689 The Application of Machine Learning in "Understanding Clouds"

Tuesday, 9 January 2018
Exhibit Hall 3 (ACC) (Austin, Texas)
Jing Zhuang, Moji Weather, Beijing, China; and L. Ding and K. Yue

Handout (3.9 MB) Handout (3.9 MB)

Clouds are a common and important weather phenomenon. The shape, quantity and variety of different clouds are important in weather forecasting, especially in short-term predictions.

In the past, cloud information was mainly recorded by artificial observation.

This article is based on the real-time image of MoJi weather app, through multiple machine learning models to deal with and analyze the features of the cloud, and get a set of algorithms for recognition of cloud.

MoJi weather is taking this algorithm to the product experience which will be one of the function of the weather APP in the near future. Through this function, users can deepen their understanding of the cloud, and help them keenly aware of the weather changes, so as to understand our climate and enhance the awareness of disaster prevention. It also increases the interest of the product and attracts more users to participate and interact with us. Users can also give us feedback to identify cloud specific information.

In the process of identification, the machine will calculate the maximum probability of three kinds of cloud. The user will also select a cloud that he thinks is closest to the image. According to the feedback and we will test the data for learning, to improve the accuracy of recognition.

At the same time, a large number of user’s feedback results and picture information, which can be with other weather data, will become an important auxiliary data source for weather identification, especially in the lack of observational data, it will also have a greater value.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner