Analysis of the Enhanced Negative Correlation between Electron Density and Electron Temperature Related to Earthquakes

Ionospheric perturbations in plasma parameters have been observed before large earthquakes, but the correlation between different parameters has been less studied in previous research. The present study is focused on the relationship between electron density (N e) and temperature (T e) observed by the DEMETER (Detection of ElectroMagnetic Emissions Transmitted from Earthquake Regions) satellite during local nighttime, in which a positive correlation has been revealed near the equator and a weak correlation at mid-and low latitudes over both hemispheres. Based on this normal background analysis, the negative correlation with the lowest percent in all N e and T e points is studied before and after large earthquakes at mid-and low latitudes. The multiparameter observations exhibited typical synchronous disturbances before the Chile M8.8 earthquake in 2010 and the Pu'er M6.4 in 2007, and T e varied inversely with N e over the epicentral areas. Moreover, statistical analysis has been done by selecting the orbits at a distance of 1000 km and ±7 days before and after the global earthquakes. Enhanced negative correlation coefficients lower than −0.5 between N e and T e are found in 42 % of points to be connected with earthquakes. The correlation median values at different seismic levels show a clear decrease with earthquakes larger than 7. Finally, the electric-field-coupling model is discussed; furthermore, a digital simulation has been carried out by SAMI2 (Sami2 is Another Model of the Ionosphere), which illustrates that the external electric field in the iono-sphere can strengthen the negative correlation in N e and T e at a lower latitude relative to the disturbed source due to the effects of the geomagnetic field. Although seismic activity is not the only source to cause the inverse N e –T e variations, the present results demonstrate one possibly useful tool in seismo-electromagnetic anomaly differentiation, and a comprehensive analysis with multiple parameters helps to further understand the seismo–ionospheric coupling mechanism.


Introduction
The DEMETER (Detection of Electro-Magnetic Emissions Transmitted from Earthquake Regions) satellite was launched on 29 June 2004 in France and operated until 10 December 2010, with an inclination of 98 • (Cussac et al., 2006).It was a solar-synchronous-orbit satellite, with the local time of each orbit passing the equator being the same: 10:30 in the daytime and 22:30 in the nighttime.From its records, many ionospheric perturbations related to seismic cases have been detected (Parrot et al., 2006;Sarkar et al., 2007Sarkar et al., , 2012;;Zhang et al., 2009Zhang et al., , 2010aZhang et al., , b, 2012a;;Zeng et al., 2009;Pisa et al., 2011;Liu et al., 2011).Furthermore, statistical analysis of different parameters has been developed to ascertain the relationship between ionospheric perturbations and large earthquakes (Némec et al., 2009;He et al., 2010He et al., , 2011;;Zhang et al., 2011Zhang et al., , 2012b;;Parrot, 2012;Pisa et al., 2012;Li and Parrot, 2012).Based on a great amount of data, statistical research can help us to understand the general distribution features of ionospheric perturbations in time and space and to provide more information on their occurrence probability related to earthquakes.
Many papers have shown the correlation characteristics at different local times and conditions of the in situ electron density and temperature (N e and T e ) observed by the satellite (Balan et al., 1997;Oyama et al., 1996;Su et al., 1996;Rich et al., 2003;Lin et al., 2007a, b;Liu et al., 2007;Ren et al., 2008;Venkatraman and Heelis 1999;Li et al., 2011;Kakinami et al., 2011a, b).The negative correlation between Published by Copernicus Publications on behalf of the European Geosciences Union.

