Multi-wavelength radio observations
as proxies
for upper atmosphere modeling
T. Dudok de Wit (University of Orléans)
Sean Bruinsma (CNES, Toulouse)
Kiyoto Shibasaki (Nobeyama Solar Radio Observatory, Nagano)
With special thanks to the instrument teams: Penticton, Nobeyama,
SORCE, TIMED, Greenwich Observatory, …
f10.7
…on the shoulders of giants
Take home message 1
Take home message 1
Do not disrupt these historic observations
Look at what former generations have done...
Radio vs UV observations
There is an intimate connection between solar irradiance
in the 310
UV and in centimetric radio emissions…
S. M. WHITE
S. White (1999)
SORCE 1/2014
6 cm (VLA)
soft X-ray (Yohkoh)
Figure 1. Radio (left, 4.6 GHz image made with the VLA using mosaic techniques) and soft X-ray
(right, Yohkoh SXT image) images of the east limb of the Sun on 7 November 1993. Features labeled
5
The radio spectrum in a nutshell
altitude
interplanetary
medium
upper corona
low corona
transition
region
chromosphere
1 mm
SORCE 1/2014
1 cm
10 cm
1m
10 m
100 m
λ
6
The radio spectrum in a nutshell
altitude
interplanetary
medium
plasma
emissions
thermal
bremsstrahlung
upper corona
low corona
gyro-resonance
emission
transition
region
gyro-resonance
emission (high B)
& bremsstrahlung
chromosphere
1 mm
SORCE 1/2014
1 cm
10 cm
1m
10 m
100 m
λ
6
The radio spectrum in a nutshell
altitude
interplanetary
medium
affected by
terrestrial
atmosphere
plasma
emissions
thermal
bremsstrahlung
upper corona
low corona
gyro-resonance
emission
transition
region
gyro-resonance
emission (high B)
& bremsstrahlung
chromosphere
1 mm
SORCE 1/2014
1 cm
10 cm
1m
10 m
100 m
λ
6
The radio spectrum in a nutshell
altitude
interplanetary
medium
affected by
terrestrial
atmosphere
plasma
emissions
thermal
bremsstrahlung
upper corona
low corona
gyro-resonance
emission
transition
region
gyro-resonance
emission (high B)
& bremsstrahlung
chromosphere
1 mm
SORCE 1/2014
1 cm
10 cm
absorbed by
ionosphere
1m
10 m
100 m
λ
6
The radio spectrum in a nutshell
altitude
interplanetary
medium
f10.7
index
affected by
terrestrial
atmosphere
plasma
emissions
thermal
bremsstrahlung
upper corona
low corona
gyro-resonance
emission
transition
region
gyro-resonance
emission (high B)
& bremsstrahlung
chromosphere
1 mm
SORCE 1/2014
1 cm
10 cm
absorbed by
ionosphere
1m
10 m
100 m
λ
6
Physical picture
Simplified picture
gyroresonance ≈ essentially sunspots
bremsstrahlung ≈ quiet Sun + plages, faculae, …
The relative contribution of gyroresonance/
bremsstrahlung is wavelength-dependent
SORCE 1/2014
7
Physical picture
Simplified picture
gyroresonance ≈ essentially sunspots
bremsstrahlung ≈ quiet Sun + plages, faculae, …
The relative contribution of gyroresonance/
bremsstrahlung is wavelength-dependent
Could synoptic radio observations inform us about
the filling factors of different solar features?
…and help reconstruct/constrain the SSI?
SORCE 1/2014
7
This picture may be too simple
Beware: the radio spectral variability is far more complex
low altitude emissions may be absorbed higher up in the corona
optically thick vs optically thin emissions
non-thermal emissions (e.g. flares) are even more complex
SORCE 1/2014
8
What measurements are there ?
from C. Marqué (2013)
SORCE 1/2014
9
Synoptic observations
T. Dudok de Wit et al.: Synoptic radio observations
Observations from Ottawa/Penticton and Toyokawa/
Nobeyama are exceptionally good
Table 1. The data sets used in this study and their main characteristics. The n
daily calibration, same instruments = excellent stability
from red noise model. The number of gaps refers to the number of missing d
daily
without
flares service
(mostly)interruptions,
= excellent continuity
files, not to
thevalues
number
of actual
which is smaller. Av
adjusted at 1 AU.
