SGS00: Site none Messages, 6 Entries..

Attached 4 CSAT3 sonics to cosmos.  Each sonic is Y'd into 4 serial ports,
in order to simulate 16 sonics into one adam.  The sonics are configured
for 20hz,  (parameter Ac).


project SGS00, ops0

Channel grouping
sonic	channels	variable suffix
#1	200,201,204,205  a,b,e,f
#2	202,203,206,207	 c,d,g,h
#3	208,209,212,213	 i,j,m,n
#4	210,211,214,215  k,l,o,p


cosmos: matrix  sn 940211101
goldenrod: ironics 3234, sn 129043
datel: sn 400190
VME chassis #4

Breezecom radio: unit A, 16db antenna

archive files

cos000610.235245: started test
cos000611.000000  OK
cos000611.080000  OK
cos000611.160000  died at 22:07  (16:07 MDT)
cos000611.220818  restarted itself, but ironics did not come up

cos000612.032846  did an MXRESET. ok
cos000612.080000  ran until 15:57 (09:57 MDT)

So, there are 2 unexplained adam crashes.  The other adams kept
running (though they only had 1 1hz analog port enabled).

A data_dump of channel 0 showed no problems before the crash:
	data_dump -f cos000612.080000 -c 0 -A
syslog also showed no errors, except for ingest timeouts after adam crashes.


Covars look good during this time.  In 23 hours of data from Jun 10 2352 to
Jun 11, 22:07 GMT the results are identical for the four ports connected
to a sonic.

The only differences seen are when a sample is put into a
different 5 minute average.  Here are some u averages:

x[44:45,c(1,2,5,6)] 
00 06 10 213230.000 0.05689919 0.05690063 0.05689919 0.05690063
00 06 10 213730.000 0.06431438 0.06431170 0.06431438 0.06431170

> attr(x,"weights")[44:45,c(1,2,5,6)]
     [,1] [,2] [,3] [,4] 
[1,] 6001 6000 6001 6000
[2,] 6000 6001 6000 6001

On two channels, 6001 samples are summed the first average and 6000 in
the second average, and vice-versa for the other two channels.

How closely can we "pack" PAM tripods for SGS?

The PAM tripods have an equilateral triangle footprint, L=115" on each
side.  Thus if we place them side-by-side in the same orientation, the
spacing of the masts would be L = 2.92m.

  /\  /\  /\  /\  /\
 /__\/__\/__\/__\/__\

However the mast itself is only a distance of L/3 from one side of the
tripod footprint, so alternating the orientations of the tripods by
180 degrees allows a spacing of the masts equal to 2/3 L = 1.95m.

  /\  __  /\  __  /\
 /__\\  //__\\  //__\
      \/      \/

Note that the tripods are also staggered in order to line up the masts
in a straight line.  The tripods could perhaps be nested even a little
closer, since the only structure on the outer edges of the footprint is
the three legs themselves.

The PAM tripod structure without the electronics box (recall that we
will be using ADAMs to ingest the sonic data) are actually very open,
so I don't expect the flow distortion with PAM masts to be any worse
than with some sort of horizontal beam to support the sonics.

How to calculate 2D spatial averages ala Tong et al?
The devil is in the details!

Let
w = arbitrary flow variable, e.g u,v,w, etc,
number of sensors in single crosswind filter = 2n+1,
sensor spacing = S,
mean wind speed = U,
mean wind direction (relative to array normal) = D,
	with D defined positive for wind blowing from x_{-n} to x_n 

Then effective sensor spacing = S' = S cos(D)

Crosswind average = 

	 = Sum_{j=-n}^{j=n} C_j w(x_j,t+jdt)

where dt = S sin(D)/u with u the instantaneous wind speed.

Does Taylor's hypothesis need to hold over the time +/-n dt required
for a single filter operation (or is it 2n dt?) or the time +/-2n dt 
required for a second order filter (or is it 4n dt?)?  I have my own
opinion, but would like some independent assessments.

Streamwise average =

	w~(t) = 1/(2T) Int_{-T}^T C(t') w(t+t') dt'

where T = nS cos(D)/U gives a streamwise filter with the same
width, 2UT = 2nS', as the crosswind filter.