X. H. Shen et al.: Analysis of the enhanced negative correlation
N e or N i (ion density) and T e during local daytime is widely accepted.Some positive correlations have been found during periods of high solar activity (Kakinami et al., 2011a) and also in the equatorial area at sunset in December (Liu et al., 2007).In earthquake research, N e and N i are frequently used in case and statistical studies.Only a few papers have focused on T e and ion temperature (T i ).For example, Oyama et al. (2008) studied the variations in T e observed by the Hinotori satellite before and after three earthquakes during 1981-1982 and found that T e significantly decreased over the epicenters in the afternoon in the 5 days before and after the earthquakes.Sharma et al. (2013) summarized T i and T e variations related to seismic activity during 1995-1998 over India by using SROSS (Stretched Rohini Satellite Series)-C2 satellite data in the altitude range of 430-630 km, in which significant enhancement had been detected, with 1.2-1.5 times the average normal values of T e and 1.1-1.3times the normal T i within a 5 • window over the epicenters.Sarkar et al. (2012) studied the ionospheric anomalies related to the Haiti earthquake on 12 January 2010, and their results from the DEMETER satellite showed the most important variations in N e and T e 1 day before the main shock, with T e exceeding the upper bound of ∼ 100 K and N e exceeding the upper bound by 20 and 9 % during the day-and nighttime, respectively.By using DEMETER satellite data, Zeng et al. (2009) found that N e and T e decreased to a value above 20 % during local daytime near the epicenter in the 4 and 5 days prior to the Wenchuan M7.9 earthquake on 12 May 2008 in China.
The big problem in earthquake prediction is the variety of anomalies before and after different earthquakes; a single kind of precursor is hard to detect in all events.Also, ionospheric perturbations cannot be detected for all earthquakes.Thus, the sensitivity estimation of precursors is also an important topic in seismological studies.Here, based on the study of the correlation between N e and T e recorded by the Langmuir probe onboard the DEMETER satellite, some short time disturbances with inverse N e and T e are recorded for the more than 6 years of the DEMETER satellite's operating time.Statistical analysis related to large earthquakes is also carried out.Finally, discussion and conclusions are provided in the last section of the paper.

The ISL onboard the DEMETER satellite
The specific scientific objectives of the Langmuir probe instrument (ISL: Instrument Sonde de Langmuir) are designed to map the bulk plasma parameters (primarily N e and T e ) and to study their variations associated with solid-earth events and other sources of perturbations (Lebreton et al., 2006).The Langmuir probe sweeps at a voltage of ±3.81 V.A com-plete voltage sweep is performed in 1 s to obtain a complete current-voltage (I-V) characteristic, corresponding to about a 7 km spatial resolution at 710-660 km altitude from the sun-synchronous DEMETER orbit.From the analysis of the I-V characteristics, the plasma parameters are extracted with 1 s time resolution; these are N e , T e , N i and spacecraft potential.
3 Correlation background between N e and T e Due to the strong effects of solar activity during local daytime, local nighttime data is generally employed in earthquake research to avoid solar and space disturbances.Therefore, here only the correlation between N e and T e during local nighttime is analyzed.Figure 1 shows the projection of N e (x axis) and T e (y axis) at the different geomagnetic latitudes (color scale) in March, June, September and December 2008 (four images labeled 2008-03, 2008-06, 2008-09 and 2008-12).It can be seen that there is seasonal variation in N e and T e , with the images from March and September being the same and those from June and December being very similar to each other.Generally, all four seasons' pictures show that at low latitudes N e changes a lot but T e varies little, while at midlatitudes N e and T e are dispersed, especially in June and December.In a previous study, the absolute values of N e and T e from DEMETER were not very accurate, with T e being much higher than it is supposed to be (Kakinami et al., 2013); therefore, in this paper only the relative variations in N e and T e are taken into account.To reveal the relationship of N e and T e at different latitudes, Fig. 2 provides the correlation coefficients for March 2008 from a linear function fitting with the equation , with the six diagrams separated by a latitude interval of 5 • in the Northern and Southern Hemisphere (Fig. 2a and b).The results show the weak connection between N e and T e during local nighttime, with a maximum R of only about 0.4 in the equatorial area of latitudes 5 • S-5 • N.With the increase in latitude, R decreased quickly, but it still maintained a positive correlation over both hemispheres.Compared with Hinotori satellite data, after 20:00 LT, both N e and T e followed the usual nighttime decay and a positive correlation was detected (Kakanami et al., 2011b), which was also observed using the incoherent scatter radar at Saint-Santin (Zhang et al., 2004).Therefore, the weak positive correlation between N e and T e is the normal state in the nighttime topside ionosphere.It can be seen that, along orbit 30 168_1, N e and N i (O+) diverged from their normally flat variational trends and decreased over the seismic region of Chile (the disturbances are framed by the blue rectangle), whereas T e increased in the same region (Fig. 3).Along orbit 30 256_1, N e and N i (O+) modulated over this seismic region and T e also showed an inverse pattern (Fig. 4).Another anomalous phenomenon in Fig. 4 is that, at the northern conjugate epicenters of the Chile earthquakes, a step-increasing variation was detected in N e and N i (O+), while T e decreased.This kind of variation was also observed before some moderate earthquakes.Figure 5 gives an example along orbit 15 572_1, observed to the north of the epicenter a few hours before the Pu'er M6.4 earthquake in China on 2 June 2007 (23.03 • N, 101.05 • E), in which a reversed step variation is illustrated between T e and N e , while N i (O+) shows variations of a similar shape as N e .All of these observations demonstrate that some coupling processes during the preparation processes of earthquakes lead to the simultaneous variations in N e and T e .
In order to test the sensibility of this kind of anomaly, a statistical analysis needs to be done based on large amounts of data from the DEMETER satellite and many seismic events.

