the total HAI on the QTP based on the AHP method: H A I = ( L a n d   u s e × 0.11 ) + ( R o a d   d i s t r i b u t i o n × 0.26 ) + ( P o p u l a t i o n   d e n s i t y × 0.41 ) + ( G r a z i n g   d e n s i t y × 0.22 ) (4)

Based on the above methods, we obtained the individual HAI maps of four categories of variables (Figure 3). The statistical results showed that the mean values of land use, road distribution, population density, and grazing density were 0.06, 0.08, 0.10, and 0.18, respectively (Table 3), which indicated that the HAI values of the four variables were low on the QTP during the study period. The high values of these individual HAIs were mainly distributed in the eastern and southern QTP. Two regions with high HAI values could be easily identified in Xining, Lhasa, and the surrounding areas (Figures 1and 3). Low HAI values were mainly distributed in the western QTP.

Through the weighted average of the four categories of individual HAI datasets, we obtained the total HAI map of the QTP for 1995, 2005, and 2015 (Figure 4). The statistical results showed that the mean HAI values were 0.10, 0.11, and 0.12, respectively. Since the theoretical maximum was 1.00, the QTP had a relatively low HAI for 1995–2015. In addition, during the study period, the percentages of grids where the HAI value was lower than the mean value were 59.12%, 58.89%, and 58.85%, respectively. These conclusions can be seen in Figure 3.

In terms of the spatial patterns of the HAI, the largest value was mainly located around the city of Xining, followed by Lhasa (city locations can be seen in Figure 1). Overall, the eastern and southeastern QTP had high HAI values for all 3 years studied. In the central part of the QTP, the extent of human activities was mainly concentrated in the QTH/QTR and other transportation corridors. In the western QTP, the HAI value was low. The main reason for this was that the region is one of the least populated regions of China due to high altitude, thin air, and permafrost. Relatively wilder areas existed on the Hoh Xil and Changtang Plateau, which were once known as “no-man's land” (Figure 1).

From 1995 to 2015, the mean value of the HAI increased by 15.35%. Grids with an increased HAI value accounted for 40.64% of the total grids, while grids with decreased HAI values accounted for 10.09%. The increased HAI regions were mostly located in the eastern and southern QTP. The detailed reasons for the increase in HAI can be obtained from Table 3. During the study period, roads and railways had the largest growth rate (28.6%); followed by population density with a growth rate of 11.1%; the smallest change was the land use, and its value remained basically stable from 1995 to 2015. No obvious HAI changes were detected in the mid-western QTP. The changes in the above data illustrated the development trend of human activities in the different areas of the QTP, and the new human footprint map can well illustrate these characteristics.

The spatial overlay of the frozen ground map and HAI map (Figure 5) showed that the permafrost region generally occurs in the hinterland and northwest of the QTP with an average altitude of more than 4,000 m. Its distribution corresponds well with the low-value regions of the HAI, except for the G219 and G109 highways and the surrounding areas. The HAI statistical results under the different thaw settlement hazard levels in the permafrost regions showed higher value of 0.09 in medium-hazard areas, while the HAIs of low-hazard areas and high-hazard areas were 0.06 and 0.03, respectively (Table 4). In contrast, seasonally frozen ground regions had relatively large HAIs, and their mean values were more than twice that of the permafrost area.

According to Figure 6, the mean value of the HAI in seasonally frozen ground areas was not only the highest but also the most dispersed. This indicated that the overall level of development in the region was relatively high but unevenly distributed. The main reason for this is that the region has favorable natural conditions and is suitable for human habitation, working, grazing, and farming compared with other parts of the QTP (Li et al., 2018). In the different thaw settlement hazard levels of the permafrost, the HAI in the high-hazard areas was the lowest, while that of the medium-hazard areas was the highest, which was mainly because most sections of the roads and railways pass through the medium-hazard areas. Overall, the HAI in the seasonally frozen ground areas was higher; however, this region only experienced periodic seasonal thaw settlement, and the hazard was relatively lower. For the permafrost areas, although the HAI value was relatively low, the roads and railways were more affected because they were more sensitive to permafrost thawing.

The map (Figure 7) of the thawing disaster warning index of the frozen ground of the QTP integrates two kinds of information: the location information related to the thawing hazard of the frozen ground and the intensity information related to human activities. We can see that the warning index in the eastern part of the QTP was obviously higher than that in the western part. The warning areas in the permafrost region were mainly distributed in the Qilian Mountains, the G219 section in the northwest, and the G109 section from Xidatan to Tanggula in the central part. Among these three typical areas, the warning indices varied between different road sections, but the main levels were high risk and warning area. In addition, the Qilian Mountains had a high or warning risk level due to the higher grazing density and road density (Figure 3). It should be noted that the previous study on thawing risk by Ni et al. (2021) was aimed at permafrost regions, so this study focused on those regions. Moreover, the warning map considered seasonally frozen ground regions due to the high intensity of human activity in these regions. However, the accuracy of the warning index in seasonally frozen ground regions needs to be further verified.

To ensure the accuracy of our results, we compared our results of the HAI map of the QTP to those of previous studies, including the global human footprint (HF, 2009; Venter et al., 2016b) and the human influence intensity (HII, 2010; Li et al., 2018) maps of the QTP. Since these two results were cumulative HAI measures, normalization was undertaken before comparison. The results revealed that the distribution patterns of the three maps were basically consistent, with high values in the eastern QTP and low values in the western QTP (Figure 8).

However, there were still some region-specific differences among the three results. In terms of our results, the mean HAI on the QTP increased by 15.35% in 1995–2015, while the corresponding increases in the HF and HII on the QTP were 30.47% and 28.52% in the 1990s to 2010s, respectively. The main reasons for these differences were the following factors: 1) The results of previous studies were a direct summation of individual layers of human activities. We comprehensively analyzed the impact of different human activity factors on the QTP and ultimately used the AHP model to assign weights to each layer, which inevitably led to different results. 2) There was a gap between the research periods. In our study, the human activity variables we considered were updated at different time scales. 3) The pasture, road, and railway data in the previous studies were static and could not reflect the dynamic information well. However, in our study, all the datasets were dynamic, which was more likely to reflect changes in human activity over time. 4) For the grazing intensity data, our study used the cattle and sheep density. Based on the statistical yearbook, we found that the grazing intensity on the QTP increased slowly (with an annual change rate of 0.45%) from 1995 to 2015. However, the 2010 HII dataset used the county-scale statistical data to represent pixel-scale data, and the 2009 HF dataset used the percentage of the pasture area in the unit pixel to represent the grazing intensity. In fact, the grazing intensity in the 2009 HF dataset was not a good representation of the grazing density, which may have led to a large bias in the results.

Accurate understanding of the spatial distribution of human activities and their changes over time is the first step to protecting human beings from natural threats. The above comparison showed that our methodology and results were reliable. This approach can be applied in other case studies as long as their characteristics are considered at a regional scale that is consistent with previous studies (Li et al., 2018).

(1) The implications of thawing disaster warning map

In a previous study, the high-hazard areas of thawing frozen ground were mainly distributed in the northwest and central parts of the QTP (Supplementary Figure S1) and were mainly affected by the distribution pattern of the ground ice content, ground temperature levels, and changes in the active layer thickness (Ni et al., 2021). However, these areas generally had low HAI values, and the potentially high-hazard areas would not cause much trouble for human activities. Therefore, the potential thaw settlement hazard map still has certain limitations for actual risk management. In this study, the thawing disaster warning map can shield high-hazard areas of thawing frozen ground without human activities and can provide a more accurate early warning. The analysis of the four individual HAI maps, combined with the thawing of frozen ground, can provide important decision support for different departments. The results from Result section showed that permafrost thawing had the greatest impact on roads and railways. Hence, early warning is needed to take effective measures, such as air-cooled ripraps, rubble slopes, ventilated pipes, and thermosyphons, to reduce economic losses (Wu and Niu, 2013).

