Sudden rain can spoil outdoor recreation or make you reconsider your plans. Therefore, people follow weather forecasts. However, even forecasters are not always able to accurately predict the behavior of the elements. According to representatives of Google, the use of neural networks will help humanity get a more accurate summary of the weather forecast in almost real time.
According to Google researchers, during the test they were able to generate precipitation forecasts with an accuracy of 1 kilometer six hours earlier than using traditional methods. Moreover, if the AI predicted rain, then the time of precipitation coincided up to a minute.
In multiple experiments, radar data were used to predict precipitation. The main advantage of the Google approach is the speed of forecasting. In addition, unlike the neural method of analysis, the traditional one requires the daily processing of hundreds of terabytes of data on super powerful computers, which makes increasing the frequency of making forecasts extremely costly.
Google’s algorithm allows you to get a weather forecast on demand in minutes. To train the neural network, radar data was collected from 2017 to 2019 in several US states. The authors of the project are sure that it is the future that is behind this forecasting method, and gradually large companies will abandon the classic calculation of the weather forecast due to its inefficiency and slowness.