Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Recent Advances in Prediction of EarthquakesShunji Murai, Professor Emeritus, Univ. of Tokyo &
CTO, Japan Earthquake Science Exploration Agency
GNSS Training Course T151-30
GIC/AIT
17 January 2019
+ Introduction
+ Part 1: Review of prediction
+ Part 2: Flow of prediction with
GNSS data
+ Part 3: Additional RS &GIS
techniques for prediction
+ Part 4: Mini-plate theory
+ Conclusions
ⒸJESEA2019
Introduction
+ I started research on the prediction of earthquakes
since 2002 with my partner; Dr. Harumi Araki.
+ The basic prediction method depends on
multi-temporal GNSS data in which abnormal signals
or pre-cursors must be involved.
+ In Japan free access is available to daily GNSS
data of 1,300 CORSs all over Japan.
+ I did realized very abnormal signals just before the
East Japan Great EQ occurred on the 11th March
2011 with 18,000 victims mainly due to Tsunami.
+ In order to rescue human lives I invested to
establish a private company namely JESEA in 2013.
ⒸJESEA2019
Part 1: Review of Prediction of Earthquakes
+ No 1: Nobody has succeeded the prediction of earthquakes in
the human history. It is a challenging theme in the latter half
of my life.
+ No 2: Though so many people died in the past history of
Japan due to huge earthquakes; 30 times with more
than one thousand victims in the past 400 years, Japanese
government and seismic scientists gave up the prediction of
earthquakes in 2012 after many unreliable trials spending
huge budget.
+ No 3: Conventional prediction method in Japan only relies on
information about seismographs, active faults and the past
record of giant earthquakes, which are analyzed with
statistical and provability model. The method is unreliable!
ⒸJESEA2019
ⒸJESEA 2019ⒸGoogle
Large EQ’s since1923
and distribution of
Nuclear Power Plants
Japan is EQ prone
country under risky
condition everywhere
and every time
Fukushima NP
X
What should be the key for prediction?
+ No.1: Remote sensing method should be applied to detect
macroscopic anomalies or pre-signals in advance to
occurrence of earthquakes. We need time series of scientific
observation data covering whole area. The existence of the
pre-signals has been verified by my research in 2007.
+ No.2: Quantitative correlation between anomalies of
observation data and the occurrence of earthquakes should
be developed.
+ No.3: The anomalies should be analyzed and visualized using
GIS techniques and artificial intelligence for better
understanding for risky areas.
ⒸJESEA2019
Macroscopic Phenomena before earthquakes
Epicenter
GNSS
Infrasound
sensor
1. Crustal Change
2. Infrasound anomaly
3. ELF emission anomaly
4. Temperature anomaly
5. Ionospheric anomaly
ⒸJESEA2019
Part 2: Flow of prediction with GNSS data
Download of GNSS data from GSI database
Data filing in EXCEL and graphic
representation
Detection of anomalies
Visualization of anomalies
Judgement and mapping of risky areas
Preparation of documents
Dissemination of MEGA EQ Prediction ⒸJESEA2019
East Japan Great EQ: 2011/3/11 M9.0, SI=7
Miyagi AreaYamagata Area
M9.0, SI=7
Damage due to Tsunami
All Japan Islands quaked
ⒸJESEA2019 ⒸTenki.jp
What were the movements at 2011/3/11 EQ?
©海上保安庁 ©NASA
1.1m sank
Sea Bottom
3m rose
Just after
ⒸJESEA2019
Ojika
5.3m moved
-120
-100
-80
-60
-40
-20
0
20
2010/1/9 2011/1/9 2012/1/9 2013/1/9 2014/1/9 2015/1/9 2016/1/9 2017/1/9
Miyagi Area 胆沢
栗駒2
宮城東和
高清水
南方
小野田
河北
丸森
女川
利府
気仙沼
栗駒
鳴子
志津川
涌谷
宮城川崎
亘理
七ヶ宿
牡鹿
白石
Ojika
2011/3/11EQ
Height Change in Miyagi Area, Tsunami hit area: 2010-2017
110cm sank!
