Aiming at the research and application needs in the field of meteorological observation, a thunder database for meteorological observations is designed and constructed, including stunned thunder, sputum and interference sounds, with 28, 33 and 61 samples, respectively, a total of 122 samples. Firstly, samples are collected, naming rules are formulated and labeled, and then the samples of the database are evaluated by using Support Vector Machine(SVM) and Group Sparse Least Squares Regression(GSLSR) model as a benchmark. The experimental results show that the establishment of the thunder database provides an important application basis and premise for the research of the meteorological field.