Towards the creation of a climate database for Catalonia and Andorra (18th - 21th centuries)

Towards the creation of a climate database for Catalonia and Andorra Towards the creation of a climate database for Catalonia and Andorra (18th - 21th centuries)(18th - 21th centuries)

Marc Prohom (1) and Pere Esteban (2)Marc Prohom (1) and Pere Esteban (2)(1)(1) Area of Climatology – Meteorological Service of CataloniaArea of Climatology – Meteorological Service of Catalonia

(2)(2) Snow and Mountain Research Centre (CENMA) – Andorran Research InstituteSnow and Mountain Research Centre (CENMA) – Andorran Research Institute

The projectThe project

During the last two years, the SMC has initiated a project of Identification, cataloguing and digitization of instrumental climate data from Catalan documentary sources, encompassing the period between 18th century and the present. The final goal is to create a complete and high quality database of climate series. The similar project is being initiated by the Snow and Mountain Research Centre (CENMA) in Andorra for Andorran series.

MethodologyMethodology

1. To analyze the climatic documentary sources, identifying as many as possible meteorological stations or observatories, and the meteorological series associated to these sites.

2. To create a database of the metadata for each of the sites detected: METADEM.

3. To construct a database of the series associated to each one of the meteorological sites (temperature and precipitation): BDSCLIM.

4. Quality control and homogeneity analysis of the series.

To identify new

sources

To cataloguethose sources

Climatic dataBDSCLIM

MetadataMETADEM

Homogeneityanalysis

Qualitycontrol

To extract the climatic

information

METADEMMETADEM

The metadata information of each site detected is introduced into METADEM (Database of Metadata). Only temperature and rainfall information is by now taken into account.

Geographicalinformation

Who is/was in charge?

Observers

Temperature:period covered,screens used,instruments,

units,documentary

sources,...

Documentarysources

Locationdescription(images)

Precipitation:period covered

instruments used, documentary

sources

Additionalinformation

METADEMMETADEM

By now, the SMC and CENMA are paying attention on trying to obtain as much information as possible from those observatories with complete and good temporal coverage series.

Fabra Observatory: location and tipology of the instruments

during the early period (1905-1912)

Engolasters WS (Andorra): location of the screen (2007)

Quality controlQuality control

A quality control process has been defined, according to the bibliography, into four levels: gross errors, tolerance tests, internal consistency, temporal coherency and spatial coherency.

Results of the quality control for the maximum and minimum temperature series of the Ebre

Observatory.

The final homogeneity testing process will be defined following the conclusions of the COST action HOME.

First improvements of the projectFirst improvements of the project

a. New data has been detected and incorporated to the database.

Temporal coverage has been improved: for the period previous the Spanish Civil War, 150 new thermopluviometric series have been identified and 200 series has now a wider temporal coverage.

Number of series per decade before the project (in grey) and after the project (in green). Period 1860-2004.

PRECIPITATION

Daily Monthly

TEMPERATURE

Daily

First improvements of the projectFirst improvements of the project

• Evolution of the spatial coverage

Meteorological sites previous to the project New meteorological sites detected (blue) Meteorological sites that has a wider temporal coverage (green)

Data for HOME’s benchmarkData for HOME’s benchmark

• For the HOME Cost action the SMC and CENMA provides the following datasets:

• Daily temperature and rainfall series from 17 sites encompassing the period 1905-2007 (although most of the series begin in 1920s), and with a good metadata and spatial coverage.

• 13 daily rainfall series from a very dense area (within a same county) encompassing the period 1915-2007.

• CENMA provides 3 daily temperature and rainfall series from Andorra encompassing the period 1934-2007, without any gap.

Analysis of Catalan, Andorran and French temperature series from the early 20th century to the present Analysis of Catalan, Andorran and French temperature series from the early 20th century to the present using different homogenisation approaches *using different homogenisation approaches *

M.J. Prohom (1), P. Esteban(2), M. Herrero(1), O. Mestre (3), E. Aguilar(4), F.G. Kuglitsch (5)

(1) Meteorological Service of Catalonia, Area of Climatology, Barcelona Catalonia, Spain ([email protected])

(2) Snow and Mountain Research Center (CENMA) Andorran Research Institute. St. Julià de Lòria, Principality of Andorra ([email protected])

(3) Météo France, Tolouse, France ([email protected])

(4) Climate Change Research Group, Geography Unit, Universitat Rovira i Virgili, Tarragona, Spain ([email protected])

