water detection... a suggestion.

[tlohde]

hi.

i have had a little play with the preprocessor (see code block below) that gives the option of only masking with nan areas that are connected and large enough (pick a threshold, any threshold).

difference when plotting the DEM (values ) with/without identifying the connected groups

the result: fewer (spurious) line breaks in the final ridgeplot.

cheers,

from scipy.ndimage import label  # requires additional import

def preprocess(
    values=None,
    water_ntile=10,
    lake_flatness=2,
    vertical_ratio=40
    filter_groups=True,
    sq_size=3,
    grp_size=9,
    ):

    if values is None:
        values = self.get_elevation_data()
    nan_vals = np.isnan(values)

    values[nan_vals] = np.nanmin(values)
    scaled_values = (values - np.min(values)) / (np.max(values) - np.min(values))

    is_water = scaled_values < np.percentile(scaled_values, water_ntile)
    is_lake = rank.gradient(img_as_ubyte(scaled_values),
                                         square(sq_size)) < lake_flatness

    masked = (nan_vals | is_water | is_lake)

    # identify connected regions of nans / water / lake
    if filter_groups:

        structure = np.ones((3,3), dtype=int) # i.e. fully connected

        labelled, _ = label(masked,
                            structure)

        unique, counts = np.unique(labelled,
                                   return_counts=True)

        # only set to nan connected regions that 
        # are bigger than grp > size
        connected = np.in1d(labelled,
                            unique[counts > grp_size]
                            ).reshape(labelled.shape)

        scaled_values[connected & masked] = np.nan
    else:
        scaled_values[masked] = np.nan

    scaled_values = vertical_ratio * scaled_values[-1::-1]  # switch north and south

    return scaled_values

[ColCarroll]

This is interesting, thank you! I'll try to incorporate it when I have some time (unless you or someone else wants to make a PR...)