Prior to this, predicting the short-term weather events was more of a tough challenge. The numerical methods that record atmospheric dynamics, ocean effects, thermal radiation, and other processes become limited because of the computational resources. For example, even giants like the National Oceanic and Atmospheric Administration (NOAA) gets to collect a data of around 100 terabytes per day.
The numerical method is also slow as it takes a number of hours to compute one round of forecast. In more common scenarios, experts get to commute the forecast after 6 hours, which further leads them to only 3-4 runs a day and 6+ hour old data of forecasts.
By tackling the similar issues, the search giant hopes to help people when it comes to making the important immediate decisions like traffic routing, logistics, or evacuation planning as well.
With their method, Google is planning to use radar data and deal with weather prediction as a computer vision problem. The team working at it has set up a neural network that will learn about atmospheric physics with real examples and with no inclusion of previous knowledge regarding how the atmosphere operates.
In one of the examples related to the kind of predictions the system can generate, a sequence of radar images throughout the past one hour gets used to predict the radar image of N hours from now and N basically ranges from 0-6 hours.
There is no doubt in the fact that Google’s machine learning powered rain forecasting method outclasses all of the three popular forecasting models that have been used till date. The predictions coming out are instantaneous and hence the forecasts turn out to be based on fresh data for short term.
Google is also looking forward to join its system together with High Resolution Rapid Refresh (HRRR) with an aim to make long-term forecasts better as well since HRRR would go with 3D physical model.
Photo: 400tmax via Getty Images
Read next: Google Maps Introduces Hyperspace Animation When Switching Between Planets
Credit: Google News