Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig...

104
Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany

Transcript of Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig...

Page 1: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Milankovic Theory and

Time Series Analysis

Mudelsee M

Institute of MeteorologyUniversity of LeipzigGermany

Page 2: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis

Data (“sample”)

Climate system (“population”, “truth”, “theory”)

Page 3: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis

Data (“sample”) STATISTICS

Climate system (“population”, “truth”, “theory”)

Page 4: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), i = 1, ..., n

Page 5: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), i = 1, ..., nUNI-VARIATE TIME SERIES

Page 6: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nBI-VARIATE TIME SERIES

Page 7: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nTIME SERIES: DYNAMICS

Page 8: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nTIME SERIES: DYNAMICS

[ t(i), x(i), y(i), z(i),..., i = 1 ]TIME SLICE: STATICS

Page 9: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nCLIMATE TIME SERIES

Page 10: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nCLIMATE TIME SERIES

o uneven time spacing*

Page 11: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Low reso lu tion H igh resolu tionIce coreD irect observations,

A rch ive, sam plingD epth

Sedim ent core Sedim ent core

l( i+1)L ( i )

C lim ateAge, T

docum ents,c lim ate m odel

R ecent P ast

Top Bottom

Arch ive, sam plingEstim ated age, t

d ( i+1)D ( i )

A rch ive, tim e series, t( i )Estim ated age, t

D iffusion

Arch ive, tim e series, t'( i )"U psam pling", t'

"D ow nsam pling", t'

Strong in troduced dependence

N oW eak

D ' ( i )

Mudelsee M (in prep.) Statistical Analysis of Climate Time Series: A Bootstrap Approach. Kluwer.

UNEVEN TIME SPACING

Page 12: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

0 2 00 4 00t ( i ) (k a)

0

0 .2

0 .4

d (i )(k a)

0 5 ,0 00 1 0 ,0 00t ( i ) (a B .P .)

1

10

1 00

d (i )(a)

2 ,0 00 6 ,0 00 1 0 ,0 00t ( i ) (a B .P .)

0

10

d (i )(a)

Mudelsee M (in prep.) Statistical Analysis of Climate Time Series: A Bootstrap Approach. Kluwer.

ICE CORE

(Vostok δD)

TREE RINGS

(atmospheric Δ14C)

STALAGMITE

(Qunf Cave δ18O)

UNEVEN TIME SPACING

Page 13: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nCLIMATE TIME SERIES

o uneven time spacingo persistence*

Page 14: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Mudelsee M (in prep.) Statistical Analysis of Climate Time Series: A Bootstrap Approach. Kluwer.

PERSISTENCE

Low reso lu tion H igh resolu tionIce coreD irect observations,

A rch ive, sam plingD epth

Sedim ent core Sedim ent core

l( i+1)L ( i )

C lim ateAge, T

docum ents,c lim ate m odel

R ecent P ast

Top Bottom

Arch ive, sam plingEstim ated age, t

d ( i+1)D ( i )

A rch ive, tim e series, t( i )Estim ated age, t

D iffusion

Arch ive, tim e series, t'( i )"U psam pling", t'

"D ow nsam pling", t'

Strong in troduced dependence

N oW eak

D ' ( i )

Page 15: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

ICE CORE

(Vostok δD)

TREE RINGS

(atmospheric Δ14C)

STALAGMITE

(Qunf cave δ18O)

PERSISTENCE

Mudelsee M (in prep.) Statistical Analysis of Climate Time Series: A Bootstrap Approach. Kluwer.

-2 0 0 2 0 4 0

D ( t( i)) [‰ ]

-2 0

0

2 0

4 0D( t( i - 1 ))[‰ ]

-3 0 0 3 0

14C (t(i)) [‰ ]

-3 0

0

3 014C( t( i - 1 ))[‰ ]

- 1 0 1

18O ( t( i)) [‰ ]

- 1

0

118O(t(i - 1))[‰ ]

Page 16: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate: Statistical analysis:

Time series analysis

Sample: t(i), x(i), y(i), i = 1, ..., nCLIMATE TIME SERIES

o uneven time spacingo persistence

Page 17: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Milankovic theory

Theory: Orbital variations influenceEarth‘s climate.

Page 18: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Milankovic theory

Data: Climate time series

Theory: Orbital variations influenceEarth‘s climate.

