A method to visualize and categorize temperature, precipitation and wind speed data nationwide using the past annual meteorological data hourly observed by AMeDAS was investigated. First, by imaging the annual meteorological data of AMeDAS, the features that can be visually read were organized. Next, based on the idea of a deep neural network, a concrete method for dimensionally compressing the imaged meteorological data using auto encoder was shown. Furthermore, process of cluster analysis of the results of dimensional compression was shown. Similarity of data between measurement points nationwide was visualized, and its validity was also evaluated.