Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://star.arm.ac.uk/preprints/395.ps Äàòà èçìåíåíèÿ: Tue Aug 12 15:10:09 2003 Äàòà èíäåêñèðîâàíèÿ: Mon Oct 1 21:34:20 2012 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: ï ï ï ï ï ï ï ï ï ï ï ï ï

COMPARISON OF TOTAL SOLAR IRRADIANCE WITH NASA/NATIONAL SOLAR OBSERVATORY
SPECTROMAGNETOGRAPH DATA IN SOLAR CYCLES 22 AND 23
Harrison P. Jones
NASAGoddard Space Flight Center, Laboratory for Astronomy and Solar Physics, Southwest Solar Station, c/o National Solar Observatory, 1
P.O. Box 26732, Tucson, AZ 85726; hjones@noao.edu
Detrick D. Branston
National Solar Observatory, P.O. Box 26732, Tucson, AZ 86726; dbranston@noao.edu
Patricia B. Jones
University of Arizona, Center for Computing and Information Technology, Tucson, AZ 85721; pjones@jemez.rc.arizona.edu
and
Miruna D. Popescu 2
Astronomical Institute of the Romanian Academy, RO­75212 Bucharest 28, Romania; mdp@star.arm.ac.uk
Received 2002 December 9; accepted 2003 January 30
ABSTRACT
NASA/National Solar Observatory Spectromagnetograph (SPM) data are compared with spacecraft
measurements of total solar irradiance (TSI) variations for 8 yr beginning with the declining phase of solar
cycle 22 and extending into the maximum of cycle 23. Previously reported conclusions based on a similar
comparison for a shorter time period appear to be robust: three factors (sunspots, strong unipolar regions,
and strong mixed­polarity regions) describe most of the variation in the SPM record, but only the first two
are associated with TSI. Additionally, the residuals of a linear multiple regression of TSI against SPM obser­
vations over the entire 8 yr period show an unexplained, increasing, linear time variation with a rate of about
0.05 W m #2 yr #1 . Separate regressions for the periods before and after 1996 January 1 show no unexplained
trends but di#er substantially in regression parameters. This behavior may reflect a solar source of TSI varia­
tions beyond sunspots or uncompensated nonsolar e#ects in one or both of the TSI and SPM data sets.
Subject headings: Sun: activity --- Sun: magnetic fields
1. INTRODUCTION
Accurate understanding of global solar variability is of
clear astrophysical interest and is also vital to distinguish
natural from anthropogenic causes of long­term changes in
terrestrial climate. Daily averages of total solar irradiance
(TSI) observations from several spacecraft radiometers over
the past two decades show clear rotational and solar­cycle
variations (Pap & Fro ˜ hlich 1999; Fro ˜ hlich & Lean 1998;
Fro ˜ hlich 2000). The modeling of this variability through
comparison of the spacecraft measurements with spatially
resolved solar observations from both ground­ and space­
based instruments is highly refined (Chapman, Cookson, &
Dobias 1996; Foukal & Lean 1988; Fligge, Solanki, &
Unruh 2000; Preminger, Walton, & Chapman 2002). Two
classes of solar features, dark sunspots and bright faculae,
account for about 90% of the TSI variance. Although it is
not yet clear whether the remaining discrepancies are obser­
vational or require additional sources of irradiance variabil­
ity, the simple observation (de Toma et al. 2001) that TSI at
the current solar maximum is very similar to the previous
maximum while photospheric indicators of solar activity
are lower in cycle 23 than in cycle 22 hints that more than
solar activity is involved. This paper presents more detailed
evidence to support this idea.
2. OBSERVATIONS
The results discussed here are based on two data sets. One
of these is the set of daily full­disk magnetograms from the
NASA/National Solar Observatory (NSO) Spectromagne­
tograph (SPM) that consist of strictly cospatial and cotem­
poral full­disk images of line­of­sight (LOS) magnetic flux,
LOS velocity, continuum intensity, equivalent width, and
central line depth derived from long­slit spectral polarime­
try of the Fe i 868.8 nm line. The observations have
1>14 # 1>14 spatial pixels, approximately 42 mA š spectral
pixels, and a magnetic noise level of about 5 G. Data for
2194 days from 1992 November 21 to 2000 September 30
are reduced and analyzed here. A more complete descrip­
tion of the instrument and data is given by Jones et al.
(2000, hereafter JBJW).
