Abstract: With the development of high-resolution numerical model, ensemble forecast and using of artificial intelligence, modern weather forecast has entered the era of fine, seamless and digital intelligent grid forecast. In order to study the training needs of using numerical prediction, this paper analyzes the operational requirements, the current ability of forecasters, and the gap between them. The results show that in the refined wisdom grid forecast, it is critical for forecasters to improve the skills of using numerical forecast and multi-source meteorological information. On the basis of understanding the basics of new numerical technologies and their products, the training needs on operation-oriented research and targeted services are also shown. The most required training on this aspect includes the principle of new numerical technologies, their evaluation and correction, and the comprehensive strategy of using multi-source meteorological information. The training needs vary with the level of forecasters, the new forecasters have relatively basic needs, the senior forecasters focus more on practical technologies, and the chief forecasters concern more about research. In the end, suggestions are made to give support to curriculum design on the numerical weather prediction.