File size: 2,160 Bytes
1ca0208
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
{
  "model_name": "SEEDS",
  "model_type": "seeds",
  "architectures": [
    "SEEDS",
    "SEEDSModel"
  ],
  "framework": "PyTorch",
  "domain": "atmosphere",
  "task": "conditional-diffusion-ensemble-weather-forecasting",
  "implementation": {
    "entry_point": "model/seeds.py",
    "scope": "Conditional score network for cubed-sphere weather fields; the local data contract is compatible with the project scripts and does not replace official GEFS/ERA5 preprocessing"
  },
  "architecture": {
    "family": "conditional diffusion score network with axial attention over spatial patches, variables, and forecast snapshots",
    "input_format": "BCFHW",
    "seed_input_format": "BSCFHW",
    "output_format": "BCFHW",
    "channels": 8,
    "faces": 6,
    "face_grid_shape": [
      48,
      48
    ],
    "seed_count": 2,
    "target_ensemble_count": 16,
    "patch_size": 12,
    "embed_dim": 768,
    "spatial_layers": 6,
    "field_layers": 4,
    "sequence_layers": 6,
    "mlp_ratio": 4,
    "dropout": 0.0,
    "prediction": "score_noise",
    "noise_schedule": {
      "sigma_min": 0.01,
      "sigma_max": 100.0,
      "sigma_formula": "sigma(t) = sigma_min * (sigma_max / sigma_min) ** t"
    },
    "sampling": {
      "default_steps": 16,
      "configured_solver": "euler_maruyama",
      "local_implementation": "explicit reverse sigma-schedule update without additional stochastic noise between steps"
    }
  },
  "data": {
    "dataset": "GEFS-conditioned weather ensemble data with ERA5 evaluation reference",
    "variables": [
      "mean_sea_level_pressure",
      "temperature_2m",
      "eastward_wind_850hpa",
      "northward_wind_850hpa",
      "geopotential_500hpa",
      "temperature_850hpa",
      "total_column_water_vapour",
      "specific_humidity_500hpa"
    ],
    "lead_time_days": 7,
    "normalization_note": "Scientific evaluation requires the matching anomaly/climatology normalization and denormalization; local synthetic data only checks tensor and pipeline contracts."
  },
  "configuration_sources": [
    "conf/config.yaml",
    "model/seeds.py",
    "scripts/common.py",
    "scripts/data_loader.py"
  ]
}