The other data set is the composite measure of TSI com­
piled from various spacecraft from 1979 through the present
by Fro ˜ hlich & Lean (1998) and Fro ˜ hlich (2000), who
describe the individual observations and the techniques
used to combine them into a single data stream. For this
study we use version 23 of the composite measure of TSI
from the Physikalisch­Meterologisches Observatorium
Davos World Radiation Center, Davos, Switzerland, which
includes unpublished data from the Variability of Solar
Irradiance and Gravity (VIRGO) experiment on the coop­
erative ESA/NASA Solar and Heliospheric Observatory
(SOHO). Data from both the composite TSI and the SPM
were available for 2123 days during the interval from 1992
November 21 to 2000 September 30.
1 The National Solar Observatory is operated by the Association of
Universities for Research in Astronomy under coperative agreement with
the National Science Foundation.
2 Also at Armagh Observatory, College Hill, Armagh BT61 9DG,
Northern Ireland.
The Astrophysical Journal, 589:658--664, 2003 May 20
# 2003. The American Astronomical Society. All rights reserved. Printed in U.S.A.
658

3. ANALYSIS
No adjustments or filtering were performed on the com­
posite TSI data other than to select days for which both
SPM and TSI observations were available. The reduction of
the SPM data was carried out exactly as described in JBJW.
Low­order, least­squares spatial fits of the daily data were
applied to remove spurious instrumental and telluric e#ects
as well as unwanted solar center­to­limb variations. Daily
multidimensional histograms were computed from the five
input images, and solar `` features '' were defined by re­
stricted domains in the histogram variable space as
described by JBJW. Area and contrast measures of each of
the features were computed from the histograms as
described below and in more detail in JBJW. Factor analysis
using principal component extraction revealed the most
important linear combinations of the original feature meas­
ures for explaining the variability of the SPM data, and a
multiple regression of the TSI observations as a function of
these factors was performed. This analysis procedure was
meant to be exploratory, allowing e#cient multiple passes
through our data set to learn what important features in the
SPM data are related to TSI without building in strong prej­
udices. We did indeed experiment with domain limits di#er­
ent from JBJW for the various features, but this did not
a#ect the substantive results of our study.
Specifically, we considered nine categories of solar fea­
tures: weak field, sunspot penumbra, sunspot umbra, bright
(and dark) strong­field unipolar features in the central
portion of the disk, bright (and dark) strong­field mixed­
polarity features in the central portion of the disk, and
bright (and dark) strong­field features near the solar limb.
Bright and dark features are distinguished by continuum
(not bolometric) intensity contrast at 868.8 nm with refer­
ence to the fitted limb­darkening curve. The defining param­
eter domains for these features both as used by JBJW and
an alternative classification are given in Table 1, where B is
in gauss, i and q are intensity and equivalent width contrasts
with respect to local center­to­limb variation, and l is the
cosine of the heliocentric angle from disk center. `` Unipo­
larity,'' u, is defined here as the fraction of flux of one sign in
a roughly supergranular area surrounding a given point on
the disk and is divided into four bins in the multidimen­
sional histograms. The revised feature classification is based
on that used by Jones (1998) to search for possible thermo­
dynamic di#erences between unipolar and mixed­polarity
features. The revised and original classifications di#er
mostly in the division of unipolar and mixed­polarity fea­
tures. Roughly speaking, strong unipolar regions corre­
spond to active regions, including some of the enhanced
network that trails from older active regions, while strong
mixed­polarity regions correspond to magnetic network
outside of active regions. The revised feature definitions
count more enhanced network as unipolar.
An important pedagogic distinction between the two his­
togram feature classifications is that the original definitions
contain some overlap between features. Pixels in the over­
lapping domains are counted more than once in the subse­
quent analysis procedure, which influences the correlations
between variables and resulting factor structure. Although
this e#ect has proved to be small, the revised feature defini­
tions contain no overlapping regions and are used for this
paper. Note as well that neither set of classifications are for­
mally complete. Practically, however, the neglected regions
of the histograms are sparse, and the total number of pixels
in the included domains equals the actual total number of
pixels within the accuracy of the interpolation techniques
that are used to extract this information from the variably
binned histograms (see JBJW).