-30
-25
-20
-15
-10
-5
0
5
10
2010/1/9 2011/1/9 2012/1/9 2013/1/9 2014/1/9 2015/1/9 2016/1/9 2017/1/9
Yamagata Area 真室川
鶴岡
大蔵
山形
飯豊
酒田
山形新庄
天童
飛島
立川
朝日
山形小国
米沢
白鷹
遊佐
村山
上山
西川
山辺
2011/3/11EQ
Height Change in Yamagata Area near Japan Sea: 2010-2017
25cm sank!
Keeping sunk after EQ
Scientific findings for prediction with GNSS data
+No.1: The Earth is moving always about 5mm to 1cm vertically
and horizontally in normal condition but moves abnormally in
advance to large earthquake without knowing the reason.
+No.2: Abnormal movement and variation will not be only one
pattern but vary case by case. Earthquake is very complicate
phenomenon.
+No.3: Sinking tendency would be more risky than rising
tendency to induce earthquakes.
ⒸJESEA2019
+No.4: The time span between pre-signal abnormality and
actual occurrence of earthquake ranges a few weeks to few
months.
Prediction method by crustal change with GNSS data
GSI’s CORS: Daily XYZH
+ Short term anomaly# Weekly vertical change
# Monthly horizontal change
+ Long term anomaly
# Two year up-down tendency
# Accumulated stress
GSI CORSDistribution of CORS
JESEA’s own CORS
+ Real time anomaly
# hourly XYZH change
ⒸJESEA2019
ⒸGSI
Prediction examples with correct judges
Example 1: West Shimane Pref. 2018/4/9 M6.1 SI=5+
X
X
2 months before: Weekly vertical anomaly
1 month before: Weekly vertical anomaly
Osaka
OsakaShimane
Shimane
ⒸJESEA2019
Prediction examples with correct judges
Example 2: North Nagano Pref. 2018/5/25 M5.2 SI=5+
Japanese SI= 5+ 3 months before: Weekly vertical anomaly with
more than 6cm
XX
NaganoNagano
ⒸJESEA2019
ⒸTenki.jp
Prediction examples with incorrect judges
Example 3: North Osaka 2018/6/18 M6.1 SI=6-
Japanese SI= 6- 2 weeks before: Weekly vertical anomaly with
more than 4cm
X
Weekly vertical anomaly
with more than 4cm was
recognized but data
were released just on
the day of occurrence.
As the result the
prediction was unable in
time.
ⒸJESEA2019
ⒸTenki.jp
Prediction examples with correct judges
Example 4: East Chiba 2018/7/7 M6.0 SI=5-
Japanese SI= 5- 10 days before: Weekly horizontal anomalies
were concentrated
X
Tokyo
ⒸJESEA2019
ⒸTenki.jp
Prediction examples with correct judges
Example 5: Hokkaido 2018/9/6 M6.7 SI=7
Japanese SI= 7 1 month before: Monthly vertical anomalies
of down tendency
Sapporo
Hokkaido
Sendai
X
6 months before 1 month before
Sapporo
Down
ⒸJESEA2019
Erimo
X
ⒸTenki.jp
Anomaly appeared on the 5th September, 14 hours before EQ!
