Stormtime dynamics of the global thermosphere and equatorial ionosphere

During magnetic storms the development of equatorial plasma bubbles (EPBs) and distributions of thermospheric densities are strongly influenced by the histories of imposed magnetospheric electric ( εM) fields. Periods of intense EPB activity driven by penetration εM fields in the main phase are followed by their worldwide absence during recovery. A new method is applied to estimate global thermospheric energy ( Eth) budgets from orbit-averaged densities measured by accelerometers on polar-orbiting satellites. During the main phase of storms Eth increases as long as the stormtimeεM operates, then exponentially decays toward quiet-time values during early recovery. Some fraction of the energy deposited at high magnetic latitudes during the main phase propagates into the subauroral ionospherethermosphere where it affects chemical and azimuthal-wind dynamics well into recovery. We suggest a scenario wherein fossils of main phase activity inhibit full restoration of quiettime dayside dynamos and pre-reversal enhancements of upward plasma drifts near dusk denying bottomside irregularities sufficient time to grow into EPBs.


Introduction
Recent growth in understanding the electromechanical forces that couple the magnetosphere-ionosphere-thermosphere system has practical consequences for calculating trajectories of objects in low Earth orbit (LEO) and improving transionospheric communications.Here we examine physical processes that commonly occur during magnetic storms Correspondence to: W. J. Burke (afrl.rvb.pa@hanscom.af.mil) with consistent effects on these disparate space applications.Stormtime dynamics of the thermosphere are effects of complex external and internal forcing.In most instances measurements of critical parameters are absent.Thus, we must construct synthetic pictures from piecemeal measurements acquired over the course of many magnetic storms.Our present goal is to illustrate that understanding in one area affects our grasp of the other and points to further questions suggesting still deeper unity.
The paper has three main parts.The first describes stormtime properties of equatorial plasma bubbles (EPBs) as observed by satellite-borne sensors.Intense EPB activity in the main phase is followed by prolonged absences during recovery.The second introduces a new method for using measurements from accelerometers on polar-orbiting satellites to predict the evolution of stormtime thermospheric energy budgets E th .On a globally-averaged scale the thermosphere returns to baseline values ∼10 h after stormtime electric fields abate.The third section investigates local thermosphere-ionosphere coupling mechanisms responsible for prolonged absences of EPBs at topside altitudes during the recovery phases of magnetic storms.Specifically, we consider how changed chemical composition, plasma density and neutral wind distributions may continue to affect the redevelopment of pre-reversal enhancements (PREs) of vertical plasma drifts near the dusk terminator for several days into recovery (Fejer et al., 2008).

Stormtime equatorial plasma bubbles
EPBs are plasma density depletions found in the topside ionosphere at low magnetic latitudes after sunset.They originate as small-amplitude irregularities in the bottomside of the equatorial F-layer and grow via the generalized Rayleigh-Taylor (R-T) instability (Scannapieco and Ossakow, 1976).Electric fields with eastward components accelerate the evolution of initial density perturbations into full grown bubbles that penetrate the F-layer peak and rise through the topside ionosphere (Ott, 1978).Three sources of eastward electric fields have been identified: (1) quiettime dynamos responsible for PREs of upward plasma drifts near sunset, (2) penetration of polar electric fields to low latitudes, and (3) wind fields associated with atmospheric gravity waves.Solar heating drives eastward thermospheric winds in the afternoon sector that support the buildup of polarization charges near the dusk meridian responsible for the PRE (Farley et al., 1986;Haerendel and Eccles, 1992) observed by the Jicamarca incoherent scatter radar (Scherliess and Fejer, 1997) and sensors on the ROCSAT satellite (Fejer et al., 2008).The relative contributions of Sq current leakage into the nightside ionosphere and F-layer azimuthal winds remain a subject of debate (Eccles, 1998).Sustained penetration electric fields (ε M ) have been measured in the inner magnetosphere during the main phase of magnetic storms (Wygant et al., 1998;Burke et al., 1998) and inferred from upward plasma drifts observed at Jicamarca after southward turnings of the interplanetary magnetic field (Kelley et al., 2003).Gravity-wave activity is also a plausible source of EPBs.Singh et al. (1997) provided observational evidence from the Atmospheric Explorer E satellite suggesting that long wavelength gravity waves seed bottomside density fluctuations that grow into EPBs.Alternately, Hysell and Kudeki (2004) argued that the growth rate of bottomside density variations is significantly enhanced by collisional Kelvin-Helmholz instabilities driven by bottomside azimuthal winds that change from westward to eastward with increasing altitude.
