Microsoft's AI Weather Prediction Model Update

TapTechNews May 24th news, Microsoft's Start team today released a blog post through the official Bing blog, announcing that the AI weather prediction model has had another major improvement and can more accurately predict the weather conditions in the next 30 days. Now it predicts when clouds and precipitation may occur.

In the blog post, the Microsoft Start team said that it has started to improve the 'precipitation forecast' system worldwide since the end of 2021.

The model will not only use the data collected by radar devices, but also combine satellite data and analyze through AI algorithms to predict precipitation conditions more accurately.

However, the Microsoft Start team also frankly admitted that there are deficiencies in this prediction model. The daily receiving time ratio of satellite weather data is 85% to 95%, but it depends on the source of the data and the acceptable delay time.

Microsoft said:

There is evidence that each task requires a separate decoder and each prediction channel requires a separate discriminator. Therefore, the model established by WeatherfromMicrosoftStart is 4 times larger than the previous model that only predicted simulated radar reflectivity. Finally, the new model jointly predicted satellite and simulated radar reflectivity, enabling its prediction to fill the gap in data availability.

According to Microsoft, in the artificial intelligence training, the weight of the radar channel model is six times that of the satellite model, because the precipitation information provided by the radar is considered more important than the satellite task. The end result is that compared to the base model that only uses radar prediction, the new precipitation and cloud model 'has a significant improvement in the F1 score'.

TapTechNews attached the reference address