Star*Net Input File - Why did I fail?
> Can you run it holding only point 101 fixed and the bearing you specified.
Yes, that is how I ran the adjustment shown above. The asterisks following one of the coordinates means that those values were considered to be without weight in the adjustment. The exclamation point means the value was held fixed and error-free.
Generally, where you're using a record bearing basis, you want to use the bearing between the two monuments most widely separated and hold that fixed as a condition of the adjustment. That typically gives the smallest relative position uncertainties. If i knew the record bearing of 107-177, for example, I would want to try running the adjustment holding that condition fixed to orient the survey.
SurvNET and StarNET do not agree?
Here is what I get when I do that:
Input observations only holding point 101 and the azimuth between 101 and 109:
Here are my resulting coordinate standard errors, 95% error ellipses and Relative Positional Precisions:
Here are my residuals:
And my statistics:
I notice quite a difference with the error ellipses. Why is this?
I run my LSA with Columbus, and it's not uncommon for raw data files to record a single record (FS angle, and single distance) when you turn a D&R. The multiple foresight and backsight distance info will be stored in the raw data, but then the DC computes the averages, ignores the backsight info, and computes the foresight point coordinates based on averages. If your LSA is reading the raw file directly, it's also probably ignoring additional data that can help sure-up your network. If you can manually edit your LSA input file, try adding a single backsight reading, and run it again, it should boost or DoF by 1 (or 2 if you type in a VA).
SurvNET and StarNET do not agree?
> I notice quite a difference with the error ellipses. Why is this?
I'd say that Bill93 probably has correctly identified the problem with SurvNet. It is inflating the standard errors by a factor of 1.43 from the standard error unit weight computed from the standard errors as input by you, even though with so few degrees of freedom the chi square test did not definitely identify a problem (your adjustment passed the chi square test without inflating the standard errors) and such inflation isn't necessarily warranted.
Here are the standard errors and 95% confidence error ellipses that Star*Net produced using the standard errors that I specified above.
[pre]
Station Coordinate Standard Deviations (FeetUS)
Station N E
101 0.000000 0.000000
109 0.010395 0.000838
103 0.055796 0.012627
102 0.011226 0.010996
105 0.011745 0.011852
106 0.021604 0.011872
107 0.011107 0.010539
108 0.011193 0.011436
177 0.015365 0.010741
178 0.011687 0.011135
219 0.017150 0.016496
Station Coordinate Error Ellipses (FeetUS)
Confidence Region = 95%
Station Semi-Major Semi-Minor Azimuth of
Axis Axis Major Axis
101 0.000000 0.000000 0-00
109 0.025527 0.000000 175-23
103 0.136577 0.030896 0-21
102 0.027774 0.026611 149-29
105 0.029631 0.028110 49-58
106 0.053031 0.028786 5-07
107 0.027856 0.025075 150-01
108 0.028013 0.027376 79-35
177 0.037611 0.026288 179-11
178 0.029247 0.026569 29-50
219 0.043656 0.038556 35-50
[/pre]
The problem with point 103 is the angle tie to it. The centering error at 102 is magnified over the longer distance to 103 and, when that uncertainty is added to the uncertainty in the instrumental measurement, the result at 487 ft. is a bit rough looking.
The way to have gotten around that would have been to have measured 103-101-109.
Thanks Bill. Very helpful.
SurvNET and StarNET do not agree?
Yeah I was expecting 103 to be not so good north-south, but the rest I was figuring would be OK.
Also, I notice that StarNET basically is scaling up the the standard errors by that 2.448 figure to get the 95% confidence numbers (which I believe assumes infinite degrees of freedom), while SurvNET only lets me have 2 degrees of freedom which really hammers my 95% confidence intervals.
SurvNET and StarNET do not agree?
> Also, I notice that StarNET basically is scaling up the the standard errors by that 2.448 figure to get the 95% confidence numbers (which I believe assumes infinite degrees of freedom), while SurvNET only lets me have 2 degrees of freedom which really hammers my 95% confidence intervals.
Yes, if you are using the same instrument on an ongoing basis, you should expect to find that the same standard errors work on every project when the same methods and procedures are used. In other words, the standard errors are based upon a great enough number of measurements to be reliable. It's hardly as if the instrument's performance is some unknown quantity to be determined exclusively from the adjustment.
For surveys with well tested instruments and procedures, the chi square test is best used as a red flag that something is abnormal and needs to be examined more closely.
