TR32 time series comparison Victor Venema
Content Jan Schween –Wind game: measurement and synthetic –Temporal resolution of 0.1 seconds Heye Bogena –Wind, air pressure, water temperature –Temporal resolution of 10 minutes –Rollesbroich Global Runoff Data Centre –Runoff Rhine Cologne –Daily, years: 1817 to 2001
Wind - Measurement and synthetic
Wind - distribution – normal plot
Increment distribution Measurement: (t) Increment time series for lag l: (x,l) = (t+l) - (t) Distribution jumps sizes Width of the distribution is the mean variance at scale l
Wind - Increment distribution
Daubechies wavelet family
Wind - Daubechies wavelet (db6)
Wind – Haar vs. Daubechies (db6)
Intermittency / Intermittence On-off intermittency –Rain, eddy in laminar flow Operationalisation: variance of variance (at a certain scale) Intermittence is typically strongest at small scales Time series modelling: Autoregressive conditional heteroskedasticity (ARCH, GARCH) Multi-fractal models (not all)
Wind - Increment distribution
Structure functions Increment time series: (x,l)= (t+l)- (t) SF(l,q) = (1/N) Σ | | q SF(l,2) is equivalent to auto-correlation function Correlated time series SF increases with l Higher q focuses on larger jumps For large l, SF equivalent to the moments
Wind – Structure functions
Fourier decomposition Decompose a time domain signal in sinuses of varying wavelength Wavelength -> scale Fourier coefficients -> variance as function of scale
Wind – power spectrum
Wind speed (Heye Bogena; 10 min.)
Wavelet - Wind speed (10 min.)
Air pressure (10 min.)
Air pressure (10 min.) - Wavelets
Water temperature
Water temperature - Wavelets
Discharge all data and zoom
Discharge Rhine - Wavelets
Discharge Rhine
Slope power spectrum vs. smoothness
Conclusions Some signals showed annual, diurnal cycle Except for this no frequency was special –Variability on all scales –Large scales: white noise or even correlated variance is never gone All signals showed intermittence –Typical for complex systems