As in JBJW, for each histogram feature we compute the
fractional area (number of pixels relative to the total num­
ber of pixels) and di#erential contrasts in intensity, equiva­
lent width, and central line depth relative to the weak­field
pixels in the same domains of u and l. As basis sets for sub­
sequent factor analysis, we use two di#erent subsets of these
quantities. In both cases, we include fractional area and
di#erential contrasts for each of the feature classes except
weak fields, where the di#erential contrasts are zero by
TABLE 1
Histogram Feature Domains
Class
|B|
(G) i q l u
Weak ......................................... [0, 8] All All [0, 1] [0, 1]
Penumbra .................................. (128, 768] [#0.3, 0.0] >0.05 [0, 1] [0, 1]
Umbra ....................................... >128 <#0.3 All [0, 1] [0, 1]
Disk center (original):
Bright mixed polarity ............. (8, 256] >0.0 All [0.5, 1.0] <0.75
Bright unipolar....................... (8, 256] >0.0 All [0.5, 1.0] >0.75
Dark mixed polarity ............... (8, 256] <0.0 All [0.5, 1.0] <0.75
Dark unipolar ........................ (8, 256] <0.0 All [0.5, 1.0] >0.75
Limb (original):
Bright ..................................... (8, 256] #0.0 All [0, 0.5) [0, 1]
Dark ...................................... (8, 256] <0.0 All [0, 0.5) [0, 1]
Disk center (revised):
Bright mixed polarity ............. (8, 256] [0.0, 0.1] [#0.1, 0.05] [0.5, 1.0] <0.5
Bright unipolar....................... (8, 256] [0.0, 0.1] [#0.1, 0.05] [0.5, 1.0] >0.5
Dark mixed polarity ............... (8, 256] [#0.1, 0.0) [#0.1, 0.05] [0.5, 1.0] <0.5
Dark unipolar ........................ (8, 256] [#0.1, 0.0) [#0.1, 0.05] [0.5, 1.0] >0.5
Limb (revised):
Bright ..................................... (8, 256] #0.0 All [0, 0.5) [0, 1]
Dark ...................................... (8, 256] [#0.3, 0.0] <0.05 [0, 0.5) [0, 1]
SOLAR IRRADIANCE AND THE NASA/NSO SPM 659

definition. In the `` original '' basis set used by JBJW, we also
include weak­field fractional areas for the whole disk, disk­
center unipolar and mixed­polarity regions, and limb. In the
`` new '' set, we include instead the weak­field fractional area
and contrasts with respect to limb darkening for intensity,
equivalent width, and line depth for the whole disk. Our
substantive conclusions do not depend on which feature
definition or basis set is used.
Altogether, 36 variables are computed for either basis set.
Factor analysis (see, for example, Harman 1976), which
models the observed variables as a linear combination of
unknown `` common '' factors plus unique variability
(including noise), is used to reduce the dimensionality of the
original basis by accounting for cross­correlations. Rather
than iteratively solving the complete factor analysis model,
we fix the unique variability at zero and identify the com­
mon factors as the principal components (eigenvectors) of
the correlation matrix, select from inspection of the eigen­
value `` scree '' plot the factors with the six largest eigenval­
ues, and perform an orthogonal `` varimax '' rotation of
these to maximize the column variance of the factor loading
matrix while maintaining zero cross­correlation between
factors. The rotated factors, which are linear combinations
of the original variable set, account for most of the variance
in the SPM data and have relatively unambiguous interpre­
tations in terms of solar features. A more detailed descrip­
tion of this process is given in JBJW. Finally, we perform a
multiple regression with TSI as dependent and SPM factors
as independent variables.
Since the work of JBJW, the accuracy of the algorithm
for computing the weak­field reference contrasts for inten­
sity, equivalent width, and line depth has been improved. In
this paper, we report results from applying the improved
algorithm to both the time period spanned by JBJW and
that spanned by the complete data set. The factor loadings
(correlations of the original variables with the factors) for
the original and improved algorithms are highly correlated,
and our conclusions would remain unchanged had we used
the original algorithm. Although any combination of fea­
ture definitions and basis sets that we tried leads to the same
conclusions, computations using the `` new '' basis variables
extracted for the `` revised '' features show, by a small mar­
gin, the best combination of clear interpretability of the fac­
tor loadings and continuation of the factor structure over
the entire time period and will be displayed in this paper.