3:08 2018/9/6
Hokkaido EQ occurred
13:00 2018/9/5
Anomaly appeared
at Erimo, Hokkaido
ⒸJESEA2019
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
20
16
/1/9
20
16
/1/1
62
01
6/1
/23
20
16
/1/3
02
01
6/2
/62
01
6/2
/13
20
16
/2/2
02
01
6/2
/27
20
16
/3/5
20
16
/3/1
22
01
6/3
/19
20
16
/3/2
62
01
6/4
/22
01
6/4
/92
01
6/4
/16
20
16
/4/2
32
01
6/4
/30
20
16
/5/7
201
6/5
/14
20
16
/5/2
12
01
6/5
/28
20
16
/6/4
20
16
/6/1
12
01
6/6
/18
20
16
/6/2
52
01
6/7
/22
01
6/7
/92
01
6/7
/16
20
16
/7/2
32
01
6/7
/30
201
6/8
/62
01
6/8
/13
20
16
/8/2
02
01
6/8
/27
20
16
/9/3
20
16
/9/1
02
01
6/9
/17
20
16
/9/2
42
01
6/1
0/1
20
16
/10
/82
01
6/1
0/1
52
01
6/1
0/2
22
01
6/1
0/2
92
01
6/1
1/5
20
16
/11
/12
20
16
/11
/19
20
16
/11
/26
20
16
/12
/32
01
6/1
2/1
02
01
6/1
2/1
72
01
6/1
2/2
42
01
6/1
2/3
1
3.9cm
6.3cm
B
A
A
B
Prediction examples with correct judges
Example 6: North Ibaragi 2016/12/28 M6.3 SI=6-
It is risky if height difference exceeds 6cm
ⒸJESEA2019
ⒸTenki.jp
Sample of AI based risk map
As of 2018/9/1
X
Hokkaido East
Iburi EQ occurred
on 2018/9/6
(M6.7, SI=7)
in the risky level 4
of AI risk map
East Iburi EQ
ⒸJESEA2019
Ratio of correct prediction of earthquakes
2013-2018
Correct = 29/54= 53.7%
Almost = 17/54= 31.5%
Not correct= 8/54= 14.8%©JESEA2019
Definition
Correct: Within 3 months from anomality, EQ occurred
Almost: Within 6 months from anomaly, EQ occurred
Incorrect: In spite of anomality, EQ did not occur
2013年 2014年 2015年 2016年 2017年 2018年 Total
Correct ○ 3/9=33.3% 5/8=62.5% 5/10=50.0% 6/10=60.0% 5/8=62.5% 5/9=55.5% 29/54=53.7%
Almost △ 5/9=55.5% 2/8=15.0% 4/10=40.0% 4/10=40.0% 2/8=25.0% 0/9=0% 17/54=31.5%
Incorrect × 1/9=11.1% 1/8=12.5% 1/10=10.0% 0/10=0.0% 1/8=12.5% 4/9=44.4% 8/54=14.8%
Part 3: Additional RS &GIS techniques
for prediction
+ New findings for prediction with RS data
# Infrasound sensor: abnormal wave pattern
# Thermometer: pseudo temperature change
# Disturbance of ionosphere: time delay of GNSS
# Atmospheric low pressure: sudden fall
+ My policy
All possible geosphere phenomena should be
scientifically validated on correlation between the
phenomena and occurrence of earthquakes.
Above phenomena have been already validated and
can be used as supplemental tools for prediction. ⒸJESEA2019
Prediction method by infrasound anomaly
Ultra low frequency sound: 0.0004~0.001Hz
Infrasound sensor made in China
Anomaly was recognized
2 days before Kobe EQ
1995/1/17 M7.3, 6400 victims
ⒸBeijing University of Technology
ⒸJESEA2019
Prediction example: Hokkaido EQ with provable judges
About 1 month before About 20 days beforeⒸJESEA2019
Hokkaido EQ: 2018/9/6 M6.7
Prediction example: Kuichi-no-Erabu Volcano
2019/12/18
Volcanic Eruption at
Kuchi-no-Erabu Island Infrasound Anomaly 4 days before
GNSS H Anomaly: 10 days before
Kyushu
H>4cm
ⒸJESEA2019
Prediction method by temperature anomaly
Temperature is measured with platinum resistance sensor
by detecting electric resistivity from weak electric current
Platinum resistance thermometerAnomaly will be recognized
about within a month
before EQ due to EM emissionR
esis
tivit
y
Temperature (K)
Pt
ⒸJESEA2019
0
5
10
15
20
25
30
0:1
0
1:1
0
2:1
0
3:1
0
4:1
0
5:1
0
6:1
0
7:1
0
8:1
0
9:1
0
10
:10
11
:10
12
:10
13
:10
14
:10
15
:10
16
:10
17
:10
18
:10
19:1
0
20
:10
21:1
0
22
:10
23
:10
20180905
日高 日高門別
0
5
10
15
20
25
20
160
411
0
:10
1:3
0
2:5
0
4:1
0
5:3
0
6:5
0
8:1
0
9:3
0
10
:50
12
:10
13
:30
14
:50
16
:10
17
:30
18
:50
20
:10
21
:30
22
:50
阿蘇山 熊本 菊池 甲佐
ⒸJESEA2019