Method and statistical analysis
In order to reflect this reversed variation feature between N e and T e during local nighttime before earthquakes, the floating coefficients are calculated for 20 observation points along orbits 15 572_1 and 30 168_1 (Fig. 6).The results show that the normal coefficients are above 0 during local nighttime, which means that the positive correlation between N e and T e is limited in time and space.Along 15 572_1, the maximal coefficients were near 1.0; however, a quick decrease was exhibited, with R lower than −0.9, around a latitude of 30 • N (Fig. 6a), which was just to the north of the Pu'er epicenter.By contrast, along orbits 30 168_1 and 30 256_1, the coefficients varied significantly, with a few points lower than −0.5 (Fig. 6b and c), in the southern part of the orbits at 10-50 • S; this illustrates the extensive modulation due to the combined effect of the major earthquake and following aftershocks along the whole Chile seismic rupture fault.
On the basis of the negative N e -T e correlation related to earthquakes shown above, a statistical analysis is carried out.First, an automatic software is developed to calculate the R of N e and T e with 12-13 observation points at 1 • of latitude during January 2005 to December 2010.Taking account of the distribution of global main quakes and the effects from polar electrojets at high latitudes, the studied area is limited to ±45 • of latitude.Figure 7 shows the number of points corresponding to different R values.It can be seen that most points have an R higher than 0.5, which means that N e and T e vary positively within a 1 • latitudinal scale.The counts coincide with an exponential attenuation, with the decrease of R to the minimum number corresponding to the maximum negative correlation of −1.0 and the exponential fitting of R being larger than 0.98.The points with R less than −0.5 only occupy 0.5 % of all points, which illustrates that good negative correlations in N e and T e are rare events.While strong earthquakes are sudden events with a small probability, earthquake preparation may be a factor that induces this inverse relationship between N e and T e .
Having selected earthquakes with a magnitude greater than 5.0, we select those points for which the R of N e -T e is less than 5.0.These points are at a distance of 1000 km to the epicenters and were observed within 7 days before and after the earthquake occurrences, suggesting that they are related to the earthquakes.The results show that of the total of 2 357 126 observation points, there are 13 101 points with strong negative correlations between N e and T e .In order to reduce the effects of solar activity, the data when Kp > 3+ is eliminated, so 6058 points are left in the end.Of these points, 2556 can be connected with earthquakes; this constitutes 42 % of global observation points with R ≤ −0.5 in N e -T e .It should be noted that earthquake is only one kind of factor that leads to a negative correlation between N e and T e , and there still exist many others in space or ground VLF (very low frequency) transmitters that may produce similar variations.Another problem is that not all earthquakes can excite the same phenomenon in the ionosphere.In the areas studied, there were 8513 earthquakes with M ≥ 5.0 during 2005-2010, while 7659 earthquakes were in the 5.0 ≤ M < 6.0 range, 611 in the 6.0 ≤ M < 6.5 range, 168 in the 6.5 ≤ M < 7.0 range and 75 had M ≥ 7.0, as listed in Table 1.In order to illustrate the relationship of N e and T e with earthquakes, the median and averaged R values are calculated for all the data together, and Table 1 exhibits a median of 0.812 and an average of 0.694.Then the median and average R of the points within 3 days of earthquakes and at a distance of 1000 km from the epicenter of the quakes are computed at different magnitude levels to reduce the normal background effects.As shown in Table 1, the averaged R values are similar to each other at four magnitude levels and are also near the average of all global data taken together.However, the median R values decrease compared to the first median R of all points, reducing to 0.683 at magnitudes greater than 7.0.This phenomenon reveals that there might exist much lower and even negatively correlated N e -T e points before and after these destructive earthquakes during quite a short time period, causing the decrease in the median R.
Table 1.The statistical results of R with all observation points and of those related to earthquakes at different magnitude levels.

