Posters

Abstract

Historically, weather forecasting has been carried out using Numerical Weather Prediction models (NWP). These models, based on observed data collected worldwide, can predict the future state of the atmosphere by solving complex mathematical equations that represent the physical laws governing atmospheric processes. This method of predicting atmospheric conditions makes these models not only complex but also cost-intensive. In recent years, there has been remarkable progress in the field of data-driven weather forecasting. This progress has been driven largely by increased observations and the development and refinement of advanced machine learning algorithms, which have enabled models to make highly accurate weather predictions. The main advantage of these models is that they can predict the future state of the atmosphere without having to solve complex physical equations, which makes them considerably less expensive. In addition, AI-based models can establish complex connections between variables that traditional models may not consider, which is very valuable in the field of meteorology. They also offer better performance and, therefore, computational savings, as well as being highly flexible in their evolution, since they can be continuously adjusted using real-time meteorological observations. Moreover, their potential for continuous learning is noteworthy. In the face of rare or extreme events, these models have the advantage of being able to retrain themselves. In this way, the model can learn about what has happened quickly and efficiently. For this reason, we have developed a new deterministic model focused on improving weather prediction in the Iberian Peninsula. This model runs four times per day, with a spatial resolution of 0.25º x 0.25º and hourly forecasts up to 72 hours. The variables it analyzes include wind at 10 and 100 meters (zonal and meridional components), temperature at 2 meters, surface solar radiation, and reduced mean pressure at sea level.