Ionospheric turbulence from ground-based and satellite VLF / LF transmitter signal observations for the Simushir earthquake ( November 15 , 2006 )

Signals from very low frequency (VLF)/ low frequency (LF) transmitters recorded on the ground station at Petropavlovsk-Kamchatsky and on board the French DEMETER satellite were analyzed for the Simushir earthquake (M 8.3; November 15, 2006). The period of analysis was from October 1, 2006, to January 31, 2007. The ground and satellite data were processed by a method based on the difference between the real signal at night-time and the model signal. The model for the ground observations was the monthly averaged signal amplitudes and phases, as calculated for the quiet days of every month. For the satellite data, a two-dimensional model of the signal distribution over the selected area was constructed. Preseismic effects were found several days before the earthquake, in both the ground and satellite observations.


Introduction
Seismo-induced electromagnetic effects in the ionosphere were first reported in Russia [Migulin et al. 1982] from observations by satellite Intercosmos-19. A lot of studies were then carried out on satellite recordings of waveplasma disturbances that were possibly associated with individual earthquakes or with several strong earthquakes [for reviews, see Parrot et al. 1993, Hayakawa 1997, Molchanov et al. 2002.
A new opportunity to investigate ionospheric perturbations that might be associated with seismic activity arose with the DEMETER satellite mission. The first study that showed examples of unusual ionospheric observations made by the DEMETER satellite over seismic regions at the beginning of the mission was published in 2006 ]. Later on, statistical investigations using data from more than 3.5 years confirmed the existence of very small, but statistically significant, decreases in wave intensity a few hours before earthquakes, at a frequency around 1.7 kHz [Nemec et al. 2009].
These DEMETER observations also provided the very interesting possibility to analyze ground transmitter signals detected by the satellite above seismic regions. The first results from these analyses were reported by Molchanov et al. [2006]. Clear drops in the subionospheric transmitter signals around the time of several strong earthquakes were recorded in 2004, including for the catastrophic Sumatra earthquake. The method applied estimated the changes in the reception zone of the transmitter signals using the signalto-noise ratio (SNR). Further analyses of the Sumatra earthquakes were made on the data collected over one and half years of observations [Solovieva et al. 2009]. Evident effects before and during the great Sumatra earthquakes were confirmed, with long-term durations of about one month. These results led to the conclusion that the size of the perturbation area in the ionosphere was of the order of several thousand kilometers. After these first publications, the effects of this influence of seismic activity on very low frequency (VLF)/low frequency (LF) signal propagation were further reported from observations of the DEMETER satellite , Muto et al. 2008, Slominska et al. 2009].
In our earlier studies [Rozhnoi et al. 2007[Rozhnoi et al. , 2010, we presented a correlated analysis of VLF/LF signals radiated by ground transmitters and collected both at ground receivers and by the DEMETER satellite. In these analyses, we used the night-time data from the alternative electric field receiver of Instrument Champ Electrique (ICE) ] in LF (20 Hz to 20 kHz) and high frequency (3 kHz to 3.3 MHz) ranges. The discretization of the power spectrum density in the LF range was 19.53 Hz, and in the high frequency range it was 3.255 kHz. Effects in the VLF/LF signal were found in both the ground and satellite observations.
In the present study that analyzes the Simushir earthquake (M 8.3;November 15, 2006), we used the signal from the Australian NWC transmitter (19.8 kHz) for the satellite observations, and in addition, for the ground analysis, we used signals from two Japanese transmitters: JJI (22.2 kHz) and JJY (40 kHz). The NWC transmitter signal is the most powerful in the VLF range, so that we can analyze the signal in a large area for satellite observations. Signals from the Japanese transmitters are local and can be used for satellite analysis only if the epicenters of the earthquakes are located in the maximum signal zone.