Statistic
All points Points related to Points related to Points related to Points related to earthquake earthquake earthquake earthquake 5.0 ≤ M < 6.0 6.0 ≤ M < 6.5 6.5 ≤ M < 7.0.Figure 5.The observation of plasma parameters from DEMETER along orbit 15 572 on 2 June before the Pu'er M6.3 earthquake in China on 3 June 2007 (the order of the five panels is the same as in Fig. 3).

Discussion and conclusion
From the observations of ionospheric perturbations associated with large earthquakes, several models have been proposed to explain the seismo-ionospheric coupling mechanism; these include an electromagnetic-wave-penetrating model from the lithosphere to the ionosphere (Molchanov et al., 1995), an acoustic-wave-propagating model (Hegai et al., 1997) and electrical-field-coupling models related to radon emission, aerosol accumulation, rock current and surface charge (Pulinets, 2004(Pulinets, , 2009;;Sorokin et al., 2007;Kuo et al., 2011).The last model plays an important role in explaining the ionospheric disturbances related to seismic activities in plasma parameters, such as the perturbations in GPS TEC (total electron content), foF2 and electron density;   the drift effects from overlapped E × B at the epicentral region coincide with the spatial distribution of perturbations in the ionosphere as shown in GPS TEC (Kuo et al., 2011).By observing the electric field and the plasma parameters simultaneously with the DEMETER satellite, the Excomponent waveforms of the ULF electric field (DC-15Hz) are analyzed along the orbits as shown in Figs.3-5.Combined with the ULF electric field analysis before and after earthquakes in a previous study (Zhang et al., 2014), synchronous 3-1 5mV m −1 disturbances were always detected in the electric field, which illustrates the existence of an external electric field in the topside ionosphere.To demonstrate the enhanced negative correlation associated with earthquakes, SAMI2 (Sami2 is Another Model of the Ionosphere) (Huba et al., 2000) is employed to simulate the effects of overlapped external electric fields in a seismic region.Here, the Chile earthquake region (36 • S, 287 • E) is taken as an example, and the sinusoidal E × B drift model is used to calculate the E × B effects as 5 m s −1 and 10 m s −1 , respectively.The computed results are shown in Fig. 8.With the increase of E × B effects, the electron density shows a typical maximum around 10 • S and two minima at 30 • S and 5 • N at an altitude of 700 km at 03:29 UT (in Figs. 3 and 4, this time is 22:10 LT in Chile), while the maximum of T e occurs at the equator and two minima occur at 10-25 • S and 10-20 • N, respectively.Comparing the two parameters, N e increases at 30-20 • S, while T e still decreases and maintains a low level.At 10-0 • S N e decreases, while T e increases quickly, and at 5-10 • N N e increases, while T e decreases quickly.It illustrates that the overlapped E × B effect from the Chile seismic region (∼ −36 • S) will strengthen the negative correlation feature between N e and T e over some specific latitudes, such as at the north of the source region and even around the equator.It should be noted here that the assumption in SAMI2 of infinite parallel conductivity may lead to nearly the same feature over both hemispheres and a smaller mod-ified area than expected.Therefore, the disturbed region in Fig. 8 might actually be larger.
In this paper, the correlation of N e and T e is studied during local nighttime on the basis of the ISL observations from the DEMETER satellite.By taking account of the normal background with positive and weak correlations between N e and T e at different latitudes, the relationship of the negative correlation between N e and T e with strong earthquakes is analyzed and discussed.The following can be concluded: 1.During local nighttime, log10(N e ) and T e show a positive linear correlation, especially at low latitudes over the Northern Hemisphere.At midlatitudes, N e exhibits almost no correlation with T e .
2. Synchronous perturbations in plasma parameters such as N e , T e , N i (O+) and the electric field have been detected before a few strong earthquakes.Over the seismic regions, a negative correlation between N e and T e is exhibited at mid-and low latitudes, which is significantly different from their normal background.
3. The statistical analysis of global strong earthquakes and negative N e -T e correlations shows that the median R of N e -T e correlations only reduces with earthquakes of a magnitude greater than 7.0, illustrating the enhancement of a negative correlation between N e and T e during large earthquakes.
4. According to results calculated by SAMI2, the overlapped E × B effects will enhance the negative correlation between N e and T e , which proves that an earthquake is one possible source that can cause reversed variations in N e and T e in the topside ionosphere.

