Small demo of how you might cluster mapped features, ignore (x, y) location and accounting for some NaNs around the edge.

Nonspatial_clustering_of_map_features.py

import numpy as np

# Make some fake data
# Make array with 100 rows, 100 columns, and 6 'features' (different maps)
shape = (100, 100, 3)
data = np.random.random(shape)

# Pretend it has NaNs around edge.
data[:10] = np.nan
data[-10:] = np.nan
data[:, :10] = np.nan
data[:, -10:] = np.nan

# If arr has some NaNs in it (e.g. around edge)
# Make a 2D array that has nans where your data has at least one nan.
summ = np.sum(data, -1)
mask = ~np.isnan(summ)
X = data[mask]

# Now you can cluster the features:
from sklearn.cluster import KMeans
clu = KMeans(n_clusters=4)
yhat = clu.fit_transform(X)