Frequency analysis of heavy rainfall
and associated synoptic weather
patterns in Kyushu, Japan using self-
organizing map
Koji Nishiyama: Kyushu Univ
Kenji Wakimizu: Kyushu Univ
Cintia Uvo: Lund Univ
Jonas Olsson: SMHI
Background
Period A
568 631 821
YEAR
0
40
80
120
160
1979 1989 1999
Fre
qu
ency
2008
Period B Period C
129 130 131 132
LONGITUDE
31
32
33
34
35
LA
TIT
UD
E
Many kinds of weather patterns make up
decadal trend of heavy rainfall frequency
Kyushu, JapanAll areas in Japan
Recent high frequency
Frequency of R
>= 50mm/h
Year
Fre
quen
cy
What kinds of patterns highly contribute to the formation of
decadal variation of heavy rainfall frequency ??
Complicated !!
The aim of this study
Pattern recognition using the
Self-Organizing Map(SOM)
Heavy rainfall freq for pattern 1
Heavy rainfall freq for pattern 2
Heavy rainfall freq for pattern N
YEAR
0
40
80
120
160
1979 1989 1999
Fre
quen
cy
2008
What kinds of patterns
cause high frequency of
heavy rainfall ??
Main topic
Methodology
U16
P
Q
R
Sample 76
Sample 51
Sample 17
Sample 34
Sample 1
Sample 5
Unit9
Unit32
Unit16
U9
U32Different features
(distant)
Similar features
(close)
Complicated high dimensional data
Visually-understanding patterns in the two dimensional array
‘unit’ means ‘pattern’(1) Reference vector showing a pattern
(2) Samples classified by SOM training
Each unit (pattern)
Self-Organizing Map (SOM) : Kohonen (1995)
are non-linearly classified into
),.....,1())()(),(()( 21 Tttxtxtxt n x
ic (BMU)
i=1 i=6
i=36i=31
)(tim
)()( tmtx i
P
Q
R
(a) Input of sample vector for
SOM training
(b) Determination of BMU( Best Matching Unit )
(c) Modification of reference
vector mi(t) depending on
Neighboring function hci
BMU
smallModifylarge
ttthtt ci iicii mxrrmm ,1
The increase of iteration step smaller modification
ic rr ,thci
Min (i=c(BMU))
SOM training algorithm
x=(PW1~PW16 , U1~U16 , V1~V16)
Input vector for the SOM training
129 130 131 132
LONGITUDE
31
32
33
34
35
LA
TIT
UD
E
NCEP/NCAR
Reanalysis
Synoptic weather for the SOM Rainfall observation
(AMeDAS)
Synoptic weather and rainfall obs area
Linking
1979~2008 (30 years)
(June~September)14648 samples
(4 times per day)
120E 130E 140E 150E20N
30N
40N
50N
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
Pacific Ocean
Japan
Korea•Peninsula
(1)PW
(2)U850
(3)V850
40mm
50mm
Target area
: Kyushu
NCEP reanalysis grid i
Feature Index
Moisture
inflow into
Japan
PW
(Precipitable
Water)
Low level Jet u, v (850hPa)
3 JST 9 JST 15 JST 21JST
24 JST181260
NCEP/NCAR reanalysis
AMeDAS AMeDAS AMeDAS AMeDAS
Event 1 Event 2 Event 3 Event 4
AMeDAS:Automated Meteorological Data Acquisition System
Field 2 Field 3 Field 4Field 1
01 02 03 04 05 06
07 08 09 10 11 12
13 14 15 16 17 18
19 20 21 22 23 24
25 26 27 28 29 30
31 32 33 34 35 36
Unit 1 Unit 30
Unit 900Unit 87130 (x) by 30 (y) patterns
x-axis
y-a
xis
2000 8/31 06UTC
2000 8/31 12UTC
2002 8 /7 12UTC
:
Heavy rainfall frequency
27N
30N
33N