4. RESULTS
4.1. Factor Analysis
Figure 1 shows the fraction of explained variance for both
the unrotated and rotated factors over the entire time period
of the SPM observations. The six retained factors explain
about 84% of the SPM variance, while the dominant three
factors alone account for about 69%. We assume that the
remaining 30 components, whose eigenvalues slowly decay
to zero and which both individually and cumulatively
account for a small fraction of the SPM variance, are pri­
marily indicative of noise in the data. Note that only the first
three factors show significant changes in explained variance
as a result of the orthogonal rotation.
Figure 2 shows the rotated factor loadings (i.e., the corre­
lations of each factor with each of the original basis varia­
bles) in bar graph form for the entire observing period (1991
November 21--2000 September 30). Factors are numbered
in decreasing order of explained variance. Over this period,
the most important factor, factor 1, is highly correlated with
fractional area and di#erential contrasts of intensity, equiv­
alent width, and line depth for both bright and dark
unipolar, disk­center features and bright limb features (with
the exception of equivalent width for dark disk­center
features) and fractional area of dark limb features and
weak­field structures. This loading pattern represents spa­
tial structures associated with active regions outside of
sunspots together with enhanced unipolar network and at
least roughly corresponds to faculae in the standard model
of TSI variations. Factor 2 is associated with area and con­
trasts in sunspots, while factor 3 depends on areas and
contrasts for both bright and dark mixed­polarity, disk­
center features (again with the exception of equivalent width
for dark structures). One consequence of our revised feature
definition is the cleaner separation of unipolar and mixed­
polarity regions in factors 1 and 3 than the original
classification of JBJW while factor 1, representing primarily
faculae, shows stronger correlation with sunspots. As will
be shown more quantitatively below, the dominant three
factors closely correspond to the factor structure described
by JBJW for the limited time period.
Although we do not yet understand how the loading pat­
terns for the three minor factors relate to conventionally
recognized solar features, they consistently appear with the
Fig. 1.---Proportion of explained variance as a function of factor num­
ber. The upper panel shows contributions for each factor individually, and
the lower panel shows cumulative contribution. Open circles show the non­
rotated factors (principal components), while plus signs denote the six
rotated factors.
660 JONES ET AL. Vol. 589

improved analysis algorithm, new basis variable set, and
revised feature definitions regardless of sampling period.
Di#erential contrasts of intensity, equivalent width, and line
depth in dark limb features are related to factor 4; factor 5 is
primarily correlated with di#erential contrast of equivalent
width for dark disk­center structures, regardless of unipo­
larity; and factor 6 is associated with contrasts of intensity
(primarily) and equivalent width (secondarily) of weak
magnetic elements.
Figure 3 shows the rotated factor loadings for the original
period treated by JBJW (1991 November 21--1994 March
30). Except for minor interchanges of order (factors 1 and 3
as well as 4 and 5), Figures 2 and 3 are visually very similar.
The loading patterns are similar for all six of the retained
factors and therefore are likely to reflect intrinsic properties
rather than an accidental result of the sampling period. The
di#erent ordering over the two time periods reflects partly
the varying importance of solar properties over di#erent
phases of the sunspot cycle and partly the near equality of
explained variance among the relevant factors.
The stability of the factor patterns is more quantitatively
verified in Table 2, which shows the correlation matrix of
the two factor patterns. That is, letting c w Ïi; k÷ and c s Ïi; k÷
denote the correlations over time (factor loadings) of the ith
factor with the kth basis variable for the whole and short
time periods, respectively, the correlations over basis
Fig. 3.---Factor loadings for the initial SPM period treated by JBJW in
the format of Fig. 2.
Fig. 2.---Bar graph of the correlation coe#cients between the rotated
factors and original variables (factor loadings) for the entire SPM observa­
tion period. The x­axis is labeled by original variable: i, q, and d denote con­
trasts of intensity, equivalent width, and central line depth; n denotes
fractional number of pixels (area); `` pen '' refers to sunspot penumbra,
`` umb '' to umbra, `` unicb(d) '' to unipolar disk­center bright (dark) pixels,
`` mixcb(d) '' to mixed­polarity disk­center bright (dark) pixels, `` lb(d) '' to
limb­bright (dark) pixels, and `` wk '' to weak­field pixels.