(5) Climatology and Meteorology Research Group Institute of Geography, University of Bern, Switzerland

* EGU-2008. CL44. Climate data homogenization and climate trend/variability assessment

FrameworkFramework

• Cooperative effort between 5 different institutions

• Homogenization of a multi-country dataset– METEOCAT AND CENMA efforts Metadata & data rescue; qc

• Comparison of different homogenization approaches– SNHT weighted average reference series – Cassinus-Mestre pairwise comparisons – RHTest no references

• Contribution to COST-HOME action benchmark dataset

• Ongoing project

The datasetThe dataset

• 17 Catalan stations• 3 Andorran stations• 11 French stations

(Languedoc-Roussillon)• Although some series go back

to the 1880s, 1921-2006 period was chosen for bettter comparison

Quality control of the datasetQuality control of the dataset

• 17 Catalan + 3 Andorran stations– QC applied to daily data (T)

• Calendar and duplicates; gross errors; statistical limits; tn<= tx; interdiurnal differences; identical consecutive values

• One entire station was dropped from the analysis• Around 12.000 individual values where flagged as suspicious;

around 2/3 where validated and the remaining where corrected from original sources or set to missing

• 11 French stations QC’d monthly values where provided by Météo France

Homogeneity methods: SNHTHomogeneity methods: SNHT

Reference Series

q-series (data-reference)

z-series (standarized q-series)

Original Data

k

jj

k

jjjj

r

YXyr

1

2

1

2 ))((

Reference Series

q-series (data-reference)

z-series (standarized q-series)

Original Data

k

jj

k

jjjj

r

YXyr

1

2

1

2 ))((

Most Probable Breakpoint: Max of 22

21 )( zjnzjT j

Correction Factor:12 qqf

Most Probable Breakpoint: Max of 22

21 )( zjnzjT j

Correction Factor:12 qqf

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

INITIAL PHASE

SESION 2

Series A2Series B2Series C2Series D2Series E2

...Series X2

SESION n

Series AnSeries BnSeries CnSeries DnSeries En

...Series Xn

Iterationuntil no

more breakpoints

are found

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

INITIAL PHASE

SESION 2

Series A2Series B2Series C2Series D2Series E2

...Series X2

SESION n

Series AnSeries BnSeries CnSeries DnSeries En

...Series Xn

Iterationuntil no

more breakpoints

are found

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

FINAL PHASE

An ... Xnare used as references

HOMOGENEOUS SERIESSeries AhSeries BhSeries ChSeries DhSeries Eh

...Series Xh

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

FINAL PHASE

An ... Xnare used as references

HOMOGENEOUS SERIESSeries AhSeries BhSeries ChSeries DhSeries Eh

...Series Xh

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

INITIAL PHASE

SESION 2

Series A2Series B2Series C2Series D2Series E2

...Series X2

SESION n

Series AnSeries BnSeries CnSeries DnSeries En

...Series Xn

Iterationuntil no

more breakpoints

are found

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

INITIAL PHASE

SESION 2

Series A2Series B2Series C2Series D2Series E2

...Series X2

SESION n

Series AnSeries BnSeries CnSeries DnSeries En

...Series Xn

Iterationuntil no

more breakpoints

are found

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

FINAL PHASE

An ... Xnare used as references

HOMOGENEOUS SERIESSeries AhSeries BhSeries ChSeries DhSeries Eh

...Series Xh

SESION 1

Series A1Series B1Series C1Series D1Series E1

...Series X1

FINAL PHASE

An ... Xnare used as references

HOMOGENEOUS SERIESSeries AhSeries BhSeries ChSeries DhSeries Eh

...Series Xh

Automated Software by Enric Aguilar

Homogeneity methods: SNHTHomogeneity methods: SNHT

Automated method by Enric Aguilar

Complete diagram of detectedbreak-points for all the series:

Yearly max. temp, Tx

Yearly min. temp, Tn

Yearly mean temp, Tm

Yearly temperature range, TRG

Homogeneity methods: Cassinus-MestreHomogeneity methods: Cassinus-Mestre

Synthesis of the detected changepoints and outliers in the Lleida Tn series (rawdata): the stations are ordered from top to bottom with respect to decreasing values of the standard errors of the residuals (STD); hence, in practice, the reliability of the comparisons increases from top to bottom ( , position of the detected changepoints in the difference series for Lleida versus the other stations; , outliers, missing years in the difference series). Vertical red stripped lines are the most likely change points.