36N
124E 127E 130E 133E
U871All samples for 30
years (14648 fields)
900 patterns
Specification of the SOM structure
SOM training
01 02 0304 05 06
0708
09
10 11 12
13 14
15
16 17 18
19 20 2122 23 24
2526
2728
29 30
31 32 33 34 35 36
y-a
xis
x-axis
K-means
Clustering
36 groups
U900
3 8 1 11 21 2 1 1 2 3 24 36 7 4 1
1 2 4 1 1 12 7 5 1
2 17 5 12 2 3 1 9 3 2 1
2 1 2 2 6 2 10 1 1 5 1 1
6 2 1 7 4 5
3 1 5 1 1 2
8 1 7 3 1 2 1
2 2 1
1 2 2 1 4 3 3 1
1 1 2 1 1 2 4 5 1 1
2 1 1 1 2 1 6 1
1 2 2 1 3 1 1 1 1
1 1 1 3 1
1 1 2 1 1 1 1
2 1 1 1 2
1 1 1 2
1 1 2 4 3 3
1 2 2 4 2 2 7 1 1 6
1 1 3 1 26 17 3 2 4 2 25 33 3 6 1
20 2 1 4 4 10 1 1 3 1 3 3 18 13 2
7 9 7 4 2 2 2 7 2 1 2 3 1 10 1 3 3 5 2 1
5 2 1 2 1 2 2 2 7 8 1 7
6 2 1 1 1 1 1 1 1 1 3 3 7 4 1 1
2 2 7 3 2 7 1 1 1 2 9 5 5 4 4 6 2 3 1 4
5 1 3 34 3 9 2 4 16 6 1 1 1 2 3 1 12 4 4 2
3 8 29 44 20 16 7 8 2 2 5 3 5 5 6 22 2 1 1 1
2 3 8 20 14 14 13 5 8 2 1 2 2 3 16 6 3 13 10 21 3 2 5 2
4 3 15 9 19 21 11 9 13 2 2 1 1 1 1 4 3 11 12 1 5 1 1 2 1 2
3 6 1 11 14 8 5 9 19 14 9 3 2 2 1 8 3 22 6 7 5 1 3
15 4 4 19 13 22 8 29 23 12 27 12 9 4 1 20 61 7 22 11 14 1
Heavy rainfall frequency per each pattern
R ≥ 50mm/h
5 ≤ f < 10
10 ≤ f < 20
20 ≤ f < 100
2 ≤ f < 5
1 ≤ f < 2
G31 G32 G34 G35
G01 G03
G25
G26
G28
G22
Each unit = Each pattern
Frequency
Heavy rainfall frequency per each group
R ≥ 50mm/hPeriod A: 1979-1988, Period B: 1989-1998
Period C: 1999-2008
Heavy rainfall groups (selected top 10 groups )
: 71.6% of all heavy rainfall records of R >= 50mm/h
Group Number Period A Period B Period C Frequency
1 23 25 36 84
2 8 10 17 35
3 16 30 39 85
4 8 22 16 46
5 1 0 1 2
6 0 0 0 0
7 1 16 7 24
8 7 6 9 22
9 7 4 2 13
10 18 2 10 30
11 0 0 0 0
12 0 1 0 1
13 8 7 5 20
14 7 3 4 14
15 3 3 1 7
16 5 8 6 19
17 0 0 3 3
18 0 0 0 0
Group Number Period A Period B Period C Frequency
19 13 4 5 22
20 7 32 35 74
21 5 9 11 25
22 7 39 58 104
23 12 6 12 30
24 1 0 0 1
25 81 67 58 206
26 35 18 23 76
27 19 6 9 34
28 35 43 57 135
29 15 13 27 55
30 2 0 6 8
31 75 59 47 181
32 70 75 183 328
33 24 19 27 70
34 35 63 71 169
35 8 37 34 79
36 12 4 2 18
total 568 631 821 2020
27N
30N
33N
36N
124E 127E 130E 133E
G01
27N
30N
33N
36N
124E 127E 130E 133E
G03
27N
30N
33N
36N
124E 127E 130E 133E
G22
27N
30N
33N
36N
124E 127E 130E 133E
G28
27N
30N
33N
36N
124E 127E 130E 133E
G25
27N
30N
33N
36N
124E 127E 130E 133E
G26
27N
30N
33N
36N
124E 127E 130E 133E
G31
27N
30N
33N
36N
124E 127E 130E 133E
G32
27N
30N
33N
36N
124E 127E 130E 133E
G34
27N
30N
33N
36N
124E 127E 130E 133E
G35
Synoptic weather
patterns constructed by
the SOM
(heavy rainfall
groups :10 groups) plots :
Average reference vector in each group
PW ( Precipitable Water):・An index of convective activity
・large value
(1) strong convective activity
(2) ample water vapor
WIND850 (u, v):・Low Level Jet (LLJ)
・Monsoon
15m/s
10m/s
27N
30N
33N
36N
124E 127E 130E 133E
27N
30N
33N
36N
124E 127E 130E 133E
60mm 40mm
No circle : PW < 30mm
Decadal variation in heavy rainfall weather patterns (10 groups)
