A very strong and shallow M6.7 (M6.9 according to JMA) earthquake was registered near the east coast of Honshu, Japan at 23:06 UTC on February 16, 2015. JMA is reporting depth of 10 km while USGS is reporting M6.8 at a depth of 23 km.


Figure 1: Earthquake epicenter and Dobrovolsky area.

JMA is describing this quake as an aftershock of the M9.0 that hit Japan on March 11, 2011 and caused the meltdown of three reactors at the Fukushima Daiichi Nuclear Power Plant.

The earthquake was followed by many aftershocks, the largest of which (M5.7) occurred on February 17 at 04:46 UTC at a depth of about 50 km.

Another significant aftershock (M 6.3) occurred on February 20, 2015, with a hypocentre at 13.5 kilometre depth, just east of the mainshock epicentre.







  1. Accelerated Moment Release (AMR) & Revised-AMR

The Catalogue earthquake analysis for this event started from January 2009 with a maximum distance of 750 km from the epicentre. In the next step of the study a temporal cut was applied starting from May 2012 in order to exclude the large Tohoku earthquake (11 March 2011, M9). Figure 2 shows AMR results with the new catalogue starting from May 2012.


Figure 2: AMR curve corresponding to Catalogue starting from May 2012.

In R-AMR analysis the maximum distance of 900km was progressively restricted to lower radius (600 till 300 km Figure 3); then, with this distance fixed, we moved to small changes of the smoothing parameters related to the region dumping (R0 and g1).

In this case, it is evident that no acceleration appears in catalogue data.



Figure 3: The search of the best Rmax: it does not seem there exist a best candidate maximum distance (C-factors always greater than 1).





Two different approaches were developed to search EQ precursor using geomagnetic Swarm data: MASS method and Wavelet method. Both analyses are based on Level 1B MAGxLR Swarm product.


  1. MASS algorithm (Magnetic Swarm anomaly detection by Spline analysis)

The algorithm MASS (Magnetic Swarm anomaly detection by Spline analysis) was applied to M6.7 Japan 2015 earthquake with different thresholds (kt=2 and kt=2.5), while the moving window was fixed at 3.0°. The algorithm analyses all tracks in DbA one month before and one after the EQ. The tracks are marked as “anomaly” only if the centre of the moving window is in DbA and if geomagnetic conditions are quiet. Figure 4 shows an example of anomalous tracks detected by MASS method for Japan EQ.


Figure4:Example of anomalous track in Y magnetic component (Satellite A- January 28, 2015).

The cumulative number of anomaly tracks detected by MASS one month before and after the Japan EQ (threshold  kt= 2.0) is shown in Figure 5. For the Y magnetic field component the number of anomalies detected before EQ is compatible with an acceleration. In addition, the nEQ and C factors are good, indeed the number of the anomalies before EQ is greater than the number of the anomalies after EQ, so this could be a good indication that some of the anomalies before EQ could be a seismic precursor.


Figure5: Cumulative number of anomaly detected by MASS one month before and one month after the M6.7 Japan EQ. Threshold is kt = 2 and the anomalies are selected only with geomagnetic quiet time (|Dst| ≤ 20 nT and ap≤ 10 nT).



  1. Wavelet Analysis


The Wavelet spectral analysis has evidenced the existence of different anomalous families, each characterized by some particular features that altogether do not clarify whether they are linked to LAIC or not.

Figure 6 shows examples of wavelet analysis results for Japan event: this kind of anomaly appears as a phenomenon originating at higher latitudes and it seem to begin just above the epicenter latitude.



Figure 6:Wavelet analysis results few days before the Japan M6.7 mainshock (February 7, 2015).

Finally, Figure 7shows  a track on the day of the mainshock; it reports a very quiet signal (background) inside the DbA and a single spike at higher latitude, even when quite large seismicity affected the region.

Unfortunately, also for this event the discrimination of the source of emerging signals from the background by applying this method alone may need the adoption of some other more complex scheme.



Figure 7: The figure show a track  for the day of the mainshock, but  there are no evidences of anomalies in the signals.



  1. SWARM – IONOSPHERE from Satellite

Satellite-based data for the ionospheric characterization of the EQ-related events are mainly those referred to the LP (Langmuir Probe) instrument aboard the SWARM satellites. The electron density Ne is the relevant parameter used for the ionospheric characterization in the frame of SAFE project. Two different methods were developed to analyze Swarm ionospheric data: NeSTAD and NeLOG.


  1. Method I: NeSTAD

The NeSTAD analysis has been applied to the Swarm constellation data (LP and IBI data) available in the period from 17 January to 18March 2015 have been used to derive the track anomaly parameters.The NeSTAD has been initialized with the mild outliers mode (k=1.5) and with an “excess area” parameter equal to 0.1.

Then, to tag the interesting track anomalies for this particular event, the following criteria have been applied:

  • R>Rthr=0.7 and standard deviation of the filtered track below σthr=0.1 or standard deviation of the filtered track above σthr=0.1 independently of R value.
  • Only night time tracks have been selected (18-06 LT), because during night time and at mid latitudes the ionosphere is expected to be less turbulent.
  • Quiet ionospheric conditions (absolute value of Dst in the considered day not exceeding 20 nT).

An example of tagged anomaly is provided in figure 8.


Figure 8: Identified anomaly with the NeSTAD algorithm and tagged as interesting for the Japan N6.7 EQ event. Tagged anomaly refers to Swarm Alpha satellite on 10 February  2015.