TABLE 2
Correlation of Factor Patterns
Factors for 1992 Nov 21--2000 Sep 30
Factors for 1992
Nov 21--1994 Mar 30 1 2 3 4 5 6
1.................................. 0.490 0.233 0.921 #0.207 #0.217 0.184
2.................................. 0.478 0.953 0.116 #0.091 0.073 0.040
3.................................. 0.920 0.507 0.019 #0.014 0.118 0.069
4.................................. #0.055 0.085 #0.529 #0.034 0.890 #0.313
5.................................. #0.043 #0.055 #0.170 0.955 #0.004 #0.402
6.................................. 0.058 #0.012 #0.011 #0.166 0.310 0.790
No. 1, 2003 SOLAR IRRADIANCE AND THE NASA/NSO SPM 661

variables,
CÏi; j÷ ¼ X 36
k¼1
c s Ïi; k÷ # c s Ïi÷
h i
½ #
# ½c w Ï j; k÷ # c w Ï j÷
h i#= # s Ïi÷# w Ï j÷
½ # ;
are shown in Table 2, where, for example, c w Ïi÷
h i and # w Ïi÷
denote the mean and standard deviation over basis variables
of c w Ïi÷. Note that since c w and c s are di#erent, CÏi; j÷ is not
symmetric. Allowing for the di#erence in ordering noted
above, the three major factor patterns have correlation coef­
ficients of greater than 0.92 with each other and less than
0.51 with other factors. Similar but slightly weaker results
apply to the three minor factors.
4.2. Multiple Regression
Table 3 shows the coe#cients and their standard errors
along with the fraction of explained variance (multiple R 2 )
for three regressions of composite TSI as a function of the
six rotated factors. The first covers the total time period
1992 November 21--2000 September 30; the second and
third regressions, following de Toma et al. (2001), are inde­
pendent fits of the periods before and after 1996 January 1,
a division date corresponding roughly to the minimum sep­
arating cycles 22 and 23 as well as the beginning of observa­
tions from the VIRGO radiometers. The bottom row of
Table 3 shows the slopes in units of W m #2 yr #1 of linear
least­squares fits of the residuals (TSI#regression) for the
three multiple regressions as a function of time together
with the formal errors and fractions of the residual varian­
ces accounted for by the fits. The composite TSI (version
23) together with regression fits and residuals are plotted as
functions of time in Figure 4.
For the total period, the six factors account for about
76% of the TSI variance, somewhat more than the similar
analysis of JBJW but less than the best traditional analyses
using the Photospheric Sunspot Index (PSI) and Photo­
spheric Facular Index (PFI) (e.g., Chapman et al. 1996).
Moreover, as in JBJW, unipolar regions (faculae) and sun­
spots account for almost all the explained variance, while
strong­field mixed­polarity (quiet network) features,
although accounting for a significant fraction of the SPM
variance, are just barely correlated with TSI at the 3 # level
and account for negligible amounts of TSI variance. For the
declining phase of cycle 22, the second multiple regression
accounts for noticeably less (70%) TSI variance, while the
third multiple regression accounts for about 88% of the TSI
variance in the growth and maximum phase of cycle 23. For
the latter two regressions, the major di#erences between TSI
and SPM prediction occur on timescales of days to a rota­
tional period, and the explained variance for cycle 23 is com­
parable to the best current fits obtained from other data.
Although quiet network (factor 3) has coe#cients in the lat­
ter two regressions that are statistically di#erent from zero
at better than 3 #, the feature accounts for less than 1% of
the TSI variance in either case. Although the present
analysis is in qualitative agreement with JBJW, the use of
improved analysis algorithms and particularly revised fea­
ture definitions (see Table 1) changes details; unipolar
regions and sunspots account for comparable amounts of
TABLE 3
Multiple Regressions
1992 Nov 21--2000 Sep 30 1992 Nov 21--1995 Dec 31 1996 Jan 1--2000 Sep 30
Factor
(SPM R 2 ) Coe#cient Error R 2 Coe#cient Error R 2 Coe#cient Error R 2
1 (0.30) ........................ 0.346 0.004 0.681 0.205 0.007 0.482 0.376 0.004 0.765
2 (0.21) ........................ #0.111 0.004 0.071 #0.190 0.008 0.205 #0.134 0.004 0.100
3 (0.18) ........................ #0.012 0.004 0.001 #0.025 0.006 0.007 #0.040 0.004 0.007
4 (0.06) ........................ 0.015 0.004 0.001 0.001 0.005 0.000 0.009 0.005 0.000
5 (0.06) ........................ 0.028 0.004 0.005 0.009 0.006 0.001 0.043 0.004 0.009
6 (0.04) ........................ 0.011 0.004 0.001 0.004 0.004 0.000 #0.032 0.007 0.002
Total (0.84) ............. 0.760 0.702 0.883
Trend .......................... 0.050 0.002 0.295 #0.006 0.006 0.001 0.006 0.003 0.003
Fig. 4.---TSI (top), multiple regression fits (middle), and residuals
(TSI#fits; bottom) as a function of time in units of years from 1992 Novem­
ber 21 to 2000 September 30. All abscissae are in units of Wm #2 . The thick
line in the upper curve of lower panel shows linear least­squares fit to the
residuals for the entire period. The lower curves of middle and bottom pan­
els show fits and residuals for regressions before (after) 1996 January 1,
displaced by 1 Wm #2 , as thin (thick) lines.
