Pages 895-905

Mapping Ionospheric Substorm Response

D. J. Knipp1 and B. A. Emery2

1Department of Physics, Suite 2A6, Fairchild Hall, U. S. Air Force Academy, CO 80840, USA,
  E-mail: knipp@ncar.ucar.edu
2High Altitude Observatory, National Center for Atmospheric Research, Boulder CO 80307-3000, USA

ABSTRACT

In this article we briefly review the Assimilative Mapping of Ionospheric Electrodynamics (AMIE) procedure and discuss responses to inferred dayside and nightside reconnection events from an ionospheric viewpoint. Output from the AMIE procedure is used to illustrate our discussion. We focus on events occurring just prior to and during the passage of a non-compressive density enhancement, a type of solar wind transient, on 3 November 1993, and during a more normal substorm event on 10 November 1993. We present instantaneous convection and current maps for several times during the events and 'difference maps', which show how the currents and convection change over short periods of time. The 'differencing' capability, especially when combined with other information, such as particle injection data and magnetometer pulsation data can provide a powerful diagnostic for identifying substorm onset and investigating substorm and reconnection response.

INTRODUCTION

Instantaneous ionospheric convection and current maps which incorporate multiple data sources are making significant contributions to the body of knowledge regarding substorms and reconnection events [Kamide et al., 1994, Lu et al., 1996, submitted, Siscoe and Lu, 1996, submitted, and Taylor et al., 1996, submitted]. These maps are generally synthesized from observations made by ground-based radar, ground magnetometers and low-earth orbiting satellites. One means of combining such diverse types of observations into "snapshots" of ionospheric features is the Assimilative Mapping of Ionospheric Electrodynamics (AMIE) procedure [Richmond and Kamide, 1988]. We have used the AMIE procedure to assimilate observations from 116 high-latitude northern hemisphere ground magnetometer stations to produce the convection and current maps in Figures 1, 2 and 3. Satellite particle precipitation data, as well as ground magnetometer data, were also used to produce distributions of ionospheric conductance (not shown) which link the current maps to the electric field maps. Although other data sources are available for the current and convection mappings, we have chosen in this study to use only ground magnetometer observations as the data input in order to eliminate artifacts in the mapping which sometimes appear when data sources, such as satellites, enter and exit the mapping region .

The AMIE procedure is an optimally constrained, weighted, least-squares fit of coefficients to observed data. Estimated Pedersen and Hall conductivities allow the ionospheric electric fields and currents to be related though Ohm's law. The first data fit within the procedure is one that estimates the height integrated Pedersen and Hall conductivities by modifying statistical models. Assuming those patterns and values have been reliably determined, the procedure goes on to fit ionospheric convection patterns from direct observations of ion drift observations made by satellites, radars, and digisondes and from indirect observations associated with the inversion of magnetic perturbations. These patterns from statistical data bases can be used as information to insure physical solutions to the fit in data sparse regions. The procedure also has an objective means of estimating the uncertainty in the electric field maps. These uncertainties are typically 50% or less [see for example Richmond and Kamide, 1988]. During intervals of good data coverage much of the field is estimated with an uncertainty of ~30%.

THE ASSIMILATIVE MAPPING OF IONOSPHERIC ELECTRODYNAMICS PROCEDURE

TECHNIQUE

The procedure determines internally consistent patterns of the electrostatic potential  the electric field E, the height integrated horizontal ionospheric electric current density I, the field-aligned current density J||, and the magnetic perturbations B. By performing a weighted, least squares fit the procedure insures the best possible pattern match to the available observations. Constraints are placed on the fit by a priori information, typically statistical patterns, that force the fit to a physical solution in areas of sparse data. Each of the quantities is expanded in a series of basis functions in magnetic latitude  and longitude  as

(,) = iai i(,)

E(,) = iaiEi(,)

I(,) = iaiIi(,)

J||(,) = iaiJ||i(,)

B(,) = iai Bi(h,,)

where h is height. For simplicity , E, I, and J|| are treated as being mapped to an ionospheric reference height of 110 km, but the height dependence of B must be kept explicitly, since it does not simply map along geomagnetic field lines.

