Rapid Update Nowcasting with AI fusion of NWP and Other Data Sources

Numerical weather predictions (NWP) models are mathematical models based on physical principles that predict many atmospheric variables, such as temperature or accumulated precipitation for a relative long time into the future. However, they are quite limited with the technology we have at this time, as well as the method of their calculation. Because of this, they offer a low granularity of their output, such as every 3 or 6 hours. The trend nowadays seems to point towards Artificial Intelligence (AI), as it has the advantage of being much faster to produce outputs as well as being fairly robust. Meteopress as a company has long history with utilizing AI forecasting models for various uses, such as nowcasting or automated severe warning systems, amongst many others. Now we have developed a model, that is able to overcome the granularity issue of NWP models. We are able to feed the model recent data and recalculate its outputs more frequently. At first, we started with using AI as a post-processing method of NWP models that was able to increase its performance. Secondly, we were able to add weather station data to the model and we observed that its outputs were twice as accurate. It also allowed us to increase the granularity of the outputs to once per hour. And as a last step, we added recent radar observation data and started recalculating its outputs every 10 minutes. This allows for very accurate local recalculation of many atmospheric variable where a radar is present and is able to help responsible people, such as meteorologists, with giving our warnings and monitoring the weather situation in real time with very frequent updates.