(3) Spectral estimation and xy-plots

In this example we will show how to use the GMT programs fitcircle, project, sample1d, spectrum1d, psxy, and pstext. Suppose you have (lon, lat, gravity) along a satellite track in a file called sat.xyg, and (lon, lat, gravity) along a ship track in a file called ship.xyg. You want to make a cross-spectral analysis of these data. First, you will have to get the two data sets into equidistantly sampled time-series form. To do this, it will be convenient to project these along the great circle that best fits the sat track. We must use fitcircle to find this great circle and choose the L2 estimates of best pole. We project the data using project to find out what their ranges are in the projected coordinate. The gmtinfo utility will report the minimum and maximum values for multi-column ASCII tables. Use this information to select the range of the projected distance coordinate they have in common. The script prompts you for that information after reporting the values. We decide to make a file of equidistant sampling points spaced 1 km apart from -1167 to +1169, and use the UNIX utility awk to accomplish this step. We can then resample the projected data, and carry out the cross-spectral calculations, assuming that the ship is the input and the satellite is the output data. There are several intermediate steps that produce helpful plots showing the effect of the various processing steps (example_03[a-f].ps), while the final plot example_03.ps shows the ship and sat power in one diagram and the coherency on another diagram, both on the same page. Note the extended use of pstext and psxy to put labels and legends directly on the plots. For that purpose we often use -Jx1i and specify positions in inches directly. Thus, the complete automated script reads:

# Purpose:	Resample track data, do spectral analysis, and plot
# GMT modules:	filter1d, fitcircle, gmtconvert, gmtinfo, project, sample1d
# GMT modules:	spectrum1d, trend1d, pshistogram, psxy, pstext
# Unix progs:	echo, rm
# This example begins with data files "ship.xyg" and "sat.xyg" which
# are measurements of a quantity "g" (a "gravity anomaly" which is an
# anomalous increase or decrease in the magnitude of the acceleration
# of gravity at sea level).  g is measured at a sequence of points "x,y"
# which in this case are "longitude,latitude".  The "sat.xyg" data were
# obtained by a satellite and the sequence of points lies almost along
# a great circle.  The "ship.xyg" data were obtained by a ship which
# tried to follow the satellite's path but deviated from it in places.
# Thus the two data sets are not measured at the same of points,
# and we use various GMT tools to facilitate their comparison.
# The main illustration (example_03.ps) are accompanied with 5 support
# plots (03a-f) showing data distributions and various intermediate steps.
# First, we use "gmt fitcircle" to find the parameters of a great circle
# most closely fitting the x,y points in "sat.xyg":
gmt set GMT_FFT kiss
cpos=`gmt fitcircle sat.xyg -L2 -Fm --IO_COL_SEPARATOR=/`
ppos=`gmt fitcircle sat.xyg -L2 -Fn --IO_COL_SEPARATOR=/`
# Now we use "gmt project" to project the data in both sat.xyg and ship.xyg
# into data.pg, where g is the same and p is the oblique longitude around
# the great circle.  We use -Q to get the p distance in kilometers, and -S
# to sort the output into increasing p values.
gmt project  sat.xyg -C$cpos -T$ppos -S -Fpz -Q > sat.pg
gmt project ship.xyg -C$cpos -T$ppos -S -Fpz -Q > ship.pg
# The gmtinfo utility will report the minimum and maximum values for all columns. 
# We use this information first with a large -I value to find the appropriate -R
# to use to plot the .pg data. 
R=`gmt info -I100/25 sat.pg ship.pg`
gmt psxy $R -UL/-1.75i/-1.25i/"Example 3a in Cookbook" -BWeSn \
	-Bxa500f100+l"Distance along great circle" -Bya100f25+l"Gravity anomaly (mGal)" \