Data processing
Data from the VLF/LF station in Petropavlovsk-Kamchatsky and data from the ICE receiver collected by the DEMETER satellite were used for the analysis. The monitoring of the VLF/LF signals was carried out from October 1, 2006, to the end of January, 2007. A very strong earthquake with M 8.3 occurred near Simushir Island of the Central Kuril region (Russia) on November 15, 2006. Following this, a series of strong aftershocks (M 5.0-6.5) was observed over several months.
The position of the station at Petropavlovsk-Kamchatsky and the VLF/LF transmitters, together with the epicenter of this earthquake and its aftershocks, are shown in Figure 1. The earthquake epicenter was in the sensitivity zone of wave paths JJY-Petropavlovsk-Kamchatsky, JJI-Petropavlovsk-Kamchatsky and NWC-Petropavlovsk-Kamchatsky. For the DEMETER data, we analyzed the signals in the part of the night-time orbits when the satellite passed above the earthquake area. The zone of analysis had a width of 25˚, which provided one orbit every day. The time averaging of the dynamic spectrum was about 2 s, and the space resolution along the orbit was about 10 km to 15 km.
The ground and satellite data were processed by a method based on the difference between the real signal in the night-time and the model signal [Rozhnoi et al. 2004[Rozhnoi et al. , 2007. The model for the ground observations was the monthly averaged signals of amplitudes and phases calculated for the quiet days of every month. For our analysis we use the residual signals of phase dP or amplitude dA: (1) where A and P are the amplitude and phase for the current day, and <A> and <P> are the corresponding averages.
For the satellite observations, we calculated the model of the signals for every real orbit based on a two-dimensional model with regular signal distributions over the selected area. The modeling consists of the following procedure: a) Computation of a polynomial expression for the surface as a function of the longitude and latitude. We applied the method of local polynomial interpolation, which uses multiple regression methods on data within a localized window to fit a set of trends. The window can be moved around and the surface value at the center of the window, n 0 (x, y), is estimated at each point, based on a weighted least squares fit to the data points Z(x i ,y i ) as follows: , where n is the number of points within the window, and w i is a weight defined as: where d i0 is the distance of the point from the center of the window, and a is a parameter that controls how fast the weights decay as a function of distance. Finally, n 0 (x i ,y i ) is the value of the polynomial. For the first-order polynomial: For the second-order polynomial: , (5) and so on. The minimization occurs for the parameters {b i }. The parameters are re-estimated whenever the center point, and consecutively, the window move [Gandian 1963]. b) Construction of a regular latitude and longitude grid of 0.32˚. c) Computing of a net point model. The models were calculated for the SNR defined as the ratio of the signal spectrum density near the transmitter frequency F 0 with the minimum value in the frequency band (SNR = A F 0 /A min ). Using this model, we can define the instantaneous variations in the VLF/LF signal intensity in the active region at any time and for any longitude and latitude as the difference between the measured SNR of the signal (longitude, latitude, t) and the reference signal (longitude, latitude).

Results
Results of ground observations for the wavepaths JJY-Petropavlovsk-Kamchatsky and NWC-Petropavlovsk-Kamchatsky are shown in Figure 2. For both wave paths, there were disturbances in the signals (Figure 2, yellow and brown colors). These started about two weeks before the earthquake and continued during the aftershock activity, to the middle of December, 2006. The effects were more evident in the 40 kHz signal. Figure 3a shows the spectra of the VLF signals recorded by the DEMETER satellite along part of the night-time orbits that passed above the earthquake area on the disturbed day of November 2, 2006. For comparison, we give here the quiet day of October 29, 2006 (Figure 3b). The signals from several powerful VLF transmitters (the Russian Alfa [11.8,12.6 and 14.88 kHz], and the Australian NTS [18.6 kHz] and NWC [19.8 kHz] transmiters) are clearly observed in Figure  3. A signal at almost 18 kHz can be seen in Figure 3b, although this was out-of-operation in Figure 3a. Changes in the NTS and NWC signals (decreases in the amplitudes, and spectral broadening of the signals) occurred in the region of 149-152˚E, 45-50˚N, just above the area of the seismic activity ( Figure 1, pink rectangle along the DEMETER pass). The same effects were observed in the dynamic spectra of the VLF signals (Figure 4). The signals of the same transmitters are easily noted here as the horizontal lines. Spectral broadening of the NWC and NTS signals can be seen above the earthquake region. Finally, a comparison of the results of the satellite and ground observations is shown in Figure 5. Here, we used the differences averaged over the night-time for the ground reception, and the differences averaged along the part orbit crossing the seismic area for the satellite data. There is an evident decrease in the amplitude of the VLF/LF signals in both the ground and satellite data that is associated with the seismicity. The amplitude anomalies are always negative for both magnetic storms and seismic activity, because of the loss of the signal in the ionosphere irregularity during the propagation. Phase anomalies can be both positive and negative; these depended on the length of the path. In the present case, the anomalies in the phase of the JJY signal were positive. ROZHNOI ET AL. 190

Discussion and conclusions
We have presented here a further comparison of ground and satellite data during periods of seismic activity. These simultaneous analyses provide cross-validation of the results, and they are more reliable for earthquake precursor studies.
As a mechanism of the observed effects, we suggest the following: Such long-term and large-scale perturbations in the ionosphere cannot be produced by the seismic shock itself (durations of minutes), and so we need to assume a longlasting agent that influences the ionosphere around the date of an earthquake.
We believe that this initial agent is an upward energy flux of atmospheric gravity waves that are induced by gaswater release in the earthquake preparatory zone.
Penetration of atmospheric gravity waves into the ionosphere leads to modifications in the natural (background) ionospheric turbulence, especially over space scales of ~1 km to 3 km and wave numbers kT of ~10 -4 -10 -3 m -1 . These weak, but reliable, effects can be revealed by direct satellite observations [e.g. Molchanov and Hayakawa 2007].
Resonant scattering of the VLF signals is possible under the following conditions of frequency-wave number synchronism: ~0 = ~s + ~T, k 0 = k s + k T , where ~0 and k 0 are for the incident VLF waves, ~T and k T are for the turbulence, and ~s and k s are for the scattered waves. Here, the amplitude of the incident wave A 0 decreases exponentially during the course of propagation through the perturbed medium: A 0~e xp(-a n A T H), where an is the coefficient of nonlinear interaction, which depends on A s and A T , and H is the length of the interaction region. In our case with VLF signals: ~T <<~0~ ~s, and the interaction is especially efficient because k 0~ k s~ k T [Molchanov 1985]. Therefore, even though the amplitude of the turbulence A T is small, the scattering can be significant if the length H is large.