Figure 9: - Cumulative number of anomalies identified through the NeSTAD algorithm for Swarm satellite Alpha.


Figure 9 shows the cumulative number of the tagged anomalies through the NeSTAD algorithm for Alpha satellite. The black dashed line indicates the day in which the Japan M6.7 EQ event occurred. The red boxes indicate the days in which disturbed geomagnetic conditions have been recorded and that are not included in the tagging process. The two blue dashed lines indicate the time interval during which the given Swarm satellite covered the DbA of the event at nighttime. It is interesting to note that the period was characterized by several disturbed days from the geomagnetic point of view , even if quiet days are in correspondence with the week before the EQ and then satisfying for the LAIC-related anomaly tagging in the proximity of the event.



  1. Method II: NeLOG

The automatic search NeLOG of ionosphere electron density anomaly from Swarm data is applied Japan M6.7case study.A sample is classified as anomaly if the residual value exceeds a threshold kt times the RMS of the residual after the spline fit (in this case kt=2.5). A track is selected as an “anomaly” if it has more than 10 anomaly samples in DbA and if geomagnetic indices are |Dst|<20 nT and ap<10 nT. Tracks are selected within a mean local time between 22 and 6.



Figure10: Anomaly detected by NeLOG (satellite Alpha, 16 February 2015)few hours before the Japan EQ.

Figure 10 shows an example of anomaly track in the date of the mainshock, while Figure 11 reports the cumulative number of anomaly windows detected by NeLOG. It is interesting to note that all the anomalies are before EQ except one, so the nEQ* factor is very good, but it is not possible to see an S-shape behaviour just around and before EQ, but at 15-20 days before, as possible precursory pattern.


Figure 11:Cumulative number of anomaly samples detected by NeLOG 1 month before and 1 month after M6.7 Japan EQ. Threshold is kt = 2.5, the anomalies are selected only with geomagnetic quiet time (|Dst| ≤ 20 nT and ap ≤ 10 nT) and in night time (22 ≤ LT ≤ 6).


  1. IONOSPHERE Ground-based


By means of the analysis on the ionosonde data (Wakkanai Ionospheric observatory, Latitude: 45.2° N, Longitude: 141.8° E) a single anomaly has been identified, occurred on 11 February 2015 at 02 UT. During such UT day, geomagnetic conditions were mostly quiet. According to the top panel Figure 12, an increase of auroral activity was recorded in that day. In particular, the AE index reached the value of 464 nT at ~3UT and of 683 nT at ~13UT.

Middle and bottom panels of Figure 12 report the behaviour of the foF2 variations in two stations in the same longitude sector, both at middle latitude. The Wakkanai ionosonde (middle panel) is at higher latitude with respect to the other in Kokubuni (bottom panel). The first maximum of AE index occurred at 3:37 UT in this longitude sector appears during daytime hours. This auroral activity has determined a positive ionospheric storm at Wakkanai at 04 UT and at Kokubuni at 06UT. So the identified ionospheric anomaly is due to an increase of auroral activity. By consequence, this ionospheric anomaly seems not to be related to LAIC.


Figure 12: The ionospheric anomaly of the 11 February 2015: in the upper panel the AE index variations and in the others panels the ΔfoF2 variations in different stations.




The “single-station GNSS analysis” is applied to vTEC values from data acquired by different IGS GNSS stations located in (or just outside) the area of Dobrovolsky from 10th February to 16th February (see Figure 13).


Figure 13: IGS GNSS stations used to perform the single station GNSS analysis. Red point represents the EQ epicenter and blue circle shows the DbA.

Considering each station separately, the EEMD (Ensemble Empirical Mode Decomposition)is applied to obtain the IMF composing the vTEC signals. Then, the statistical test µ/σ is applied to separate the IMF contributions to the signal trend and oscillation.The CWT is then used to extract the time/frequency representation of the oscillation component. Figure 14 shows the wavelet spectrum of the oscillating signal from 10th February to 16th February for each of the stations considered. From these panels can be argued that “oscillations” are attenuated moving away from the epicenter.


Figure 14: Continuous Wavelet spectrum of the oscillation component of vTEC over kgni (panel a), mizu (panel b), stk2 (panel c) and yssk (panel d).



The M6.7 Japan EQ was analyzed using both Swarm geomagnetic and ionospheric data searching for earthquake-related anomalies in the frame of LAIC theory.

MASS analysis revealed that the nEQ and C factors values obtained for this event are good indicators in the search for EQ-related effects.  In particular, the number of the anomalies before EQ is greater than the number of the anomalies after EQ and this could indicate that some of the anomalies before EQ could be a seismic precursor.

NeLOG algorithm found a concentration of anomalies 19 days before the EQ. The anticipation time is too large with respect to magnitude of EQ to assert that this could be a precursor, but it is not possible exclude this possibility.

The cumulative number of anomalies derived by tagging procedure on the NeSTAD output on Alpha and Charlie Swarm satellites allowed identifying 3 anomalies per satellite before the EQ and 3 in total (2 for Alpha and 1 for Charlie) after the EQ.

Finally, the result of the GNSS analysis seems to reveal a dependence of the intensity of the anomalies on the station-epicenter distance. In particular, a couple of peaks are captured analysing the data from the closest station (MIZU) during 11 February (5 days before the EQ) corresponding to a frequency of about 0.35 mHz (period≈47 minutes), occurring at 04.00 UT and 10.00 UT. These peaks seem to decrease in intensity moving away from the epicenter and disappear in the vTEC over YSSK, being located just outside the DbA.