We can currently use up to 244 electric potential (i) basis functions for the estimation (although fewer, typically, 79, can be used for a work-station version of AMIE output, as is the case for this paper). The functions are generated by combining generalized associated Legendre functions at high latitude with low latitude functions that satisfy a zero Laplacian requirement and are capable of representing any general pattern of electric potential poleward of 40° magnetic latitude. The 244 basis functions have effective wavelengths, down to about 6° in the north-south direction, signifying effective distance scales down to 6°/(2) or roughly 1°. The effective distance for fewer coefficients increase as the square root of the ratio of 244 to the number of new coefficients, so 79 coefficients results in effective distances scales of about 2° in latitude and 8° in longitude. In longitude the minimum wavelength is 30°, corresponding to a distance scale of about 5°. The other basis functions for the electric field Ei, the horizontal current density Ii, and the field-aligned current density J||i, are derived from i by the following relationships:

Ei=-i

Ii=Ei

J||i=Ii

where  is the ionospheric conductance tensor. A basis function i for the equivalent current function is obtained by solving:

2hi=rIi

where h2 is the component of the Laplacian differential operator involving only horizontal derivatives and r is the unit radial vector. This relation is valid under the simplifying assumption used in AMIE that all electric currents above the ionosphere are purely radial, thus ignoring the tilt of the magnetic field lines, and the existence of the ring current, the magnetopause currents and the magnetotail currents.

Relating the magnetic deviations to the three dimensional current patterns is conceptually and computationally simplified if the height integrated horizontal ionospheric current is considered as two parts: a toroidal current IT, which circulates without divergence in the ionosphere, and a poloidal current IP, which is the ionospheric horizontal closure of field-aligned currents generated in the distant magnetosphere. Both of these components make a contribution to the height integrated horizontal current I

I = (x)-=IT+IP

The ground-based magnetic variations can be associated with only the divergence-free (toroidal) component of the height integrated horizontal current, the so-called 'equivalent current.' The ionospheric horizontal equivalent sheet current, Iequiv, is related to the equivalent current function  by

This fictitious current would produce the same ground magnetic effects as the actual three-dimensional overhead current including the field-aligned currents. The basis functions for the equivalent current are used to calculate those for the ground-level magnetic potential Vi, by assuming the sheet current to be at an altitude of 110 km. Such an overhead current will induce currents in the earth which in turn give rise to ground magnetic effects. AMIE uses the approximation of a perfectly conducting layer at a depth of 250 km, overlain by a perfect insulator to estimate the surface magnetic effects of the induced currents. The basis functions Bi for the magnetic perturbations on the ground are then calculated from


The procedure incorporates both horizontal and vertical components of magnetic perturbations as data input. The most direct use of the vertical information is in determining the locations and intensity of the auroral electrojets, although the procedure gains collateral information about field-aligned currents and electric field distributions from this data.

The effects of this three dimensional (poloidal) current system on a satellite magnetometer are calculated in AMIE by assuming that the poloidal contribution to height-integrated horizontal current comes from the closing of the field-aligned currents across the ionosphere. The local divergence of the height-integrated current arises from the entry and exit of the vertical field-aligned currents. Since these closing currents are curl-free they can be expressed in terms of a current function, . We express the field-aligned current density as


Information about the field-aligned currents is derived from the divergence of the height integrated current and from the measurements of current flow above the ionosphere (i.e. satellite magnetic perturbations). We assume that these currents produce no magnetic effect at the ground. Satellite magnetic perturbations measured above the ionosphere can be expressed as:


PRIOR INFORMATION

In an ideal situation the observations would provide most if not all of the information about the coefficients. In some situations or locations, however, observations may be very limited, unreliable, or simply unavailable. In data sparse regions the AMIE technique relies partly on statistical information to reconstruct reasonable electrodynamic patterns. In this manner we bypass the typical non-uniqueness problems usually associated with data poor regions. We accept a priori information about both the electric potential distribution and the conductance distribution. A 10-level Hemispheric Power Index (HPI) computed from NOAA and/or DMSP satellite particle precipitation data, can be used to parameterize the electric potential [Foster et al., 1986] and conductance [Fuller-Rowell and Evans, 1987] distributions. These precipitation data are interpolated in time to provide HPI values throughout the period of interest. Statistical patterns of electric potential have been derived by Foster et al. [1986] who used seven years of Millstone Hill radar plasma drift measurements and binned the measurements according to the HPI of auroral precipitating particles, the latitude and magnetic local time, and then derived statistical models of the electric potential from the binned data values. The electric potential distribution can also be parameterized by IMF orientation [Heppner and Maynard, 1987] if solar wind data are available.