There are two stormtime energy sources recognized as being capable of generating electric fields with westward components at low latitudes in the dayside and evening sectors that act to suppress EPB growth.These are (1) over-shielding electric fields driven by enhanced Region 2 field-aligned currents (Kikuchi et al., 2008) and (2) disturbance dynamos generated by winds that result from Poynting and energetic particle fluxes into the high-latitude atmosphere (Blanc and Richmond, 1980;Scherliess and Fejer, 1997;Fejer andScherliess, 1995, 1997).
Figure 1 is a plot of the D st index during an 11-day interval that included the magnetic superstorm of March 1991 when D st approached −300 nT.Vertical lines mark EPB detections by the Defense Meteorological Satellite Program (DMSP) Flight 9 spacecraft at ∼840 km near 21:00 LT (Huang et al., 2001).Attention is directed to three empirical aspects of EPB detections: 1. Regular patterns of pre-storm EPB detections exemplify the season-longitude distributions found near solar maximum intervals (Gentile et al., 2006).
2. Intense EPB activity was detected during the main phase while the D st trace had a negative slope.At the same time the double probe experiment on the Combined Release and Radiation Effects Satellite (CRRES) detected strong and continuous ε M in the equatorial plane of the inner magnetosphere (cf.Fig. 2 of Wygant et al., 1998).
3. Except for two isolated detections, EPBs were absent for 4 days before the quiet-time pattern returned.These recovery-phase exceptions occurred during brief reactivations of the ring current signified by the reappearance of ε M (cf.Fig. 3 of Wygant et al., 1998).
The prolonged absence of EPBs throughout the 4-day recovery phase is problematic.Evidence presented in Fig. 7 of Kikuchi et al. (2008) indicates that westward, over-shielding electric fields occur but only last for a few hours.On the other hand, Scherliess and Fejer (1997) estimated that disturbance dynamo action should only be effective for ∼30 h after stormtime driving ends.Clearly, better understanding of recovery-phase mechanisms and time scales is needed.We return to this point in the discussion below.Similar EPB patterns appear during all major (D st <−100 nT) magnetic storms of the past two solar cycles for which DMSP data are available (Huang et al., 2002;Gentile et al., 2006).A comparison of ROCSAT and DMSP observations showed that EPBs reach topside altitudes after 20:00 LT during quiet times but at ∼19:30 LT during magnetic disturbances (Burke et al., 2004).Repeated detections of intense EPBs by multiple DMSP satellites throughout the main phases of storms confirm the persistence of penetration electric fields.A lack of EPBs in DMSP plasma density data does not imply that radiowave scintillations should also vanish during recovery.They persist.Rather, driving eastward electric fields in the post-sunset ionosphere lack the strength and/or duration needed for irregularities to grow into EPBs, before the field polarity reverses to stabilize bottomside plasma.

