Written by 4:00 pm AI, GraphCast

**Enhancing Weather Forecast Accuracy: DeepMind AI Outperforms Microprocessors**

DeepMind’s new machine learning algorithm takes less than a minute to make its forecasts and …

According to Google DeepMind, a machine learning algorithm it has developed can predict the weather more accurately than current supercomputer-based forecasting methods.

In comparison to the European Centre for Medium-Range Weather Forecasts (ECMWF) High Resolution Forecast (HRES) system, Google’s model, GraphCast, generated a more precise 10-day forecast in minutes rather than hours. The latest gold-standard climate simulation system developed by Google DeepMind is referred to as HRES.

As per a study published on November 14 in the journal Science, GraphCast, which is compatible with a personal computer system, outperformed ECMWF in 90% of the 1,300 tested weather elements.

However, researchers note that the model is not flawless as the results are produced in a black box, preventing the AI from explaining its methodology or demonstrating its functionality. They suggest that these AI tools should complement existing tools rather than replace them.

The question arises: Is the wind exacerbating due to climate change?

Current forecasting methods rely on complex natural models that integrate data and run simulations on supercomputers. These simulations are resource-intensive and costly to operate, with their accuracy contingent on intricate model details.

On the other hand, machine learning climate models are more cost-effective as they require less computational power and execute tasks more rapidly. Researchers trained GraphCast on 38 years of global Earth weather data up to 2017 to develop the new AI model. Despite this, researchers were unable to decipher the relationships the model established between variables like air pressure, temperature, weather conditions, and humidity.

Following this training, the model swiftly generated 10-day forecasts using 2018 global climate data. It was found that GraphCast provided more precise estimates for over 90% of the 12,000 data points considered, compared to ECMWF’s high-resolution forecast, which relies on traditional physical principles for predictions.

By focusing on the lower atmospheric levels, specifically the stratosphere where weather events affecting people occur, GraphCast successfully predicted extreme temperatures such as heatwaves, cold spells, and tropical storms with over 99% accuracy.

Rémi Lam, a DeepMind research expert, stated that a live version of GraphCast deployed on the ECMWF website accurately predicted Hurricane Lee’s landfall in Nova Scotia nine days in advance in September. In contrast, traditional forecasts only pinpointed Nova Scotia about six days prior.

Despite its impressive performance, scientists believe that the model will not immediately replace conventional tools. Traditional forecasts are necessary to validate and provide initial data for any prediction, and machine learning algorithms may encounter issues or “hallucinations” by producing inexplicable results.

AI models could be utilized alongside existing forecasting techniques to expedite predictions and offer insights into evolving weather patterns, according to analysts. The utilization of AI in weather forecasting is projected to benefit billions of individuals, aiding in a better understanding of climate trends and potentially addressing significant environmental challenges through innovative tools and accelerated research.

Visited 1 times, 1 visit(s) today
Last modified: December 1, 2023
Close Search Window