Scaling in long term data sets of geomagnetic indices and solar wind epsilon as seen by WIND spacecraft
ABSTRACT We study scaling in fluctuations of the geomagnetic indices (AE, AU, and AL) that provide a measure of magnetospheric activity and of the epsilon parameter which is a measure of the solar wind driver. Generalized structure function (GSF) analysis shows that fluctuations exhibit self-similar scaling up to about 1 hour for the AU index and about 2 hours for AL, AE and epsilon when the most extreme fluctuations over 10 standard deviations are excluded. The scaling exponents of the GSF are found to be similar for the three AE indices, and to differ significantly from that of epsilon. This is corroborated by direct comparison of their rescaled probability density functions.
Article: Gaussian Self-A nity and Fractals
- [show abstract] [hide abstract]
ABSTRACT: The probability distribution functions (Pdfs) of the AE-index fluctuations at different time scales have been investigated using a time series covering a period from January 01, 1978 to December 31, 1985. The Pdfs are always non Gaussian for time scales in the range 1-120 min both in quiet and disturbed periods. The scale dependence of the Pdfs indicates that AE-index is not characterized by a global time self-similarity, indicating that an intermittency phenomenon characterizes both phases. The results on Pdfs are compared with functional form, proposed by Castaing et al. , to characterize intermittency phenomena in ordinary turbulent fluid flows. Moreover the relevance of these observations to the understanding of the magnetospheric dynamical configuration is pointed out.Geophysical Research Letters - GEOPHYS RES LETT. 01/1998; 25(21):4087-4090.
- [show abstract] [hide abstract]
ABSTRACT: The behavior of substorms as sudden transitions of the magnetosphere is studied using the Bargatze et al.  data set of the solar wind induced electric field vBs and the auroral electrojet index AL. The data set is divided into three subsets representing different levels of activity, and they are studied using the singular spectrum analysis. The points representing the evolution of the magnetosphere in the subspace of the eigenvectors corresponding to the three largest eigenvalues can be approximated by two-dimensional manifolds with a relative deviation of 10-20%. For the first two subsets corresponding to small and medium activity levels the manifolds have a pleated structure typical of the cusp catastrophe. The dynamics of the magnetosphere near these pleated structures resembles the hysteresis phenomenon typical of first-order phase transitions. The reconstructed manifold is similar to the ``temperature-pressure-density'' diagrams of equilibrium phase transitions. The singular spectra of vBs, AL, and combined data have the power law dependence typical of second-order phase transitions and self-organized criticality. The magnetosphere thus exhibits the signatures of both self-organization and self-organized criticality. It is concluded that the magnetospheric substorm is neither a pure catastrophe of the low-dimensional system nor a random set of avalanches of different scales described by the simple sandpile models. The substorms behave like nonequilibrium phase transitions, with features of both first- and second-order phase transitions.Journal of Geophysical Research 01/2000; 105:12955-12974. · 3.17 Impact Factor
Scaling in long term data sets of geomagnetic indices and solar wind ? ? ? ? ?
as seen by WIND spacecraft
B. Hnat,1S. C. Chapman,1G. Rowlands,1N. W. Watkins,2and M. P. Freeman2
Received 20 July 2003; revised 7 October 2003; accepted 14 October 2003; published 28 November 2003.
