I don't see how you can objectively test RPP standards.
> I have noticed that this topic with the differences in approach to scaling the 95% confidence error ellipses has been discussed before on this board (see [msg]75990[/msg] and [msg]242017[/msg]).
> The scaling of 95% confidence error ellipses is required to see if you have met Relative Positional Precision (RPP) standards such as the ALTA 0.07'+50ppm standard.
> If there is no definitive method of computing 95% confidence error ellipses, I don't see how you can objectively test whether a RPP standard has been met or not.
There is only one method of computing 95% confidence error ellipses that I'm aware of.
There are, however, two different approaches to estimating the standard errors of observations. The standard errors are the measures of the uncertainties in observations from which the error ellipses are derived.
One approach assumes that reliable apriori estimates of standard error of observations (implemented in Star*Net) have been used as input and uses the chi square test to confirm that no blunders exist and that the standard error estimates appear to be realistic (or more exactly put: are not unrealistic). This is generally the most efficient approach.
The other approach assumes that the standard errors are to be worked out from the adjustment (implemented in SurvNet) and automatically inflates the standard errors by a factor determined from the standard error of unit weight of the adjustment residuals. That approach only works well if you have a superabundance of observations and a high value for degrees of freedom.
The SurvNet approach is fairly impractical and inefficient for many land survey projects such as the one you posted above as an example. Star*Net's approach, when the estimates of standard errors are realistic, is much more satisfactory on small projects such as the typical small parcel survey, where there will be few observations and the upper and lower bounds of the chi square test will be spread quite a ways apart.
I don't see how you can objectively test RPP standards.
In any event, the answers that you get for the 95% confidence intervals are different depending the software you are using. For the given example yesterday, SurvNET told me I failed the ALTA 0.07'+50ppm standard on all of my specified monument pairs. On the other hand, it looks like if I had run it through Star*NET, I would have passed most of my specified monument pairs. All the while, using the same data. To me, this puts the objectivity of RPP standards into question.
Difference explained in old thread
The nugget in those old threads is Dennis Milbert's post about 80% of the way down [msg=242017]this thread[/msg]
I don't see how you can objectively test RPP standards.
> In any event, the answers that you get for the 95% confidence intervals are different depending the software you are using. For the given example yesterday, SurvNET told me I failed the ALTA 0.07'+50ppm standard on all of my specified monument pairs. On the other hand, it looks like if I had run it through Star*NET, I would have passed most of my specified monument pairs. All the while, using the same data. To me, this puts the objectivity of RPP standards into question.
I don't think that is as correct a statement as simply observing that some least squares survey adjustment software does a better job of handling the standard errors in typical land survey measurements than other software does. SurvNet looks like a bit of a mess to me.
Difference explained in old thread
> The nugget in those old threads is Dennis Milbert's post about 80% of the way down [msg=242017]this thread[/msg]
And the key bit in Dennis Milbert's post is the point that I've been trying to make. Star*Net implements the approach used by Mikhail and SurvNet follows Gilhani and Wolf:
>What is happening is that two somewhat different statistical problems are being considered in Mikhail vs. G&W. In Mikhail, the binormal distribution parameters are considered known. That is, we know sigma X, sigma Y and the covariance of X and Y. A typical way of knowing these values is if we know the sigmas of our observations, and we compute linear error propagation. In this version of the problem, we can do the error propagation in a LS adjustment, or not. We can do the propagation without any measurments at all. k=2.447.
>In Gilhani and Wolf, they are also looking at scaling an error ellipse to 95%. But, they assume that the variances and covariance are not known. Rather, they are subject to a variance scaling factor (formally: the "a posteriori variance of unit weight"). In this version of the problem, we have observations, we do a LS adjustment, we estimate our "a posteriori variance of unit weight", we scale all of our input observation variancs. and we do error propagarion to our scaled output variances and covariances,
>In this problem [the Gilhani and Wolf approach] the observations have to do double duty. They have to estimate coordinates (and sometimes other parameters), and they also have to estimate the variability of the observations. Because the coordinates can slosh around, the residuals underestimate the error of the observations. Quite a bit of error can be hidden in coordinates for low degrees of freedom -- hence the larger error estimates in Table 19.2. k>2.447 (depending on redundancy).
