StormScopeGOES#

class earth2studio.models.px.StormScopeGOES(
model_spec,
means,
stds,
latitudes,
longitudes,
variables=array(['abi01c', 'abi02c', 'abi03c', 'abi07c', 'abi08c', 'abi09c', 'abi10c', 'abi13c'], dtype='<U6'),
conditioning_variables=array(['z500'], dtype='<U4'),
conditioning_means=None,
conditioning_stds=None,
conditioning_data_source=None,
glm_mask=None,
conditioning_glm_mask=None,
topo=None,
nexrad_proximity=None,
sampler_args={'S_churn': 10, 'num_steps': 100},
input_times=array([0], dtype='timedelta64[h]'),
output_times=array([1], dtype='timedelta64[h]'),
y_coords=None,
x_coords=None,
input_interp_max_dist_km=12.0,
conditioning_interp_max_dist_km=26.0,
amp=True,
compile=False,
)[source]#
NANWC202680 GB

StormScope model forecasting GOES data on the HRRR grid.

This model supports multiple variants at different spatiotemporal resolutions, selected by passing model_name to load_model (default: "3km_10min"). The primary focus is CONUS nowcasting at 3km resolution; coarser 6km nearcasting variants are retained as legacy checkpoints. Variant names are semantic (<resolution>_<cadence>):

  • 3km_10min : 3km resolution, 10 minute timestep (CONUS nowcasting)

  • 6km_1hr : 6km resolution, 60 minute timestep (legacy nearcasting)

Use list_available_models() to inspect the variants in a given package (including any added after this release). Legacy training-style names are still accepted as aliases.

Variants whose input cadence is finer than their output cadence use a sliding window of input timesteps and predict one output timestep; others use a single input timestep and predict one output timestep.

Parameters:

  • model_spec (list[dict[str, Any]]) – Sequence of stage specifications; see StormScopeBase.

  • means (torch.Tensor) – Per-variable mean for normalization, shape [1, C, 1, 1].

  • stds (torch.Tensor) – Per-variable std for normalization, shape [1, C, 1, 1].

  • latitudes (torch.Tensor) – Latitudes of the grid, expected shape [H, W].

  • longitudes (torch.Tensor) – Longitudes of the grid, expected shape [H, W].

  • variables (np.ndarray, optional) – GOES input variables. Default is [“abi01c”, “abi02c”, “abi03c”, “abi07c”, “abi08c”, “abi09c”, “abi10c”, “abi13c”].

  • conditioning_variables (np.ndarray, optional) – Auxiliary conditioning variables. Default is [“z500”].

  • conditioning_means (torch.Tensor | None, optional) – Means to normalize any external conditioning data. Default is None.

  • conditioning_stds (torch.Tensor | None, optional) – Stds to normalize any external conditioning data. Default is None.

  • conditioning_data_source (Any | None, optional) – Data source for external conditioning. Default is None.

  • sampler_args (dict[str, Any] | None, optional) – Default sampler arguments passed to the diffusion sampler. Default is {“num_steps”: 100, “S_churn”: 10}.

  • input_times (np.ndarray, optional) – Input timesteps, of type timedelta64. Default is [0 m] (i.e., the current time).

  • output_times (np.ndarray, optional) – Output timesteps, of type timedelta64. Default is [60 m] (i.e., 1 hour from the current time).

  • y_coords (np.ndarray | None, optional) – Y coordinates of the grid, expected shape [H, W]. Default is None, in which case the model uses the enumerated indices inferred from the latitude and longitude grid shapes.

  • x_coords (np.ndarray | None, optional) – X coordinates of the grid, expected shape [H, W]. Default is None, in which case the model uses the enumerated indices inferred from the latitude and longitude grid shapes.

  • input_interp_max_dist_km (float, optional) – Maximum distance in kilometers for nearest neighbor interpolation of input data. Points beyond this distance are masked as invalid. Default is 12.0.

  • conditioning_interp_max_dist_km (float, optional) – Maximum distance in kilometers for nearest neighbor interpolation of conditioning data. Points beyond this distance are masked as invalid. Default is 26.0.

  • glm_mask (Tensor | None)

  • conditioning_glm_mask (Tensor | None)

  • topo (Tensor | None)

  • nexrad_proximity (Tensor | None)

  • amp (bool)

  • compile (bool)

Note

To have a unified coordinate system over CONUS for convenience, the model uses the HRRR grid. As a result, there are portions of the domain which go beyond the extent of the GOES-East data, so these portions are masked as invalid (set to NaN).

__call__(x, coords)[source]#

Runs the prognostic model one step. Assumes the last two dimensions of the input tensor are the spatial dimensions.

Parameters:

  • x (torch.Tensor) – Input tensor.

  • coords (CoordSystem) – Input coordinate system.

Returns:

Output tensor and coordinate system.

Return type:

tuple[torch.Tensor, CoordSystem]

create_iterator(x, coords)[source]#

Creates an iterator to perform time-integration of the prognostic model.

Parameters:

  • x (torch.Tensor) – Input tensor.

  • coords (CoordSystem) – Input coordinate system.

Yields:

Iterator[tuple[torch.Tensor, CoordSystem]] – Iterator that generates time-steps of the prognostic model containing the output data tensor and coordinate system dictionary.

Return type:

Iterator[tuple[Tensor, OrderedDict[str, ndarray]]]

classmethod load_default_package()[source]#

Load the default StormScope package from Hugging Face.

Return type:

Package

classmethod load_model(
package,
model_name='3km_10min',
conditioning_data_source=None,
amp=True,
compile=False,
)[source]#

Load model from package.

Parameters:

  • package (Package) – Package to load model from

  • model_name (Literal["3km_10min", "6km_1hr"], optional) –

    Variant to load, by default "3km_10min" (the recommended CONUS nowcasting variant). Available variants (see list_available_models()):

    • "3km_10min": 3km resolution, 10 minute timestep (CONUS nowcasting)

    • "6km_1hr": 6km resolution, 60 minute timestep (legacy nearcasting)

    Legacy training-style names are accepted as aliases.

  • conditioning_data_source (DataSource | ForecastSource | None, optional) – Data source to use for conditioning, by default None.

  • amp (bool, optional) – Enable automatic mixed precision (autocast) for the sampler’s network forward passes. Default is True.

  • compile (bool, optional) – Compile each staged expert with torch.compile (“reduce-overhead”). Default is False.

Returns:

Instantiated StormScopeGOES model

Return type:

PrognosticModel

Examples using earth2studio.models.px.StormScopeGOES#

StormScope Satellite and Radar Nowcasting

StormScope Satellite and Radar Nowcasting