How to use machine learning to predict the weather?

At present, the colorful cloud weather is mainly deduced by using the radar map on the website of the National Meteorological Administration, and then the moving trend of echo intensity is predicted by the algorithm, and then some weather phenomena are predicted. In fact, it is also a machine version of the "extrapolation method" commonly used by forecasters. This algorithm is generally accurate in forecasting the development trend of weather system. After all, in the era when there was no numerical forecast, weather charts were also analyzed and predicted by extrapolation. But if the weather in Cai Yun is a short-term weather forecast made by machine learning, I think it is not comprehensive, mainly for the following reasons: Weather radar can only detect and predict some weather phenomena. Weather radar is the main means to detect precipitation system, and it is also one of the main tools to monitor and warn severe convective weather. Moreover, with the development of technology, traditional weather radar can only measure echo intensity, that is, it can only provide reflectivity factor products. On this basis, the new generation Doppler weather radar can also provide the radial velocity product and the velocity spectrum width product which represent the fluctuation degree of velocity. In other words, radar maps are mainly used for detecting, monitoring and early warning precipitation phenomena and severe convective weather, and other weather phenomena can not be effectively monitored at present, let alone predicted by radar products. At present, only the basic reflectivity is dealt with. At present, only the mosaic of basic reflectivity can be seen on China Weather Network, and the weather in Cai Yun is only using this data. Radar reflection map is not the real-time distribution (or forecast map) of precipitation, and the radar principle is very complicated, so I won't introduce it. Generally speaking, the greater the reflectivity, the greater the water content in the cloud, the stronger the updraft in the cloud, and the easier it is to form strong cumulus clouds (that is, clouds that thunder and rain in summer). Moreover, the radar echo map is generally judged by the forecaster by looking at the moving direction of the system and some remarkable characteristics of the echo (such as hanging echo, hook echo, etc.). The radar reflectivity factor of your picture is very small, maybe there is only a little cloud. In fact, if you see an echo above 45dBZ (that is, red) on the echo map, it indicates that there may be strong convective weather in a few hours. Of course, this is only possible, not necessarily, and it needs to be judged by combining many conditions. For large-scale precipitation, such as stratiform clouds, the reflectivity factor will be much smaller, but you can see patches of distribution areas on the map.