2.1. Data sources
1) Ground temperature data
The MAGT is an important factor that reflects the thermal state of permafrost, and is defined as the ground temperature at the zero annual amplitude depth (ZAA, i.e., the depth at which the annual temperature variation < 0.1°C) (Qin, 2016). Due to the harsh environment of the QTP, some boreholes are measured manually using a multimeter once each year (Qin et al., 2017). Most MAGTs, however, are not easily accessible from the ZAA. In these cases, the temperature at or closest to 10 m below the ground surface was used (Nan et al., 2002; Liu et al., 2017). All disturbed measurement sites (e.g., sites submerged by the rising waters of a lake) were removed. Ultimately, 84 MAGT sites (Figure 1) were selected from both field station observations (Cryosphere Research Station on the Qinghai-Tibet Plateau, Chinese Academy of Sciences, available athttp://www.crs.ac.cn/) and the related literatures (Wu et al., 2012a; Qin et al., 2017; Wang et al., 2017). We selected the period from 2000 to 2015 as the reference period, and all observations obtained were during this period. Some sites were based on one year of observation, while others were based on the average of several years, from which we calculated the long-term average value.
2) Active layer thickness data
In order to better fit the ALT, we attempted to collect a large amount of observed data from relevant literatures (Wu et al., 2012a; Qin et al., 2017; Wang et al., 2017). An additional portion of the active layer data came from field pit detection. A total of 77 ALT observation sites (Figure 1) were selected. The time node selection and disturbance data processing for ALT were the same as those used for the MAGT. Based on the distribution of MAGT and ALT observation sites, we divided them into five typical regions, the Wenquan typical region (WQIR), Xikunlun typical region (XKLIR), Gaize typical region (GZIR), Aerjin typical region (AEJIR) and Qinghai-Tibet Highway typical region (G109IR), which represent the permafrost regions of the eastern, western, southern, northern and central areas of the QTP, respectively.
3) Meteorological data
In order to obtain climate data for the reference periods (2000–2015), the China Meteorological Forcing Dataset (CMFD) (available athttp://www.tpedatabase.cn/; Yang et al., 2010b; Yang et al., 2010b; He et al., 2020) with temporal and spatial resolutions of 3 hours and 0.1° × 0.1°, respectively, was utilized in this study. The time scale of the dataset covered the studying period. According to the study of He et al. (2020), the CMFD was established by merging Princeton reanalysis data, GLDAS data, GEWEX-SRB radiation data, and TRMM precipitation data, as well as the regular meteorological observations made by the China Meteorological Administration. The accuracy of CMFD is between the observation data and the remote sensing data (Yang et al., 2010b), and it has been widely used due to its high reliability (Xue et al., 2013; Xu et al., 2017a; Wang et al., 2019a).
In the study, we used air temperature and precipitation data from the CMFD to calculate the two key predictors, including the thawing indices (thawing degree days, TDD) and the freezing indices (freezing degree days, FDD), which play essential roles in the studies of the frozen ground. As useful indicators, they have been widely applied in the permafrost region to predict the ALT (Zhang et al., 2005; Nelson et al., 1997; Peng et al., 2018; Shiklomanov and Nelson, 2002) and permafrost distribution (Nelson and Outcalt, 1987). In addition, we also calculated the other two predictors, including the solid precipitation (i.e., precipitation with a temperature below 0°C, Sol_pre), and liquid precipitation (i.e., precipitation with a temperature above 0°C, Liq_pre).
For future conditions, the BCC-CSM 1.1 climate change modeling data was used (available athttp://www.worldclim.org/). It was downscaled GCMs data from CMIP5 (IPCC Fifth Assessment). BCC-CSM1.1 is the version 1.1 of the Beijing Climate Center Climate System Model, the coupling was realized using the flux coupler version 5 developed by the National Center for Atmosphere Research (NCAR) (Wu et al., 2019). It was a fully coupled model with ocean, land surface, atmosphere, and sea-ice components, and was often used to simulate the response of global climate to rising greenhouse gas concentrations, the performance is satisfactory in China (Zhang and Wu, 2012b; Xin et al., 2018). In this study, we chose the monthly average air temperature and precipitation over the time period 2061–2080 under three Representative Concentration Pathways (RCPs): RCP2.6, RCP4.5, and RCP8.5 (Moss et al., 2010; Taylor et al 2012). The four predictors (TDD, FDD, Sol_pre, and Liq_pre) were recalculated in the same way for each time period and RCP scenario.
4) Geospatial environmental predictors
The geospatial environmental predictors were mainly derived from topographic data and regional environmental data. The Shuttle Radar Topography Mission (SRTM) data for a 1-km spatial resolution digital elevation model (DEM) (Reuter et al., 2007) were selected to calculate the predictors of elevation (Ele) and potential incoming solar radiation (PISR) (McCune and Keon, 2002). Soil organic matter is also an important factor affecting the ALT of permafrost. Due to the low decomposition rate of organic matter, high soil organic carbon has been accumulated in the permafrost regions (Ping et al., 2008). The adiabatic properties of organic matter relative to minerals will reduce the heat exchange between ground and air (Mölders and Romanovsky, 2006; Nicolsky et al., 2007; Paquin and Sushama, 2015). Moreover, the organic matter can also affect the thermal properties and the amount of unfrozen water of soil (Romanovsky and Osterkamp, 2000; Nicolsky et al., 2009). In order to consider the influence of the regional organic matter content (Wu et al., 2012b), soil organic carbon content information (SOC, ton·ha-1) from global SoilGrids 1-km data (available at https://soilgrids.org; Hengl et al., 2014) was also used in our prediction analysis. Finally, all of the data layers were resampled to the matching spatial resolution (0.1°×0.1°) and cropped to the study area (QTP).
5) Glacier and lake data
The spatial distributions of the glaciers and lakes on the QTP were collected from the Second Glacier Inventory Dataset of China and the Chinese Cryosphere Information System provided by the Cold and Arid Regions Science Data Center (http://westdc.westgis.ac.cn).