wavelength
[cm]
frequency
[GHz]
origin of
observations
beginning of
measurements
3.2
8.0
10.7
15.0
30.0
9.4
3.75
2.8
2.0
1.0
Toyokawa/Nobeyama
Toyokawa/Nobeyama
Ottawa/Penticton
Toyokawa/Nobeyama
Toyokawa/Nobeyama
May 1, 1956
Nov. 6, 1951
Feb. 14, 1947
June 1, 1957
March 1, 1957
SORCE 1/2014
number of
gaps since
beginning
195
203
311
231
163 10
A new dataset
We now have a homogeneous dataset with
daily values from 6 November 1957 till today
5 wavelengths: 3.2, 8, 10.7, 15 and 30 cm
all values regridded to noon UT
without flares, data gaps filled in by expectation-maximization
one month latency (can be reduced to < 24 hrs)
Download from http://projects.pmodwrc.ch/solid/
SORCE 1/2014
11
The observations
600
flux [sfu]
500
400
3.2cm
8cm
10.7cm
15cm
30cm
300
200
100
0
1940
SORCE 1/2014
1950
1960
1970
1980
year
1990
2000
2010
2020
12
Analysis procedure
Decompose the radio fluxes into
[Kundu, 1965]
a baseline component (B)
a rotationally
modulated component (S, or SVC)
T. Dudok de Wit et al.: Synoptic radio observations
10.7 cm flux [sfu]
250
B component
S component
B+S
200
150
observation
baseline
100
50
0
2003
SORCE 1/2014
2004
year
rotational
modulation
2005
13
Two components
Different wavelengths show different slopes
= driven by different physical processes
30cm flux [sfu]
200
150
100
50
0
0
S comp
B comp
50
100
150
10.7 cm flux
SORCE 1/2014
200
250
14
Two components
Different wavelengths show different slopes
= driven by different physical processes
30cm flux [sfu]
200
150
100
50
0
0
S comp
B comp
50
100
150
10.7 cm flux
SORCE 1/2014
200
250
14
Synoptic observations
T. Dudok de Wit et al.: Synoptic radio observations
flux [a.u.]
30cm
3
2
1
0
-1
10.7cm
3.2cm
B component
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015
flux [a.u.]
15
S component
10
5
0
2003
2004
year
2005
3. Excerpt of the flux at three wavelengths with the background component, averaged ov
All quantities have been rescaled
15
hsSORCE (top),
and
the
S
component
(bottom).
All
values
have
been
rescaled,
and
baselines
h
1/2014
Synoptic observations
T. Dudok de Wit et al.: Synoptic
radio observations
Long-term
variability
flux [a.u.]
30cm
3
2
1
0
-1
is mostly
driven by the same process:
10.7cm
3.2cm
thermal Bremsstrahlung
B component
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015
flux [a.u.]
15
S component
10
5
0
2003
2004
year
2005
3. Excerpt of the flux at three wavelengths with the background component, averaged ov
All quantities have been rescaled
15
hsSORCE (top),
and
the
S
component
(bottom).
All
values
have
been
rescaled,
and
baselines
h
1/2014
Intermediate conclusion
Very similar variations at all wavelengths
➞ a few contributions may explain most of the
variability
Linear combinations of different wavelengths may be
used to reconstruct the SSI
e.g. [Schmahl & Kundu, 1995].
SORCE 1/2014
16
Part I
Can we isolate the different physical
contributions to these synoptic radio
fluxes?
Blind source separation
SORCE 1/2014
18
Blind source separation
Objective : extract the original constituents (“sources”)
that appear mixed in the radio fluxes
SORCE 1/2014
18
Blind source separation
Objective : extract the original constituents (“sources”)
that appear mixed in the radio fluxes
Assumptions
linear mixing
the sources are mostly independent in time
positivity: the sources and their concentrations must be ≥ 0
SORCE 1/2014
18
Blind source separation
Objective : extract the original constituents (“sources”)
that appear mixed in the radio fluxes
Assumptions
linear mixing
the sources are mostly independent in time
positivity: the sources and their concentrations must be ≥ 0
Solution: use blind source separation
used recently in acoustics, chemometry, cosmology, …
consider Bayesian Positive Source Separation [Moussaoui et al., 2003]
SORCE 1/2014
18
How many sources are there ?
