Researchers have developed IRENE, a deep learning model designed for short-range precipitation forecasting in Italy. This model utilizes an encoder-forecaster architecture with multi-scale Convolutional Gated Recurrent Units (ConvGRUs) and is trained on data from the Italian Civil Protection Department. IRENE incorporates an importance-sampling scheme to focus on precipitation events and uses a probabilistic loss function. Variants of IRENE, including an adversarial version (IRENE-GAN) and a spectrally constrained version (IRENE-GAN-RAPSD), were evaluated against existing methods like STEPS and DGMR, demonstrating improved probabilistic skill and ensemble calibration. AI
IMPACT This model offers improved probabilistic skill and ensemble calibration for short-range precipitation forecasting.
RANK_REASON The cluster describes a new deep learning model presented in an arXiv paper. [lever_c_demoted from research: ic=1 ai=1.0]
- Continuous Ranked Probability Score
- ConvGRUs
- Convolutional Gated Recurrent Units
- DGMR
- IRENE
- IRENE-GAN-RAPSD
- Italian Civil Protection Department
- Italian Radar Ensemble Nowcasting Experiment
- STEPS
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