Page 19: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Milankovic theory

Data: Climate time seriesTIME SERIES ANALYSIS: TEST

Theory: Orbital variations influenceEarth‘s climate.

Page 20: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Milankovic theory and

time series analysis

Part 1: Spectral analysis

Part 2: Milankovic & paleoclimate —back to the Pliocene

Page 21: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Acknowledgements

Berger A, Berger WH, Grootes P, Haug G, Mangini A, Raymo ME, Sarnthein M, Schulz M, Stattegger K, Tetzlaff G, Tong H, Yao Q, Wunsch C

Page 22: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Alert!

Mudelsee-bias

Page 23: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis

Sample: t(i), x(i), y(i), i = 1, ..., n

Simplification: uni-variate, only x(i),equidistance, t(i) = i

Page 24: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis

Sample: x(t) Time series

Page 25: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis

Sample: x(t) Time series

Population: X(t)

Page 26: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis

Sample: x(t) Time series

Population: X(t) Process

Page 27: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

X(t)TIME DOMAIN

Page 28: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

X(t)TIME DOMAIN

FOURIER TRANSFORMATION: FREQUENCY DOMAIN

Page 29: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

X(t) +T

GT(f) = (2π)–1/2∫–T XT(t) e–2πift dt,

XT= X(t), –T ≤ t ≤ +T,0, otherwise.

TIME DOMAIN

FOURIER TRANSFORMATION: FREQUENCY DOMAIN

Page 30: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

h(f) = limT→∞ [ E {|GT(f)|2 / (2T)} ]NON-NORMALIZED POWER SPECTRAL DENSITY FUNCTION,

“SPECTRUM”

Page 31: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

h(f) = limT→∞ [ E {|GT(f)|2 / (2T)} ]NON-NORMALIZED POWER SPECTRAL DENSITY FUNCTION,

“SPECTRUM”

“ENERGY” (VARIATION) AT SOME FREQUENCY

Page 32: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

Discrete spectrum

Harmonic process

Astronomy

0Fre q u e n cy, f

0

h (f)

0Fre q u e n cy, f

0

h (f)

0Fre q u e n cy, f

0

h (f)

Page 33: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

Discrete spectrum

Harmonic process

Astronomy

0Fre q u e n cy, f

0

h (f)

0Fre q u e n cy, f

0

h (f)

0Fre q u e n cy, f

0

h (f)

Continuous spectrum

Random process

Climatic noise

Page 34: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Process level

Discrete spectrum

Harmonic process

Astronomy

0Fre q u e n cy, f

0

h (f)

0Fre q u e n cy, f

0

h (f)

0Fre q u e n cy, f

0

h (f)

Continuous spectrum

Random process

Climatic noise

Mixed spectrum

Typical climatic

Page 35: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis

The task of spectral analysis is to estimate the spectrum.

There exist many estimation techniques.

Page 36: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Harmonic regression

X(t) = Σk [Ak cos(2πfk t) + Bk sin(2πfk t)] + ε(t)

HARMONIC PROCESS

Page 37: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Harmonic regression

X(t) = Σk [Ak cos(2πfk t) + Bk sin(2πfk t)] + ε(t)

If frequencies fk

known a priori:

Minimize Q = Σi {x(i) – Σk [Ak cos(2πfk t) + Bk sin(2πfk t)]}2

to obtain Ak and Bk.

HARMONIC PROCESS

Page 38: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Harmonic regression

X(t) = Σk [Ak cos(2πfk t) + Bk sin(2πfk t)] + ε(t)

If frequencies fk

known a priori:

Minimize Q = Σi {x(i) – Σk [Ak cos(2πfk t) + Bk sin(2πfk t)]}2

to obtain Ak and Bk.

HARMONIC PROCESS

LEAST SQUARES

Page 39: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

If frequencies fk notknown a priori:

Take least-squaressolutions Ak and Bk, fk = 0, 1/n, 2/n, ..., 1/2,

to calculate P(fk) ~ (Ak)2 + (Bk)2.

Page 40: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

If frequencies fk notknown a priori:

Take least-squaressolutions Ak and Bk, fk = 0, 1/n, 2/n, ..., 1/2,

to calculate P(fk) ~ (Ak)2 + (Bk)2. PERIODOGRAM

Page 41: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

If frequencies fk notknown a priori:

Take least-squaressolutions Ak and Bk, fk = 0, 1/n, 2/n, ..., 1/2,

to calculate P(fk) ~ (Ak)2 + (Bk)2.