indices (AE, AU, and AL) that provide a measure of
magnetospheric activity and of the ? parameter which is a
measure of the solar wind driver. Generalized structure
function (GSF) analysis shows that fluctuations exhibit
self-similar scaling up to about 1 hour for the AU index
and about 2 hours for AL, AE and ? when the most
extreme fluctuations over 10 standard deviations are
excluded. The scaling exponents of the GSF are found
to be similar for the three AE indices, and to differ
significantly from that of ?. This is corroborated by
direct comparison of their rescaled probability density
INDEX TERMS: 2159 Interplanetary Physics:
Plasma waves and turbulence; 2704 Magnetospheric Physics:
Auroral phenomena (2407); 2784 Magnetospheric Physics: Solar
wind/magnetosphere interactions; 2788 Magnetospheric Physics:
Storms and substorms; 3250 Mathematical Geophysics: Fractals
and multifractals. Citation: Hnat, B., S. C. Chapman,
G. Rowlands, N. W. Watkins, and M. P. Freeman, Scaling in
long term data sets of geomagnetic indices and solar wind ? as
seen by WIND spacecraft, Geophys. Res. Lett., 30(22), 2174,
We study scaling in fluctuations of the geomagnetic
 The statistical properties of fluctuations in geomag-
netic indices and their relation to those in the solar wind, is a
topic of considerable interest (see, e.g., [Sitnov et al., 2000;
Tsurutani et al., 1990; Ukhorskiy et al., 2002; Vo ¨ro ¨s et al.,
1998]). Scaling has been identified as a key property of
magnetospheric energy release in the form of bursty bulk
flows in the magnetotail [Angelopoulos et al., 1992],
‘‘blobs’’ in the aurora [Lui et al., 2000], non-Gaussian
fluctuations in geomagnetic indices [Hnat et al., 2002,
2003a; Consolini et al., 1996] and in single station magne-
tometer data [Kova ´cs et al., 2001; Vo ¨ro ¨s et al., 1998].
Models include Self-Organized Criticality (SOC) [Chang
et al., 2003] (see also the review [Chapman and Watkins,
2001]) and multi-fractal models [Kova ´cs et al., 2001]
related to those of turbulence [Consolini et al., 1996; Vo ¨ro ¨s
et al., 1998].
 These measures of scaling and non-Gaussian fluctua-
tions in magnetospheric output need to be understood in the
context of the system’s driver, the solar wind, which is
turbulent and thus also scaling. Other work has focussed on
comparing properties of input parameters such as ? and the
indices (AE, AU, and AL) to establish whether they are
directly related. However, these studies have not provided a
consistent answer. While Freeman et al.  found that
both the ? and the AU and AL indices exhibited nearly
identical scaling of burst lifetime probability density func-
tions (PDFs), Uritsky et al.  obtained quite different
scalings for AE and the solar wind quantity vxByzusing
spreading exponent methods motivated by SOC. Hnat et al.
[2002, 2003a] used a PDF rescaling technique to character-
ize the fluctuation PDF of 4 years ? data from WIND and a
1 year data set of AE indices with fluctuations over a few
standard deviations. Direct comparison of the PDF’s func-
tional form suggested close similarity to within statistical
 In this paper we use a larger 10-year data set for the
AE indices to obtain a more accurate statistical determina-
tion of the functional form of the PDF of fluctuations over a
more extensive dynamic range, including characterization
of extremal events up to 10 standard deviations for the first
time. We apply structure functions to characterize and
compare both the low and higher order moments for all
quantities. A 4-year subset of the index data, corresponding
to the same period in the solar cycle as that used to produce
?, is used to facilitate this comparison. We then verify these
results by direct examination of the fluctuation PDF using
the full 10-year AE indices dataset.
2. Data Sets
 The AL, AU, and AE index data sets investigated here
comprise over 5.5 million, 1 minute averaged samples from
January 1978 to December 1988 inclusive. The ? data set is
identical to that used in Hnat et al. [2002, 2003a] and
extends from January 1995 to December 1998 inclusive. It
includes intervals of slow and fast speed streams. ? is
defined (see [Hnat et al., 2002]) in SI units as ? = v(B2/
was calculated from the WIND spacecraft key parameter
database [Lepping et al., 1995; Ogilvie et al., 1995]. The
indices and ? are from different time intervals and here we
assume statistical stability over these long time intervals.
2sin4(?/2), where l0? 7REand ? = arctan(jByj/Bz), and
3.Generalized Structure Functions
 Generalized structure functions (GSF), or general-
ized variograms, can be defined in terms of an average
over time of a differenced variable dx(t, t) = x(t + t) ?
x(t) as Sm(t) = hjdx(t, t)jmi [Rodrı ´guez-Iturbe and Rinaldo,
1997]. If dx exhibits scaling with respect to t, then Sm/
tz(m). A log-log plot of Smversus t should then reveal a
straight line for each m with gradients z(m). If z(m) = am
(a constant) then the time series is self-similar with single
scaling exponent a.
