Just a few weeks ago, Microsoft announced that its Start weather team had developed new AI prediction models that seem to improve the forecasting of weather during a 30-day time period. Today, the company announced another big improvement in its prediction models, this time for forecasting when clouds and precipitation might occur.
In a post on its Bing blog, Microsoft stated the Start team had launched improvements in what it called 'precipitation nowcasting" worldwide in late 2021. The model took data from not only local radar installations but also combined it with satellite data. This combination was needed as many parts of the world don't have weather radar hardware set up.
However, Microsoft Start felt the precipitation prediction model still had some flaws. This was due to the satellite weather data being available 85 to 95 percent of the time, depending on where it came from, with acceptable latency.
The blog post stated:
With evidence suggesting the need for a separate decoder per task and a separate discriminator for each predicted channel, Weather from Microsoft Start built a model 4X bigger model than the previous one that only predicted simulated radar reflectivity. Finally, the new model jointly predicts both satellite and simulated radar reflectivity, enabling its predictions to fill data availability gaps.
The radar channel model was given six times more weight in AI training than the satellite model because the precipitation info from the radar is considered to be more important than the satellite task, according to Microsoft.
The final result was that the new precipitation and clouds model had "a marked improvement in F1-score:" compared to the base model, which only used radar-based predictions. The blog added:
Additionally, it was observed that predicted satellite images score better than a persistence forecast after 15 minutes, meaning these predictions can be used when satellite outages last longer than 15 minutes.
Microsoft's Satellite and Radar forecasting model has now been fully integrated into the company's Weather from Microsoft Start's weather models.
6 Comments - Add comment