Where fk ≈ true f P(fk) has a peak.

PERIODOGRAM

Page 42: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

0 fk

0

P (fk)

1n_ 2

n_ 1

2_

Page 43: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

Original paper:

Schuster A (1898) On the investigation of hidden periodicities with application to a supposed 26 day period ofmeteorological phenomena.Terrestrial Magnetism 3:13–41.

Page 44: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

Hypothesis test (significance of periodogram peaks):

Fisher RA (1929) Tests of significance in harmonic analysis.Proceedings of the Royal Society of London, Series A, 125:54–59.

Page 45: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

A wonderful textbook:

Priestley MB (1981) Spectral Analysis and Time Series.Academic Press, London, 890 pp.

Page 46: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

Major problem with the periodogram as spectrum estimate:

Relative error of P(fk) = 200% for fk= 0, 1/2,

100% otherwise.

Page 47: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

0 fk

0

P (fk)

1n_ 2

n_ 1

2_

Page 48: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Periodogram

“More lives have been lost looking at the raw periodogram

than by any other action involving time series!”

Tukey JW (1980) Can we predict where ‘time series’ should go next? In: Directions in time series analysis (eds Brillinger DR, Tiao GC). Institute of Mathematical Statistics, Hayward, CA, 1–31.

Page 49: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Smoothing

0 fk

0

h

0 fk

0

h

0 fk

0

h

0 fk

0

h

Page 50: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

0 fk

0

h

0 fk

0

h

0 fk

0

h

0 fk

0

h

Page 51: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

0 fk

0

h

0 fk

0

h

0 fk

0

h

0 fk

0

h

0 t(i)

x(i)

1stSegment

2ndSegment

3rdSegment

WELCH OVERLAPPEDSEGMENT AVERAGING(WOSA)

Page 52: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

0 fk

0

h

0 fk

0

h

0 fk

0

h

0 fk

0

h

0 t(i)

x(i)

1stSegment

2ndSegment

3rdSegment

WELCH OVERLAPPEDSEGMENT AVERAGING(WOSA)

ERROR REDUCTION <

√3

Page 53: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Smoothing

Tapering: Weight time series

Spectral leakage reduced

(Hanning, Parzen,triangular windows, etc.)

*

Page 54: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Smoothing problem

Several segments averaged

Spectrum estimate more accurate :-)

Fewer (n‘ < n) data per segment

Lower frequency resolution :-(

Page 55: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Smoothing problem

0 fk

0

h

Page 56: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Smoothing problem

Subjective judgement is unavoidable.

Play with parameters and be honest.

Page 57: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

100-kyr problem

Δt = 1 fk = 0, 1/n, 2/n, ...

Δt = d fk = 0, 1/(n·d), 2/(n ·d), ...Δf = (n·d)–1

Page 58: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

100-kyr problem

Δt = 1 fk = 0, 1/n, 2/n, ...

Δt = d fk = 0, 1/(n·d), 2/(n ·d), ...Δf = (n·d)–1

[ BW > (n·d)–1 SMOOTHING

]

Page 59: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

100-kyr problem

n·d ≈ 650 kyr Δf = (650 kyr)–1*

Page 60: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

100-kyr problem

n·d ≈ 650 kyr Δf = (650 kyr)–1

(100 kyr)–1 ± Δf = (118 kyr)–1 to(87 kyr)–1

*

Page 61: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

100-kyr problem

n·d ≈ 650 kyr Δf = (650 kyr)–1

(100 kyr)–1 ± Δf = (118 kyr)–1 to(87 kyr)–1

[ ± BW wider SMOOTHING

]

*

Page 62: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

100-kyr problem

The 100-kyr cycle existed not long enough to allow a precise enough frequency estimation.

Page 63: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

]

h = Fourier transform of ACV

Page 64: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

E [ X(t) · X(t + lag) ]

h = Fourier transform of ACV

Page 65: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

PROCESS LEVEL E [ X(t) · X(t + lag) ]

h = Fourier transform of ACV

Page 66: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

PROCESS LEVEL E [ X(t) · X(t + lag) ]

h = Fourier transform of ACV

SAMPLE Σ [ x(t) · x(t + lag) ] / n

h = Fourier transform of ACV

Page 67: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

Fast Fourier Transform:

Cooley JW, Tukey JW (1965) An algorithm for the machine calculationof complex Fourier series.Mathematics of Computation 19:297–301.