GEOPHYSICAL RESEARCH LETTERS, VOL. 30, NO. 22, 2174, doi:10.1029/2003GL018209, 2003
1Space and Astrophysics Group, University of Warwick Coventry, UK.
2British Antarctic Survey, Natural Environment Research Council,
Copyright 2003 by the American Geophysical Union.
SSC6 - - 1
 In order to compare the scaling properties of the
non-contemporaneous ? and AE indices time series, we
select a 4-year subinterval 1984–1987 from the AE indices
at the same phase in the solar cycle as the ? data. Figure 1
shows the second order GSFs as measured by the standard
deviations s(t) = [S2(t)]1/2of the fluctuation dx(t, t). A
scaling region is apparent between 27and 212s where
s(t) / tH, where H is the Hurst exponent [z(2)/2]. The R2
goodness of fit analysis was performed to select the
optimal power law region and gradient and results are
summarized in Table 1. The upper limits of the scale
regions tmaxare in good agreement with values reported
previously [Consolini and De Michelis, 1998; Takalo et
al., 1993; Takalo and Timonen, 1998].
 Any such single estimate of the H, whilst establishing
the region of t over which there is scaling, does not fully
characterize the properties of the time series. For example,
a fractional Brownian motion (fBm) can be constructed
to share the same H value as AE, but the fBm series
has Gaussian distributed increments dx by definition
[Mandelbrot, 2002] whereas those of AE are non-Gaussian
[Consolini and De Michelis, 1998; Hnat et al., 2002]. As
discussed by Mandelbrot  the similar values arise
because H aggregates two sources of scaling in monofractal
random walks: persistence (the ‘‘Joseph’’ effect) and heavy
tails in the increments (the ‘‘Noah’’ effect). In the above
example the anomalous value of H for fBm comes just from
the Joseph effect, whilst for AE the Noah effect must be at
work. Furthermore, estimating H by only one method may
not distinguish a fractal time series from a discontinuous
one [Watkins et al., 2001; Katsev and L’Heureux, 2003]. We
thus turn next to the higher order m values of z(m).
 Figure 2 shows scaling exponents z(m) derived from
raw GSFs with m varying between ?1 and 8 for the d? and
AE indices fluctuations. These suggest the departure of
higher orders from self-similarity, i.e., z(m) departs from a
straight line. The inset of this figure shows the origin of
these z(m) values for d AU and m = 1, ..., 7. Only the first
four orders exhibit clear linear behavior expected in the
scaling region. For higher orders, the value of z very
strongly depends on the assumed extent of the scaling
region to which one fits a straight line. In principle, z(m)
can be obtained for any m. However, errors do not contrib-
ute uniformly over m, for example, the largest fluctuations
that affect large m, are statistically poorly resolved, whereas
the smallest fluctuations (dx ! 0) are dominated by instru-
ment thresholds. For the latter reason we will exclude m =
?1 for d? as d? ! 0 is not well determined through its
 Conditioned GSFs quantify the impact of intermit-
tency on fluctuations of different sizes by imposing a
threshold A on the event size [Kova ´cs et al., 2001]. Here,
this threshold will be based on the standard deviation of the
differenced time series for a given t, A(t) = 10s(t). This
procedure allows us to exclude rare extreme fluctuations
with large statistical errors which, for large m, could lead to
a spurious departure from self-similar behavior. Alterna-
tively, conditioning with different thresholds estimates a
maximum size for the fluctuations for which self-similarity
is still valid.
 Following conditioning, log-log plots of Sm
good correspondence with straight line fits, shown for dAU
in the inset of Figure 3. This power law dependence holds
between times already obtained from the R2analysis
performed for s(t). The main plot then shows z(m) obtained
from the conditioned Sm
fitted for moments between ?1 (0 for ?) and 6 and then
extended to the entire range of data. Scaling exponents
c(t). All lines in the figure were
Figure 1. Scaling of the standard deviation of the PDFs of:
?-?, 6-AU index, 4-AL index and 5-the AE index. The
plots have been offset vertically for clarity. Error bars are
estimated assuming Gaussian statistics for the binned data.
