Firstly, the newly released QTP Permafrost Thermal State Comprehensive Dataset provides 65 boreholes with soil temperatures at depths of 10 and 20 meters. Therefore, the determination of whether there is permafrost at the drilling locations before and after 2010 is based on the annual average soil temperature of the two depths mentioned above, to determine whether there is permafrost at the drilling locations before and after 2010. These boreholes are further divided into three categories: boreholes in stable permafrost, boreholes in unstable permafrost, and boreholes with seasonal frost; Secondly, 7 boreholes were collected from existing literature, which provided information on the presence of permafrost in the Yellow River source area, which is a key area in the eastern part of QTP. The ground temperatures of these boreholes were measured in the summer of 2013 and 2014, and it is assumed that they reflect the thermal mechanism of the Yellow River source area. If the soil temperature at a depth of 15 meters is higher than 0.5 degrees Celsius, the drilling location is classified as a seasonal frost zone. Otherwise, it will be classified as permafrost.

   Apply the FROSTNUM/COP mapping method developed by Hu et al. to draw the distribution map of permafrost on QTP. It is based on an extended surface frost quantity model provided by satellite temperature data (FROSTNUM), which requires sub regions of the permafrost distribution map as optimization constraints. This method takes into account the factors of the local model parameter E, whose values have been optimally determined for all spatial units following spatial clustering, parameter optimization, and decision tree procedures. DDF and DDT are products calculated based on Level 11 MOD1A11 and MYD1A3 (version 6). Due to turbidity, there is a gap in MODIS LST data, resulting in uncertainty in drawing permafrost maps based on these data. Select the region in the MODIS LST data gap to insert the geometric (SCSG) effect based on the sun cloud satellite stepwise interpolation method. The Particle Swarm Optimization (PSO) algorithm is used to find the optimal value of E related to each soil cluster. In this population-based heuristic method, candidate solutions are guided towards the most well-known position in the search space, achieving very fast convergence to the optimal value. In previous studies, the only objective function was Cohen's Kappa coefficient, which quantified the simulated map and the survey based distribution map of permafrost in the subregion. Despite the performance benefits achieved in the field of experimental research, this relatively simple objective function inevitably leads to equivalence in larger regions such as QTP. Recognizing that the kappa coefficient is the overall consistency between the simulation results and the sub regional map, we retained the kappa coefficient and created an objective function to make it more stringent by adding specially defined boundary consistency.