C_j and C(t') are weighting functions used to define filter shape:

	Sum_{j=-n}^{j=n} C_j = 1

	1/(2T) Int_{-T}^T C(t') dt' = 1

Then the resolved and sgs flow variables are

	w^r = ~ 

	w' = w - w^r

Note that both w^r and w' remain time-dependent, stochastic variables.
It would appear to be straightforward to apply this process again to
obtain second moments such as

	~ = < ~ (w - ~) >~

This is all fine for a stationary flow field.  But how do we define U
and D for a non-stationary flow as will be the case for much of the
data?  We need to define or simply choose a procedure for our real-time
data analysis in the field.  Do we just pick an arbitrary averaging
time, such as 5, 15, 30, or 60 minutes?  There is perhaps some logic to
make this averaging time on the order of T = nS'/U.  But note the
difficulty of calculating the latter if U=f(T).

12/3/99

I have looked briefly at the issue of sonic spacing:

For a height of z=6m and a filter width of W=12m=(4*3m), Tong et al
calculate a filter cut-off at a wavelength lc = 2*pi*z/1.67.

Thus lc = 2*pi*(6m)/1.67 = 22.6m = 1.88*W,
where W = width of spatial filter (of Tong design)

We want lc to match lw, the wavelength at the peak in the vertical
velocity power spectrum.

From Kaimal and Finnigan, Fig. 2.9, this peak occurs
at:

	z/L	z/lw	W(z=6m)
        ---     ----   	-------
	-1	0.2	16m 	
	-0.5	0.3	11m
	 0	0.6	 5.3m
	 0.5	1	 3.2m
	 1	1.5	 2.1m

Note that the peak wavelength of the u spectrum is 5-10 times larger
than that for the w spectrum.

Combining these two, W = z/(1.88*z/lw), which is in the 3rd column
above.

Thus it would appear that we might choose perhaps 12m (Tong's spacing),
6m and 3m for the filter width (sonic spacing of 3m, 1.5m and 0.75m!)
in three different arrays that cover the range of stability.  Note
that the closer spacings would require wind directions very close to
normal to the array in order to minimize flow distortion of one sonic
by another.  The spacing is proportional to z, so the higher we can
place the array, the better.

Comments?  I infer that practical LES simulations do not use filter
widths as small as 3m, so perhaps we do not want to use lw as the
criterion for stable conditions?  I recall that Chin-Hoh and Peter
wanted to use perhaps 12m and 24m as two filter widths.  Obviously this
is not compatible with the lw criterion, which was not apparent to me
at our meeting.

12/17/99

Peter suggests that we investigate two cases, 
Kc = z/lc = 0.067 and 0.333.

	Kc = z/(1.88*W) = z/(7.5*S) , where S = sonic spacing

Thus for Kc = 0.067:  S/z = 2 and a height of 5m implies S = 10m

         Kc = 0.333:  S/z = 0.4 and a height of 5m implies S = 2m

Could we keep the spacing the same and just change the height?
Let S=4m, then for Kc = 0.067: z = S/2 = 2m
               for Kc = 0.333: z = 2.5S = 10m
It's almost doable!

We could get Kc as high as z/(7.5*S) = 9m/(7.5*2m) = 0.6

As Don noted during our meeting, the validity of Taylor's hypothesis
depends on sigw/U << 1 and sigu/U << 1.

sigw/U = sigw/u* x u*/U = sigw/u* x k/[ln(z/zo - psi(z/L)]

sigu/U = sigu/u* x u*/U = sigu/u* x k/[ln(z/zo - psi(z/L)]

sigw/u* ~ 1.3, sigu/u* ~2, so the validity of Taylor's hypothesis
depends more critically on sigu/u*.  Note that there is no explicit
dependence on wind speed and only a weak dependence on stability.

For neutral stability, z=6m and zo=3cm,

sigu/U = 2*0.4/ln(6/0.03) = 0.15

I have plotted sigma_u/U versus U for the SJVAQS field program, held in
the San Joaquin valley of CA in 1991.  For 5m wind speeds above 3 m/s,
sigma_u/U ~ 0.2.  At lower wind speeds, the range of sigma_u/U
increases with decreasing wind speed, from values as small as 0.05
around 2 m/s up to values approaching 1 below 1 m/s.  

Although the possible values for sigma_u/U increase at low wind speeds,
there is still plenty of data at low wind speeds with values equal to
or below 0.2.  The values below 0.2 are associated with low values of
u*/U.  These occur during the stably-stratified conditions (at night)
when the drag coefficient is low.  Thus I anticipate no problem with
satisfying Taylor's hypothesis if sigma_u/U < 0.2 is satisfactory.

initial log file /net/aster/projects/SGS00/logbook/tklog.log created by USER maclean