Figure 2. Dependence of the scaling exponent z(m) of the
raw GSF on moment order m. Inset shows the GSF Sm
versus time lag t for AU.
Table 1. Scaling Indices Derived from P(0, t), s(t) and GSF
a from P(0, t)
a from s(t)
a from GSF
–0.31 ± 0.04
0.46 ± 0.02
0.41 ± 0.02
0.44 ± 0.02
0.25 ± 0.04
0.40 ± 0.02
0.37 ± 0.02
0.36 ± 0.03
?0.47 ± 0.03
?0.46 ± 0.03
?0.45 ± 0.03
SSC6 - 2HNAT ET AL.: SCALING OF AE INDICES
obtained from this technique were unchanged for thresholds
A(t) between 6s and 12s.
 Firstly, our analysis suggests that the statistics of the
fluctuations for all four quantities are self-similar for times
between 2 and ?100 minutes and fluctuations of size dx ?
10s(t). Secondly, the scaling exponent a in z(m) = am that
characterize this self-similar behavior, are identical within
errors for fluctuations in the AE indices but different to that
in ? at the 1s level.
4.Rescaling of Fluctuation PDFs
 Scaling of the GSFs can be related to scaling proper-
ties of the fluctuation PDFs [Hnat et al., 2002, 2003a]
using the generic, model-independent rescaling method
(e.g., [Mantegna and Stanley, 1995; Hnat et al., 2003b])
based on the rescaling of the PDFs P(dx, t) of dx(t, t) on
differenttime scales t. If atime seriesexhibits statistical self-
similarity, a single argument representation of the PDF can
be found that is given by P(dx, t) = t?aPs(dxt?a), where a
is the rescaling exponent. We now express Smusing the
fluctuations’ PDF, P(dx, t) as follows:
Sm t ð Þ ¼
j jmP dx;t
Expressing the integral in (1) in terms of rescaled variables
Psand dxs= dxt?ashows that the scaling exponent z(m) is a
linear function of m, z(m) = ma, for a statistically self-
similar process, as suggested here by Figure 3.
 The exponent a is ideally obtained from the scaling
of the peaks of the PDF P(0, t). However, the finite
accuracy of the measurement may discretize the amplitude
leading to errors in the peak values. Table 1 gives all scaling
exponents, obtained by different methods. These yield
consistent values of a, to within the errors. We will use a
from the scaling of s(t) versus t. If the fluctuations are
statistically self-similar, as suggested by our GSF analysis,
then the unscaled PDFs P(dx, t) should collapse onto a
single curve Ps(dxs). We applied PDF rescaling to the
fluctuation PDFs of all quantities and obtained satisfactory
collapse of the curves within the scaling regions. The c2test
applied to all quantities revealed that, for the scaling regions
given above, the collapsed curves lie within 5–7% error
 Figure 4 shows the re-scaled fluctuation PDFs for the
indices alone for t ? 15 min. The dx variable has been
normalized to the rescaled standard deviation ss(t ?
15 min.) of Psin each case to facilitate this comparison.
The inset of this figure shows the comparison for AU, AE
and –AL fluctuations and these PDFs are nearly identical.
These results are consistent with conclusions of the GSF
analysis at the 1s level.
 Figure 5 shows the normalized PDFs Ps(dxs) for dx =
d?, dAE and t ? 15 min overlaid on a single plot. We can
clearly distinguish between the PDFs of the d? and AE
Figure 3. Dependence of the scaling exponent of the
conditioned GSF on moment order. Inset shows the
conditioned GSF Sm
cversus time lag t for AU.
Figure 4. Direct comparison between the fluctuation PDFs
for AE(5), AU(6) and AL(4), again at t = 15 minutes.
Inset shows overlaid PDFs of AU, AE and –AL fluctuations.
Error bars as in Figure 1.
Figure 5. Direct comparison, for the particular choice t =
15 minutes, of the fluctuation PDFs for ? (?) and AE index
(5). Error bars as in Figure 1.