Page 68: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

Some paleoclimate papers:

Hays JD, Imbrie J, Shackleton NJ (1976) Variations in the Earth's orbit: Pacemaker of the ice ages. Science 194:1121–1132.

Imbrie J Hays JD, Martinson DG, McIntyre A, Mix AC, Morley JJ, PisiasNG, Prell WL, Shackleton NJ (1984) The orbital theory of Pleistocene climate: Support from a revised chronology of themarine δ18O record. In: Milankovitch and Climate (eds Berger A,Imbrie J, Hays J, Kukla G, Saltzman B), Reidel, Dordrecht,269–305.

Page 69: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Blackman–Tukey

Ruddiman WF, Raymo M, McIntyre A (1986) Matuyama 41,000-year cycles: North Atlantic Ocean and northern hemisphere ice sheets. Earth and Planetary Science Letters 80:117–129.

Tiedemann R, Sarnthein M, Shackleton NJ (1994) Astronomic timescale for the Pliocene Atlantic δ18O and dust flux records of Ocean Drilling Program Site 659. Paleoceanography 9:619–638.

Page 70: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Multitaper Method (MTM)

Spectral estimation with optimal tapering

Thomson DJ (1982) Spectrum estimation and harmonic analysis.Proceedings of the IEEE 70:1055–1096.

MINIMAL DEPENDENCE AMONG AVERAGED INDIVIDUAL SPECTRA

MINIMAL ESTIMATION ERROR

Page 71: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Multitaper Method (MTM)

0 500 1000

Age t ( i ) [kyr]

2223242526O bliqu ity

x ( i ) [°]

-0 .08-0 .0400.040.08

Taper va lue

0 500 1000

Age t ( i ) [kyr]

-0 .08

0

0.08Tapered,detrendedx( i ) [°]

0 500 1000

Age t ( i ) [kyr]0 0.02 0.04

Frequency fk [kyr-1 ]

04080120160 M ultitaper

spectrum

k = 0

k = 1

k = 1

Average(k = 0 , 1)

a b

c d (41 kyr)-1

Page 72: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Multitaper Method (MTM)

0 500 1000

Age t ( i ) [kyr]

2223242526O bliqu ity

x ( i ) [°]

-0 .08-0 .0400.040.08

Taper va lue

0 500 1000

Age t ( i ) [kyr]

-0 .08

0

0.08Tapered,detrendedx( i ) [°]

0 500 1000

Age t ( i ) [kyr]0 0.02 0.04

Frequency fk [kyr-1 ]

04080120160 M ultitaper

spectrum

k = 0

k = 1

k = 1

Average(k = 0 , 1)

a b

c d (41 kyr)-1

[ BETTER: DIRECTLY VIA ASTRONOMY EQS.]

Page 73: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Multitaper Method (MTM)

Some paleoclimate papers:

Park J, Herbert TD (1987) Hunting for paleoclimatic periodicities in a geologic time series with an uncertain time scale. Journal ofGeophysical Research 92:14027–14040.

Thomson DJ (1990) Quadratic-inverse spectrum estimates: Applications to palaeoclimatology. Philosophical Transactions of the RoyalSociety of London, Series A 332:539–597.

Berger A, Melice JL, Hinnov L (1991) A strategy for frequency spectra ofQuaternary climate records. Climate Dynamics 5:227–240.

Page 74: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Uneven time spacing

Page 75: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Uneven time spacingUse X(t) = Σk [Ak cos(2πfk t) + Bk sin(2πfk t)] + ε(t)

Lomb NR (1976) Least-squares frequency analysis of unequallyspaced data. Astrophysics and Space Science 39:447–462.

Scargle JD (1982) Studies in astronomical time series analysis. II.Statistical aspects of spectral analysis of unevenly spaceddata. The Astrophysical Journal 263:835–853.

HARMONIC PROCESS

Page 76: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Red noise

0Fre q u e n cy, f

0

h (f) PERSISTENCE

Page 77: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Red noise

AR1 process for uneven spacing:

Robinson PM (1977) Estimation of a time series model from unequally spaced data. Stochastic Processes and their Applications 6:9–24.

