In the last few decades, weather forecasts have become highly advanced and capable of accurately predicting if rain is expected in the foreseeable future. However, there is still a wide area of opportunity for further development. At present, superconductors are used for forecasting wherein they crunch a large amount of atmospheric data. Most weather forecasters and other experts agree that these systems can correctly predict long-term weather patterns. Even so, short-term forecasting remains underdeveloped.
This scenario might soon change as a research team has developed a method to predict the weather of a given place for the coming two hours or 90 minutes. They accomplished this by applying their concepts of deep learning to the science of "nowcasting." The new tool may revolutionize the Weather Forecasting Systems and Solutions Market as it will enable agencies to predict short-term weather, something which traditional tools have yet to accomplish.
Short-term forecasting is still in its initial stages. The subject of forecasting whether it will rain and how much rain will fall in a specific area in the next two hours is of particular importance. To be sure, some short-term forecasts are straightforward—when enormous rainclouds span hundreds of miles, everyone will get wet. The forecasting of thunderstorms is challenging because the amount of water they contain varies over time, and their forms change as they move across land. This has made the 'nowcasting' a significant challenge hitherto.
As per the new research, the team applied DGMR (Deep Generative Model of Rainfall), a deep-learning network, to the ongoing problem. The algorithm uses a concept known as generative modeling. Like other deep-learning systems, it accomplishes the task by analyzing the describing patterns (here, the weather patterns) as they have evolved through time. Then with the help of the information received, it makes predictions for the next 90 minutes. For the present study, the data was revealed by the U.K.'s national weather service and the Met Office.
The system, DGMR, was evaluated for its accuracy wherein its forecasts were taken by 56 weather forecasters and compared with traditional forecasting tools. It was noted that 89% of the forecasters chose DGMR's predictions over others as they were much more reliable. The team has also suggested that their research is an excellent example of the presence of AI in weather forecast systems as it helps develop powerful new tools.