HNAT ET AL.: SCALING OF AE INDICES
SSC6 - 3
indices’ fluctuations. We obtain the same result repeating
this comparison for several values of t, within the scaling
range tmax. We have also verified that the functional form of
the PDF are insensitive to the solar cycle within errors. The
use of a larger, 10 year data set for the indices has reduced
statistical scatter and expanded the dynamic range of the
considered fluctuations as compared to the analysis given in
[Hnat et al., 2002, 2003a], and would lead us to draw the
opposite conclusion, that on time scales less than ?1 hour
the AE index amplitude fluctuations are not driven linearly
by those of the solar wind. We would also conclude that the
difference seen at the 1s level in the scaling of the ? and the
indices is significant, even though they agree at the 2s level
[Freeman et al., 2000].
 In this paper we have addressed an open question of
the possible connection between the scaling properties of
fluctuations in the solar wind driver and those observed in
global measures of magnetospheric dynamics. We applied
two statistical methods, generalized structure functions and
PDF rescaling, to study the scaling of fluctuations in the ?
parameter and the magnetospheric indices AU, AL and AE.
We find that, statistically, fluctuations in all four quantities
are approximately self-similar when their size is limited to
?10s. This self-similarity extends to ?1–1.5 hours. The
scaling exponents of the AE indices are close to each other
and are appreciably different to that of the ? parameter.
 The fluctuation PDFs of the AE indices, unlike that
of d?, are asymmetric. Direct comparison of the PDFs for
the fluctuations in the AU, AE and –AL index indicates that
they are nearly identical. Whilst the low frequency behavior
of the solar wind and the indices may be well correlated
[Tsurutani et al., 1990], here we have concluded that, on
time scales smaller than 1 hour the properties of the
fluctuations in the solar wind and the indices differ in both
amplitude and persistence. If the underlying physical origin
of the auroral scaling is turbulence, then different scaling
behavior implies a different type of turbulence, i.e., different
dimensionality/topology or different relevant physics
[Frisch, 1995]. If the underlying physics is SOC or similar
[Chang et al., 2003] then similar conclusions would still be
drawn (c.f. [Uritsky et al., 2001]). However, at this point we
also can not rule out the possibility that the way in which
the indices are constructed ‘‘burns’’ information still present
in the magnetometer data about the solar wind scaling, here
possibly by changing either or both of the degree of
persistence (power spectral slope) and the heavy-tailed
property (see [Edwards et al., 2001] for a related prelimi-
GR the Leverhulme Trust. We thank R. P. Lepping and K. Ogilvie for
provision of data from the NASA WIND spacecraft and the World Data
Center C2, Kyoto for geomagnetic indices.
SCC and BH acknowledge the PPARC and
Angelopoulos, V., et al., Bursty bulk flows in the inner central plasma sheet,
J. Geophys. Res., 59, 4027–4039, 1992.
Chang, T., S. W. Y. Tam, and C. C. Wu, Complexity induced anisotropic
bimodal intermittent turbulence in space plasmas, Phys. Plasmas, in
Chapman, S. C., and N. W. Watkins, Avalanching and self-organised criti-
cality: A paradigm for magnetospheric dynamics?, Space Sci. Rev., 95,
Consolini, G., and P. De Michelis, Non-Gaussian distribution function of
AE-index fluctuations: Evidence for time intermittency, Geophys. Res.
Lett., 25, 4087–4090, 1998.
Consolini, G., M. F. Marcucci, and M. Candidi, Multifractal structure of
auroral electrojet index data, Phys. Rev. Lett., 76, 4082–4085, 1996.
Edwards, J. W., A. S. Sharma, and M. I. Sitnov, Spatio-temporal Structure
of Geomagnetic Activity Triggered by Dynamic Pressure Pulses: Mutual
Information Functional Analysis, Bull. Amer. Phys. Soc., 27, 156, 2001.
Freeman, M. P., N. W. Watkins, and D. J. Riley, Evidence for a solar wind
origin of the power law burst lifetime distribution of the AE indices,
Geophys. Res. Lett., 27, 1087–1090, 2000.
Frisch, U., Turbulence, The legacy of A.N. Kolmogorov, Cambridge Uni-
versity Press, Cambridge, 1995.
