E-mail from Patrick Wallace on 18 March recommended a change to the longitude and latitude of the telescope:

I've just updated obs.f in the repository with the JCMT telescope
position that is actually used by the JCMT TCS.

  *  James Clerk Maxwell 15 metre mm telescope, Mauna Kea             JCMT
 -*  (IfA website, Richard Wainscoat, height from I.Coulson)
 +*  (as used by the JCMT telescope control system; R Kackley)
   570  CONTINUE
        NAME='JCMT 15 metre'
 -      W=WEST(155,28,37.20)
 -      P=NORTH(19,49,22.11)
 +      W=WEST(155,28,47.00)
 +      P=NORTH(19,49,33.00)
        H=4111D0

If you go to http://www.vlba.nrao.edu/astro/obstatus/current/node5.html 
you will see that NRAO think the WGS (=ITRS) coordinates of the Mauna Kea 
VLBA antenna are

  +19 48 04.97 -155 27 19.81      VLBA antenna, NRAO

The resolution, 10 mas, corresponds to about 31 cm.  Given the extremely
demanding application, I think these coordinates can be trusted down to 
the last digit, and in fact are probably known to rather better than that.

If you then measure the position of the antenna using Google Earth 
(while trying to allow for the slanted viewpoint) you will get something 
like

  +19 48 05.00 -155 27 20.80      VLBA antenna, Google Earth

This provides a calibration of Google Earth in that region, suggesting
corrections of S 0.03 and E 0.99 arcsec need to be added.

If you then use Google Earth to measure the JCMT coordinates, you should
get roughly

  +19 49 22.34 -155 28 38.44      JCMT, Google Earth

Applying the above corrections gives

  +19 49 22.31 -155 28 37.45      JCMT, Google Earth + corrections

This position compares well with the Richard Wainscoat figures used in 
the existing sla_OBS code...

  +19 49 22.11 -155 28 37.20      JCMT, IfA

But the JCMT TCS figures that you quote...

  +19 49 33.00 -155 28 47.00

...are clearly wrong:  perhaps a different reference spheroid.

Per Friberg commented upon this last thought:

Pat's last line hints at an often forgotten issue - there are many 
different coordinate systems. The JCMT antenna position likely refers to 
something like the Old Hawaiian Reference System and might very well be 
precise in that system.

and later added

I do not doubt GPS will be the preferred system in the future, I just 
wanted to point out that the old number likely not was an error. I assume 
they just referred to another reference system. Good surveying even 150 
years ago was amazingly accurate. There is no reason to think there was an 
error.

See http://en.wikipedia.org/wiki/Geodetic_system

Tim Jenness's subsequent inquiry re the values used in the eSMA work, brough this response from Remo Tilanus:

This is what we got from the GPS measurements on the JCMT dish. Note that
the gps receiver was sitting on the edge of the JCMT dish and that between 
measurement 1 and 3 the telescope was rotated by 180 deg. I.e. the position 
of the vertical axis should be the average of the two measurements.


 JCM1       LAT   07APR11     19.822765272  +-     0.0402   0.761633 -0.925637
 JCM1       LON   07APR11   -155.477030200  +-     0.0593  -0.825422
 JCM1       RAD   07APR11        4124.8212  +-     0.1977
------------------------------------------------------------------------
 JCM3       LAT   07APR11     19.822910680  +-     0.0211   0.446184 0.389126
 JCM3       LON   07APR11   -155.477026019  +-     0.0377   0.134328
 JCM3       RAD   07APR11        4124.6859  +-     0.0805


Mean lon/lat:
lon0 = (lon1+lon3)/2 = -155.4770281095
lat0 = (lat1+lat3)/2 =   19.8228379760

These have an  accuracy of 2-8 centimeters. The height came out a bit more 
uncertain with an accuracy of 10-20cm.

heigth: (h1+h3)/2    =  4124.75355

Iain Coulson contributed:

The largest parts of the changes would fall into the TPOINT terms
AN & AW (currently +9.00" and -15.38"). As a first approximation,
adoption of the new values would amount to a relocation of the
telescope by 11"S and 10"E, so AN and AW would become approx +20.00" and 
-5.38".

- - and Patrick Wallace agreed

 . . . more or less.

Using the erroneous position, it looked to the TCS as though the telescope
was tilted S and E.  So for nominally zero true tilts we would expect the
model to claim negative AN and negative AW.  There is the separate effect
that the longitude was wrong for the HA calculation, and this has an 
effect on IA in addition to the AW component (and provides yet another 
opportunity to get a sign wrong).

I think we need to add to the existing terms the following:

  IA    +3.29
  AN   +10.78
  AW    +9.13

So the new values (expressed to excessive precision) will be:

  IA   current value plus 3.29
  AN   +19.78
  AW    -6.25

The disagreement in the AW change might be a cos(phi) factor.  The
longitude difference is more than the EW tilt change except at the 
equator.

I would expect AN and AW to reflect the deflection of the vertical by the
pull from Mauna Loa, so that the azimuth axis is tilted away from the
latter.  As a guide to how much, the values for Keck 1 are about +20, -13
respectively.  The revised JCMT figures imply a compass bearing of 198,
which can't be far off.

By the way, of the various "true" positions, I'd definitely use your own
GPS determination.

The JCMT coordinates and pointing model parameters were therefore adjusted (to the 'GPS" values above) at about 9pm HST 22 March 2008:

param TELESCOPE.LONG       DBR_DOUBLE -2.713633073   to  -2.713586047
param TELESCOPE.LAT        DBR_DOUBLE  0.346026069   to   0.345973806
param TELESCOPE.HT         DBR_DOUBLE  4111.0        to   4120.0

param POINTING.IA          DBR_DOUBLE -255.20        to   -251.91
param POINTING.AW          DBR_DOUBLE  -15.38        to     -6.25
param POINTING.AN          DBR_DOUBLE   +9.00        to    +19.78

Several RxA pointing data were subsequently obtained with the new model. No huge problems were revealed so observing was allowed to continue as normal and pointing data were collected serendipitously thereafter. The first 16 data in the new regime are plotted below:

Mean collimation offsets (CA,IE) were (-5.7",-0.2"), and rms scatters in (dS,dZ) were (4.0",3.4"). There are many strong systematics, so a full, 7-parameter optimization was performed:

               original                 new
       coeff    value       change     value    sigma

  1     IA    -251.9100    -11.701   -263.61   10.230
  2     IE     -59.2916     +7.012    -52.28    2.102
  3     NPAE   +54.4341     -0.877    +53.56   13.322
  4     CA     -87.9949     +2.996    -85.00   15.482
  5     AN     +19.7800     -2.017    +17.76    0.866
  6     AW      -6.2500     +1.349     -4.90    0.978
  7     TF     -19.3        +9.247    -10.05    2.969

On the basis of the adjusted model the data show distributions of residuals as follows:

predicting future rms scatters of (2.2",1.7").

The new model was installed at ~09:00 HST 23 March 2008, although the subsequent five nights were mostly too good for much further use of RxA. Five RxA pointing data were logged on UT20080328, with (CA+dS,IE+dZ) = (+2.3"+2.2",-2.1"+2.0"). The combined 58 HARP pointing data obtained in those 5 nights are shown below:

They have mean collimations of (CA,IE)=(+16.9",+31.8") -- compared with the nominal (+15",+35") (see /jac_sw/itsroot/src/harpb/XML/instrument_harpb.ent ) -- and rms scatters of (2.2",3.3"). This seems no better or worse than previous HARP data, uncorrected, as they aren't, for the K-mirror misalignments.