P ROBLEMS IN DETECTING TREND IN HYDROMETEOROLOGICAL SERIES FOR CLIMATE CHANGE STUDIES Jasna Plavšić 1 and Zoran Obušković 2 1 University of Belgrade –

PROBLEMS IN DETECTING PROBLEMS IN DETECTING TREND IN TREND IN

HYDROMETEOROLOGICAL HYDROMETEOROLOGICAL SERIES FOR CLIMATE SERIES FOR CLIMATE

CHANGE STUDIESCHANGE STUDIES

Jasna Plavšić1 and Zoran Obušković2

1University of Belgrade – Faculty of Civil Engineering2Energoproject – Hydroengineering

16. naučno savetovanje SDHI/SDH, 22-23. oktobar 2012, Donji Milanovac

Climate changeClimate change

• Global warming and increased concentrations of greenhouse gases

Hansen et al, Proc. Natl. Acad. Sci., (2006)

Copenhagen Diagnosis (2009)

Climate change – we knowClimate change – we know

Radionica - Klimatske Promene - 2010 www.slobodansimonovic.com

Copenhagen Diagnosis (2009)

Climate change – we knowClimate change – we know

Radionica - Klimatske Promene - 2010 www.slobodansimonovic.com

Climate change – we knowClimate change – we know

Radionica - Klimatske Promene - 2010 www.slobodansimonovic.com

Church and White , Geophysical Research Letters, (2006)Cazenave et al, Global and Planetary Change, (2009)

Climate change impactsClimate change impacts

• Questions:– Change projections?– Impact on water

resources?

IPCC (2007)

Impact of climate change on water Impact of climate change on water resourcesresources

Estimation of climate change impacts

Future climate scenarios + hydrologic

models

Statistical trends

fairly complicated approach; propagation of uncertainty

simple calculations; but:

How to prove presence of a trend? How to interpret the trend?

Trend detectionTrend detection

• Starting point: hydrometeorological series are considered stationary– stationarity is well defined and departures from

stationary indicate changes• Trend detection vs. identification of non-stationarities– trend in mean is just one type of non-stationarities– false trend detection in time series where other non-

stationarities are present• slow changes (long memory) can look like trend when observed

in shorter periods– significance of trends can decrease in series with long

memory and high serial correlation

Practical aspects of trend analysis – Practical aspects of trend analysis – choice of variableschoice of variables• Runoff– mean flows, floods, low

flows – annual and monthly

values– time of occurrence of

annual maximum flood– ice start and end dates,

number of days with ice

• Precipitation– annual and monthly

precipitation– daily precipitation

annual maxima– number of rainy days– etc.

Practical aspects of trend analysis – Practical aspects of trend analysis – choice of stationschoice of stations• Trend analysis is valid if performed on adequate

series– time series should be long enough for reliable

statistical analysis• WMO recommends 30-year statistics for describing climate

(eg. standard climatological period 1961-1990)• series used for analysis of change in climate should be

much longer than 30 years– series should reflect natural flow regime with no

human interventions within the basin– data from a station should be checked for accuracy

and consistency (rating curves etc.)

Tests for trendTests for trend

• Linear regression: X = a + bt

slope significance?

Tests for trendTests for trend

• Non-parametric tests– data need not be drawn from a (normal) distribution– some test assume data independence

• Most popular: Mann-Kendall test

– H0: no monotonic decreasing or increasing trend

– H0 is rejected when S significantly departs from 0

– serial correlation decreases detection power

Other test for detecting changes in Other test for detecting changes in time seriestime series

Tests for change in the mean Z-test, t-test, Pettitte test

Tests for change in variance F-test

Tests for change in distribution

Mann-Whitney, Kolmogorov-Smirnov

Tests for randomness Run test

Tests for serial correlation Bartlett’s test

Tests for trend Mann-Kendall, Spearman rho, linear regression slope

ExampleExample

• Runoff, precipitation and temperatures in the Drina Basin – Brodarevo/Lim– Drina/Radalj

Energoprojekt- Hidroinženjering

2011, 2012

ExampleExample• Precipitation and runoff cycles– cumulative standardized deviation from the

mean

ExampleExample

• Runoff– no

significant trend

MEAN ANNUAL FLOWS

ANNUAL MAXIMUM FLOODS

LOW FLOWS (annual minimum monthly flows)

Radalj

ExampleExample

• Runoff– Significant

decreasing trend in mean annual flow

MEAN ANNUAL FLOWS

ANNUAL MAXIMUM FLOODS

LOW FLOWS (annual minimum monthly flows)

Brodarevo

ExampleExample

• Temperatures– 8 met.

stations

Results of trend Results of trend analysisanalysis• Temperatures– change in 2035

– in accordance with other studies

BeraneKolašin

Bijelo Polje

Brodarevo

Prijepolje

Bajina Bašta

Loznica

Radalj

Zlatibor

Pljevlja

Žabljak

1.5

1.5

1.3

1.1

1.0

1.01.3

1.5

ExampleExample

• Precipitation– 10 stations

Results of trend Results of trend analysisanalysis• Precipitation:– % change in 2035

– other studies: absence of trend or weak increasing or decreasing trends

– change in seasonal distribution of precipitation, with opposite tendencies for summer and winter seasons Berane

Kolašin

Bijelo Polje

Brodarevo

Prijepolje

Bajina Bašta

Loznica

Radalj

Zlatibor

Pljevlja

Žabljak

14.7%

-16.5%

22.7%

-0.6%

5.9%0.4%

15.2%

9.9%5.0%

6.5%

ConclusionsConclusions

• Trend detection – problems:– Series of different lengths can exhibit different,

even opposite, trends– Spatial inconsistency of the stations are

considered separately– Presence of non-stationarities makes trend

detection more difficult– Opposite changes in different seasons result in

insignificant changes at annual level

ConclusionsConclusions

• River basins with heavily modified flow regime (such as reservoirs) require detailed and careful analysis based on climate and hydrologic modelling with consideration of water management practices

THANKS FOR THANKS FOR ATTENTIONATTENTION