Una metodologia di radar nowcast probabilistico basata sull'accoppiamento di una rete neurale PredNet con una procedura di dowscaling.

A novel probabilistic nowcasting technique for meteorological radar images based on a generative neural network (Pred- Net) coupled with a process of stochastic noise generation for the continuous dowscaling of the PredNet outputs is presented. The PredNet has been trained, tested and verified using a public domain data set of radar images, covering an area of about 104 km2 over Japan, and a period of three years with a sampling frequency of 5 minutes. An extensive analysis of the results in terms of CRPS, ROC, rank histograms and reliable diagrams shows that the proposed procedure is able to generate reliable and sharp ensemble rainfall forecasts with a quality comparable or superior to the state of the art of the available open-source alternatives.

@TechReport{MM22,
  author       = {Marrocu, M. and Massidda, L.},
  title        = {Una metodologia di radar nowcast probabilistico basata sull'accoppiamento di una rete neurale PredNet con una procedura di dowscaling.},
  institution  = {crs4},
  month        = {june},
  year         = {2022},
  keywords     = {probabilistic nowcast; meteorological radar data; deep learning; generative neural network; optical flow },
  url          = {https://publications.crs4.it/pubdocs/2022/MM22},
}