Sample data set overview

An overview of the coordinated sample data set.

import matplotlib.pyplot as plt
import astropy.units as u

import sunpy.map
import sunpy.timeseries
import sunpy.data.sample as sample_data

On 2011 June 7, various solar instruments observed a spectacular solar eruption from NOAA AR 11226. The event included an M2.5 flare, a filament eruption, a coronal mass ejection, and a global coronal EUV wave (IAU standard: SOL2011-06-07T06:24:00L045C112). This event was spectacular because it features the ejection of a large amount of prominence material, much of which failed to escape and fell back to the solar surface. This event received some press coverage (e.g. National Geographics, Discover Magazine) and the literature contains a number of a papers about it (e.g. Li et al., Inglis et al.)

The following image of the flare is now fairly iconic.

aia_cutout03_map = sunpy.map.Map(sample_data.AIA_193_CUTOUT03_IMAGE)
fig = plt.figure()
ax = fig.add_subplot(111, projection=aia_cutout03_map)
aia_cutout03_map.plot()
plt.show()
../../../_images/sphx_glr_2011_06_07_sampledata_overview_001.png

Let’s take a look at the GOES XRS data.

goes = sunpy.timeseries.TimeSeries(sample_data.GOES_XRS_TIMESERIES)
fig = plt.figure()
goes.plot()
plt.show()
../../../_images/sphx_glr_2011_06_07_sampledata_overview_002.png

Out:

/home/docs/checkouts/readthedocs.org/user_builds/sunpy/conda/stable/lib/python3.7/functools.py:827: UserWarning: Discarding nonzero nanoseconds in conversion
  return dispatch(args[0].__class__)(*args, **kw)

Next let’s investigate the AIA full disk images that are available. Please note that these images are not at the full AIA resolution.

aia_131_map = sunpy.map.Map(sample_data.AIA_131_IMAGE)
aia_171_map = sunpy.map.Map(sample_data.AIA_171_IMAGE)
aia_211_map = sunpy.map.Map(sample_data.AIA_211_IMAGE)
aia_335_map = sunpy.map.Map(sample_data.AIA_335_IMAGE)
aia_094_map = sunpy.map.Map(sample_data.AIA_094_IMAGE)
aia_1600_map = sunpy.map.Map(sample_data.AIA_1600_IMAGE)

fig = plt.figure(figsize=(6, 28))
ax = fig.add_subplot(611, projection=aia_131_map)
aia_131_map.plot(clip_interval=(0.5, 99.9)*u.percent)
aia_131_map.draw_grid()

ax = fig.add_subplot(612, projection=aia_171_map)
aia_171_map.plot(clip_interval=(0.5, 99.9)*u.percent)
aia_171_map.draw_grid()

ax = fig.add_subplot(613, projection=aia_211_map)
aia_211_map.plot(clip_interval=(0.5, 99.9)*u.percent)
aia_211_map.draw_grid()

ax = fig.add_subplot(614, projection=aia_335_map)
aia_335_map.plot(clip_interval=(0.5, 99.9)*u.percent)
aia_335_map.draw_grid()

ax = fig.add_subplot(615, projection=aia_094_map)
aia_094_map.plot(clip_interval=(0.5, 99.9)*u.percent)
aia_094_map.draw_grid()

ax = fig.add_subplot(616, projection=aia_1600_map)
aia_1600_map.plot(clip_interval=(0.5, 99.9)*u.percent)
aia_1600_map.draw_grid()

fig.tight_layout(pad=6.50)
plt.show()
../../../_images/sphx_glr_2011_06_07_sampledata_overview_003.png

We also provide a series of AIA cutouts so that you can get a sense of the dynamics of the in-falling material.

aia_cutout01_map = sunpy.map.Map(sample_data.AIA_193_CUTOUT01_IMAGE)
aia_cutout02_map = sunpy.map.Map(sample_data.AIA_193_CUTOUT02_IMAGE)
aia_cutout03_map = sunpy.map.Map(sample_data.AIA_193_CUTOUT03_IMAGE)
aia_cutout04_map = sunpy.map.Map(sample_data.AIA_193_CUTOUT04_IMAGE)
aia_cutout05_map = sunpy.map.Map(sample_data.AIA_193_CUTOUT05_IMAGE)

fig = plt.figure(figsize=(6, 28))
ax = fig.add_subplot(511, projection=aia_cutout01_map)
aia_cutout01_map.plot()

ax = fig.add_subplot(512, projection=aia_cutout02_map)
aia_cutout02_map.plot()

ax = fig.add_subplot(513, projection=aia_cutout03_map)
aia_cutout03_map.plot()

ax = fig.add_subplot(514, projection=aia_cutout04_map)
aia_cutout04_map.plot()

ax = fig.add_subplot(515, projection=aia_cutout05_map)
aia_cutout05_map.plot()

fig.tight_layout(pad=5.50)
plt.show()
../../../_images/sphx_glr_2011_06_07_sampledata_overview_004.png

There are a number of other data sources available as well, such as SWAP.

swap_map = sunpy.map.Map(sample_data.SWAP_LEVEL1_IMAGE)
fig = plt.figure()
swap_map.plot()
plt.show()
../../../_images/sphx_glr_2011_06_07_sampledata_overview_005.png

And also RHESSI.

rhessi_map = sunpy.map.Map(sample_data.RHESSI_IMAGE)
fig = plt.figure()
rhessi_map.plot()
plt.show()
../../../_images/sphx_glr_2011_06_07_sampledata_overview_006.png

Out:

/home/docs/checkouts/readthedocs.org/user_builds/sunpy/conda/stable/lib/python3.7/site-packages/sunpy-1.0.3.dev1+gece6362cd-py3.7-linux-x86_64.egg/sunpy/map/mapbase.py:601: SunpyUserWarning: Missing metadata for observer: assuming Earth-based observer.The following sets of keys were checked:
('hgln_obs', 'hglt_obs', 'dsun_obs')
('crln_obs', 'crlt_obs', 'dsun_obs')
  warnings.warn(warning_message, SunpyUserWarning)

Total running time of the script: ( 0 minutes 14.086 seconds)

Gallery generated by Sphinx-Gallery