662 JONES ET AL. Vol. 589

TSI variance in JBJW, while in this analysis the regressions
are dominated by unipolar features.
It is worth noting that our regression analysis is purely
correlative (indeed, all the basis variables, being normalized
to zero mean and unit standard deviation, are dimension­
less) and should not be considered physically predictive.
This may partially account for the tendency of multiple R 2
in the above regressions to be somewhat lower than tradi­
tional analyses, particularly on rotational timescales and
shorter. For example, although the division of the spatial
domain into disk­center and limb domains is a coarse
attempt to take into account the marked center­to­limb dif­
ferences in facular contrast, we make no attempt to translate
these contrasts to physical units by multiplying them by a
standard limb­darkening curve. (In fact, it is not clear how
to transform contrasts in equivalent width and line depth to
physically relate to TSI variability.) Thus we are less likely
to model accurately the passage of faculae across the visible
disk.
A new feature appears in the current regression analysis
for the total time period, which can be seen both in the bot­
tom row of Table 3 and in the bottom panel of Figure 4. In
addition to noise on short timescales (days--weeks), the plot
shows systematic linear variation with a slope of about 0.05
W m #2 yr #1 . The fit accounts for 30% of the variance in the
residual curve, i.e., about 7% of the TSI variance, suggesting
that if this trend were removed from the TSI data, multiple
R 2 would increase to about 0.83. Over an 8 yr period this
trend is a significant fraction of the TSI variation from solar
minimum to maximum. One can also see from Table 3 and
Figure 4 that there are no significant residual trends if the fit
is segmented at 1996 January 1. However, the regression
coe#cients of the dominant factors for the two time periods
di#er from each other well beyond their formal errors in the
sense that the composite TSI after 1996 January 1 depends
more strongly on factor 1 (faculae) and less strongly on fac­
tor 2 (sunspots). As discussed below and in agreement with
de Toma et al. (2001), these results imply either that there
are systematic observational errors in one or both the TSI
or SPM data sets or that an additional solar source of irradi­
ance variation exists that is not detected in the SPM data.
5. DISCUSSION
The two major conclusions of JBJW are shown to extend
to a decadal timescale. First, unipolar magnetic areas asso­
ciated with active­region and active­network faculae and
sunspots dominate the correlation of SPM observations
with TSI. Second, strong­field, mixed­polarity regions
(quiet network), although contributing substantially to the
total variance of the SPM record, are e#ectively uncorre­
lated with TSI. The duration of our observations is not
extensive enough to exclude quiet network as an important
long­term source of solar irradiance variations. However,
our data do span most phases of the sunspot cycle, and the
correlation of TSI with mixed­polarity network is low
enough to imply that any such contribution is likely to occur
only on temporal scales considerably exceeding a solar
cycle. Our analysis does not provide further insight into
why TSI is poorly correlated with mixed­polarity network.
We note that Jones (1998) was unable to find any indica­
tions of thermodynamic di#erences between unipolar and
mixed­polarity regions with comparable LOS fluxes and
speculate, as did JBJW, that quiet network is associated
with a `` magnetic carpet '' (Title & Schrijver 1998), which
disappears and renews over times much shorter than a solar
rotation period.
The linear temporal trend in the residuals of the multiple
regression over the entire interval spanned by our data is
consistent with a nonmagnetic solar source of TSI variation
or with long­term, systematic, instrumental e#ects. We are
unaware of any likely source of long­term instrumental
trends in the SPM measurements of LOS field but plan to
check on this possibility by comparing SPM observations
with other measurements. Even if such a trend were found,
however, we note that magnetic flux enters into our analysis
only through the division of histograms into features and is
not otherwise used in determining the measures that com­
prise the basis set for the factor analysis and multiple regres­
sion. Thus, a change in SPM sensitivity over time would
modify the temporal variation of each factor but, to the
extent that our representation is complete, would not a#ect
the overall regression. There may, of course, be irradiance
changes associated with magnetic features below the
sensitivity or resolution limits of the SPM that would be
undetected in our measurements.