Kumamoto EQ(2016/04/16)
Anomaly detected 5 days before
Hokkaido EQ(2018/9/6)
Anomaly detected 1 day before
Validation example: Kumamoto and Hokkaido EQ
Asosan
HidakaMombetsu
20180411
Japanese patent approved
2016/4/15
1 day before2016/4/11
5 days before2016/4/7
9 days before
2016/4/3
13 days before
MashikiAsosan
X
Kumamoto EQ
(2016/04/16: M7.3)
From 2 weeks before
anomaly is observed
at Mashiki & Asosan
near the epi-center
X X
X
Mashiki
Hokkaido EQ
2018/9/6:M6.7, SI=7
2018/9/5
1 day before
X
Monbetsu
Hidaka
Very new innovation Japanese patent approved
Prediction method by ionospheric anomaly
GNSS
GNSS antenna
EM wave
ⒸJESEA2019
Comparison between the exist and new AI method
ここが赤色曲線からのズレなのか、緑色の曲線の一部なのかを見分けることができるか?
02120700_obs_#15.xlsx
It is hard to detect anomaly
from high order curve
TEC Method AI New Method
with peak curve
This prediction method would be my final goal
to enable early warning a few hours in advance
to the occurrence of huge earthquake.
ⒸJESEA2019
East Japan EQ(2011/03/11)
Anomaly just about 4 hours before
EQ
5 4 3 2 16
Validation example: East Japan Great EQ
0
5 GPS data
at 2 CORS
Kumamoto EQ, Fore shock
M6.5, SI=7 2016.04.14
Kumamoto major EQM7.3, SI=7 2016.04.16
Anomaly before 2
hours and 15min.
Anomaly before 2
hours and 25 minutes
Validation example: Kumamoto EQ
ⒸJESEA2019
Validation of low pressure: East Japan Great EQ
Fore shock(M7.3):2011/3/9: Main shock(M9.0):2011/3/11
Fore shock
Main shock
Atmospheric pressure at near epi-center
Pre
ssure
Fall
Fore
Main
Pre-slip
ⒸJESEA2019
ⒸTenki.jp
Validation of low pressure
Kobe Great EQ
1995/1/17(M7.3)
Pre
ssure
Fall
Pre
ssure
Fall
Atmospheric pressure at near epi-center
January 8-17
Validation of low pressure
Kumamoto EQ
2016/4/16(M7.3)
990
995
1000
1005
1010
1015
1020
1 5 9 131721 1 5 9 131721 1 5 9 131721 1 5 9 131721 1 5 9 131721
Kumamoto
12 16151413
Pre
ssure
Fall
2 days before
4 days before
ⒸJESEA2019
Part 4: Mini-plate theory
+ New findings of mini-plates
# The hint of mini-plate theory was obtained from
the change analysis of Kumamoto EQ occurred
2016/4/16
# Mini-plates was clustered with time sequence
GNSS data variation in 2016
# Mini-plates are more related to occurrence of
EQs rather than faults or geotectonic map
# About 70% of large Eqs are located near
the boundary of mini-plates
# Mini-plates are useful for understanding
geodynamic characteristics of the moving Earth
ⒸJESEA2019
The hint of mini-plate theory from Kumamoto EQ
M7.3 SI=7
・Green/red: rose
・Blue: sank・Arrows: Horizontal
changeⒸJESEA2019
2016.4.16
ⒸTenki.jp
Clustering of mini-plates with GNSS data
GNSS data (XYZ) in 2016
Convert to (NEU)
PCA analysis (PC1,PC2)
Eliminate errors/noises
Select cluster number
Nearest neighbor interpolation for image map
Comparison with geotectonic map
ⒸJESEA2019
Correlation with earthquakes
PCA analysis (PC1,PC2)
Pointwise cluster map
Determine Cluster Number
Clustering of PC1 and PC2 domain
ⒸJESEA2019
Comparison between
Mini-plate map and Faults
ⒸJESEA2019
a) National Atlas: Geotectonic Map b) Clustering Map
Comparison between geotectonic map and mini-plate map
ⒸJESEA2019
Comparison between geological map and mini-plate map
ⒸJESEA2019
Ⓒ地質調査総合センター
Distribution of larger Eqs than SI=5
10 years from 2008 to 2018
Depth of
Epi-center
Less than 15km
15km-30km
More than 30km
Size of marker
SI=7
SI=6+ SI=5-
SI=5+ SI=5-
ⒸJESEA2019ⒸGoogle
Buffering radius= 25km
Boundary check
Near boundary
Not boundary
Does EQ occur near mini-plate boundary?