UNCERTAINTIES

The method of determining the uncertainty in the estimated fields is discussed by Richmond and Kamide [1988]. This uncertainty arises both from the total effective errors in the observations (which includes the effects of errors in the assumptions used by the AMIE procedure, as well as, measurement error) and from the incomplete coverage of the observations. Data-poor regions like the Arctic Ocean, for example, have large uncertainties in the mapped fields. The weight for any given datum is the inverse square of its expected error. The error includes not only measurement errors, but also errors due to finite truncation of the series of coefficients, errors in the assumptions employed in the AMIE procedure and errors introduced by accepting data from a window of time about the period of interest. In general the observations windows are plus or minus 15 minutes for satellite data, and approximately 5 minutes for ground-based data. We expect greater errors to be associated with longer temporal mismatches. We impose exponentially decreasing weights on observations temporally distant from the estimation time. In addition, we now compare the mean square differences between computed fields and observations with what the error analysis of AMIE predicts for these differences, and when necessary we adjust the effective data errors to improve the agreement in this comparison. This adjustment increases the objectivity of the data error specifications, which are a necessary input to the AMIE procedure.

CONDUCTANCE

The ability of the AMIE procedure to accurately relate the currents to the electric field via Ohm's Law requires knowledge of the ionospheric conductance distribution. The AMIE technique can incorporate direct and indirect information about the height integrated conductivities. The procedure has an explicit estimation of the distributions of auroral energy flux and mean particle energy, based on the direct observations from satellites and indirect information from ground magnetometers. The energy flux and mean energy of auroral particles are closely related to the height-integrated ionospheric Pedersen and Hall conductivities, which are estimated in the first stage of the AMIE procedure. Our initial estimates of the auroral energy flux are made by the statistical auroral conductivity model of Fuller-Rowell and Evans [1987], parameterized by the 10-level Hemispheric Power Index (HPI). Energy fluxes and average energies of the auroral precipitating electrons measured by DMSP, and most recently UARS [Lu et al, 1996], are used to calculate Hall and Pedersen conductance modifiers. The conversion from flux to conductance for DMSP data are based on empirical formulae from Robinson et al. [1987] and a UV conductance model is added. The electron transport code of Fuller-Rowell and Evans [1987] estimates conductivities from the NOAA satellites which are also used as modifiers for the statistical patterns. Simultaneous observations from both hemispheres are often incorporated into the conductance fitting routine, assuming the auroral precipitation is approximately conjugate. But such observations are typically given only half the weight of observations from the hemisphere of interest. More indirect information about the conductance distribution can also be obtained from empirical relations of conductance and ground magnetic perturbations [Ahn et al., 1983]. Additional discussion of the technique can be found in Richmond [1992] and Richmond et al. [1996, submitted].

MAPPING CURRENT AND CONVECTION VARIATIONS ASSOCIATED WITH SUBSTORMS

In the past much of the emphasis in substorm research has focused on the AE index and its components AU and AL. With the maturation of global mapping schemes and the advent of new auroral observing techniques and instruments, it will be possible to be far more quantitative about the spatial and temporal development of substorms. As we show below, it is possible to locate the peak current, and the latitude of substorm onset. As suggested by Cooper et al. [1995], it is also possible to use gridded AMIE output to recalculate the AE index during intense geomagnetic activity when the electrojets have moved beyond the locations of the AE observatories. This further suggests that it will be possible to estimate, from the gridded output, the size of the mid-latitude bays associated with substorms and to compare these estimates with individual and observatory chain estimates. By using the differencing scheme, that we illustrate below, it is possible to study the evolution of the convection current enhancements and the development of the surge or wedge current. It is also possible to estimate the energy dissipated in Joule and particle heating. Finally, when auroral images are available it will be possible to estimate the intensity area moment of the auroral bulge and trace the bulge development along with the development of the current systems.

EXAMPLE 1. MODERATE AND STRONG GEOMAGNETIC ACTIVITY

We illustrate an event which begins with a near classic substorm generated by a southward turning of the IMF. However, shortly after the recovery phase began a large density enhancement in the solar wind caused considerable compression of the magnetosphere. A new substorm developed almost immediately upon arrival of the density enhancement. We provide a snapshots of both disturbances in Figure 1.

