Read and inspect

Discover the supported interface

import hyperproc as hp
print(hp.summary())
hp.list_readers()  # prints the supported reader table; returns None

hp.sniff(path) identifies a provider product from its filename pattern without reading the spectral cube. hp.open(path, sensor=None, level=None, **kwargs) resolves the registry entry, opens the product, and passes keyword arguments to its reader. Read the sensor-specific signature before supplying options; not every reader accepts chunks, geometry, or good_bands_only.

Inspect before processing

ds = hp.open("/path/to/provider_product")
hp.describe(ds)
var = hp.main_var(ds)
print(ds[var].dims, ds[var].dtype)
print(ds.attrs.get("units"), ds.attrs.get("level"))
print(ds.attrs.get("crs"), ds.attrs.get("transform"))

Confirm units, scaling, bad-band handling, source metadata, expected geometry, and the amount of valid data. describe() reads sampled values; it is not purely a metadata operation. Scalar angles may be expanded spatially without gaining additional physical information.

Subset explicitly

window = ds.isel(y=slice(3000, 3500), x=slice(400, 900))
visible = hp.open("/path/to/provider_product", wl_range=(400, 900))

The spatial example is only appropriate when those indices exist and intersect valid observations. Wavelength subsets change available spectral features. Removing necessary bands before atmospheric inversion or a vegetation mask can make a later operation impossible.

Error interpretation

Unknown naming/level information normally raises ValueError; missing paths raise FileNotFoundError; registered but unwritten readers raise NotImplementedError. Explicit sensor/level parameters are appropriate for genuine naming ambiguity, not for forcing unsupported data through the wrong reader.

See the core API, registry, and sensor matrix.