Hnat, B., S. C. Chapman, G. Rowlands, N. W. Watkins, and M. P. Freeman,
Scaling in solar wind epsilon and the AE, AL and AU indices as seen by
WIND, Geophys. Res. Lett., 29(10), doi:10.1029/2002GL016054, 2002.
Hnat, B., S. C. Chapman, G. Rowlands, N. W. Watkins, and M. P. Freeman,
Correction to ‘‘Scaling in solar wind epsilon and the AE, AL and AU
indices as seen by WIND’’, Geophys. Res. Lett., 30(10), doi:10.1029/
Hnat, B., S. C. Chapman, and G. Rowlands, Intermittency, scaling, and the
Fokker-Planck approach to fluctuations of the solar wind bulk plasma
parameters as seen by the WIND spacecraft, Phys. Rev. E, 67, 056404,
Katsev, S., and L’Heureux, Are Hurst Exponents Estimated from Short and
Irregular Time Series Meaningful?, Computers and Geosciences, in press,
Kova ´cs, P., V. Carbone, and Z. Vo ¨ro ¨s, Wavelet-based filtering of intermit-
tent events from geomagnetic time series, Planet. Space Sci., 49, 1219–
Lepping, R. P., et al., The WIND Magnetic Field Investigation, Space Sci.
Rev., 71, 207, 1995.
Lui, A. T. Y., et al., Is the Dynamic Magnetosphere an Avalanching Sys-
tem?, Geophys. Res. Lett., 27, 911–914, 2000.
Mandelbrot, B. B., Gaussian Self-Affinity and Fractals: Globality, The
Earth, 1/f Noise and R/S, Springer-Verlag, Berlin, 2002.
Mantegna, R. N., and H. E. Stanley, Scaling Behavior in the Dynamics of
an Economic Index, Nature, 376, 46, 1995.
Ogilvie, K. W., et al., SWE, a comprehensive plasma instrument for the
wind spacecraft, Space Sci. Rev., 71, 55–77, 1995.
Rodrı ´guez-Iturbe, I., and A. Rinaldo, Fractal River Basins: Chance and
Self-Organization, Cambridge University Press, Cambridge, 1997.
Sitnov, M. I., et al., Phase transition-like behavior of the magnetosphere
during substorms, J. Geophys. Res., 105, 12,955–12,974, 2000.
Takalo, J., J. Timonen, and H. Koskinen, Correlation dimension and affinity
of AE data and bicolored noise, Geophys. Res. Lett., 20, 1527–1530,
Takalo, J., and J. Timonen, Comparison of the dynamics of the AU and PC
indices, Geophys. Res. Lett., 25, 2101–2104, 1998.
Tsurutani, B. T., et al., The nonlinear response of AE to the IMF Bsdriver:
A spectral break at 5 hours, Geophys. Res. Lett., 17, 279–282, 1990.
Ukhorskiy, A. Y., M. I. Sitnov, A. S. Sharma, and K. Papadopoulos, On the
origin of the power-law spectra in magnetospheric dynamics during sub-
storms, J. Geophys. Res., submitted, 2002.
Uritsky, V. M., A. J. Klimas, and D. Vassiliadis, Comparative study of
dynamical critical scaling in the auroral electrojet index versus solar wind
fluctuations, Geophys. Res. Lett., 28, 3809–3812, 2001.
Vo ¨ro ¨s, Z., P. Kova ´cs, A´. Juha ´sz, A. Ko ¨rmendi, and A. W. Green, Scaling
laws from geomagnetic time series, Geophys. Res. Lett., 25, 2621–2624,
Watkins, N. W., M. P. Freeman, C. S. Rhodes, and G. Rowlands, Ambi-
guities in determination of self-affinity in the AE-index time series, Frac-
tals, 9, 471–479, 2001.
? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?
S. C. Chapman, B. Hnat, and G. Rowlands, Space and Astrophysics
Group, University of Warwick Coventry, CV4 7AJ, UK. (hnat@astro.
M. P. Freeman and N. W. Watkins, British Antarctic Survey, Natural
Environment Research Council, Cambridge, CB3 0ET, UK.
SSC6 - 4HNAT ET AL.: SCALING OF AE INDICES