    Fig. 1. AMIE maps viewed from the north geomagnetic pole, which is at the center of each plot. The outer ring is 50° magnetic. The time and IMF values are in the upper left corner of each plot. Column 1 contains horizontal current vectors with vector length representing 0.5 A as indicated in the lower right corner. Column 2 contains electric potential contours at 10 kV intervals, the potential difference between the maximum value and minimum value is printed in the upper right corner. column 3 contains the field-aligned currents. The contour interval is 0.3 µA/m2, beginning at ± 0.15 µA/m2. The downward currents have solid contours. The hemispherically integrated value of the downward field-aligned current is shown in the upper right corner of the plot.

Each row of Figure 1 contains a horizontal current, electric potential and field-aligned current map for the times designated in the upper left corner of each map. Also in the upper left corner of each map are the interplanetary field components observed by the GEOTAIL spacecraft which was 200 Re downtail of the earth (the observations have been "backed" to time of earth passage). Figure 2 shows 'difference plots', which are made by subtracting the present 5-minute snapshot from that of the previous 5-minute interval. We use these difference maps to illustrate the changes occurring in convection and currents on the 5-minute time-scale.

  Fig. 2. Difference plots for column 1) horizontal currents, column2) electric potential and column 3) field-aligned current. The differences are created by subtracting the gridded values for the previous 5-minute interval from those at the time shown in the upper left of the map. Note that for clarity the current difference vectors have been rescaled to 0.25 A/m as indicated in the lower right corner. The potential contours are 5 kV and the field-aligned current contours are 0.15 µA/m2, starting at 0.075 µA/m2.


Figures 1a-c show the ionospheric state of horizontal currents, electric potential and field aligned currents, respectively at 2155 UT on 3 November. The maps are indicative of quiescent ionospheric conditions which had prevailed for approximately one hour. The cross polar cap voltage in Figure 1b was 27 kV and the area-integrated downward field-aligned current (Figure 1c) was 1.0 MA. Within the subsequent 5 minute interval the IMF turned southward. Figures 1d-f show an immediate dayside response in terms of both enhanced currents and convection. The cross polar cap voltage rose by 19 kV and the field aligned current strength increased by 30% in the 5-minute interval. The elements of Figure 2, row 1 shows the 5-minute ìdifference mapsî produced by numerically subtracting the maps of Figure 1, row 1 from Figure 1, row 2. Consistent with the ideas presented in Lockwood et al. [1990], these maps indicate enhancements of the dayside horizontal and field aligned current systems and increasing electric field in the region whose field lines should map to dayside reconnection sites. Between 2155 UT and 2200 UT all of the new potential was added on the dayside. The nightside region, not yet affected by the addition of newly reconnected flux, remained unchanged.

Figures 1g-i show the ionospheric state some 30 minutes later. The convection system had extended nightward and equatorward as new flux entered the polar cap in response to the increasingly southward IMF. The cross polar cap voltage rose to 65 kV. An intense new post-midnight field-aligned current system (Fig. 1i) and strengthening horizontal currents (Figure 1g) developed in the post-midnight region as the first substorm occurred. The more tailward position of the electric potential maximum at 2235 UT (Figure 2h), as well as the intense midnight field-aligned current system are in agreement with the ideas of substorm expansion phase presented by Kamide et al. [1994].

In order to more clearly delineate the changes associated with substorm expansion, we show 5-minute difference plots (2235 UT minus 2230 UT) of the horizontal currents in Figure 2, row 2. The growing convection currents are evident near the dawn-dusk line in Figure 2d. Even more evident is the substorm current surge across the midnight meridian covering most of the latitude interval between 60 and 70 magnetic. Figure 2e indicates that nightside electric fields were increasing rapidly between 2230 and 2235 UT, while the dayside had reached an equilibrium state. Figure 2f shows not only the enhancement of the global field aligned current system, but the intensification of the upward and downward currents in the vicinity of the midnight current wedge.

