ChevalSzentimrey_Seminar2008ok

HOMOGENIZATION OF MONTHLY TEMPERATURE
SERIES IN ROMANIA (1901-2005)
USING METADATA
Sorin CHEVAL*, Tamás SZENTIMREY**, Ancuţa MANEA***
*National Meteorological Administration, Bucharest, Romania and
Euro-Mediterranean Centre for Climate Change, Venice, Italy
** Hungarian Meteorological Service, Budapest, Hungary
***National Meteorological Administration, Bucharest, Romania
Rationale:
• The meteorological network suffered multiple major
changes along the 20th century: transfers between
different administrative or political regimes, relocations,
wars, changing of instruments/shelters etc.
• Previous studies approached the variability of the annual
temperatures in Romania based on raw data, and paid
little attention to homogenization
• No homogenized climatic datasets for Romania
Objectives:
• To investigate the homogeneity level of the monthly
temperature series recorded at weather stations from
Romania in the period 1901-2005
• To test MASH
• Significance of metadata
Meteorological
Network –
Romania 2008
160 weather stations,
most of them cover the
interval 1961-present,
cca. 90 stations
automatic
Data:
Monthly average
temperatures 1901-2005
31 stations Romania +
4 stations Hungary
(homogenized)
Metadata:
-relocation
-turning from manually
operated to automatic
weather stations
30.0
%
25.0
1
2
3
20.0
4
5
6
15.0
7
8
9
10.0
10
11
12
Rate of missing data / month
NY
DE
TU
SE
TR
TM
TJ
SU
SB
RV
RS
RM
PT
PL
OC
MS
IS
GR
GL
DT
CT
CR
CP
CL
BZ
BV
BT
BS
BM
BF
BD
BC
AX
AR
0.0
CC
5.0
Rate of missing data / dataset
2005
2001
1997
1993
1989
1985
1981
1977
1973
1969
1965
1961
1957
1953
1949
1945
1941
1937
1933
1929
1925
1921
1917
1913
1909
1905
1901
missing data/dataset
18
16
14
12
10
8
6
4
2
0
relocations or automatization
8
Temporal distribution
of relocations or
automatization of
stations
7
6
5
4
3
2
1
0
1901
1911
1921
1931
1941
1951
1961
1971
1981
1991
2001
year
CC
CT
SU
2005
Automatization of
stations
RM RS
2004
2003
BC
2002
BM BS
BV
CR
GL
AX
MS OC PL
GR
TJ
PT
RV
2001
AR
BT
BZ
CL
DT
IS
SB
TM
2000
1999
1998
1997
1996
1995
0
5
10
15
20
25
30
• The homogenization of the dataset used the Multiple
Analysis of Series for Homogenization (MASH) v3.02,
and took benefit of the metadata existing in the archives
of the National Meteorological Administration
• The MASH system can use the metadata information –
in particular the probable dates of break points –
automatically
• Homogenized temperature series from Hungary near the
border were also used during the procedure in order to
increase the power.
PROBLEM of HOMOGENIZATION
•
Basis: DATA
•
•
•
•
Tools:
MATHEMATICS : abstract formulation
META DATA : historical, climatological information
SOFTWARE : automatization
•
SOLUTION = MATHEMATICS + META DATA + SOFTWARE
• (i) without SOFTWARE:
MATHEMATICS + META DATA = THEORY WITHOUT BENEFIT
• (ii) without META DATA:
MATHEMATICS + SOFTWARE = GAMBLING
• (iii) without MATHEMATICS:
META DATA + SOFTWARE = “STONE AGE” + “BILL GATES”
BASIC PRINCIPLES of MASH Procedure
•
•
Relative homogeneity test procedure
Step by step procedure: the role of series (candidate or reference
series) changes step by step in the course of the procedure
• Additive or cumulative model can be used depending on the climate
elements
• Monthly, seasonal or annual time series can be homogenized together
(SAM procedure: Seasonal Application of MASH)
• META DATA (probable dates of break points) can be used
automatically
• The actual or the final stage of the homogenization can be verified.
Methodology:
• Homogenized for 0.05 level of confidence
• Three homogenization instances:
1. no metadata considered
2. metadata1: relocation of the stations
3. metadata2: relocation of the stations and
automatization
I. TEST STATISTICS FOR SERIES INHOMOGENEITY
Null hypothesis: the examined series are homogeneous.
Critical value (significance level 0.05): 21.73
Test statistics (TS) is compared to the critical value.
The larger TS values are more suspicious!
