[SVC45-P36] The performance of the back elevation angle estimation of 3-dimensional infrasonic small array observation
Keywords:Infrasound, Array observation, volcano
The infrasonic array technique has been widely used to detect an infrasonic signal and to constrain its source. Yamakawa et al. (GRL, 2018), hereafter referred to as YMK2018, reported that a small array with a 20-m aperture had a practically sufficient back azimuth (BAZ) resolution, but did not resolve the back elevation angle (BEL). The BEL estimation is important in identifying the source from multiple mechanisms. In order to improve the BEL resolution, we tried a small 3D array observation with three sensors on the ground and one sensor 1.9 meters above the ground. Here we examined the performance of the small 3D array of this experiment.
The experiment was conducted at the summit of Stromboli, the same place as YMK2018, from 20 to 21 on June 2019 (Fig. a). We compared the performances for three sensor combinations (Fig. b): (A) the 3D array using all the four sensors, (B) the 2D array the plane of which was inclined to the ground surface by the use of the elevated sensor, and (C) the 2D array using only the sensors on the ground similar to YMK2018.
The MUSIC method [Schmidt, 1986] was used in the array processing. We analyzed 0.5–40 Hz band dividing into narrow bands (central frequency ±10 %) in steps of 0.25 Hz. The infrasound speed was assumed to be 340 m/sec. The directions of arrival (DOAs), which are BAZ and BEL, were picked from the MUSIC spectrum peak.
First, we examined the frequency dependence of the accuracy of DOAs using the data during the observation with a handy camera. We selected 42 explosion signals the source vents of which were confirmed by the images, and 24 one-minute time windows containing only puffing signals with no explosions. A mirror peak was observed at the symmetric BEL with respect to the array plane in each frequency band with the 2D arrays. The 3D configuration effectively reduced the mirror effect in >15 Hz. On the other hand, in BAZ the spatial aliasing effect became significant above 15 Hz. By comparing high and low frequencies, BAZs and BELs were adequately estimated to the vent direction in >8 Hz and >20 Hz, respectively. Then, we selected the peak of each MUSIC spectrum closest to the vent direction as the signal DOAs.
For statistical processing, we analyzed additional 63 explosion signals and 48 one-minute time windows for puffing signals with no images. Of the 105 signals and 72 time windows, we selected 64 signals and 62 time windows the sources of which were estimated either of the three active vents. The standard deviation (STD) of DOAs was calculated for each set of signals from the same vent in each frequency band. Although the STDs significantly increased in < 3 Hz, we could constrain DOAs in ±20° down to 1 Hz (Fig. c). Of all the frequency band above 3 Hz and for all of the three source vents, 90 % of STDs of A were in <2.2° for BAZ and in <3.9° for BEL. B had similar but a bit larger STDs, while those of C were larger (<2.8° for BAZ and <7.4° for BEL) (Fig. d). Besides, there is a notch in the signal spectra, in which the signal power is very low, in which STDs increased up to 6° with A and B, and 10° with C. We also found that the BEL estimations with the 2D configurations (B and C) sometimes concentrated in the dip direction of their array planes.
We also found systematic shifts of DOAs depending on the frequency. Particularly in 3–5 Hz, DOAs were always in the same direction more than 10° off the vents regardless of the source. The shifts were observed in all of A, B, and C, and larger than the STDs of either BAZs and BELs, and such shifts were not produced in the numerical tests which used a signal with ideal time lags even with overlapping random noises. We consider the shifts are significant, though the cause needs to be identified.
The performances of A and B were similar and better in BEL resolutions than C. The reason was that their array planes were inclined against the BEL. We conclude that putting a sensor at a high position by only 2 m improved the BEL estimation.
The experiment was conducted at the summit of Stromboli, the same place as YMK2018, from 20 to 21 on June 2019 (Fig. a). We compared the performances for three sensor combinations (Fig. b): (A) the 3D array using all the four sensors, (B) the 2D array the plane of which was inclined to the ground surface by the use of the elevated sensor, and (C) the 2D array using only the sensors on the ground similar to YMK2018.
The MUSIC method [Schmidt, 1986] was used in the array processing. We analyzed 0.5–40 Hz band dividing into narrow bands (central frequency ±10 %) in steps of 0.25 Hz. The infrasound speed was assumed to be 340 m/sec. The directions of arrival (DOAs), which are BAZ and BEL, were picked from the MUSIC spectrum peak.
First, we examined the frequency dependence of the accuracy of DOAs using the data during the observation with a handy camera. We selected 42 explosion signals the source vents of which were confirmed by the images, and 24 one-minute time windows containing only puffing signals with no explosions. A mirror peak was observed at the symmetric BEL with respect to the array plane in each frequency band with the 2D arrays. The 3D configuration effectively reduced the mirror effect in >15 Hz. On the other hand, in BAZ the spatial aliasing effect became significant above 15 Hz. By comparing high and low frequencies, BAZs and BELs were adequately estimated to the vent direction in >8 Hz and >20 Hz, respectively. Then, we selected the peak of each MUSIC spectrum closest to the vent direction as the signal DOAs.
For statistical processing, we analyzed additional 63 explosion signals and 48 one-minute time windows for puffing signals with no images. Of the 105 signals and 72 time windows, we selected 64 signals and 62 time windows the sources of which were estimated either of the three active vents. The standard deviation (STD) of DOAs was calculated for each set of signals from the same vent in each frequency band. Although the STDs significantly increased in < 3 Hz, we could constrain DOAs in ±20° down to 1 Hz (Fig. c). Of all the frequency band above 3 Hz and for all of the three source vents, 90 % of STDs of A were in <2.2° for BAZ and in <3.9° for BEL. B had similar but a bit larger STDs, while those of C were larger (<2.8° for BAZ and <7.4° for BEL) (Fig. d). Besides, there is a notch in the signal spectra, in which the signal power is very low, in which STDs increased up to 6° with A and B, and 10° with C. We also found that the BEL estimations with the 2D configurations (B and C) sometimes concentrated in the dip direction of their array planes.
We also found systematic shifts of DOAs depending on the frequency. Particularly in 3–5 Hz, DOAs were always in the same direction more than 10° off the vents regardless of the source. The shifts were observed in all of A, B, and C, and larger than the STDs of either BAZs and BELs, and such shifts were not produced in the numerical tests which used a signal with ideal time lags even with overlapping random noises. We consider the shifts are significant, though the cause needs to be identified.
The performances of A and B were similar and better in BEL resolutions than C. The reason was that their array planes were inclined against the BEL. We conclude that putting a sensor at a high position by only 2 m improved the BEL estimation.