The last row of Figure 1 shows the ionospheric response at 2320 UT to the leading edge of a non-compressive density enhancement, a type of solar wind transient. (N. Crooker, personal communication, 1996). The passage began at 2310 UT on 3 November and lasted for approximately 45 minutes. During the 45 minute interval the IMP-8 satellite, which was ~ 30 Re downtail, recorded Bz component values as large as -35 nT and solar wind densities in excess of 50 amu cm-3. The GEOTAIL satellite was enveloped by the magnetosheath during the transient passage and did not make IMF observations. Figure 2 row 3 shows the 5-minute difference plots and illustrates the dayside response to the dayside magnetospheric compression and the onset of a second substorm.

At 2320 UT, just ten minutes into the passage, the cross-polar-cap-voltage had risen to 81 kV and intense currents flowed both horizontally and vertically. The 5 minute potential difference plots (2320 UT minus 2315 UT) shown in Figure 2 row 3, indicates continuing growth of the dayside convection with nearly 15 kV of new potential in the dayside negative cell and 10 kV of new potential in the dayside positive cell (Fig 2h). The current difference plot (2220 UT minus 2315 UT) shows clearly defined and separate regions of current enhancement. Dayside current enhancements were the most extreme, perhaps owing to the extraordinary compression associated with the high densities. (The magnetopause crossed geosynchronous orbit at approximately 2325 UT, J. Borovsky, personal communication, 1996). A simultaneous large current enhancement occurred across the midnight meridian. Figure 1j shows horizontal currents in excess of 1 A/m occurred across much of the midnight region while the solar wind density enhancement and strong southward field influenced the ionospheric behavior. Consistent with a new expansion phase at 2320 UT, geosynchronous satellites indicated a strong particle injection event (G. Reeves, personal communication, 1996). During the 60 minute interval after 2320 UT the entire current system was so excited that it was not possible to separate the convection current system from the substorm system.

EXAMPLE 2. A MORE "NORMAL" SUBSTORM

The substorm that we show in Figure 3 is more illustrative of those mapped during typical solar wind conditions. This event took place on 10 November 1993. The IMF was slowly varying in orientation and solar wind flow speed was near 500 km/s.


  Fig. 3. Same as for Figure 1 except that row 4 contains difference plots as described in Figure 2 with the electric potential contours at 3 kV.

Figures 3a-c show a well-developed convection and current system representative of the conditions for the previous hour. The IMF had been steadily southward with a magnitude of ~3 nT. Between 1225 UT (not shown) and 1230 UT, as the expansion phase began, a new local nightside minimum in the evening convection cell developed and the morning convection cell migrated further across the midnight meridian into the pre-midnight region. The field-aligned current flow in the pre-midnight region also strengthened. Despite this rather substantial reconfiguration, the total cross polar cap voltage was nearly steady. In the plots for 1235 UT (Fig. 3g-i) these features further intensified and the horizontal current flow across the midnight meridian increased. All channels of the low energy particle detectors on the midnight geosynchronous satellite registered a slight increase in particle flux at 1235 UT, indicative of a small particle injection.

In the bottom row of Figure 3 we show the 'difference plots' for 1235 UT minus 1230 UT There is evidence of a horizontal current surge in the midnight region (Figure 3j) , and enhancement of the nightside electric field (Figure 3k) and enhancements of the near midnight field-aligned current system (Figure 3l). By 1245 UT (not shown) these features were absent.

SUMMARY

We have shown three different substorm conditions, each illustrative of a different type of solar wind and forcing. In the order of presentation, these substorm examples generally typify moderately, greatly, and slightly disturbed geomagnetic conditions. In Figures 1 and 2 we provided hemispheric snapshots and difference plots that illustrate the electrodynamics associated with the addition of open flux to the polar cap and the onset of substorm expansion phase. The difference plots in Figures 2 and 3 provide a method by which we can identify substorm onset and track the progress of substorm development. For these events we illustrated here auroral imagery was not available as a means of identifying substorms, however, the substorm expansion phases were verified with geosynchronous particle injection data. We also provided a snapshot of the combined response to a magnetospheric compression and a simultaneous substorm expansion. The difference map for that event clearly illustrated the separate dayside and nightside response. A more detailed treatment of the compression and ensuing storm will be the subject of a future publication.

We anticipate that a more complete view of substorm dynamics will be possible with the combination of data assimilation procedures and global auroral coverage like that soon to be available from the POLAR satellite. In short order we expect to be able to provide snapshots of the spatial and temporal development of the peak and total substorm current and the total energy dissipated by individual substorms. We will also be able to determine the latitude and longitude of substorm onset, and the size of the mid-latitude bay. When auroral imagery are available for assimilation we anticipate making estimates of the temporal and area integral of the auroral bulge associated with a given substorm. Further we anticipate making useful estimates of the temporal and spatial variations of the field-aligned current contributions to substorm dynamics.