71% of larger EQs occurred
near mini-plate boundary
ⒸJESEA2019ⒸGoogle
No.dN
N-S
dE
E-W
dU
H
dNEU
Move
azNEU
Azimuth
Hori./Vert.
move
1 -4.01 3.966 0.815 5.698 135.32 SE+: Down - -
2 -2.137 1.361 0.311 2.553 147.5 SE+: Down -
3 1.378 -2.117 0.152 2.531 303.06 NW-: UP/Down -
4 -6.044 6.927 -0.301 9.199 131.1 SE ++: Down -
5 -8.802 9.17 -0.052 12.711 133.83 SE++: UP/Down-
6 -13.336 15.733 0.193 20.626 130.28 SE++:Up/Down -
7 -26.764 27.556 -0.487 38.417 134.17 SE+++* Down - -
8 -25.265 32.97 11.144 43.006 127.46 SE+++: UP+++
Geo-dynamism of mini-plates with GNSS data
ⒸJESEA2019
Naming of Mini-Plates of Japan Name of Mini-Plate
1: Central belt MP
with SE moves/rising
2: Southern central belt MP
with SE moves/little up/down
3: South coast MP
with NW moves/little sinking
4: Northern central belt MP
with SE moves/little up/down
5: Wet to East central belt MP
with SE moves/little up/down
6: West to East northern belt MP
with SE moves/little up/down
7: West Tohoku MP
with big SE moves/sinking
8: East Tohoku MP
with SE big moves /rising
❶
❷
❸
➍
❺
❻
❼
❻❺
➍❶
❷❸
❺
❽
➍❶
❷❺
➍
❸
❸ ❸
❶➍❺
❷
ⒸJESEA2019
Comparison between existing geo-tectonics map
and mini-plate map with GNSS data
Existing
geo-tectonics map
Mini-plate map
Information
Analog
and
Qualitative
Digital
and
Quantitative
Status
Static
and
Time independent
Dynamic
and
Time dependent
Productivity
Professional
and
Not reproductive(geological survey needed)
Reproductive
and
Repeatable
(clustering needed)
Relation
to
Earthquakes
Weakly related Strongly related
ⒸJESEA2019
Conclusions
+ I realized that the main stream of the prediction of
earthquakes should be along with remote sensing but not
seismic science.
+ Geospatial techniques and artificial intelligence are strong
tools to support advanced data analysis and prediction.
+ All possible macroscopic phenomena should be used or
supplemented in order to increase the accuracy.
+ Alert level prediction would be possible with ionospheric
anomalies in a few years after validation and verification are
to be implemented.
+ Mini-plate theory would be an innovation principle for
better understanding and predicting earthquakes.
+ I hope that collaboration with GIC/AIT and UN will contribute
to the reduction of earthquake disasters in Asian region.
ⒸJESEA2019
Thank you for your attentionⒸJESEA2019