Stormtime thermospheric energy budgets
The US Space Surveillance Network (SSN) is responsible for tracking and predicting the trajectories of active satellites Ann.Geophys., 27,[2035][2036][2037][2038][2039][2040][2041][2042][2043][2044]2009 www.ann-geophys.net/27/2035/2009/and space debris.The destruction of Feng Yun 1C by an antisatellite weapon in January 2007 created more than 2500 new debris fragments in LEO (B.R. Bowman, private communication, 2008).The potential for Feng Yun debris and pre-existing objects to inflict severe to catastrophic damage on spacecraft operating in LEO has refocused attention on the need to enhance collision-avoidance capabilities (Wright, 2007).This in turn requires improved predictions/specifications of the atmosphere though which space objects fly.The variability of atmospheric drag forces exerted during severe magnetic storms poses a significant challenge to meeting this need.Burke et al. (2007) showed that during the magnetic storms of November 2004 models used by SSN to predict stormtime thermospheric densities (ρ) underestimate those inferred from accelerometers on the polar-orbiting Gravity Recovery and Climate Experiment (GRACE) satellites by more than 100%.GRACE's orbital plane was in the noonmidnight meridian at an altitude of ∼487 km. Figure 2a compares variations of local ρ (black) and orbit-averaged ρ (blue) mass densities with −D st (red).Figure 2b compares ρ with magnetospheric electric fields ε V S ≈ PC /2L Y R E estimated from parameters observed by the Advanced Composition Explorer (ACE) satellite near the L 1 point (Burke, 2007).The symbols PC and 2L Y R E represent the polar cap potential and magnetospheric width, respectively.Quasisinusoidal changes in pre-disturbance ρ (Fig. 2a) reflect daynight and apogee-perigee orbital variations.Crossing the polar cap during the storm GRACE experienced large positive and negative "ρ" spikes, as the spacecraft encountered head and tail winds driven by anti-sunward convecting ions.We note four features of the data: (1) while stormtime ρ varied widely, ρ evolved systematically as magnetic activity progressed, (2) during the main and early recovery phases D st and ρ were well correlated, (3) ε V S anticipated ρ variations with lead times of 3 to 5 h, and (4) ρ relaxed toward prestorm status at a much faster rate than D st .
Even allowing for "local turbulence" evident in stormtime traces of ρ, the first result suggested that on a global scale the thermosphere evolves like a large thermodynamic system that never strays far from diffusive equilibrium (Wilson et al., 2006).To the degree that this assessment is accurate the stormtime evolution of the global thermosphere can be understood in terms of the first law of thermodynamics: dE th =C V dT +dW g , where dE th , and dT signify changes in the thermosphere's total energy and temperature, dW g represents work done against gravity and C V is the local heat capacity of thermospheric gases.As a matter of convenience in providing quantitative estimates we turned to the J77 atmospheric model (Jacchia, 1977).
J77 consists of a set of stable density and temperature profiles, each uniquely specified by the exospheric temperature T ∞ .The modeled atmosphere consists of a fixed number of diatomic (N 2 , O 2 ) and monatomic (O, Ar, He, H) species (σ ). and maintain approximately the ground fractional densities.A minimum temperature of 188 K is assigned at h=90 km.Above 90 km all species are in diffusive equilibria specified by profiles T (r) that asymptotically approach T ∞ (cf.Appendix A).For calculations we define: number densities where r=R E +h is distance from the Earth's center; k B and A are Boltzmann's constant and Avogadro's number, respectively.The approximation sign in Eq. ( 1) recognizes the absence of chemical reactions in the J77 model.Figure 3a shows selected J77 density profiles appropriate for five values of T ∞ between 800 and 2000 K.These profiles indicate that at any given altitude a unique relationship exists between ρ and T ∞ .Since ρ and the orbit-averaged altitude ( h) of GRACE are known it is possible to determine T ∞ .Leastsquare testing revealed relationships between T ∞ and ρ(h) found in J77 tables that are well represented in the form of a quadratic polynomial In the 300≤ h≤500 km altitude range sampled by GRACE and CHAMP, the coefficients a i ( h) are described by 5th order polynomials of the form Regression coefficients obtained in fitting T ∞ and a i to polynomials exceed 0.999.In applying Eqs. ( 2) and (3) ρ and h are in grams/cc and kilometers, respectively.Figure 3b shows an example of T ∞ isotherm distributions for the J77 model plotted on a latitude-versus-local time (LT) grid.Note that T ∞ min and T ∞ max develop in the postmidnight and afternoon LT sectors.Their ratio is ∼1.3 (Jacchia, 1977).Thus, polar-orbiting satellites should sample about the same value of T∞ ≈1.15 T ∞ min independent of the orbital plane's LT.We have compared values of T∞ inferred via Eqs.( 2) and (3) from simultaneous ρ measurements of CHAMP and GRACE during the summer months of 2004 (not shown).Consistent with our conjecture, inferred values of T∞ agreed within 1%.During this period the two spacecraft were separated by ∼90 • in LT and >100 km in altitude.
