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