The data consists of a set of meteorological quantities including tropospheric delays (zenith wet delay, zenith hydrostatic delay, and zenith total delay), precipitable water vapor, and surface temperature and pressure. The data is retrieved from the observations of 414 globally distributed radiosonde stations from 2014 to 2019. In addition to the geographic information of radiosonde stations, the profiles of tropospheric delays and precipitable water vapor are contained in the data file. This data has a wide range of applications, e.g., validating the tropospheric delays and precipitable water vapor derived from other techniques, investigating the spatial-temporal variations of water vapor, and acting as training data of machine learning to build tropospheric delay models. In the manuscript "Machine Learning-based Model for Real-time GNSS Precipitable Water Vapor Sensing", this data is used to train a machine learning model to map the zenith total delays to precipitable water vapor. The data is split into training data and test data, where the data from 2014 to 2018 are employed for model training, and the data of 2019 are used for testing. If you use these data for scientific research, we would appreciate a cite of both the data and the paper. A suitable reference is: Zheng, Y., Lu, C., Wu, Z., Liao, J., Zhang, Y., & Wang, Q. (2022). Machine Learning‐Based Model for Real‐Time GNSS Precipitable Water Vapor Sensing. Geophysical Research Letters, 49(3), e2021GL096408.