Python: Change color of bad pixels

Question

I have an HDF file that contains satellite data and when plotted, looks like the attached image below. It has a lot of bad pixels (they're colored black when using this particular colormap). How can I change the color of bad pixels to white for example? Thank you.

Link to HDF file: https://drive.google.com/file/d/1AYhvqxNlvlqQB7otn_veLkO3t2qPogV_/view?usp=sharing

Here is a simplified code that produces the image:

import h5py
import numpy as np
import matplotlib.pyplot as plt
import cartopy.crs as ccrs

filename = 'NmHRIR1H.19660626_14-57-45_564_005.hdf'

with h5py.File(filename) as f:  
    temp = f['HRIR-Temp'][:]-273.15 # Convert K to C
    lat = f['latitude'][:]
    lon = f['longitude'][:]
    
    f.close()

fig = plt.figure(figsize=(8,8))
ax = plt.axes(projection=ccrs.Mercator())

extent = [124,139,12,27]
ax.set_extent(extent, crs=ccrs.PlateCarree())

ax.coastlines(resolution='10m', color='gold', linewidth=0.5)

ax.pcolormesh(lon, lat, temp, cmap='binary', vmin=-60, vmax=40, transform=ccrs.PlateCarree())

plt.show()

UPDATE: I checked the documentation for this dataset again, and it has a field called "Flags" that has bit flags stored in it. Bit 4 in particular seems to be the key for me to achieve what I want (when bit 4 is on, it means bad data).

Does this answer your question? [How can I plot NaN values as a special color with imshow in matplotlib?](https://stackoverflow.com/questions/2578752/how-can-i-plot-nan-values-as-a-special-color-with-imshow-in-matplotlib) — Michael Delgado, Dec 27 '22 at 00:08
Specifically, look for the use of [`cmap.set_bad`](https://matplotlib.org/stable/api/_as_gen/matplotlib.colors.Colormap.html#matplotlib.colors.Colormap.set_bad) in the accepted answer — Michael Delgado, Dec 27 '22 at 00:09
Unfortunately, I'm not able to achieve what I want with that solution (and I have already tried that before posting here). I added some new information to my post that might help. — doomhaMwx, Dec 27 '22 at 02:25
In that case you’ll need to mask your data and then use that answer. See np.where. Also, have you considered xarray? — Michael Delgado, Dec 27 '22 at 03:36

score 1 · Accepted Answer · answered Dec 27 '22 at 08:24

The Flags never seem to get a value higher than 3, so I don't think that will help you out for this specific issue. I don't see any attributes specifying the nodata value, or the documentation mentioning it, but it's quite obvious that the maximum float32 value is used for this.

The coordinates outside the swath are filled with this value, and also the extreme low (?) values in the temperature.

As @Michael Delgado already mentioned, cmap.set_bad() is a good way to assign a specific color (or alpha!) to these values. But it requires Matplotlib to know what is "bad". In the current form, you could use cmap.set_over() but that will loose the distinction with valid values that are clipped (temp > vmax).

It's usually best to parse a nodata value from the metadata of the file, but I don't see it right away (didn't look that careful). When hard coding it:

import h5py
import numpy as np
import cartopy.crs as ccrs
import matplotlib.pyplot as plt
import matplotlib as mpl

fn = 'NmHRIR1H.19660626_14-57-45_564_005.hdf'
nodata_value = np.finfo(np.float32).max

with h5py.File(fn) as f:
    temp = np.ma.masked_values(f['HRIR-Temp'][:], nodata_value) - 273.15
    lat = np.ma.masked_values(f['latitude'][:], nodata_value)
    lon = np.ma.masked_values(f['longitude'][:], nodata_value)

cmap = mpl.colormaps["binary"].copy()
cmap.set_bad("r", alpha=1)
norm = mpl.colors.Normalize(vmin=-60, vmax=40)

fig, ax = plt.subplots(
    figsize=(6,6), dpi=86, facecolor="w", layout="compressed", 
    subplot_kw=dict(projection=ccrs.PlateCarree()),
)

im = ax.pcolormesh(lon, lat, temp, cmap=cmap, norm=norm)
cb = fig.colorbar(im, ax=ax, shrink=.5, extend="both")

ax.coastlines(resolution='10m', color='gold', linewidth=0.5)

extent = [124,139,12,27]
ax.set_extent(extent, crs=ccrs.PlateCarree())

Thank you very much! So, the bad pixels are filled with the value 3.4028235e+38 (I just realized now that this value is also mentioned in the documentation). One question, when **alpha=1**, why do I get the warning this warning? **The colormap's 'bad' has been set, but in order to wrap pcolormesh across the map it must be fully transparent.** — doomhaMwx, Dec 27 '22 at 09:43
I think that only applies when your data is crossing the dateline (+/- 180 longitude), it's related to how Cartopy handles the discontinuity at that point. — Rutger Kassies, Dec 27 '22 at 10:47

Python: Change color of bad pixels

1 Answers1