Figure 1 .
Figure 1.The distribution of log10(N e ) and T e at different magnetic latitudes (color bar) during quiet geomagnetic periods with Kp < 3 during 4 months (four panels correspond to March, June, September and December) in 2008.

Figure 2 .
Figure 2. The linear correlation between N e and T e at 0-30 • at 5 • latitude intervals over the Southern (a) and Northern (b) Hemisphere (at the top of each panel, MLat represents the magnetic latitude scale, NORTH/SOUTH means Northern or Southern Hemisphere, and N e and T e are the averaged values of all the points distributed in that latitude range; the equation in each panel represents the fitted results, in which x represents N e , f (x) is T e and R is the linear correlation coefficient of N e and T e ).

Figure 3 .
Figure 3.The observation of plasma parameters from DEMETER along orbit 30 168 on 20 February 2010 before the Chile M8.8 earthquake on 27 February 2010 (first panel: N e ; second panel: T e ; third panel: the ion density of O+; fourth panel: the electric field Ex; bottom panel: the earthquakes taking place within ±30 days relative to the time of this orbit at a distance of 2000 km from the orbit; green squares represent past earthquakes, red triangles mean future earthquakes and open squares are the epicenters at geomagnetic conjugate position).

Figure 4 .
Figure 4.The observation of plasma parameters from DEMETER on 26 February 2010, 1 day before the Chile M8.8 earthquake on 27 February 2010 (the order of the five panels is the same as in Fig. 3).

Figure 7 .
Figure 7.The histogram of distribution probability at different correlation coefficients before and after global earthquakes (in the right corner of the panel, the exponential decay has been fitted between the counts (variable y) and different coefficient scales (variable x); furthermore, the correlation coefficient R has been calculated).

Figure 8 .
Figure 8.The calculation results from SAMI2 of N e (left) and T e (right) at an altitude of 700 km during local nighttime at E × B effects of 5 m s −1 (+) and 10 m s −1 (*), respectively.