Google’s New AI Models May Make for Better Weather Predictions

Predicting the weather has always been a bit of a risky business. In ancient days, if a meteorologist failed to predict the weather accurately enough over a period, he was likely to lose his head. Today’s weather prediction systems are less based on guesses and more based on technology. However, trying to accommodate the massive amounts of data that is required to predict weather patterns accurately means that there is less processing time left to offer up results. Google’s AI engineers believe they have found a way around this processing problem.

The Current Issue With Time

Currently, the systems used by traditional forecasting combine two methodologies – Optical Flow (OF) predictions, which look at present cloud formation and their movement, and detailed physics-based simulations. The problem with these methods is that they can be quite time-consuming to process, especially utilizing the entire collected data set. As a result, these systems can only complete one simulation for weather systems every six hours or so. The biggest problem here is that the results will always be dated since they’re based on data collected so far back in the past.

Google’s Solution is Much Quicker

Google’s AI believes it has a way to increase the speed of predictions. In a blog post published by the company, its researchers have stated that they have been able to accurately predict rainfall up to six hours before it happens, using only minutes of calculation. With climates and weather patterns changing rapidly, faster prediction times for weather phenomena may be a useful innovation.

Speed Improvement Through Simplification

Google’s increased speed of prediction is partially based on the AI’s ability to use simple radar data as a proxy for rainfall. The system was trained on radar data ranging between 2017 and 2019 across the continental US, collected by the National Oceanic and Atmospheric Administration (NOAA). Google claims that its AI’s predictions were as good or better than existing forecasting methods. AI hasn’t really made an impact on the field of meteorology up to now, and Google’s attempt is the most successful one thus far. Other companies such as IBM and Monsanto are researching their own methods for AI-based weather prediction.

CloudWedge