In any case, a trend produced by a continuous phenom­
enon, such as might be expected from the changing sensitiv­
ity of a single instrument or a long­term solar variation,
should be observed in regressions of subsamples of the data.
However, evidence for long­term systematic variation of the
residuals in the segmented regressions is absent, which sug­
gests that the trend for the entire time period may be an arti­
fact of forcing a continuous fit to disjoint intervals.
Although the fit for either interval would be consistent with
the two­component activity model of TSI variation, the
dominant regression parameters for cycles 22 and 23 are dis­
tinctly di#erent. It is possible to construct scenarios for
which the response of actual TSI to solar activity, as mea­
sured by SPM, is di#erent in cycle 23 than it was in cycle 22.
For example, W. Livingston (2002, private communication)
finds that sunspot umbrae in the 1.6 lm lines near the maxi­
mum of cycle 23 show weaker magnetic fields and intensity
contrasts than do those drawn from a comparable sample
during the maximum of cycle 22. These di#erences might be
undetected in SPM measurements because of substantial
photospheric stray light and would appear as a weaker sen­
sitivity of TSI to observed sunspot properties in cycle 23.
Unfortunately, it is not possible to test this idea directly
since the SPM was not operational during the maximum of
cycle 22. Similarly, if, in cycle 23, unipolar network and fac­
ulae were brighter in the UV for given photospheric proper­
ties, then an enhanced apparent response of TSI would be
observed. Such a scenario would be consistent with the
work of Unruh, Solanki, & Fligge (1999) and Preminger et
al. (2002), who suggest that lines are responsible for the bulk
of the cyclic variability of TSI as well as with observations
of the Mg ii h and k resonance lines, which tend to show
comparable levels of emission in cycles 22 and 23. However,
a source of chromospheric heating that is independent of
photospheric magnetic properties detectable by the SPM
would be required.
A simpler and perhaps more likely explanation involves
uncompensated systematic instrumental e#ects. While there
are no documented changes in SPM instrumentation or
sensitivity between cycles 22 and 23, the observed com­
posite TSI is sensitive to changes in the ensemble of opera­
tional radiometers, and one such change occurred on 1996
No. 1, 2003 SOLAR IRRADIANCE AND THE NASA/NSO SPM 663

January 18 with the onset of VIRGO observations. Thus,
the composite TSI, even though it has been carefully pre­
pared to compensate for individual instrument sensitivities,
plausibly records TSI with a di#erent `` gain '' in cycle 23
than in cycle 22. Note that the SPM data fit the cycle 23 (pre­
dominantly VIRGO) observations best. Moreover, other
composite records of TSI have been proposed. In particular,
Willson (1997), using di#erent assumptions about degrada­
tion of individual spacecraft radiometers, suggests that the
value of TSI at solar minimum has not been constant over
the past two solar cycles but has increased by about 0.036%
per decade, which agrees well with the 0.05 W m #2 yr #1
trend (0.037% per decade) of our regression for 1992--2000.
We plan to compare our results with Willson's composite
record of TSI in the near future.
6. FUTURE WORK
Our histogram­based technique tends to account for less
TSI variance than the best representations based on other
ground­based data, particularly in cycle 22. To understand
better what aspects of our analysis lead to this result, we
plan to compare the spatial structures that represent the fac­
tor patterns outlined in x 4.1 with other methods of feature
classification, particularly those developed by Harvey &
White (1999), Turmon, Pap, & Mukhtar (2002), and
Preminger et al. (2001). The work of Turmon et al. (2002)
in particular may suggest better posed multidimensional
alternatives to our histogram classification. Functional
relationships between all the SPM observables derived from
histogram analysis can be cross­checked with predictions
from extant models such as those constructed by Fontenla
et al. (1999), Solanki & Unruh (1998), Fligge et al. (1998),
and Unruh et al. (1999) to explore physical di#erences
between features.
We have not attempted to separate quantitatively rota­
tional, decadal, and other possible temporal scales in the
representation of either TSI or SPM data. However, we plan
to apply singular spectrum analysis (Pap & Varadi 1996)
and possibly other techniques, such as wavelet analysis, in
the near future. This may help us understand what parts of
the unexplained variance are attributable to di#erent scales
of variation and to clarify the lack of correlation between
mixed­polarity features and TSI.
