Precipitation modeling on HRRR64 Mini (100k samples)
0.755Extreme Event Frequency ErrorQuantile
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| QuantileNoise distribution=Learned Quantile, Network architecture=U-Net, Number of parameters=35M, Resolution=64x642025.10 | 0.755 | 0.0634 | 1.1063 | 0.0393 | 1.588 | 0.58 | |
| Student-t baselineNoise distribution=Student-t, Network architecture=U-Net, Number of parameters=35M, Resolution=64x642025.10 | 0.8859 | 0.1482 | 2.0719 | 0.1014 | 2.89 | 0.83 | |
| Gaussian baselineNoise distribution=Gaussian, Network architecture=U-Net, Number of parameters=35M, Resolution=64x642025.10 | 0.9689 | 0.2455 | 3.1836 | 0.2067 | 4.93 | 1.157 |