With 5 wavelengths we can identify at most 5 sources
Various criteria all show that we have 3 sources
SORCE 1/2014
19
3.2 cm flux
S1
200
S
100
S3
0
150
2
10.7 cm flux
100
flux 30cm [sfu]
flux 10.7cm [sfu] flux 3.2cm [sfu]
What do the sources look like ?
50
0
80
30 cm flux
60
40
20
0
2003
SORCE 1/2014
2004
year
2005
20
S1
200
S2
S3
100
0
150
100
flux 30cm [sfu]
flux 10.7cm [sfu] flux 3.2cm [sfu]
Spectral profile
50
0
80
60
40
20
amplitude [a.u.]
0
2003
2004
year
2005
1
S1
0.8
S2
0.6
S3
0.4
0.2
0
0
SORCE 1/2014
10
20
wavelength [cm]
30
21
S1
200
S2
S3
100
0
150
100
flux 30cm [sfu]
flux 10.7cm [sfu] flux 3.2cm [sfu]
Spectral profile
50
0
80
60
40
20
amplitude [a.u.]
0
2003
2004
year
2005
1
S1
0.8
S2
0.6
S3
0.4
0.2
0
0
SORCE 1/2014
10
20
wavelength [cm]
30
Source 1: only during active Sun
= gyroresonance emissions from
active regions with high B
Source 2: only with sunspots
= gyroresonance emissions
Source 3: only with plages/faculae
= Bremsstrahlung emissions
21
S1
200
S
100
S3
2
0
150
100
flux 30cm [sfu]
flux 10.7cm [sfu] flux 3.2cm [sfu]
Spectral profile
50
0
80
60
40
20
0
2003
SORCE 1/2014
2004
year
2005
Source 3 = Bremsstrahlung proxy22
Power spectral density
PSD [sfu2 * day]
Power spectral density (periodogram)
0
10
−2
10
30cm
15cm
10.7cm
8cm
3.2cm
−2
10
SORCE 1/2014
−1
10
frequency [1/day]
0
10
23
Power spectral density
PSD [sfu2 * day]
Power spectral density (periodogram)
0
10
−2
10
30cm
15cm
10.7cm
8cm
3.2cm
−2
10
SORCE 1/2014
−1
10
frequency [1/day]
0
10
Stronger harmonics at 30cm are not due to
center-to-limb effects but to longer lifetime of
emitting solar structures [Donnelly et al., 1982]
23
A Bremsstrahlung proxy ?
How do our sources correlate with the SSI ?
Data from TIMED & SORCE, 2003-2010
1
correlation R
0.8
Source 3: Bremsstrahlung
0.6
0.4
Source 2: gyroresonance
S1
S2
0.2
Source 1: gyroresonance
(active Sun)
S3
best fit
0
0
SORCE 1/2014
50
100
150
wavelength [nm]
200
250
24
A Bremsstrahlung proxy ?
How do different wavelengths correlate with the SSI ?
Data from TIMED & SORCE, 2003-2010
1
0.8
correlation R
0.6
0.4
0.2
0
−0.2
−0.4
−0.6
−0.8
0
SORCE 1/2014
30 cm
15 cm
10.7 cm
8 cm
3.2 cm
best fit
50
100
150
200
wavelength [nm]
250
300
350
25
Intermediate conclusions
We have a new Bremsstrahlung proxy that is indeed
highly correlated with the MUV-NUV bands
Use the 30 cm flux if you want a better correlation
with the SSI
because it has a higher Bremsstrahlung content
Gyroresonance emissions account for over 85% (50%) of
the rotational variability at 10.7 cm (30 cm)
SORCE 1/2014
26
Intermediate conclusions
Should we consider wavelengths longward of 30 cm ?
SORCE 1/2014
27
Intermediate conclusions
Should we consider wavelengths longward of 30 cm ?
yes, 42 cm
SORCE 1/2014
27
Part II
Reconstruction of the SSI
Reconstruction of the SSI
The SSI can be reconstructed in various ways
Simple empirical linear regression
5
X
I( , t) =
↵k (
SSI
k=1
k , t)
radio flux
Linear regression after decomposing into different scales
(much better)
5
XX
I( , t) =
↵k,⌧ W⌧ ( k , t)
⌧
SORCE 1/2014
k=1
wavelet transform
29
Reconstruction of the SSI
Excerpt: various reconstructions
(model has been trained on a different interval)
measured
10.7cm
10.7 and 30cm
HI
MgII
0.28
0.275
0.27
0.265
1
MgII index
Lyman-α line
0.8
He II
0.25
HeII line
0.2
TSI
0.15
1366
TSI
1364
1362
2003
SORCE 1/2014
2004
2005
30
What can we reconstruct ?
