daleroberts / nrt-predict Goto Github PK

Refactor to be able to operate off L1C or L2A scenes from Sinergise or Element 84 as well.

• https://registry.opendata.aws/sentinel-2/
• https://registry.opendata.aws/sentinel-2-l2a-cogs/

It would be great if this library also supported references to bands by names rather than their indices.

While a 3d array arr[y, x, band] interface works great for certain algorithms, a dict of 2d arrays interface {band_name: arr[y, x]} would make the code more readable and maintainable at the cost of little or no performance penalties for most algorithms.

Of course, this would also mean that standard ODC dask.array-backed xarray.Dataset interface would just work, without this library needing to depend on either of those libraries.

Under this scheme, for example,

def predict(self, pre, pst):
    pre_nbr = (pre[:, :, 6] - pre[:, :, 8]) / (pre[:, :, 6] + pre[:, :, 8])
    pst_nbr = (pst[:, :, 6] - pst[:, :, 8]) / (pst[:, :, 6] + pst[:, :, 8])
    return pre_nbr - pst_nbr

would look like

def predict(self, pre, pst):
    pre_nbr = (pre['nbar_B06'] - pre['nbar_B08']) / (pre['nbar_B06'] + pre['nbar_B08'])
    pst_nbr = (pst['nbar_B06'] - pst['nbar_B08']) / (pst['nbar_B06'] + pst['nbar_B08'])
    return pre_nbr - pst_nbr

note that these band names should work with ODC. Or maybe even better

def predict(self, pre, pst):
    pre_nbr = (pre['nbar_red_edge_2'] - pre['nbar_nir_1']) / (pre['nbar_red_edge_2'] + pre['nbar_nir_1'])
    pst_nbr = (pst['nbar_red_edge_2'] - pst['nbar_nir_1']) / (pst['nbar_red_edge_2'] + pst['nbar_nir_1'])
    return pre_nbr - pst_nbr

or something like that, if I recall the band names correctly. This of course needs a dictionary to translate 'nbar_B08' to 'nbar_nir_1' somewhere in this code base.

Need to decide whether to use flake8 or pylint for linting the source code.

This is required in order to import the NRT library into other applications.