Abstract: A high-quality, long-term, and real-time-updatable frozen soil depth dataset from in-situ observations collected by National Meteorological Information Centre has been developed. This dataset is crucial for evaluation of models, reanalysis products, and satellite remote sensing products, and also provides important data support for cryospheric climate monitoring. A comprehensive quality control procedure was applied to the frozen soil depth in-situ observations, including metadata check, characteristic value check, limit check, temporal consistency check, and multi-layer ground temperature-soil freeze depth collaborative check. Relative to high-quality frozen soil depth data in A-file, the bias and RMSE of the lower limit of the first frozen layer after quality control were reduced by approximately 5—20 and 20—60 cm, respectively, compared with the original data. The proposed algorithm achieves an error detection rate of ~70% and accurately identifies erroneous historical extreme values at 194 stations, thereby ensuring the data quality of ~1100 national stations since 1954. The data completeness reaches 60%—90% in Northeast China, northern Xinjiang, and the eastern Qingzang Plateau, approximately 50%—60% in North China Plain and the southern Qingzang Plateau, and about 10%—20% in the middle and lower reaches of the Yangtze River and regions to the south, with spatial differences attributed to varying regional soil freezing durations. These spatial differences in data completeness are attributed to varying durations of regional soil freezing periods. An analysis of three major soil freezing regions (Northeast China, northern Xinjiang, and the Qingzang Plateau) reveals a decreasing trend in the maximum frozen soil depth, which has been particularly significant since the mid-1980s and is closely associated with climate change.