Disentangling the origin of chemical differences using GHOST
2018-11-27 Forecasting activity of Japanese volcanoes through geophysical monitoring
1. Forecasting activity of
Japanese volcanoes through
geophysical monitoring
Yosuke Aoki
Earthquake Research Institute, The University of Tokyo
Email: yaoki@eri.u-tokyo.ac.jp
27 November 2018
Int’l Workshop on optimizing the use of
volcano monitoring database to anticipate unrest
Yogyakarta, Indonesia
2. Goal of volcano monitoring and
WOVOdat
“Just as epidemiological databases help medical researchers to identify factors in
the spatial and temporal distribution of diseases, WOVOdat is helping
volcanologists to discover new relationships between different variables, “
(Newhall et al., JVGR, 2017)
Understanding how the volcano is working
Forecasting what happens next:
Unrest or no unrest?
Eruption or no eruption?
When is the next eruption if any?
Magmatic or phreatic eruption?
Explosive or effusive?
When does unrest end? (not in this talk)
We could learn from the past, but the record from a single volcano
usually is not large enough. We thus need to learn from other
volcanoes as well.
3. Volcanoes in Japan
Tokyo
Recent major eruptions:
2018 Kusatsu-Shirane
(phreatic, VEI=1, killed 1)
2015 Hakone (phreatic, VEI=1)
2014 Ontake
(phreatic, VEI~2, killed 60+)
2013 Nishinoshima
(magmatic, VEI=2-3)
2011 Shinmoe-dake
(magmatic, VEI=2)
2011, 2014, 2016 Aso
(magmatic, phreatic, VEI=1-3)
Quasi-continuous Sakurajima
(magmatic, failed eruption in 2015)
Recent volcanic activity in Japan is lower than Indonesian volcanoes. No
VEI=5 since 18th century and no VEI=4 since 1914.
Hakone
Kusatsu-Shirane
Ontake
Nishinoshima
Shinmoe-dake
Sakurajima
Aso
4. Volcano observatories of ERI
Asama
Izu-Oshima
Kirishima
Mt. Fuji
Asama
2004 (magmatic, VEI=2)
2008-9 (phreatic, VEI=1)
2015 (phreatic, VEI=1)
Kirishima
2011 (magmatic, VEI=2)
2018 (magmatic, VEI=2-3)
Izu-Oshima
1986-1990 (magmatic,
VEI=3)
Mt. Fuji
1707 (VEI=5)
Not enough number of
eruptions to learn from the
past.
5. Possible scenarios following unrest
Moran et al. (Bull. Volcanol., 2011)
How can we know which scenario is relevant, given
unrest started?
6. Difficulty in forecasting
volcanic activity
Unrest to eruption
Seismicity faithfully takes the magma pathway in some
cases (e.g., Kilauea, 2000 Usu) but sometimes not (e.g.,
Asama, Shinmoe-dake).
Failed eruption
Unrest does not always lead to an eruption. Similar
deformation pattern can lead to either eruption (2011
Shinmoe-dake) or failed eruption (2012 Shinmoe-dake)
Blue-sky eruption (?)
Some eruptions start with an apparent lack of “short-term”
unrest (2015 Kuchinoerabu). The apparent lack of unrest could
be real or due to the lack of appropriate monitoring network.
7. The 2000 Usu eruption
Onizawa et al. (JVGR, 2007)
Time and location of the eruption was well forecasted
from seismicity, resulting in a succssful evacuation of
local residents.
8. The 2000 Usu eruption
Onizawa et al. (JVGR, 2007)
9. 2004, 2009 Asama eruptions
Aoki et al.
(Geol. Soc. London Spec. Pub., 2013)
The volcano inflated before an
eruptions, but inflations sometimes
do not result in an eruption.
Seismic activity does not seem to
correlate with volcanic activity in
long term.