0Fre q u e n cy, f

0

h (f) PERSISTENCE

Page 78: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Aliasing

0Fre qu e n cy, f

0

h (f)

12d_

Page 79: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Aliasing

Safeguards: o uneven spacing (Priestley 1981)

o for marine records: bioturbationPestiaux P, Berger A (1984) In: Milankovitch

and Climate, 493–510.

0Fre qu e n cy, f

0

h (f)

12d_

Page 80: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Running window Fourier Transform

0 t(i)

x(i)

Priestley MB (1996) Wavelets and time-dependent spectral analysis.Journal of Time Series Analysis 17:85–103.

Page 81: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Detrending*

Page 82: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Errors in t(i): tuned dating,absolute dating,stratigraphy.

Errors in x(i): measurement error,proxy error,interpolation error.

Page 83: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Bi-variate spectral analysis

For example: x = marine δ18Oy = insolation

Page 84: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Higher-order spectra (bi-spectra, ...)

Page 85: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 1: Spectral analysis:

Further points

Etc., etc.

Page 86: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Page 87: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Less ice /w arm er

M ore ice /co lder

0 1 2 3 4A ge t (i ) [M yr]

5

4

3

2

1 18O [‰ ]benth ic

OD P 659

Page 88: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Less ice /w arm er

M ore ice /co lder

0 1 2 3 4A ge t (i ) [M yr]

5

4

3

2

1 18O [‰ ]benth ic

OD P 659Northern Hemisphere Glaciation

NHG

Page 89: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Less ice /w arm er

M ore ice /co lder

0 1 2 3 4A ge t (i ) [M yr]

5

4

3

2

1 18O [‰ ]benth ic

OD P 659Northern Hemisphere Glaciation

NHG

Mid-Pleistocene Transition

Page 90: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Climate transitions, trend

Age t ( i )

X fit(t) x2

x1

t1 t2

Page 91: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Climate transitions, trend

x1, t < t1,Xfit(t) = x2, t > t2,

x1+ (t−t1) ·(x2−x1)/(t2−t1), t1 ≤ t ≤ t2.

Page 92: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Climate transitions, trend

x1, t < t1,Xfit(t) = x2, t > t2,

x1+ (t−t1) ·(x2−x1)/(t2−t1), t1 ≤ t ≤ t2.

LEAST SQUARES ESTIMATION

Page 93: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Mid-Pleistocene Transition

Less ice /w arm er

M ore ice /co lder

0 0.5 1 1.5A ge t (i ) [M yr]

5

4

3

2 18O [‰ ]benth ic

OD P 659

M IS 23/24

Page 94: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Mid-Pleistocene Transition

Less ice /w arm er

M ore ice /co lder

0 0.5 1 1.5A ge t (i ) [M yr]

5

4

3

2 18O [‰ ]benth ic

OD P 659

M IS 23/24100 kyr cycle

Page 95: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

Mid-Pleistocene Transition

Mudelsee M, Schulz M (1997) Earth and Planetary Science Letters 151:117–123.

DSDP 552DSDP 607ODP 659ODP 677ODP 806

~ size of Barents/Kara Sea ice sheets

Page 96: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Part 2: Milankovic & paleoclimate

NHG

Database: 2–4 Myr, 45 marine δ18O records, 4 temperature records

benthicplanktonic

Mudelsee M, Raymo ME (submitted)

Page 97: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

NHG:

Results

2,000 3,000 4,000

3.0

4.0

18 O

(‰

vs

. P

DB

sta

nd

ard

)

3 .0

4 .0

3.0

4.0

2.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

2,000 3,000 4,000Age (kyr )

2 ,000 3,000 4,000

3.0

4.0

2.0

3.0

2.0

3.0

3.0

4.0

3.0

4.0

-1 .0

0.0

0.0

1.0

-2 .0

-1 .0

0.0

-2 .0

-1 .0

-2 .0

-1 .0

-2 .0

-1 .0

-1 .0

0.0

2,000 3,000 4,000Age (kyr )

Mudelsee & Raymo, Figure 1

606 b G .s.

606 b P .w .

607 b

610 b

659 b

662 b

722 b

758 b

806 b

846 b

849 b

925 b

929 b

o

xo

o

o

o

o

o

oo

o o

xo

x

o

o

o

o o

o

o

o

o

o

o

o

o

o

oo

xx

o

o

o

982 b

999 b

1085 b

1143 b

1148 b

572 p

606 p

625 p

758 p

806 p

851 pG .sac.