ACKNOWLEDGMENTS

Data for this study were made available by M. Nakamura, T. Mukai, S. Kokubun, T. Terasawa, A. Lazarus, F. Rich, D. Evans, L. Cafarella, B. Clauer, E. Friis-Christensen, A. Green, J. Hughes, M. Engebretson, L. H. Luehr, T. Moretto, L. Morris, J. Olson, D. Orr and D. Milling, T. Rosenberg G. Rostoker and T. Hughes, P. Stauning, C. Szuberla, O. Troshichev, W. Worthington, K. Yumoto and K. Shiokawa, R. Lepping, R. Coles and D. Boteler. This work is partially supported by NSF Grant ATM 93-02144 and a grant from the Air Force Office of Scientific Research.

REFERENCES

Ahn, B.-H., S.-I. Akasofu, and Y Kamide, The Joule heat production rate and the particle energy injection rate as a function of the geomagnetic indices AE and AL, J. Geophys. Res., 88, 6275, 1983.

Cooper, M. L., C. R. Clauer, B. A. Emery, A. D. Richmond and J. D. Winningham, A storm time assimilative mapping of ionospheric electrodynamics analysis for the severe geomagnetic storm of November 8-9, 1991, J. Geophys. Res., 100, 19329, 1995.

Foster, J., J. M. Holt, R. G. Musgrove and D. S. Evans, Ionospheric convection associated with discrete levels of particle precipitation,, Geophys. Res. Lett., 13, 656, 1986.

Fuller-Rowell, T. J. and D. S. Evans, height-integrated Pedersen and Hall conductivity patterns inferred from the TIROS-NOAA satellite data, J Geophys. Res., 92, 706, 1987.

Heppner, J. P., and N. C. Maynard, Empirical high-latitude electric field models, J Geophys. Res., 92, 4467, 1987.

Kamide, Y., A. D. Richmond, B. A. Emery, C. F. Hutchison, B.-H. Ahn, O. de la Beaujardiere, J. C. Foster, R. A. Heelis, H. W. Kroehl, F. J. Rich and J. A. Slavin, Ground-based studies of ionospheric convection associated with substorm expansion, J Geophys. Res., 99, 19451, 1994.

Lockwood, M., S. W. H. Cowley and M. P. Freeman, The excitation of plasma convection in the high-latitude ionosphere, J. Geophys. Res., 95, 7961, 1990.

Lu, G., B. A. Emery, A. Rodger, M. Lester, J. Taylor, D. S. Evans, M. Ruohoniemi, W. F. Denig, O, de la Beaujardiere, R. A. Frahm, J. D. Winningham and D. Chenette, High-latitude ionospheric electrodynamics as determined by the Assimilative Mapping of Ionospheric Electrodynamics procedure for the ATLAS-1/GEM Period of March 28-29, 1992, J. Geophys. Res., in press, 1996.

Richmond, A. D., and Y. Kamide, Mapping electrodynamic features of the high-latitude ionosphere from localized observations: Technique, J. Geophys. Res., 93, 5741, 1988.

Richmond, A. D. Assimilative Mapping of Ionospheric Electrodynamics, Adv. Space Res., (6)59, 1992.

Richmond, A. D., G. Lu, B. A. Emery and D. J. Knipp, The AMIE Procedure: Prospects for Space Weather Specification and Prediction, Adv. in Space Res., submitted, 1996.

Robinson, R. M., R. R. Vondrak, K. Miller, T. Dabbs, and D. Hardy, On calculating ionospheric conductances from the flux and energy of precipitating electrons, J. Geophys. Res., 28, 2565, 1987.

Siscoe G. And G. Lu, Dayside Birkeland currents during substorms: an AMIE test of a substorm model, J. Geophys. Res., submitted, 1996.

Taylor, J. R., T. K. Yoeman, M. Lester, B. A. Emery and D. J. Knipp, Variations in the polar cap area during intervals of substorm activity on March 20-21 1990 deduced from AMIE convection patterns, in press Ann. Geophysicae, 1996.