10. 2004, 2009, 2015
Asama eruptions
142
0221-950268 および KVCO-TASH 基線の距離変化.東北太平洋沖地震
したトレンドを示していたが,950221-950268 基線の距離変化に地
含んでいるため,地震直後は異なるトレンドを示している.これは,
の余効変動の影響によるものと考えられる.
Monthly number of earthquakes
Low-freq. EQs
Total EQs
Volcano-tectonic Eqs
Tornillo
11. The 2011 Shinmoe-dake eruption
Nakao et al. (Earth Planet. Space, 2013)
第 142 回火山噴火予知連絡会 気象庁
新燃岳
御鉢
韓国岳
Inflation followed
by eruption
Increasing
seismicity but the
distribution is
diffuse.
14. Coming back to Shinmoe-dake unrest第 142 回火山噴火予知連絡会 気象庁
新燃岳
御鉢
韓国岳
:2017 年7月1日~2018 年9月 30 日の震源
:2013 年 11 月1日~2014 年 10 月 31 日の震源
:2009 年 11 月1日~2011 年 11 月 30 日の震源
:上記を除く期間の震源
The 2010 and 2017 inflations
resulted in an eruption, but the
2011 and 2013-2014 inflations
resulted in failed eruptions.
Deformation field between 2010
and 2011 unrest was similar.
What controls whether an
unrest results in an eruption or
not?
– Not clear or unknown.
Lab experiments suggest that
intruded volume (deformation)
is a key (Taisne et al., Bull.
Volcanol., 2011), but the
observation suggests that other
factors may be at work.
15. Blue-sky eruption? 2014 and 2015
Kuchinoerarabu eruptions
Hotta & Iguchi
(Earth Planet. Space, 2017)
Both are magmatophreatic eruptions.
Unrest for ~15 years but no short-term precursors are apparent.
It is difficult to predict how the volcano evolves when we observed unrest.
The 2014 eruption (VEI=1) damaged much of the instruments, making incapable
of closely monitoring the volcano preceding the 2015 eruption (VEI=3).
16. Blue-sky eruption?
The 2014 Ontake eruption
The eruption took
place in the worst
possible timing.
Increasing
earthquakes from
~3 weeks before
the eruption, it is
difficult to judge
whether it is
considered as
unrest.
Tilt changes
started 7 minutes
before the
eruption.
Kato et al.
(Earth Planet. Space, 2015)
17. Definition of unrest?
The 2006-2009 Mt. Fuji inflation
Accelerating seismicity
and expansion since
~2006, but the
expansion is invisible
without extensive
post-processing of the
GPS time series.
Harada et al.
(Bull. Volcanol. Soc. Jpn., 2010)
18. Scaling law for
“short-term”
precursors
Linear scaling between ”Pause Time”
and the maximum amplitude of the
explosion earthquake.
An accumulation of volcanic gases in
the shallow conduit might be the
main driver of the explosion.
Nishimura et al. (Bull. Volcanol., 2013)
21. Alternative: Introducing
Deep Learning
Feature 1 2 3 ….
Taking time series as input
Seismic waveform
(Earthquake count?)
Deformation
Gas
Electromagnetics
…….
Outcome
Explosion (magmatic, (magmato)phreatic,
large, small, tiny…)
Effusive eruption (large, small, tiny…)
Failed eruption
……
Start from a volcanoes
with a lot of dataset to
have the system learn
(Sakurajima, Etna,
Stromboli, Sinabung, …)?
Extension to multi-
volacanoes will involve a
challenge.
22. Summary
Assessing precursors to eruptions involve a lot of difficulty for various
reasons.
For example, apparent lack of precursors could be real or due to a lack
of optimum monitoring network.
However, there are some precursors which have repeating features or
obey scaling laws.
Employing data science perspectives will advance the capability of
forecasting volcanic activity.
What is not discussed here
Precursors detected by something other than seismic and geodetic
instruments (gas, electromagnetics, surface temperature…)
How can we judge whether the activity ended or not?
How to include observations in pre-instrument era (eye witness,
historical accounts…)
Data sharing