| dc.description.abstract | Predicting near-future rainfall, defined as nowcasts, is essential to various industries and sectors, including aviation safety, agriculture, flood management, and the general public. Traditional nowcasts are calculated using optical flow methods, which extrapolate radar observations. However, these methods have shortcomings and struggle to capture important non-linear events. Deep learning has shown promising results in mitigating these challenges. Especially promising are frameworks built on generative models (GM), enabling the generation of realistic precipitation scenarios. Recent years have seen a tremendous advance in a specific form of GMs called Diffusion Models (DM). To the best of our knowledge, these models have yet to be explored for use in nowcasting or weather forecasting in general. With this as motivation, this thesis presents a novel approach to precipitation nowcasting using DMs. | eng |