This experiment contains two datasets: corrected surface air temperature data and training data. The corrected surface air temperature data are generated by integrating observation temperature data from six glaciers in the high-altitude areas of the Qinghai-Tibet Plateau and thermal infrared remote sensing data. In regions with sparse observation stations, the limited distribution of stations leads to inaccuracies in the estimation of near-surface air temperature. In such cases, remote sensing data provide better spatial coverage. The L2SP dataset includes Surface Temperature (ST), and the L2SP image data are atmospherically corrected. Surface temperature is derived from the ST band based on thermal infrared remote sensing information. The Digital Number (DN) of the standard L2SP is stored in a 16-bit unsigned integer format, and ST is calculated using the Single Channel algorithm. Supervised learning regression algorithms, including the Decision Tree Regressor and Random Forest Regressor, are used to correct biases and inaccuracies, thereby updating the spatialized near-surface air temperature. The datasets is suitable for climate research, environmental monitoring, and other applications requiring relatively accurate surface air temperature data.