The total thermospheric energy density is the sum of the thermal η T (r)=C V (r)n(r)T (r)/A, and potential energy contributions φ G (r)=ρ(r)M E G/r.The symbols M E and G represent the Earth's mass and the gravitational constant, respectively.To estimate the thermosphere's total energy content E th =H T +F G , it is necessary to integrate η T (r) and φ G (r) over the volume of the thermosphere.In the present calculations we ignore local time effects and limit the range of integration from a selected altitude h 0 ≥100 km to 1000 km.Our use of orbit-averaged densities has already filtered latitudinal variations within the integrands.The total thermal energy within this altitude range is approximately Since we are only interested in changes in potential energy it is useful to set the potential energy of the thermosphere to zero at the base of the integration range (Wilson et al., 2006) and represent the total potential energy of thermospheric neutrals as With information available in J77 tables it is possible to calculate the integrands of Eqs. ( 4) and ( 5) for 700≤T ∞ ≤2000 K at increments of 100  of T ∞ .The lower limit of integration h 0 =100 km includes E-region contributions to the total thermospheric budget.Figure 4 displays E th in Joules for h≥100 km obtained by numerical integrations of Eqs. ( 4) and ( 5) plotted as a function of T∞ .A linear regression analysis of these "data" shows that E th (h ≥ 100 km) = 5.365 × 10 17 + 8.727 × 10 13 T∞ (6) with a regression coefficient R>0.998.A change of 100 • in T∞ corresponds to a thermospheric energy gain/loss of ∼8.7×10 15 J.Over the range 700≤T ∞ ≤2000 K, E th only varies by ∼15%.
Figure 5a compares the geophysical parameters D st (top) and ε V S (second) with orbit-averaged measurements of ρ (third), T∞ (fourth), and E th (bottom) acquired by GRACE between Julian days (JDs) 150 (1 June) and 230 (17 August) 2004.During this interval the orbital plane precessed from near the noon-midnight to the dawn-dusk meridian; h was ∼486.5 km.T∞ and E th traces were obtained by applying Eqs. ( 2) and ( 6) to the ρ(t) stream and ranged between ∼750 and 1200 K and E th between 6.05 and 6.4×10 17 J, respectively.The plots show ρ(t), T∞ and E th varying on two time scales: (1) low-frequency variations with the 27day solar rotation period, and ( 2 that correlate with intensifications of D st and ε V S .Structures observed between JDs 204 and 211 are discussed below.However, we note that the largest change in the E th trace of ∼3×10 16 J occurred on JD 209 in response to interplanetary forcing when D st attained a minimum value of about −200 nT.The Dessler-Parker-Sckopke equation as formulated by Stern (2005) indicates that energy of the ring current E RC increased by ∼5.16×10 15 J.This is substantially less than the E th needed to account for observed ρ variations.Thus, while the ring current contributes to the energy of the upper atmosphere through particle precipitation and Poynting flux associated with Region 2 field-aligned currents, it cannot be the main reservoir from which the stormtime thermosphere draws energy.Rather, some large fraction of the thermosphere's energy budget must come from the solar wind unmediated by the ring current.Figure 5b compares the same geophysical and GRACE parameters measured during JDs 305-330 (2-27 November) 2004.The orbital plane of GRACE was again near the noon-midnight meridian.The D st plot shows that two magnetic disturbances occurred between 7 and 10 November.Similar disturbances appear on all of the other plots.D st minima were −373 nT (06:00 UT, 8 November) and −289 nT (10:00 UT, 10 November).Corresponding E RC values are 9.62 and 7.46×10 15 J.The pre-disturbance baselines for T∞ and E th were near 900 K and 6.15×10 17 J, respectively.T∞ rose to 1390 K at 07:00 UT on JD 313 and 1350 K at 10:00 UT on JD 315.E th maxima at these times were 6.58 and 6.52×10 17 J.Again, corresponding E th values of 4.3 and 3.7×10 16 J greatly exceed estimates of E RC .