(2) The new discovery of human activity intensity data

The updated HAI map in this study more clearly showed the location information and activity intensity. It helps to track economic development and provides more useful information for science-based protection of human activities. This study also found a series of additional uses for the HAI dataset. It can serve as an important input for ecosystem models to assess human-induced losses of ecosystem services. In addition, it can be used as a proxy for human disturbance, species extinction risk and distribution analyses, and conservation science and decision-making, among others (Venter et al., 2016a). In fact, one of the interesting uses of the HAI map may be to identify places where sensitive species thrive despite high levels of human activities and determine which human behaviors permit coexistence. The HAI map provides an important foundation for understanding conservation work at regional and global scales.

In summary, climate warming will aggravate the occurrence of thawing settlement hazards in the frozen ground regions of the QTP, which would cause detrimental effects on human activities. Research on the impacts of frozen ground degradation on human activities is a topic worthy of consideration. Collaboration among permafrost scientists, hazard specialists, and policy experts is critical for a timely and effective response to the problems associated with a thawing frozen ground. As a preliminary research, this study only considered the thawing hazard of the frozen ground brought by the slow warming of the climate and did not consider some disaster incidents. The sudden thawing of permafrost and the outburst of lakes will pose a huge threat to human activities, although they may not be common and may rarely occur (Lu et al., 2020). The results provide basic data for the study of the impact of thawing frozen ground on human activities. However, the relationship between humans and nature is mutually reinforcing. While developing the economy, humans have also caused damage to the natural environment, which has exacerbated the thawing risk of frozen ground. Nevertheless, the consequences of long-term impacts of human activities on any particular location are complex, positive or negative, benign or catastrophic, depending on the history of the location, the current type of impacts, and our willingness to shoulder responsibility for our stewardship (Sanderson, et al., 2002; Redford and Richter, 1999).

There are some uncertainties in the results that need to be understood to improve HAI assessments in the QTP in future studies. First, poor data availability on the QTP is the main reason for the limited reliability of the results. For example, the road vector data for 1995 was not available, so we had no choice but to use the data from 2000 as an alternative. For some rural roads, due to the limitation of the dataset, there are still some missing data. The grazing intensity data for 1995, 2005, and 2015 were also unavailable. In this study, these data were interpolated using data from 2010, but the impacts of subsequent policy interventions were not considered. Due to the implementation of ecological restoration projects (e.g., the Grain for Green Project and the Grazing Withdrawal Program) and the establishment of nature reserves (Fang, 2013; Ren et al., 2016; Zhang et al., 2018), grazing intensity has changed locally. To some extent, the above factors may have weakened the reliability of the results at the local scale.

In addition, we did not account for all human activity data, including wind power stations, transmission lines, and gas/oil pipelines. In general, this type of infrastructure is always distributed on the periphery of the roads. To make the results more realistic, we added a buffer analysis to the road and set the maximum distance to 7 km for a conservative estimation (in Data section). Meanwhile, traffic volumes can also represent different intensities of human activity, but the unavailability of these data forced us to omit this indicator, which may lead to a conservative assessment of activity intensity. Therefore, the QTP may have a greater HAI than we have demonstrated. More human activity datasets need to be considered in future works. Additionally, with the development of new technologies such as big data, mobile internet, and crowdsourcing platforms, data acquisition means will be effectively increased. Compared with the traditional GIS (Geographic Information System)-based mapping method, these approaches have obvious advantages.

Third, the weight assignment method also requires refinement. The semiquantitative method based on the AHP is subjective, and more accurate evaluation models need to be used in the future. In addition, more evidence shows that a reasonable allocation of activity intensity within a local scale would improve the accuracy of the results (Li et al., 2018). We treated the land surface as a blank slate, but we know that is not the case in reality. The overlay of multiple layers and the omission of some intensity data result in the same HAI region not representing the same degree of importance. Therefore, we suggest that the HAI in different areas should be interpreted with caution. In general, although the abovementioned factors may affect the accuracy of the assessment, the basic pattern of HAI will not change, and our results are representative in the current study.