Series
Index
TSA
Series
AR
1
114.37 BC
RM
24
51.86 PT
BM
6
44.91 CP
PL
22
33.97 BF
CC
11
27.78 GL
CR
14
25.14 CT
TR
31
24.51 RV
BZ
10
23.09 DE
BS
7
19.32 NY
MS
20
17.77 TJ
BD
4
16.22 SU
SB
27
12.99 TU
AVERAGE:
31.07
Series
AR
PL
AX
RM
TM
BM
CT
DT
OC
SE
BD
SB
Index
TSB
1
393.26
22
164.04
2
128.02
24
87.87
30
68.43
6
54.23
15
33.94
16
25.85
21
22.27
32
19.39
4
16.53
27
13.82
AVERAGE:
72.52
Series
TR
BC
PT
TJ
BZ
RV
BS
BV
DE
MS
IS
TU
Index
TSA
3
23
13
5
17
15
26
34
35
29
28
33
Index
92.04
51.44
38.08
30.19
26.91
25.12
23.85
20.97
18.41
17.09
15.79
8.17
Series
OC
GR
RS
TM
CL
AX
BV
SE
BT
IS
DT
TSB
31
3
23
29
10
26
7
9
34
20
19
33
202.2
160.51
108.26
84.17
65.1
53.48
31.67
24.64
20.93
18.94
16.44
8.17
Series
RS
CC
BF
CL
CP
GR
GL
CR
BT
NY
SU
Index
TSA
21
18
25
30
12
2
9
32
8
19
16
Index
25
11
5
12
13
18
17
14
8
35
28
52.33
46.2
34.89
29.71
26.88
24.86
23.76
19.36
18.4
16.44
14.62
TSB
182.79
147.64
94.98
79.31
63.13
39.98
30.28
24.21
19.48
18.41
15.79
After homogenization
Before homogenization
Evaluation of metadata
IV. TEST STATISTICS
Null hypothesis: the inhomogeneities can be explained by the Meta Data.
Critical value (significance level 0.05): 21.73
Test statistics (TSM) can be compared to the critical value.
The larger TSM values are more suspicious!
Series
Index
TSM
Series
RS
25
182.79 BC
CC
11
116.77 BF
AR
1
87.28 RM
CL
12
80.74 AX
BZ
10
46.59 CP
RV
26
33.14 TM
DT
16
24.47 BM
BS
7
23.35 OC
GL
17
19.92 DE
MS
20
18.94 NY
BT
8
15.5 BD
SB
27
13.36 TU
AVERAGE:
51.81
Index
3
5
24
2
13
30
6
21
34
35
4
33
TSM
160.51
94.65
83.45
76.55
45.01
30.32
24.46
22.27
19.61
18.17
15.12
8.17
Series
TR
PL
TJ
PT
GR
CT
BV
CR
SE
IS
SU
Index
31
22
29
23
18
15
9
14
32
19
28
TSM
157.13
91.52
83.29
57.14
39.98
28.86
24.12
20.07
19.16
16.44
14.53
-2
-4
-6
-8
-10
-12
2005
2001
1997
1993
1989
1985
1981
1977
1973
1969
1965
1961
1957
1953
1949
1945
1941
1937
1933
1929
1925
1921
1917
1913
1909
1905
1901
8
6
Arad - January
4
2
0
No homogenization
No metadata
Metadata1
Metadata2
ESTIMATED BREAK POINTS AND SHIFTS
AR:
1904: -0.04/
1914: -0.02/
1926: 0.06/
1937: 0.01/
1952: 0.01/
1972: 0.08/
1999: -0.01/
1905: -0.08/
1919: 0.01/
1927: 0.12/
1939: 0.02/
1953: 0.03/
1974: 0.03/
2001: -0.03
1908: 0.09/
1922: -0.03/
1928: 0.07/
1940: -0.08/
1958: -0.01/
1975: 0.02/
1909: 0.15/
1923: -0.06/
1929: 0.02/
1941: 0.12/
1964: 0.03/
1983: 0.01/
1910:
1924:
1935:
1944:
1966:
1987:
0.13/
0.06/
0.04/
0.02/
0.01/
0.02/
ESTIMATED BREAK POINTS AND SHIFTS
(Mark M: META DATA 1)
AR:
1904: -0.03/ 1905: -0.04/ 1908: 0.08/ 1909: 0.15/ 1910: 0.18/
1921: -0.01/ 1922: -0.04/ 1923: -0.10/ 1924: 0.05/ 1926: 0.05/
1927: 0.10/ 1928: 0.13/ M1936: 0.04/ 1940: 0.03/ M1951: 0.03/
M1972: 0.07/ 1999: -0.02/ 2001: -0.03
ESTIMATED BREAK POINTS AND SHIFTS
(Mark M: META DATA 2)
AR:
1904: -0.03/ 1905: -0.04/ 1908: 0.08/ 1909: 0.15/ 1910: 0.18/
1921: -0.01/ 1922: -0.04/ 1923: -0.10/ 1924: 0.04/ 1926: 0.06/
1927: 0.10/ 1928: 0.13/ M1936: 0.04/ 1940: 0.03/ M1951: 0.03/
M1972: 0.06/ M1999: -0.09/ 2000: 0.04
year
CC
CT
SU
2005
RM RS
2004
2003
BC
2002
BM BS
BV
CR
GL
AX
MS OC PL
GR
TJ
PT
RV
2001
AR
BT
BZ
CL
DT
IS
SB
TM
2000
1999
1998
1997
1996
1995
0
5
10
15
20
25
30
Concluding remarks and further developments
• A major part of the long-term monthly temperature
variations in Romania is caused by natural factors, but in
some cases the changing in the weather stations
location and/or automatization have induced artificial
leaps in the series
• Metadata = very useful, but they cannot explain
everything!
• More metadata (very possible)
• Efforts for series reconstruction (carefully)
• Bordering problems (possible)
• More methods (very possible)