In surveying several years of stormtime ρ(t), T∞ and E th inferred from measurements by accelerometers on the CHAMP and GRACE satellites it became apparent that outcrops above solar UV-induced backgrounds coincide with increases in ε V S .Moreover, as ε V S →0, E th always relaxes at the same rate.To quantify the relaxation rate (1/τ E ) of E th , Fig. 6 shows traces of ε V S and the natural logarithm of E th plotted as functions of time through the disturbed interval on JDs [204][205][206][207][208][209][210][211]2004.Vertical lines highlight times when ε V S decreased to low values.Slanted lines with the same slopes mark thermospheric responses to the removal of ε V S driving.Numerically, the e-fold relaxation time of E th is τ E ≈6.5 h.
Within the context of J77, application of the first law of thermodynamics to accelerometer data streams allows us to infer the values of T∞ and E th needed to explain observed levels of ρ(t).However, this "kinematic" exercise tells us little about the global dynamics responsible for the thermosphere's passing from one energy state to the next.Correlations between ε V S and E th found in Fig. 5 suggest that E th ≈E U V +E SW is driven by two independent but additive sources the solar ultraviolet (UV) radiance (Bowman et al., 2006) and the solar wind.
The variability of E th SW during magnetic disturbances correlates with changes in ε V S .Rather than seek statistical relationships between the two quantities, we propose to exploit implications of information contained in the decay characteristics seen in Fig. 6.Specifically, we postulate that on a global scale the stormtime thermosphere acts like a drivendissipative system.To the degree that this conjecture is correct, E th SW should evolve as The coupling constant α E and relaxation time τ E =6.5 h must be determined empirically.Numerical solutions of Eq. ( 7) can then be compared with GRACE measurements to confirm or refute the driven-dissipative assumption.Numerical solutions take the form where t represents the time step between samples.For convenience we set t=1 h, and express ε V S in mV/m.In this case Through trial-and-error comparisons of solutions to Eq. ( 8b) with GRACE measurements during JDs 150-230, 2004 we found that α E ≈5.5×10 15 [(J/h)/(mV/m)].
Figure 7 contains plots of ε V S (black), solutions of Eq. (8b) (blue) and GRACE measurements of E th SW (red).Top and bottom panels represent the disturbed intervals in July and November 2004, respectively.In both cases we approximated E th SW as E th −6.1×10 17 J.Since the E th SW calculations in Fig. 7 are presented in units of 10 16 J, α E ≈0.55.
Although α E was determined using data from the summer of 2004, the predictions of Eq. (8b) also appear to be in good agreement with GRACE data during the November storms.
Finally, we note that Bowman et al. (2008) applied a modified version of this approach to a large number of magnetic storms for which both CHAMP and GRACE accelerometer data are available.They were able to replicate the main features of measured stormtime density variations.Weimer (2005) combined statistically determined models of high-latitude electrostatic and magnetic Euler potential responses to interplanetary changes to estimate Poynting fluxes into the global ionosphere and thermosphere.Hereafter we refer to this model as W5.The model uses solar wind/IMF measurements to predict distributions of Poynting flux into the northern and southern ionosphere.W5 requires no prior knowledge about ionospheric conductance.Integration over the affected areas provides the global rates of energy input P W5 .
On the other hand, E th values inferred from GRACE accelerometer measurements are snapshots that capture the thermosphere's total energy content averaged over orbital periods.GRACE data reflect compromises that the thermosphere strikes between power received from the solar wind and lost via radiative and heat conduction processes (Mlynczak et al., 2005).Equation (8b) provides a bridge between the two approaches.The term α E ε VS represents the rate of energy input needed to drive the stormtime thermosphere and should be directly comparable with P W5 .Since α E ≈5.5×10 15 [(J/h)/(mV/m)] and P W5 is in Watts we must divide by 3600 s/h to obtain α E ≈1.528×10 12 W/(mV/m) or 1.528 TW/(mV/m).Figure 8 shows plots of P W5 (red) at a 5-min cadence and hourly-averaged α E ε VS in TW for the storms of July and November 2004.Both the α E ε VS and P W5 traces were time shifted by 1 h to allow for transport from ACE to the magnetosphere.In general delay times used in W5 vary according to the speed of the solar wind and the tilt angle of surfaces of constant phase for the interplanetary electric field (Weimer et al., 2002).Agreement between P W5 and the energy requirements of GRACE is qualitatively good.Thus, on a global scale power inputs predicted by W5 are not inconsistent with GRACE measurements.Consequently, P W5 as well as α E ε VS can probably be used in Eq. (8a) to estimate the development of E th SW .