Time scales < ~100 days
XUV-NUV : excellent
VIS-NIR : reasonable
Time scales > ~100 days
too early to assess (SSI data are not stable enough)
unlikely to work well: long-term trends may not be the same
SORCE 1/2014
31
Spinoff: stability of composites
Reconstruction of the Lyman-α composite [LASP]
2
Lyman− [mW/m ]
7
6
Composite
Lyman-α line
5
4
3
1950
1960
1970
1980
1990
2000
2010
2020
2
residual error [mW/m ]
1
−1
residuals
source for data
−2
1950
SORCE 1/2014
Residual =
observations reconstruction
0
1960
1970
1980
1990
2000
2010
2020
32
Spinoff: stability of composites
Reconstruction of the Lyman-α composite [LASP]
2
Lyman− [mW/m ]
7
6
Composite
Lyman-α line
5
4
3
1950
1960
1970
1980
1990
2000
2010
2020
2
residual error [mW/m ]
1
−1
residuals
source for data
−2
1950
SORCE 1/2014
Residual =
observations reconstruction
0
1960
1970
1980
1990
2000
2010
2020
33
Part III
Impact on upper atmosphere
Impact on upper atmosphere
Evaluate the performance by using these proxies as
inputs to a satellite drag model.
DTM (Drag Temperature Model) [Bruinsma et al., 2012]
predicts temperature and composition as a function of location
main inputs are solar and geomagnetic forcing
Metric of performance : O/C = ratio of observed-tocomputed neutral density
SORCE 1/2014
35
Example of results
O/C ratio for two different solar inputs
Mean O/C ratio DTM2012_F10
1.3
1.2
with 10.7 cm flux
CHAMP
GRACE
Nav29
Intelsat
Cosmos1263
EDR(400-500)km
F10.7m
Mean O/C ratio DTM2012_F30
240
1.3
200
1.2
1.1
160 1.1
1.0
120 1.0
0.9
with 30 cm flux
CHAMP
GRACE
Nav29
Intelsat
Cosmos1263
EDR(400-500)km
0.9
80
0.8
1996
SORCE 1/2014
1998
2000
2002
year
2004
2006
2008
2010
0.8
1996
1998
2000
2002
year
2004
2006
2008
2010
36
Example of results
O/C ratio for two different solar inputs
Mean O/C ratio DTM2012_F10
1.3
1.2
with 10.7 cm flux
CHAMP
GRACE
Nav29
Intelsat
Cosmos1263
EDR(400-500)km
F10.7m
Mean O/C ratio DTM2012_F30
240
1.3
200
1.2
1.1
160 1.1
1.0
120 1.0
0.9
with 30 cm flux
CHAMP
GRACE
Nav29
Intelsat
Cosmos1263
EDR(400-500)km
0.9
80
0.8
1996
1998
2000
2002
year
2004
2006
2008
2010
0.8
1996
1998
2000
2002
year
2004
2006
2008
2010
error on O/C is on average 7% lower when
30 cm radio flux is used (instead of 10.7 cm)
SORCE 1/2014
36
Conclusions
A homogeneous dataset of synoptic radio observations
By statistical analysis we are able to disentangle
contributions associated with Bremsstrahlung &
gyroresonance emissions.
The Bremsstrahlung proxy is excellent for MUV-NUV
The best single proxy for EUV-NUV is the 30 cm flux
Caveats
daily averages ≠ instantaneous snapshots
flares are NOT included
etc.
SORCE 1/2014
37
Take home message 2
Take home message 2
The spectral information of centimetric radio
observations provides considerable added value
to the f10.7 index
More…
Long-term trends
4
2
amplitude [sfu]
0
−2
−4
−6
−8
30cm
15cm
10.7cm
8cm
3.2cm
−10
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
year
SORCE 1/2014
40
Contribution
S1
S2
S3
residual
100
contribution [%]
80
60
40
20
0
SORCE 1/2014
30cm 15cm 10.7cm 8cm 3.2cm ISN
DSA MWSI MPSI MgII
HI
He II
41