So, if a surveyor knows nothing about the uncertainties in his or her survey measurements, then he or she is stuck with making many redundant observations in a design from which the uncertainties can be worked out, basically a test survey on every project. For simple parcel surveys, that is a remarkably inefficient choice compared to spending the relatively small amount of time to test the performance of the instruments and procedures in advance and assess their characteristic standard errors as the Mikhail/Star*Net approach contemplates.
- See more at: http://beerleg.com/index.php?mode=thread&id=242017#sthash.YUVVOFTx.dpuf
I don't see how you can objectively test RPP standards.
> To me, this puts the objectivity of RPP standards into question.
The validity of your statistics is critical. As is the understanding of the implications of your adjustment results in a physical sense.
I can't imagine a better tool (than LS software) to analyze positional tolerances. You are simply viewing the data you have collected through the lens of a LS adjustment to predict how well your survey agrees with the "truth". By "truth" I mean the actual physical relationship between the positions at 100% certainty, which is impossible BTW , unless you have the time to make "degrees of freedom = infinity".
I don't see how you can objectively test RPP standards.
> > To me, this puts the objectivity of RPP standards into question.
>
> The validity of your statistics is critical. As is the understanding of the implications of your adjustment results in a physical sense.
Which is why it is particularly important to actually test the standard errors of the various measurements processes, i.e.
- horizontal angles,
- vertical angles,
- disances,
- instrument centering,
- target centering
and, in particular, to use methods and procedures that are reasonably expected to produce survey measurements with those same standard errors. Doing that greatly simplifies the entire problem of estimation of uncertainty in land surveys while producing realistic results.
Testing is not a merely academic exercise. It's an investment in quality that pays off on every project, particularly if the alternative is to spend large amounts of time trying to derive the standard errors from scratch on those same projects using the approach implemented in SurvNet.
I don't see how you can objectively test RPP standards.
I have to agree with you Bow Tie. In fact, based on the posts, it seems to me that Star*Net would not be an acceptable indicator of complying with the ALTA contract, whereas SurvNet would.
This is because Star*net is compromised by previous test data that can not be confirmed to be true on the adjustment in question. Sure, you say you didn't drop the gun half way through this job; but how do we know for sure?
As impractical as the SurvNet approach seems; it appears to be the only one that could be seriously used to question the results of an ALTA contract compliance as it only uses data from the job in question.
Better raise your rates to include the redundancy needed in the SurvNet approach.
I don't see how you can objectively test RPP standards.
>Sure, you say you didn't drop the gun half way through this job; but how do we know for sure?
You never know for sure, but if your error estimates are realistic, you'd see unexpectedly large residuals for observations after the drop, and the adjustment wouldn't pass the chi square test.
You can't just push the button and not analyze results.
I don't see how you can objectively test RPP standards.
> I have to agree with you Bow Tie. In fact, based on the posts, it seems to me that Star*Net would not be an acceptable indicator of complying with the ALTA contract, whereas SurvNet would.
>
> This is because Star*net is compromised by previous test data that can not be confirmed to be true on the adjustment in question. Sure, you say you didn't drop the gun half way through this job; but how do we know for sure?
That might possibly be true if there were no testing of the residuals to verify that the residuals are not inconsistent with the apriori values of standard errors based on testing and subsequent experience. However, Star*Net tests the residuals and flags unrealistic values.
In the case of Star*Net's approach, the chi square test is a realistic measure of quality, whereas in SurvNet's approach, the chi square test is the target value to fudge to. It's a big difference.
> As impractical as the SurvNet approach seems; it appears to be the only one that could be seriously used to question the results of an ALTA contract compliance as it only uses data from the job in question.
But that's obviously not true. The fact is that the standard errors of distances, angles, instrument centering, and target centering cannot be worked out from any familiar land survey design (it requires a highly sepecialized design that wouldn't be useful for any typical land survey). So both Star*Net and SurvNet require some realistic apriori estimates of the standard errors of intstrument and target centering. They won't be worked out from the adjustment and are critical values on small parcel surveys.
So, SurvNet has the schizophrenic approach of assuming that some standard errors can be reliably estimated beforehand and others such as those of angles and distances are completely unknowable and must be somehow estimated by arithmetical magic from the residuals, without, of course, really taking into account that in a typical land survey both the standard errors of distances AND angles can't be efficiently worked out.
