{
  "model_name": "GlobalSurgeML",
  "model_type": "globalsurgeml",
  "architectures": [
    "GlobalSurgeML"
  ],
  "framework": "PyTorch",
  "domain": "physical-oceanography",
  "task": "daily-maximum-storm-surge-regression",
  "implementation": {
    "entry_point": "model/globalsurgeml.py",
    "scope": "core-method and real-feature-dimension scaled end-to-end engineering reproduction",
    "train_script": "scripts/train.py",
    "inference_script": "scripts/inference.py",
    "evaluation_script": "scripts/result.py",
    "synthetic_data_script": "scripts/fake_data.py"
  },
  "architecture": {
    "family": "station-wise PCA stepwise-linear and random-forest regression ensemble",
    "configurations": [
      "LR-RS",
      "LR-RS-lag",
      "RF-RS-lag",
      "LR-AR",
      "LR-AR-lag",
      "RF-AR-lag"
    ],
    "daily_feature_count": 50,
    "lagged_feature_count": 300,
    "p_value_threshold": 0.05,
    "random_forest_trees": 50,
    "engineering_random_forest_max_depth": 4,
    "paper_pca_explained_variance": 0.9,
    "paper_post_pca_feature_range": "300-500"
  },
  "data": {
    "datasets": [
      "GESLA-2",
      "CCMP",
      "20CRV2c",
      "Microwave OI SST",
      "GPCP",
      "ERA-Interim",
      "GTSR"
    ],
    "protocol": "global_storm_surge_pca_station_daily_v1",
    "format": "NPZ",
    "input_layout": "NF",
    "target_layout": "N1",
    "input_shapes": {
      "rs_daily": ["N", 50],
      "rs_lagged": ["N", 300],
      "ar_daily": ["N", 50],
      "ar_lagged": ["N", 300]
    },
    "target": "daily maximum non-tidal residual",
    "target_unit": "m",
    "sample_interval_hours": 24,
    "lag_interval_hours": 6,
    "maximum_lag_hours": 30,
    "rs_variables": ["u", "u2", "u3", "v", "v2", "v3", "slp", "sst", "precipitation"],
    "ar_variables": ["u", "u2", "u3", "v", "v2", "v3", "slp"]
  },
  "configuration_sources": [
    "conf/config.yaml",
    "model/globalsurgeml.py",
    "scripts/fake_data.py",
    "scripts/train.py",
    "scripts/inference.py",
    "scripts/result.py"
  ]
}