Attention is directed to two particular aspects of the results shown in Fig. 8: (1) GRACE measurements and W5 predictions indicate sustained electromagnetic power input ≥1 TW during both the July and November storms.This is greater than the stormtime inputs due to energetic particle precipitation and from solar UV (Knipp et al., 2005).(2) During the third disturbance of the July period and the first one of the November storms, P W5 >α E ε VS by significant amounts.This discrepancy probably reflects the different estimates of PC .In both instances the polar cap potential used in W5 was larger than that predicted by Burke (2007).For a given system of field-aligned currents the one driven by the higher PC requires a higher power input to overcome collisional drag exerted by neutrals on E×B drifting ions.

Summary and discussion
Data shown in previous sections exemplify two common stormtime phenomena: (1) Intense EPB activity in the main phase is followed by prolonged absences during recovery (Huang et al., 2001).( 2) At topside altitudes orbit-averaged densities increase as D st and ε VS intensify in the main phase then revert to near quiet-time values by ∼10 h into recovery (Burke et al., 2007).The temporal distribution of EPB detections before and during the magnetic storm of March 1991 (Fig. 1) suggests a basis for understanding the role of magnetospheric electric fields in driving the post-sunset equatorial ionosphere.During this storm CRRES measured sustained dawn-to dusk ε M earthward of the ring current into L≈2 throughout the main phase (cf.Fig. 2 of Wygant et al., 1998).Earthward of L≈2, the nominal ±1 • uncertainty about spacecraft attitude leads to errors in measured V ×B electric fields that are greater than the ambient ε M , rendering CRRES double-probe data unusable for scientific analysis (Wygant et al., 1998).The durations of EPB detections are consistent with sustained electric field penetration throughout the main phase and during episodic reactivations of the ring current during recovery.In the low latitude ionosphere these penetration electric fields have eastward components near the dusk terminator (Wolf, 1970;Nopper and Carovillano, 1978) where they amplify normal PRE dynamics to increase the growth rate of the R-T instability (Ott, 1978).Citing CRRES observations, Kikuchi et al. (2008) argued that enhanced equatorial electrojet signatures observed during the main phase of the 6 November 2001 magnetic storm indicate sustained ε M penetration of the inner magnetosphere.This was followed by a 2-h period in which the electrojet reversed polarity.Phenomenologically a counter electrojet can be ascribed to either over-shielding by residual Region 2 currents or the disturbance dynamo (Blanc and Richmond, 1980).Here, context favors an over-shielding interpretation.The November 2001 storm was unusual in that the main-phase H only lasted ∼2 h.It is difficult to understand how disturbance dynamo winds could affect equatorial electrodynamics in such a short time (Bruinsma and Forbes, 2007).Consistent with the E th −ε V S relationship (Figs. 6 and  7), the over-shielding episode during early recovery marked the end of EPB activations and of significant magnetospheric involvement in stormtime ionosphere-thermosphere electrodynamics.Aside from the brief penetration events associated with ring-current activations, ε M remained near zero during the recovery phase (Wygant et al., 1998).
J77 provides a means to estimate E th from stormtime ρ variations through an application of the first law of thermodynamics.The evolution of E th shown in Fig. 6 suggests that on a global scale the thermosphere acts like a driven-dissipative system.Empirically established e-fold relaxation τ ≈6.5 h and coupling α E ≈5.5×10 15 (J/h)/(mV/m) coefficients reproduced E th SW variations inferred from GRACE accelerometer data during magnetic storms.In substantial agreement with independent predictions of W5 (Weimer, 2005), the electromagnetic power α E ε V S required to explain ρ variations during large storms was 1-3 TW, a factor of four larger than solar UV inputs to the dayside thermosphere during solar maximum (Knipp et al., 2005).
