Download Issues with the Application of Empirical Mode

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Global analysis of recent frequency
component changes in interannual
climate variability
Murray Peel1 & Tom McMahon1
1 Civil
& Environmental Engineering, The University of Melbourne,
Victoria, Australia
Outline

Background
 Temporal changes in frequency components
– Empirical Mode Decomposition
– Data set
– Results for dividing year = 1970
– Sensitivity of results to the dividing year

Conclusions
EGU 2006 - Session
AS1.07/CL040/CL007
Climate change impact on
climate variability

Mainly assessed at daily, monthly &
seasonal scales
– Changes in extreme event frequency
– Changes in the shape parameter of the daily
frequency distribution

Less attention has been paid to the annual
scale
EGU 2006 - Session
AS1.07/CL040/CL007
Background

Potential modification of interannual
climate variability is important
– Multi-year drought severity
– Reservoir reliability
– Ecosystem dynamics
EGU 2006 - Session
AS1.07/CL040/CL007
Interannual Climate Variability

What drives changes in the mean and
variance of an annual time series?
– Look at the components of a time series using
spectral analysis
Source: IPCC, Climate Change 2001
EGU 2006 - Session
AS1.07/CL040/CL007
Spectral Analysis (EMD)

Empirical Mode Decomposition (EMD)
– Decomposes a time series into
 Intrinsic Mode Function(s) (IMFs)
 A residual (Trend)
– Locally adaptive algorithm
 Robust to non-linear / non-stationary data
– No data pre-processing (like removal of
“trend”)
EGU 2006 - Session
AS1.07/CL040/CL007
Spectral Components

EMD spectral components
– High Frequency (<10 years; Intra-Decadal)
 Sum of IMFs with average period < 10 years
– Low Frequency (>10 years; Inter-Decadal)
 Sum of IMFs with average period >= 10 years + the
residual
– Effectively using EMD as a high/low pass filter
EGU 2006 - Session
AS1.07/CL040/CL007
EMD Example
Robe, South Australia
Annual Precipitation (mm) .
1000
800
600
Observed
High Component
Low Component
400
200
0
-200
-400
1860
1880
1900
1920
1940
1960
Year

Can assess temporal changes in component
behaviour (pre and post a dividing date)
EGU 2006 - Session
AS1.07/CL040/CL007
1980
2000
Data Set

Annual temperature and precipitation data from
the GHCN (version 2)
 Choose 1970 as the dividing year
– To maximise the number and spatial distribution of
stations with >= 15 years of unbroken record pre- and
post- the dividing year (N >= 30)
– Annual temperature

Stations = 1,524, average N = 63 years, ~1930 - 1993
– Annual precipitation

Stations = 2,814, average N = 74 years, ~1920 - 1993)
EGU 2006 - Session
AS1.07/CL040/CL007
Results
Robe Example
VarRatio1970 = Variance>=1970 / Variance<1970
– Obs. = 0.65, High = 0.80, Low = 0.64
1000
Annual Precipitation (mm) .

800
600
Observed
High Component
Low Component
400
200
0
-200
-400
1860
1880
1900
1920
1940
Year
EGU 2006 - Session
AS1.07/CL040/CL007
1960
1980
2000
Results
Temperature – 1,524 Stations
VarRatio1970
Percentiles
% Stations
Obs.
High
Low
5%
0.41
0.45
0.09
50%
0.95
1.04
0.67
95%
1.97
2.08
5.36
>=1
44.3
53.3
38.3
<1
55.7
46.7
61.7
EGU 2006 - Session
AS1.07/CL040/CL007
Temperature
Observed VarRatio1970
EGU 2006 - Session
AS1.07/CL040/CL007
No. Stations (>=1) > (<1)
No. Stations (>=1) = (<1)
No. Stations (>=1) < (<1)
Results
Precipitation – 2,814 Stations
VarRatio1970
Percentiles
% Stations
Obs.
High
Low
5%
0.46
0.45
0.11
50%
0.95
0.98
0.79
95%
2.03
2.21
4.43
>=1
44.8
48.6
40.7
<1
55.2
51.4
59.3
EGU 2006 - Session
AS1.07/CL040/CL007
Precipitation
High Component VarRatio1970
EGU 2006 - Session
AS1.07/CL040/CL007
No. Stations (>=1) > (<1)
No. Stations (>=1) = (<1)
No. Stations (>=1) < (<1)
Sensitivity to dividing year
Temperature
100%
80%
Obs
High
Low
Stations
1440
1280
70%
1120
60%
960
50%
800
40%
640
30%
480
20%
320
10%
160
0%
0
1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985
Dividing Year
EGU 2006 - Session
AS1.07/CL040/CL007
Stations
% of Stations with VarRatio (>=1)
90%
1600
Sensitivity to dividing year
Precipitation
4000
Obs
High
Low
Stations
% of Stations with VarRatio (>=1)
90%
80%
3600
3200
70%
2800
60%
2400
50%
2000
40%
1600
30%
1200
20%
800
10%
400
0%
0
1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985
Dividing Year
EGU 2006 - Session
AS1.07/CL040/CL007
Stations
100%
Conclusions – Temperature

VarRatio1970
– Observed: slight decrease
– High Frequency (intra-decadal): slight increase
– Low Frequency (inter-decadal): large decrease
 Variance moving from low to high frequency
component, over much of the last century
 Decreasing the long-term memory
 Increasing the degree of randomness
EGU 2006 - Session
AS1.07/CL040/CL007
Conclusions – Precipitation

VarRatio1970
– Observed: slight decrease
– High Frequency (intra-decadal): slight decrease
– Low Frequency (inter-decadal): large decrease
 Variance moving from low to high frequency
component, over much of the last century
 Decreasing the long-term memory
 Increasing the degree of randomness
– To a lesser extent than the temperature results
EGU 2006 - Session
AS1.07/CL040/CL007
Overall Conclusions

Recent increase in intra-decadal fluctuations
maybe due to climate change
– Consistent with other research indicating that
the degree of randomness will increase under a
warmer climate

Recent decrease in inter-decadal
fluctuations may reduce the usefulness of
teleconnection based forecasting systems
EGU 2006 - Session
AS1.07/CL040/CL007
Acknowledgements

The analysis presented forms part of a paper
under review at
– Geophysical Research Letters

Funded by
– Australian Research Council Discovery Grant

Useful Discussions
– Geoff Pegram
EGU 2006 - Session
AS1.07/CL040/CL007
Related documents