	-JX8i/5i -X2i -Y1.5i -K -Wthick sat.pg > example_03a.ps
gmt psxy -R -JX -O -Sp0.03i ship.pg >> example_03a.ps
# From this plot we see that the ship data have some "spikes" and also greatly
# differ from the satellite data at a point about p ~= +250 km, where both of
# them show a very large anomaly.
# To facilitate comparison of the two with a cross-spectral analysis using "gmt spectrum1d",
# we resample both data sets at intervals of 1 km.  First we find out how the data are
# typically spaced using gmtmath DIFF to get the delta-p between points and view it with 
# "gmt pshistogram".
gmt math ship.pg -T -i0 DIFF = | gmt pshistogram  -W0.1 -Gblack \
	-JX3i -K -X2i -Y1.5i -B0 -B+t"Ship" -UL/-1.75i/-1.25i/"Example 3b in Cookbook" \
	> example_03b.ps
gmt math sat.pg -T -i0 DIFF = | gmt pshistogram  -W0.1 -Gblack \
	-JX3i -O -X5i -B0 -B+t"Sat" >> example_03b.ps
# This experience shows that the satellite values are spaced fairly evenly, with
# delta-p between 3.222 and 3.418.  The ship values are spaced quite unevenly, with
# delta-p between 0.095 and 9.017.  This means that when we want 1 km even sampling,
# we can use "gmt sample1d" to interpolate the sat data, but the same procedure applied
# to the ship data could alias information at shorter wavelengths.  So we have to use
# "gmt filter1d" to resample the ship data.  Also, since we observed spikes in the ship
# data, we use a median filter to clean up the ship values.  We will want to use "paste"
# to put the two sampled data sets together, so they must start and end at the same
# point, without NaNs.  So we want to get a starting and ending point which works for
# both of them.  This is a job for gmt info -L -Af.
bounds=`gmt info ship.pg sat.pg -I1 -Af -L -C -i0  --IO_COL_SEPARATOR=/`
# Now we can use $bounds in gmt math to make a sampling points file for gmt sample1d:
gmt math -T$bounds/1 -N1/0 T = samp.x
# Now we can resample the gmt projected satellite data:
gmt sample1d sat.pg -Nsamp.x > samp_sat.pg
# For reasons above, we use gmt filter1d to pre-treat the ship data.  We also need to sample
# it because of the gaps > 1 km we found.  So we use gmt filter1d | gmt sample1d.  We also
# use the -E on gmt filter1d to use the data all the way out to bounds :
gmt filter1d ship.pg -Fm1 -T$bounds/1 -E | gmt sample1d -Nsamp.x > samp_ship.pg
# Now we plot them again to see if we have done the right thing:
gmt psxy $R -JX8i/5i -X2i -Y1.5i -K -Wthick samp_sat.pg \
	-Bxa500f100+l"Distance along great circle" -Bya100f25+l"Gravity anomaly (mGal)" \
	-BWeSn -UL/-1.75i/-1.25i/"Example 3c in Cookbook" > example_03c.ps
gmt psxy -R -JX -O -Sp0.03i samp_ship.pg >> example_03c.ps
# Now to do the cross-spectra, assuming that the ship is the input and the sat is the output 
# data, we do this:
gmt convert -A samp_ship.pg samp_sat.pg -o1,3 | gmt spectrum1d -S256 -D1 -W -C -T
# Now we want to plot the spectra. The following commands will plot the ship and sat 
# power in one diagram and the coherency on another diagram, both on the same page.  
# We end by adding a map legends and some labels on the plots.
# For that purpose we often use -Jx1i and specify positions in inches directly:
gmt psxy spectrum.coh -Bxa1f3p+l"Wavelength (km)" -Bya0.25f0.05+l"Coherency@+2@+" \
	-BWeSn+g240/255/240 -JX-4il/3.75i -R1/1000/0/1 -P -K -X2.5i -Sc0.07i -Gpurple \
	-Ey+p0.5p -Y1.5i > $ps
echo "Coherency@+2@+" | gmt pstext -R -J -F+cTR+f18p,Helvetica-Bold -Dj0.1i \
	-O -K >> $ps
gmt psxy spectrum.xpower -Bxa1f3p -Bya1f3p+l"Power (mGal@+2@+km)" \
	-BWeSn+t"Ship and Satellite Gravity"+g240/255/240 \
	-Gred -ST0.07i -O -R1/1000/0.1/10000 -JX-4il/3.75il -Y4.2i -K -Ey+p0.5p >> $ps
gmt psxy spectrum.ypower -R -JX -O -K -Gblue -Sc0.07i -Ey+p0.5p >> $ps
echo "Input Power" | gmt pstext -R0/4/0/3.75 -Jx1i -F+cTR+f18p,Helvetica-Bold -Dj0.1i -O -K >> $ps
gmt pslegend -R -J -O -DjBL+w1.2i+o0.25i -F+gwhite+pthicker --FONT_ANNOT_PRIMARY=14p,Helvetica-Bold << EOF >> $ps
