Microsoft has unveiled a new machine learning tool called Probabilistic Bias Correction (PBC) that promises to improve weather forecasts up to six weeks in advance. This tool could help resource-strapped weather agencies worldwide, making forecasts more reliable.

Typically, weather models become less accurate beyond two weeks, due to the complex nature of weather patterns and the data gaps that arise. However, PBC uses machine learning to correct biases in existing models, enhancing their accuracy without the need for supercomputers or powerful GPUs.

The research, published in Nature, showed that PBC improved the accuracy of a leading physics-based weather model, the ECMWF, by 98% for precipitation and 92% for temperature targets. Moreover, it doubled the accuracy of the ECMWF’s new AI forecasting system.

Genevieve Flaspohler, one of the authors, is excited about the potential of PBC, stating that it is lightweight and easy to train. This means that scientists without huge resources can work on improving specific localities, potentially making weather warnings more effective in economically developing countries.

While the tool offers hope for better weather forecasts, it also raises questions about how accurate our predictions need to be. Is a slightly less accurate but more accessible forecast better than a perfect one that requires vast resources?