Abstract: In this paper, the field observation and new generation geostationary meteorological satellite data are utilized to conduct the low visibility training experiments on Zhejiang Jinliwen Expressway with six machine learning methods, including logistic regression, linear discriminant analysis, K-nearest neighbor algorithm (KNN), classification and regression tree, Gaussian naive Bayes, and support vector machine (SVM). In addition, this paper also incorporates a variety of methods to evaluate the training results. The SVM algorithm demonstrated a good training effect, particularly for recognizing visibility of ＜1000 m. Further, the integration of the ground observation and satellite data to establish the recognition model exhibits that the modeling results are better than a single source. Generally, the KNN algorithm has improved the modeling effect. Thus, to distinguish between dense fog and strong dense fog, the model combining the ground and satellite data has enhanced the recognition results. The SVM algorithm was employed for single source data, whereas the KNN algorithm was utilized for ground as well as satellite data to distinguish the fogging process on the Jinliwen Expressway. The results demonstrate that the recognition effect of the new generation geostationary meteorological satellite data is not only equivalent to the field observation recognition but also capable of distinguishing between morning and night fog, which can be utilized as an effective functional extension to the field observation recognition. Therefore, this research will provide an auxiliary reference for the identification and prediction of low visibility, thereby eliminating the requirement for field meteorological observations in the province.