Finally, longevity and continuity are essential ingredients
for any observational study of solar irradiance. Within cal­
endar year 2003, the SPM will be retired from service to be
replaced by the Vector Spectromagnetograph (VSM), a part
of NSOs Synoptic Optical Long­term Investigations of the
Sun (SOLIS) instrumental package. The VSM will continue
the observational repertoire of the SPM, albeit using a dif­
ferent spectral line, and will provide hitherto unavailable
information regarding the vector magnetic field over the
entire solar disk. Thus, important parts of our future work
will be to complete analysis for the SPM data set, to ensure
that the transition between the two instruments produces
minimal discontinuity in the synoptic magnetogram record,
and to develop methods for including the more complete
physical description of the solar atmosphere provided by
the VSM into our irradiance analysis.
The authors acknowledge useful discussions with G.
Chapman, K. Harvey, W. Livingston, J. Pap, M. Turmon,
and S. Walton as well as use of computer facilities at the
University of Arizona for carrying out the factor and multi­
ple regression analyses. The authors also wish to thank an
anonymous referee for a very careful reading of the original
manuscript and constructive comments that led to sub­
stantial improvements. This research was partially sup­
ported by NASA Supporting Research and Technology
tasks 344­12­52­14 and 344­12­52­19. NSO/KPVT data
used here were produced cooperatively by AURA/NSO,
NASA/GSFC, and NOAA/SEC.
REFERENCES
Chapman, G. A., Cookson, A. M., & Dobias, J. J. 1996, J. Geophys. Res.
101, 13,541
de Toma, G., White, O. R., Chapman, G. A., Walton, S. R. Preminger,
D. G., Cookson, A. M., &Harvey, K. L. 2001, ApJ, 549, L131
Fligge, M., Solanki, S. K., &Unruh, Y. C. 2000, A&A, 353, 380
Fligge, M., Solanki, S. K., Unruh, Y. C., Fro ˜ hlich, C., & Wehrli, Ch. 1998,
A&A, 335, 709
Fontenla, J., White, O. R., Fox, P. A., Avrett, E. H., &Kurucz, R. L. 1999,
ApJ, 518, 480
Foukal, P., &Lean, J. 1988, ApJ, 328, 347
Fro ˜ hlich, C. 2000, Space Sci. Rev., 94, 15
Fro ˜ hlich, C., &Lean, J. 1998, Geophys. Res. Lett., 25, 4377
Harman, H. H. 1976, Modern Factor Analysis (Chicago: Univ. Chicago
Press)
Harvey, K. L., &White, O. R. 1999, ApJ, 515, 812
Jones, H. P. 1998, in ASP Conf. Ser. 140, Synoptic Solar Physics, ed.
K. S. Balasubramaniam, J. W. Harvey, & D. M. Rabin (San Francisco:
ASP), 145
Jones, H. P., Branston, D. D., Jones, P. B., & Wills­Davey, M. J. 2000,
ApJ, 529, 1070 (JBJW)
Pap, J. M., &Fro ˜ hlich, C. 1998, J. Atmos. Sol.­Terr. Phys. 61, 15
Pap, J. M., & Varadi, F. 1996, in ASP Conf. Ser. 95, Solar Drivers of Inter­
planetary and Terrestrial Disturbances, ed. K. S. Balasubramaniam,
S. L. Keil, &R.N. Smartt (San Francisco: ASP), 576
Preminger, D. G., Walton, S. R., & Chapman, G. A. 2001, Sol. Phys., 202,
53
---------2002, J. Geophys. Res., 107 (All), 10.1029/2001JA009169
Solanki, S. K., &Unruh, Y. C. 1998, A&A, 329, 747
Title, A. M., & Schrijver, C. J. 1998, in ASPConf. Ser. 154, 10th Cambridge
Workshop: Cool Stars, Stellar Systems, and the Sun, ed. R. A. Donahue
& J. A. Bookbinder (San Francisco: ASP), 345
Turmon, M., Pap, J. M., &Mukhtar, S. 2002, ApJ, 568, 396
Unruh, Y. C., Solanki, S. K., &Fligge, M. 1999, A&A, 345, 635
Willson, R. C. 1997, Science 277, 1963
664 JONES ET AL.