The generation of this dataset is based on multi-source meteorological forcing data, land surface environmental data, a high-fidelity benchmark constraint map, and in-situ observation data used for independent validation. The primary data sources include:

 1. Meteorological forcing data: The China meteorological forcing dataset (He et al., 2020) was adopted, containing 7 meteorological elements (near-surface air temperature, surface pressure, specific humidity, wind speed, downward shortwave radiation, downward longwave radiation, and precipitation rate) at a spatial resolution of 0.1° and a temporal resolution of 3 hours from 1979 to 2018. Retrieved from https://doi.org/10.11888/AtmosphericPhysics.tpe.249369.file.

 2. Land surface environmental data: Vegetation types were derived from the Vegetation Map of the People's Republic of China (1:1,000,000) (Zhang, 2007), retrieved from https://doi.org/10.12282/plantdata.0155; Leaf Area Index (LAI) climatology was sourced from the GIMMS product (Zhu et al., 2013); Soil texture data were obtained from A Multilayer Soil Texture Dataset for the Qinghai-Tibetan Plateau (Wu et al., 2016).

 3. Spatial calibration benchmark map: The 2010 permafrost distribution map over the Qinghai-Tibet Plateau (Cao et al., 2023), used as the spatial constraint target, was retrieved from https://doi.org/10.12072/ncdc.permalab.db3965.2023.

 4. Glacier and lake mask data: Used to exclude water bodies and glacier areas. The glacier mask was derived from The Second Glacier Inventory Dataset of China (version 1.0) (2006–2011) (Guo et al., 2015), retrieved from https://doi.org/10.3972/glacier.001.2013.db; the dynamic lake masks were sourced from The lakes larger than 1 km² in Tibetan Plateau (v3.1) (1970s-2022) (Zhang et al., 2019), retrieved from https://www.tpdc.ac.cn/zh-hans/data/7fee8675-d4ab-4f97-8bbf-269e20b7ac16/.

 5. Validation data: Consists of in-situ measurement data from 12 active layer monitoring sites and 84 boreholes (Zhao et al., 2021), retrieved from https://doi.org/10.11888/Geocry.tpdc.271107. .

 Based on the modified Noah-Tibet land surface model, the specific workflow is as follows: First, to establish quasi-equilibrium initial thermal and hydrological conditions, the model underwent a 500-year spin-up by repeatedly recycling the first five years of forcing data (1979–1983). Following this spin-up phase, the main simulation was driven by the transient climate forcing from 1979 to 2018. Second, we implemented the CaliMAP strategy utilizing a parameter-ensemble approach. To improve optimization efficiency and reduce parameter equifinality, the calibration was focused exclusively on "sensitive regions", i.e., transition zones encompassing over 13,000 cells where initial default simulations disagreed with the 2010 map. Within these sensitive regions, 26 heterogeneous parameters were tuned using a Bayesian optimization algorithm to maximize the Kappa coefficient of agreement between the simulated 2010 permafrost distribution and the 2010 map. Rather than selecting only the single deterministic optimal parameter set, we ranked all evaluated sets based on their final Kappa coefficients and selected 13 best-performing parameter combinations from the optimization posterior to construct a parameter ensemble. The Noah-Tibet model was then driven by these 13 optimal parameter sets to generate 13 independent historical simulations (1980–2018). The final frozen ground distribution dataset represents the ensemble mean of these 13 simulations. Finally, pixel reclassification was physically based on the simulated thermal state of the 15.2 m soil column. Specifically, a grid cell is classified as permafrost if the simulated soil temperature of at least one layer remains at or below 0 °C continuously throughout both the current and the preceding year. Terrestrial cells that do not meet this continuous freezing criterion are further differentiated based on their seasonal thermal dynamics: they are designated as SFG if the soil experiences periodic freezing during the cold season, or as non-frozen ground if the entire soil column remains unfrozen year-round. Furthermore, modeling cells corresponding to existing glaciers and large lakes are explicitly excluded from the frozen ground classification.

 The dataset’s thermodynamic reliability was evaluated against an extensive network of independent in-situ observations across the QTP. Crucially, these observation sites were strictly spatially independent from the data used in the CaliMAP procedure. The assessment focused on three key permafrost thermal state variables derived from the simulated soil temperatures: ALT, TTOP, and ground temperature at a 10m depth (GT10m). Performance was quantified primarily using the root mean square Error (RMSE). The evaluation yielded an RMSE of 0.68 m for ALT (n= 83 samples), 0.41 °C for TTOP (n=83 samples), and 1.30 °C for GT10m across a comprehensive compilation of 291 samples.

 The simulated long-term change rates were further evaluated against observed multi-year time series. The dataset achieved an RMSE of 0.08 m/10a for ALT change rates (n=8 sites), 0.36 °C/10a for TTOP change rates (n=8 sites), and 0.64 °C/10a for GT10m change rates (n=38 borehole locations).