Software by Olivier Mestre

Homogeneity methods: RHTESTv2Homogeneity methods: RHTESTv2

Regression – based

Can use reference series

Here applied to station data

(without reference series)

Software and documentation available from Xiaolan Wang and Yang Feng at http://cccma.seos.uvic.ca/ETCCDMI/software.shtml

Lleida tn anomaly series (i.e., anomalies to the mean annual cycle of the base series) along with its multi-phase regression model fit.

ProcedureProcedure

• First run of each method for blind detection– Cassinus-Mestre (notice C-M needs human input) annual– SNHT annual– RHTest monthly deseasonalized

• Breakpoint detection and validation – Metadata– Data plots and test plots– Comparison of detected breakpoints– Final breakpoints list

• New runs of C-M, SNHT, and RhTest forced with breakpoints list

Procedure: breakpoints detectionProcedure: breakpoints detection

METADATA

SNHT

C-M

RHTestv2

Final listof breakpoints

Procedure: final list of detected break-pointsProcedure: final list of detected break-points

Final list of break-points according toavailable METADATA and homogeneitytesting results:• Level 1: the three methods detect the break and metadata confirms the finding.• Level 2: two methods detect the break and metadata confirms the finding.• Level 3: at least two methods detect the break, but metadata does not confirm the finding.• Level 4: at least one method detects the break, and metadata confirms the finding.

Comparison of detected break-pointsComparison of detected break-points

TX

0

20

40

60

80

100

120

140

160

SHNT CAME RHTE LIST

TX

TN

NUMBER OF DETECTED BREAKPOINTS

Most of the breakpoints detected by SNHT are also detected by CAME. The opposite is not trueRHTest and SNHT breakpoints are quite different, specially on TX

010

20304050

607080

90100

SHNT CAME RHTE LIST

SHNT

CAME

RHTE

LIST

Good agreement for therecent period (mid-1970s

to the present), whilefor the early period more

random behaviour isreported (there are lessseries available), beingthe RHTest direct (non-

forced) the method showing more

discrepancies (Tx).

Results: aggregated annual seriesResults: aggregated annual series

Results: aggregated seasonal series. TxResults: aggregated seasonal series. Tx

Results: aggregated seasonal series. TnResults: aggregated seasonal series. Tn

Results: differences in trendsResults: differences in trends

ADJUSTED SERIES. AGGREGATED

SNDI -0.037 0.067 0.181 SNDI -0.033 0.068 0.172SNFO -0.041 0.063 0.174 SNFO 0.000 0.105 0.205CMDI -0.036 0.074 0.179 CMDI 0.014 0.122 0.207CMFO -0.010 0.087 0.199 CMFO 0.049 0.154 0.248RHDI 0.072 0.169 0.259 RHDI 0.055 0.150 0.243RHFO -0.056 0.033 0.125 RHFO 0.046 0.148 0.233

SNDI -0.088 0.027 0.133 SNDI -0.094 0.034 0.172SNFO -0.101 0.006 0.109 SNFO -0.077 0.042 0.172CMDI -0.093 0.010 0.114 CMDI -0.072 0.043 0.170CMFO -0.035 0.061 0.158 CMFO -0.028 0.095 0.210RHDI 0.004 0.076 0.150 RHDI 0.040 0.143 0.246RHFO -0.033 0.058 0.139 RHFO -0.094 0.022 0.143

SNDI 0.015 0.099 0.172 SNDI 0.023 0.124 0.241SNFO 0.006 0.090 0.165 SNFO 0.011 0.116 0.232CMDI 0.025 0.107 0.190 CMDI 0.001 0.108 0.210CMFO 0.070 0.156 0.242 CMFO 0.065 0.177 0.282RHDI 0.078 0.144 0.208 RHDI 0.138 0.231 0.320RHFO 0.069 0.139 0.210 RHFO 0.015 0.114 0.203

SNDI -0.064 0.018 0.104 SNDI -0.067 0.029 0.124SNFO -0.045 0.035 0.115 SNFO -0.067 0.025 0.120CMDI -0.039 0.047 0.130 CMDI -0.078 0.020 0.113CMFO 0.026 0.110 0.190 CMFO -0.047 0.046 0.138RHDI 0.030 0.105 0.188 RHDI 0.054 0.139 0.219RHFO 0.008 0.086 0.165 RHFO -0.076 0.015 0.104

SNDI -0.017 0.054 0.126 SNDI -0.021 0.064 0.148SNFO -0.011 0.056 0.122 SNFO -0.015 0.064 0.144CMDI 0.004 0.071 0.136 CMDI -0.017 0.058 0.143CMFO 0.048 0.120 0.185 CMFO 0.023 0.100 0.179RHDI 0.085 0.121 0.161 RHDI 0.129 0.182 0.233RHFO 0.060 0.113 0.161 RHFO -0.012 0.056 0.123

SPRING TMIN

WINTER TMIN WINTER TMAX

SPRING TMAX

ANNUAL TMIN ANNUAL TMAX

AUTUMN TMIN

SUMMER TMIN SUMMER TMAX

AUTUMN TMAX

Similar

Different

Results: individual seasonal series.Results: individual seasonal series.

