Summary
Raster skydips yield the same results for tau₈₅₀ and tau₄₅₀ as discrete skydips.

850um
Consecutive skydips - one discrete, one raster - were identified from the summit archive, and the tau at 850um from each, as determined by ORACDR, are plotted below :

The red line is the least squares fit to the data, and the green line is of slope unity passing through the origin. The green line is a perfectly good fit to the bulk of the data, so it is clearly possible that raster skydips give essentially the same answers as discrete skydips. Outliers from the green line occur only in the sense that raster skydips occasionally (14% of the time) produce larger values of tau than discrete skydips.

The differences do not appear strongly correlated with the errors in tau as determined from either the discrete data or the raster data :

We examine more closely the 20 outliers with | differences in tau | > 0.1 in order to determine which factors, if any are causing their deviance. Two of these pairs (18,20 below) actually have the discrete skydip value exceeding that of the raster skydip, although one of these (20) is of the second highest tau recorded, and the scatter in the diagram appears to increase with tau. The highest recorded value of tau (17) also exhibits a large difference and we dismiss that one also on the same grounds.

         UT        # tau_discrete      # tau_raster    diff
   1  20010401      33    0.897       34    1.073     0.176
   2  20010416       6    0.337        7    0.466     0.129
   3  20010421       1    0.730        2    1.438     0.708
   4  20010501       3    0.834        4    1.355     0.520
   5  20010504     107    0.295      108    0.530     0.234
   6  20010504      53    0.261       52    0.416     0.155
   7  20010504      87    0.219       88    0.397     0.177
   8  20010505      28    0.258       29    0.450     0.193
   9  20010505       3    0.304        4    0.563     0.259
  10  20010505      81    0.250       82    0.435     0.185
  11  20010507      15    0.428       16    0.807     0.379
  12  20010507      20    0.484       21    0.924     0.440
  13  20010507       3    0.448        4    0.890     0.442
  14  20010517       6    0.848        7    1.178     0.331
  15  20010520       5    0.608        6    1.319     0.711
  16  20010524       5    0.690        6    1.000     0.310
  17  20010610       5    1.820        6    2.044     0.224
  18  20010616      30    0.713       29    0.605    -0.109
  19  20010627       3    0.979        4    1.146     0.167
  20  20010705       3    1.647        4    1.465    -0.181

Half of the raster skydips (in pairs 3,5,6,7,8,9,10,11,12,13) had serious imperfections, such as changes of slope, that make them impossible to model. In all but one case (11) the corresponding discrete dataset also had obvious defects. Seven more (2,4,14,15,16,17,18) had lesser, but obvious, oddities. In only 3 cases (1,19,20) were the raster data apparently suitable for model fitting (A), and in one of these (19) the corresponding discrete data were 'poor'. The fitting process itself, however, generated consistently odd values of the bandwidth parameter, b, and high values of the standard deviation of the fit. Within these descriptions must be grounds for applying quality control, which we should establish by reference to well-behaved skydip data.

Quality Control
A control sample of 10 pairs was established having small differences and small errors. These data lie within the clusters of points in the above diagrams. Qualitatively, the data, both discrete and raster, were monotonic, smoothly varying, and flawless. Quantitatively, the fitted values of b ranged between 0.7 and 0.9, with differences of less than 0.05 (discrete - raster), and the s.d.s of the fits were almost always 1K or so, and never more than 2K.

In the 20 outliers above, the raster-b values were outside the normal range half the time : this was also true for one of the three cases above (A, viz 1). Five (3,4,8,17,19) discrete skydips also fail the 'b-range' test, while #20 fails the 'b-difference' test. Only 6 of the outliers (1,14,17,18,19,20) would have survived the 's.d.' test. Even without the subjective estimates of the quality of the data, all twenty cases above would be described as deviant from one or more of these tests:

Quality Control tests for raster skydips

• 0.7 < bandwidth (b) < 0.9
• | b_raster - b_discrete | < 0.05
• s.d. of fit < 2K

Applying only the restrictions on b to the remaining 102 data removes a further 32; about half by b-range, half by b-difference :

The fit now is essentially a line of slope unity passing through the origin, implying that raster skydips, suitably quality-controlled, give the same results as discrete skydips. The scatter about the line is 0.019 in tau, although when restricted to tau<0.4 this drops to 0.014.

Notes :

• It is recognized that the test #2 above cannot be applied in practice once raster skydips alone are used at the telescope.
• There has been discussion about whether the preponderance of data at low airmass biases the results of any one raster skydip and whether a weighting scheme other than the default (where equal weight is given to each datum) is needed, but at this time there would appear to be no need to seek another weighting scheme.

450um
The 450um data from this resticted sample are shown below :

Not all 70 data at 850um had 450 equivalents reduced by ORACDR : hence the smaller sample here. There are again a couple of outliers but the bulk of the data would be well fit by a straight line of slope unity passing through the origin. Without the most deviant point in the above diagram the rms error of the remaining 57 points would be 0.4 in tau.

An examination of the individual fits reveals that the conditions described above in tests 1 and 3 again correlate with good fits, although there may be a slightly larger spread in the acceptable value of the bandwidth parameter, b.

When this sample is limited to tau₄₅₀ < 2, or approximately tau_cso < 0.09 - which is the regime in which 450um data becomes available anyhow - the most deviant point is found to have an exceedingly small fitted value of b and would have been rejected on this basis. It also had an oddly small range of measured sky temperatures. With its removal these data yield the relationship shown below :

The formal least squares fit is essentially a line of slope unity passing through the origin, and the scatter in this regime is 0.14 in tau.

Conclusion
Raster skydips yield the same results for tau₈₅₀ and tau₄₅₀ as discrete skydips, provided the data are of a suitable quality as determined using the tests above.

Thanks to the observing staff who made these measures over the past several months.