Based on reading the contents of the full_cropped CloudCast dataset.
CloudCast defines 10 cloud type results, which can be merged to get a set of low, mid and high cloud time-occurence maps.
CCAST_CTYPES = {0: "No Cloud",
1: "Very Low Cloud",
2: "Low Cloud",
3: "Mid-level Cloud",
4: "High Opaque Cloud",
5: "Very High Opaque Cloud",
6: "Fractional Cloud",
7: "High Semitransparent Thin Cloud",
8: "High Semitransparent Moderately Thick Cloud",
9: "High Semitransparent Thick Cloud",
10: "High Semitransparent above Low/Medium Cloud"}
CCAST_CTYPES_Z = {"No and Fractional Cloud": (0, 6),
"Low Cloud": (1, 2),
"Mid Cloud": (3,),
"High Cloud": (4, 5, 7, 8, 9, 10)}
Time Occurrence/Frequency
No and Fractional Cloud Min 9.34051% Mean 34.97845% Max 67.87932%
Low Cloud Min 5.53692% Mean 25.37808% Max 45.81590%
Mid Cloud Min 3.50805% Mean 11.39559% Max 31.93364%
High Cloud Min 20.09595% Mean 28.24690% Max 41.37695%
Low Cloud Occurence Mid Cloud Occurence High Cloud Occurence
These can be mapped to 0-1 and then used as RGB additive color model intensities; here Low Cloud == Red, Mid Cloud == Green and High Cloud == Blue). This gives tuples like [0.7294, 0.5137, 0.5764] or if converted to 0-255 [186, 131, 147] for each grid in the map.
Low Cloud Occurence Mid Cloud Occurence High Cloud Occurence
This can be stacked/combined to get an RGB composite.
rgb_normalized = np.dstack((red_normalized, green_normalized, blue_normalized))
Which shows that high- and mid-cloud is the dominate cloud type over high topography (i.e., could be snow/ice contamination), which low cloud is common elsewhere, except of the Mediterranean which all cloud is uncommon. This image could also of course be shown as a greyscale, in which case it just documents cloud occurence.