999 p

x

x

xx

x

x

x

96–100 M 2–M G 2 96–100 M 2–M G 2

High-resolution recordsMudelsee M, Raymo ME (submitted)

Page 98: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

2,000 3,000 4,000

3.0

4.0

18 O

(‰

vs

. P

DB

sta

nd

ard

)

3 .0

4 .0

3.0

4.0

2.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

3.0

4.0

2,000 3,000 4,000Age (kyr )

2 ,000 3,000 4,000

3.0

4.0

2.0

3.0

2.0

3.0

3.0

4.0

3.0

4.0

-1 .0

0.0

0.0

1.0

-2 .0

-1 .0

0.0

-2 .0

-1 .0

-2 .0

-1 .0

-2 .0

-1 .0

-1 .0

0.0

2,000 3,000 4,000Age (kyr )

Mudelsee & Raymo, Figure 1

606 b G .s.

606 b P .w .

607 b

610 b

659 b

662 b

722 b

758 b

806 b

846 b

849 b

925 b

929 b

o

xo

o

o

o

o

o

oo

o o

xo

x

o

o

o

o o

o

o

o

o

o

o

o

o

o

oo

xx

o

o

o

982 b

999 b

1085 b

1143 b

1148 b

572 p

606 p

625 p

758 p

806 p

851 pG .sac.

999 p

x

x

xx

x

x

x

96–100 M 2–M G 2 96–100 M 2–M G 2

High-resolution records

NHG:

Results

Mudelsee M, Raymo ME (submitted)

Page 99: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

NHG was a slow global climate change (from ~3.6 to 2.4 Myr).

NHG ice volume signal: ~0.4 ‰.

Part 2: Milankovic & paleoclimate

NHG

Page 100: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Milankovic theory and

time series analysis: Conclusions

(1) Spectral analysis estimates thespectrum.

(2) Trend estimation is alsoimportant (climate transitions).

Page 101: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

G O O D I E S

Page 102: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

Climate transitions: error bars

t1, x1, t2, x2

Time series,size n

{t(i), x*(i)}

{t(i), x(i); i = 1,…, n } {t(i)}

Ramp estimation

t1*, x1*, t2*, x2*

Take standard deviation of simulated ramp

parameters

Simulated time series, x*(i) = ramp + noise

Simulated ramp parameters

Bootstrap errors

STD, PersistenceNoise estimation

Repeat 400 times

Page 103: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

NHG amplitudes: temperature2,000 3,000 4,000

20.0

25.0

Tem

pera

ture

(°C

)

1 .0

3 .0

5.0

3.0

5.0

25.0

30.0

2,000 3,000 4,000Age (kyr )

D SDP 572S ST(via ostracoda)

D SDP 607B W T(via M g/C a)

O D P 806B W T(via M g/C a)

O D P 806S ST(via foram s)

96–100 M 2–M G 2

cooling (°C) in ~3,606−2,384 kyr

0.12 ± 0.47

0.62 ± 0.29

1.0 ± 0.5

−0.85 ± 0.17

Mudelsee M, Raymo ME (submitted)

Page 104: Milankovic Theory and Time Series Analysis Mudelsee M Institute of Meteorology University of Leipzig Germany.

NHG amplitudes: ice volume

Temperature calibration: 18OT/T = −0.234 ± 0.003 ‰/°C (Chen 1994; own error determination)

Salinity calibration: 18OS/T = 0.05 ‰/°C (Whitman and Berger 1992)

DSDP 572 p 18OT = 0.03 ± 0.12 ‰ 18OS = −0.01 ‰ 18OI = 0.34 ± 0.13 ‰

DSDP 607 b 18OT = 0.15 ± 0.07 ‰ 18OS = −0.03 ‰ 18OI = 0.41 ± 0.09 ‰

ODP 806 b 18OT = 0.24 ± 0.12 ‰ 18OS = − 0.05 ‰ 18OI = 0.25 ± 0.13 ‰

ODP 806 p 18OT = −0.20 ± 0.04 ‰ 18OS = 0.04 ‰ 18OI = 0.43 ± 0.06 ‰

(DSDP 1085 b cooling by 1 °C 18OI = 0.35 ‰)

Average 18OI = 0.39 ± 0.04 ‰