0
50
100
150
200
250
300
350
400
450
500
G35
G34
G32
G31
G28
G26
G25
G22
G3
G1
Other groups
Period A Period B Period C
Fre
quen
cy o
f hea
vy r
ainfa
ll p
att
ern
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
Annual
variation in the
frequency of
heavy rainfall
weather pattern
G01 G03
G22
G28
G25
G26
G35
G31 G32 G34
0
20
40
60
80
100
120
140
160
180
G35
G34
G32
G31
G28
G26
G25
G22
G3
G1
Other groups
Decadal variation in heavy rainfall frequency of 10 HR groups
Period A Period B Period C
1993 2006
1999 2007
Fre
quen
cy o
f hea
vy r
ainfa
ll
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
Annual variation
in the frequency
of heavy rainfall
of R>= 50 mm/h
per each group
G01 G03
G22
G28
G25
G26
G35
G31 G32 G34
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
3 8 1 11 21 2 1 1 2 3 24 36 7 4 1
1 2 4 1 1 12 7 5 1
2 17 5 12 2 3 1 9 3 2 1
2 1 2 2 6 2 10 1 1 5 1 1
6 2 1 7 4 5
3 1 5 1 1 2
8 1 7 3 1 2 1
2 2 1
1 2 2 1 4 3 3 1
1 1 2 1 1 2 4 5 1 1
2 1 1 1 2 1 6 1
1 2 2 1 3 1 1 1 1
1 1 1 3 1
1 1 2 1 1 1 1
2 1 1 1 2
1 1 1 2
1 1 2 4 3 3
1 2 2 4 2 2 7 1 1 6
1 1 3 1 26 17 3 2 4 2 25 33 3 6 1
20 2 1 4 4 10 1 1 3 1 3 3 18 13 2
7 9 7 4 2 2 2 7 2 1 2 3 1 10 1 3 3 5 2 1
5 2 1 2 1 2 2 2 7 8 1 7
6 2 1 1 1 1 1 1 1 1 3 3 7 4 1 1
2 2 7 3 2 7 1 1 1 2 9 5 5 4 4 6 2 3 1 4
5 1 3 34 3 9 2 4 16 6 1 1 1 2 3 1 12 4 4 2
3 8 29 44 20 16 7 8 2 2 5 3 5 5 6 22 2 1 1 1
2 3 8 20 14 14 13 5 8 2 1 2 2 3 16 6 3 13 10 21 3 2 5 2
4 3 15 9 19 21 11 9 13 2 2 1 1 1 1 4 3 11 12 1 5 1 1 2 1 2
3 6 1 11 14 8 5 9 19 14 9 3 2 2 1 8 3 22 6 7 5 1 3
15 4 4 19 13 22 8 29 23 12 27 12 9 4 1 20 61 7 22 11 14 1
Unit patterns and heavy rainfall freq (>=50mm/h) in G32
heavy rainfall freq (>=50mm/h)
0
20
40
60
1979 1984 1989 1994 1999 2004
Fre
qu
ency
0
20
40
60
80
1979 1984 1989 1994 1999 2004
Fre
qu
ency
G32
Pattern frequency
Heavy rainfall frequency >= 50mm/h
Heavy rainfall properties of all the patterns included in G32
2006
2006
G32
27N
30N
33N
36N
124E 127E 130E 133E
G32
Notable peak of heavy
rainfall frequency in 2006
Period A Period B Period C
Period A Period B Period C
7570 183
206 166 172
09JST, JULY 22, 200621JST, JULY 5, 200609JST, JULY 2, 2006
03JST, JULY 2, 2006
Weather map with high heavy rainfall frequency in G32
09JST, JULY 30, 200503JST, JULY 6, 2005
U879 freq=17 times U881 freq=9 timesU849 freq=9 times
U848 freq=6 timesU850 freq=9 timesU878 freq=12 times
Conclusion
It was found that the SOM is available for trend analysis of
heavy rainfall frequency linking to weather patterns.
Annual variation in heavy rainfall frequency in Kyushu, Japan
can be divided into the groups of heavy rainfall patterns using
the SOM.
Patterns with Low-level jet and frontal activity affected
annual variation in heavy rainfall frequency.
0
20
40
60
80
100
120
140
160
180
G35
G34
G32
G31
G28
G26
G25
G22
G3
G1
Other groups0
20
40
60
80
100
120
140
160
180
Other groups
27N
30N
33N
36N
124E 127E 130E 133E
G32