S 0.1i T 0.07i red - 0.3i Ship
S 0.1i c 0.07i blue - 0.3i Satellite
# Now we wonder if removing that large feature at 250 km would make any difference.
# We could throw away a section of data with awk or sed or head and tail, but we
# demonstrate the use of "gmt trend1d" to identify outliers instead.  We will fit a
# straight line to the samp_ship.pg data by an iteratively-reweighted method and
# save the weights on output.  Then we will plot the weights and see how things
# look:
gmt trend1d -Fxw -Np1+r samp_ship.pg > samp_ship.xw
gmt psxy $R -JX8i/4i -X2i -Y1.5i -K -Sp0.03i \
	-Bxa500f100+l"Distance along great circle" -Bya100f25+l"Gravity anomaly (mGal)" \
	-BWeSn -UL/-1.75i/-1.25i/"Example 3d in Cookbook" samp_ship.pg > example_03d.ps
R=`gmt info samp_ship.xw -I100/1.1`
gmt psxy $R -JX8i/1.1i -O -Y4.25i -Bxf100 -Bya0.5f0.1+l"Weight" -BWesn -Sp0.03i \
	samp_ship.xw >> example_03d.ps
# From this we see that we might want to throw away values where w < 0.6.  So we try that,
# and this time we also use gmt trend1d to return the residual from the model fit (the 
# de-trended data):
gmt trend1d -Fxrw -Np1+r samp_ship.pg | gmt select -Z0/0.6 -o0,1 -Iz \
	| gmt sample1d -Nsamp.x > samp2_ship.pg
gmt trend1d -Fxrw -Np1+r samp_sat.pg  | gmt select -Z0/0.6 -o0,1 -Iz \
	| gmt sample1d -Nsamp.x > samp2_sat.pg
# We plot these to see how they look:
R=`gmt info -I100/25 samp2_sat.pg samp2_ship.pg`
gmt psxy $R -JX8i/5i -X2i -Y1.5i -K -Wthick \
	-Bxa500f100+l"Distance along great circle" -Bya50f25+l"Gravity anomaly (mGal)" \
	-BWeSn -UL/-1.75i/-1.25i/"Example 3e in Cookbook" samp2_sat.pg > example_03e.ps
gmt psxy -R -JX -O -Sp0.03i samp2_ship.pg >> example_03e.ps
# Now we do the cross-spectral analysis again.  Comparing this plot (example_03e.ps) with
# the previous one (example_03d.ps) we see that throwing out the large feature has reduced
# the power in both data sets and reduced the coherency at wavelengths between 20--60 km.
gmt convert -A samp2_ship.pg samp2_sat.pg -o1,3 | gmt spectrum1d -S256 -D1 -W -C -T
gmt psxy spectrum.coh -Bxa1f3p+l"Wavelength (km)" -Bya0.25f0.05+l"Coherency@+2@+" -BWeSn \
	-JX-4il/3.75i -R1/1000/0/1 -UL/-2.25i/-1.25i/"Example 3f in Cookbook" -P -K -X2.5i \
	-Sc0.07i -Gblack -Ey+p0.5p -Y1.5i > example_03f.ps
echo "Coherency@+2@+" | gmt pstext -R -J -F+cTR+f18p,Helvetica-Bold -Dj0.1i \
	-O -K -Wthicker -C0.1i >> example_03f.ps
gmt psxy -Bxa1f3p -Bya1f3p+l"Power (mGal@+2@+km)" -BWeSn+t"Ship and Satellite Gravity" \
	spectrum.xpower -ST0.07i -O -R1/1000/0.1/10000 -JX-4il/3.75il -Y4.2i -K -Ey+p0.5p \
	>> example_03f.ps
gmt psxy spectrum.ypower -R -J -O -K -Gblack -Sc0.07i -Ey+p0.5p >> example_03f.ps
echo "Input Power" | gmt pstext -R -J -F+cTR+f18p,Helvetica-Bold -Dj0.1i \
	-O -K -Wthicker -C0.1i >> example_03f.ps
gmt pslegend -R0/4/0/3.75 -Jx -O -DjBL+w1.2i+o0.25i -F+glightgray+pthicker \
	--FONT_ANNOT_PRIMARY=14p,Helvetica-Bold << EOF >> example_03f.ps
S 0.1i T 0.07i black - 0.3i Ship
S 0.1i c 0.07i black - 0.3i Satellite
rm -f report tmp samp* *.pg *.extr spectrum.*

The final illustration shows that the ship gravity anomalies have more power than altimetry derived gravity for short wavelengths and that the coherency between the two signals improves dramatically for wavelengths > 20 km.


Spectral estimation and x=y-plots.