The professional approach is that implemented in StarNet which assumes that the surveyor won't be completely clueless, but will have a well documented idea of what the standard errors of angles and distances are expected to be and will used the adjustment statistics to demonstrate that the apparent errors (residuals) in the adjusted measurements are entirely consistent with that expectation.
I don't see how you can objectively test RPP standards.
For example, the question is "So, how did you know that the measurements you made on Mr. Green's survey had the uncertainties you claim?"
SurvNet user : "I didn't. I just ran my work through SurvNet and it told me that I'd given the wrong values and increased them by 50%."
Star*Net user : "I used methods, procedures, and equipment that I have tested and have extensive experience with. The uncertainties were based upon my test results, subsequently verified on many similar projects."
I don't see how you can objectively test RPP standards.
Yes, I don't disagree that the Star*Net approach makes the most sense. But it seems to me that the ALTA contract specifically went the other way. There were provisions in earlier ALTA contracts that took into account previously tested and known capabilities of equipment and procedures. These were replaced in favor of a job by job proof of measurement capabilities.
I would argue that SurvNet is at least trying to comply with the contract, while StarNet is patently not. The chi test of STarNet is not based on measurements taken on the job at hand, but rather on previously estimated errors; much like allowances of earlier editions of ALTA contracts.
I don't have the expertise to debate which approach is "best"; but I argue that the contract needs a bit of work in this area. Bow Ties concern seems valid; who wants to be arguing over this in a breach of contract action.
The last thing a surveyor should want is a contract with a built in retirement argument for attorneys. They could go on for days/weeks with a mess like this and at the end the judge or jury (with a dazed and confused look) would just flip a coin.
I don't see how you can objectively test RPP standards.
> Yes, I don't disagree that the Star*Net approach makes the most sense. But it seems to me that the ALTA contract specifically went the other way. There were provisions in earlier ALTA contracts that took into account previously tested and known capabilities of equipment and procedures. These were replaced in favor of a job by job proof of measurement capabilities.
That isn't true, though. What the ALTA specification actually says is this:
>i. “Relative Positional Precision” means the length of the semi-major axis, expressed in feet or meters, of the error ellipse representing the uncertainty due to random errors in measurements in the location of the monument, or witness, marking any corner of the surveyed property relative to the monument, or witness, marking any other corner of the surveyed property at the 95 percent confidence level (two standard deviations). Relative Positional Precision is estimated by the results of a correctly weighted least squares adjustment of the survey.
The Star*Net approach is based upon correct weights being assigned in the adjustment and uses the chi square test to validate those weights.
The SurvNet approach discards the weights and generates the 95% confidence relative error ellipses by substituting other weights for those determined by the surveyor.
I don't see how you can objectively test RPP standards.
>The chi test of STarNet is not based on measurements taken on the job at hand
It DOES test the measurements taken.
The chi-sq test of Star*Net compares the actual goodness of fit of the measurement residuals against the standard error supplied by the surveyor. This relationship is summarized in the "standard error of unit weight" (often symbolized as sigma-0), and compared to chi-squared limits to see if the supplied std err values are statistically consistent with the residuals.
I don't see how you can objectively test RPP standards.
> >The chi test of STarNet is not based on measurements taken on the job at hand
>
> It DOES test the measurements taken.
>
> The chi-sq test of Star*Net compares the actual goodness of fit of the measurement residuals against the standard error supplied by the surveyor. This relationship is summarized in the "standard error of unit weight" (often symbolized as sigma-0), and compared to chi-squared limits to see if the supplied std err values are statistically consistent with the residuals.
Thanks for emphasizing that point, Bill. It's an important one.
I don't see how you can objectively test RPP standards.
Yes. So what is the correct weight?
The ones determined by SurvNet from the measurements taken on the job in question?
Or the ones finally decided to be used by the surveyor after adjusting them until the result passes the chi test and gives 95% 0.07'?
Or can it be shown that the ones the surveyor uses from previous testing are valid on the job at hand by showing the same equipment (in the same condition and adjustment) and procedures were used under the same field conditions by the same personnel?