VILA-SECA: INDIVIDUAL STATIONS SHOW LARGER DIFFERENCES

Correction factors comparison: CM vs SNHTCorrection factors comparison: CM vs SNHT

Flix (CAT): 3 b-p Central (AND): 3 b-p

-0.6

-0.4-0.2

0

0.20.4

0.6

0.8

11.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-1.0-0.8-0.6-0.4-0.20.00.20.40.60.81.01.21.4

J F M A M J J A S O N D

C-M

SNHT

-1.2-1

-0.8-0.6-0.4-0.2

00.20.40.60.8

11.21.4

J F M A M J J A S O N D

C-M

SNHT

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

J F M A M J J A S O N D

C-M

SNHT

-0.6-0.4-0.20.00.20.40.60.81.01.21.41.61.8

J F M A M J J A S O N D

C-M

SNHT

-0.6

-0.4-0.2

0

0.20.4

0.6

0.8

11.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-1.0-0.8-0.6-0.4-0.20.00.20.40.60.81.01.21.4

J F M A M J J A S O N D

C-M

SNHT

Tn Tx

-1.2-1

-0.8-0.6-0.4-0.2

00.20.40.60.8

11.21.4

J F M A M J J A S O N D

C-M

SNHT

Tn Tx

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

J F M A M J J A S O N D

C-M

SNHT

-0.6-0.4-0.20.00.20.40.60.81.01.21.41.61.8

J F M A M J J A S O N D

C-M

SNHT

Flix (CAT): 3 b-p Central (AND): 3 b-p

-0.6

-0.4-0.2

0

0.20.4

0.6

0.8

11.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-1.0-0.8-0.6-0.4-0.20.00.20.40.60.81.01.21.4

J F M A M J J A S O N D

C-M

SNHT

-1.2-1

-0.8-0.6-0.4-0.2

00.20.40.60.8

11.21.4

J F M A M J J A S O N D

C-M

SNHT

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

J F M A M J J A S O N D

C-M

SNHT

-0.6-0.4-0.20.00.20.40.60.81.01.21.41.61.8

J F M A M J J A S O N D

C-M

SNHT

-0.6

-0.4-0.2

0

0.20.4

0.6

0.8

11.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

J F M A M J J A S O N D

C-M

SNHT

-1.0-0.8-0.6-0.4-0.20.00.20.40.60.81.01.21.4

J F M A M J J A S O N D

C-M

SNHT

Tn Tx

-1.2-1

-0.8-0.6-0.4-0.2

00.20.40.60.8

11.21.4

J F M A M J J A S O N D

C-M

SNHT

Tn Tx

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

J F M A M J J A S O N D

C-M

SNHT

-0.6-0.4-0.20.00.20.40.60.81.01.21.41.61.8

J F M A M J J A S O N D

C-M

SNHT

Results: comparison of trends signalResults: comparison of trends signal

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

CMDI-CMFO

CMDI-RHDI

CMDI-RHFO

CMFO-RHDI

CMFO-RHFO

RHDI-RHFO

SNDI-CMDI

SNDI-CMFO

SNDI-RHDI

SNDI-RHFO

SNDI-SNFO

SNFO-CMDI

SNFO-CMFO

SNFO-RHDI

SNFO-RHFO

DIFF

SIGN

EQUAL

DIFFERENCES IN TRENDS.

Percentage of trends sharing the EQUAL sign and

significance; sharing the SIGN of

the estimate; DIFFerent

CONCLUSIONSCONCLUSIONS

- Different homogenization methods (and different applications) can produce different breakpoints and different adjustments, leading to different trends.

- This work has compared 3 widely used methods (SNHT, RHTEST, Caussinus-Mestre)

- Although no assumptions can be made here about which method performs better, we have highlighted obvious differences

- There is a need for extended comparison of results for detection and correction approaches

- There is a need for extending this comparisons to other elements and daily data

- HOME action can contribute to this effort