The sequential termination of ε V S and the return of E th to pre-storm values do not imply that the thermosphere simultaneously reverts to its status quo ante at all localities.Some of the energy deposited during the main phase convects to subauroral latitudes as traveling atmospheric disturbances (TADs) that alter the thermospheric density and composition (Lin et al., 2005) at mid-to low latitudes.Observationally, TADs take ∼4 h to propagate from auroral to equatorial latitudes (Bruinsma and Forbes, 2007).After contributions from ε V S and Region 2 currents turn off, residual thermospheric changes become the only agents capable of modulating an immediate restoration of the solar UV-driven dynamo that characterizes the quiet-time ionosphere.
Worldwide absences of EPBs for several days into recovery set lower bounds on the time required for full restoration of the PRE.Why should it take days for the solar UV driver to reestablish quite-time conditions at equatorial latitudes?The growth of the R-T instability responsible for EPB formation is abetted by the presence of eastward PRE electric fields and seeds of plasma irregularities that form at pre-dusk LTs.At least three broad types of stormtime thermospheric effects persist at low latitudes well into recovery.These are changes in the relative distributions of molecularatomic species, plasma densities and neutral winds.In combination all three affect the full redevelopment of the PRE that drives quiet-time EPBs.The remaining paragraphs consider a few plausible mechanisms.
Altered distributions of thermospheric species are reflected in UV radiances measured by the Thermosphere, Ionosphere, Mesosphere Energetics and Dynamics (TIMED) satellite during the magnetic storm of October 2003.Lin et al. (2005) showed that O/N 2 ratios in the equatorial F-layer decreased in the afternoon sector.Regions of low O/N 2 ratios were marked by reduced total electron content (TEC), indicating increased recombination rates in the F layer.Reported O/N 2 and TEC reductions at low latitudes persisted at least through the first full day of recovery.Lin et al. (2005) noted that late in the first day of recovery 135.6 nm emissions reappeared at off-equatorial latitudes and marked the initial restoration of the Appleton anomaly.Ionospheric uplift that produced the anomaly indicates a re-establishment of the Flayer dynamo responsible for bottomside irregularities seen in the recovery phase.
However, if plasma production in the dayside ionosphere falls below quiet-time equilibrium values, the polarization electric fields needed to maintain current continuity in the

Fig. 1 .
Fig. 1.D st trace for 19-29 March 1991.Vertical lines indicate times of EPB detections by the plasma density monitor on DMSP F9 near the 21:00 LT meridian.
Fig. 2. (a) The black trace indicates mass densities measured by GRACE on 7-10 November 2004 at 10 s resolution.The red dashed line shows variations of −D st over this period.Blue dots in both panels mark orbit-averaged densities.(b) The red trace indicates the electric field in the equatorial plane of the inner magnetosphere estimated from ACE measurements at L 1 .

Fig. 3 .Fig. 4 .
Fig. 3. (a) Mass density profiles as functions of altitude for five selected values of T ∞ .Vertical lines bound the range of altitudes of CHAMP and GRACE orbits in 2004.(b) Example of J77 T ∞ distributions plotted on a latitude and local time grid.

Fig. 5 .
Fig. 5. Geophysical parameters and thermospheric responses plotted as functions of time for (a) Julian days 150 to 230 and (b) 306 to 330, 2004.From top to bottom the traces show the histories of the D st index, ε V S , GRACE measurements of the orbit-averaged density, exospheric temperature and thermospheric energy.

Fig. 6 .
Fig. 6.Natural logarithm of E th SW and ε V S for the disturbances on JDs 206-211, 2004.Vertical lines mark periods of decreased electric fields; slanted lines give the relaxation time τ E .

Fig. 7 .
Fig. 7. Comparison of ε V S (black), modeled E th SW (blue) and values inferred from density measurements by GRACE (red) plotted as functions of time during magnetically disturbed periods of July (top) and November (bottom) 2004.Energies are expressed in units of 10 16 J.

Fig. 8 .
Fig. 8.Comparison of global power into the stormtime thermosphere predicted by the W5 model (red) and αε V S (black) plotted as functions of time during the magnetically disturbed periods of July (top) and November (bottom) 2004.