You are on page 1of 2

Name: Fan Zhang

EAS6490-Final Exam

December 12, 2012

Page 1: (a) Pseudo code & Algebra 1. SSTa = SST - seasonal cycle ! Obtain SST anomaly with respect to the seasonal cycle 2. [U,S,V] = SVD(SSTaSSTa) ! Get EOF of SSTa 3. ! EOF of mode 3 matches the given SST PATTERN 4. = correlation(P C3, T imeSeriesP ointIJ) ! Correlation coecient between PC3 and the time series of each point (i,j) 5. 2 ! Fraction of SST variance (2 ) map explained by EOF 3 Page 1: (b) Pseudo code & Algebra 1. TimeSeriesPointIJ = SLPa(i,j,:) ! Get SLP time series for each point (i,j) 2. = correlation(P C3, T imeSeriesP ointIJ) ! Correlation coecient between PC3 and the time series of each point (i,j) 3. Map the correlation coecient Page 2: (c) Determination of gamma coecient 1. T imeSeriesSST = (SST > 40 N ) ! Build the SST time series using SST whose latitude is more northern than 40N (higher latitude) 2. r(t) = autoCorrelation(T imeSeriesSST ) ! Auto-correlation of T imeSeriesSST at lag-1 3. = 1 r(t) Page 2: (d) Monte Carlo signicance test pseudo code for itime = 1 : 10000 ! Repeat 10000 times W hite = randn(T imeLength, 1) ! Build the random white noise time series, since SLP is considered as a white noise process for iRedTime =1 : TimeLength ! This loop will give a random red noise Redt = r(t) Redt1 + noiset ! Build the red noise time series by using the lag-1 autocorrelation of the real SST time series, since SST is considered as a red noise process end ! end for iRedTime CorrCoecient(itime) = correlation(White, Red) ! Get the correlation coecient between the random white noise and red noise end ! end for itime [PDF percentage,PDFx] = PDF(CorrCoecient) ! Compute the PDF of correlation coecient from 10000 random white and red time series histogram(PDF percentage) ! Histogram of the PDF ! Finally, nd the 95% percentage signicance value

Name: Fan Zhang

EAS6490-Final Exam

December 12, 2012

Page 4: (a) Pseudo code & Algebra for SLPa and SSTa EOFs SSTa = SST - SeasonalCycle; SLPa = SLP - SeasonalCycle ! Obtain the anomalies of SST an SLP with respect to the seasonal cycle SLPa = ObjectiveMapping(SLPa grids to SSTa grids) ! Do objective mapping, to update SLPa matches SSTa spatially; thus, SLPa and SSTa have the same dimensions ! Remove the land points from both SSTa and SLPa [I,J,T] = size (SSTa) SSTa(t,s) = reshape(SSTa,[TimeLength, IJ]) ! Reshape SSTa to a timespace matrix SLPa(t,s) = reshape(SLPa,[TimeLength, IJ]) ! Reshape SLPa to a timespace matrix SSTa(i) = SSTa(:,1)/std(SSTa(:,1)) ! Normalize SSTa point by point. Each point (i) is normalized by the standard deviation of the time series of point (i) SLPa(i) = SLPa(:,1)/std(SLpa(:,1)) ! Normalize SLPa Covariance = SSTaSLPa ! Build the covariance matrix between [U,S,V] = SVD(Covariance) SSTEOF = U; SLPEOF = V ! The EOFs of SST and SLP are in U and V, respectively SSTPC = SSTaU; SLPPC = SLPaV ! PCs of SST and SLP Page 4: (b) DOF & discussion of modes r(t) = autoCorrelation(SST P C(1)) ! Auto-correlation of SST PC1 at lag-1
N N (1r(t)2 ) , (1+r(t)2 ) 2 N

! Get the eective Degree of Freedom, N = 84

! The North rule. Get the errorbar of each eigenvalue in S

The 1st mode is distinct from other modes while the rest errorbars are overlapped, indicating corresponding modes are not independent.

You might also like