No Angle At 102, Then It Is Not A Traverse
In reality there is not enough information to do a good adjustment.
Go back, observer at 102 and then adjust. Least squares can then place your errors in more appropriate positions and your results should be better.
Paul in PA
No Angle At 102, Is Not A Traverse
> Go back, observer at 102 and then adjust. Least squares can then place your errors in more appropriate positions and your results should be better.
Except, the problem with the 95% confidence positional uncertainties reported by SurvNet won't go away. It's apparently a design feature of the program to automatically inflate standard errors by the standard error of unit weight from the adjustment, regardless of whether or not the adjustment passes the chi square test.
Star*Net Input File - Why did I fail?
> In case anyone would like to adjust Bowtie Surveyor's project in Star*Net, here is the input file.
Kent, I think that the data you present is derived from BowTies adjusted coordinates. You are adjusting the adjustment. Substituting raw data angles and distances from the diagram I get a very tight adjustment with the greatest angular residual under 4".
Update: In the StarNet adjustment report the adjusted angles are reported together with the residuals. In BowTie's report, the raw angles are reported together with the residuals. Hence the confusion.
Star*Net Input File - Why did I fail?
> Kent, I think that the data you present is derived from BowTies adjusted coordinates. You are adjusting the adjustment. Substituting raw data angles and distances from the diagram I get a very tight adjustment with the greatest angular residual under 4".
I went back and double-checked the data that Bowtie Surveyor posted. These are the measured angles and distances he supplied that I used in the Star*Net input file:
[pre]
Unadjusted Observations
=======================
Control Coordinates: 2 Observed Points, 0 Fixed Points, 0 Approx. Points
Sta. N: E: StErr N: StErr E:
101 4982.039 5037.077 0.000 0.000
102 4951.355 5181.797 0.000 0.000
Distances: 11 Observations
From Sta. To Sta. Dist. StErr
101 102 147.935 0.011
101 103 487.597 0.012
101 105 56.085 0.012
101 106 171.296 0.012
101 107 10.124 0.012
101 108 38.731 0.012
101 109 239.780 0.011
109 177 39.541 0.012
109 178 122.179 0.011
178 102 213.806 0.012
178 219 42.134 0.012
Angles: 10 Observations
BS Sta. Occ. Sta. FS Sta. Angle StErr (Sec.)
102 101 103 348-23'01.5" 15.4
102 101 105 217-59'32.0" 43.1
102 101 106 353-09'07.5" 18.8
102 101 107 048-03'00.0" 215.5
102 101 108 157-36'51.0" 60.0
102 101 109 073-25'08.0" 17.5
101 109 177 207-47'08.0" 57.1
101 109 178 097-09'21.0" 20.5
109 178 102 088-20'42.5" 20.9
109 178 219 226-08'24.0" 56.3
[/pre]
If I'd used the adjusted values in Star*Net, the residuals should have been very small, with no residual over about 4", I'd expect. :>
No Angle At 102, Is Not A Traverse
My own preferred philosophy is to inflate the standard errors by the std err of unit weight IF that increases them, but never deflate them. This moves in the safe direction in case the apparent increase was due to real conditions and not just a random variation.
But I use the Rayleigh calculation like Star*Net and not the F-distribution which can increase the values much more.
No Angle At 102, Is Not A Traverse
> My own preferred philosophy is to inflate the standard errors by the std err of unit weight IF that increases them, but never deflate them. This moves in the safe direction in case the apparent increase was due to real conditions and not just a random variation.
That would certainly be a conservative approach. However, it's understood that even if the standard errors of the measurements are correctly assigned, there is a reasonable likelihood on small samples that the standard error of unit weight will vary considerably from unity.
I think I'd prefer to examine the upper bound of the chi square test interval to see how close the residuals are to exceeding the limit even if the adjustment nominally passes the test. The closer the standard error of unit weight it to exceeding that limit, the less hesitant I'd be to increase the standard errors.
In land surveying practice, I find that actually testing and evaluating as many of the standard errors as possible is the better road to assured quality, rather than relying upon the adjustment to provide the weights.
Star*Net Input File - Why did I fail?
I apologize, apparently it is me that is adjusting the adjustment then.
No Angle At 102, Is Not A Traverse
> In land surveying practice, I find that actually testing and evaluating as many of the standard errors as possible is the better road to assured quality, rather than relying upon the adjustment to provide the weights.
:good:
I applaud your commitment to better understanding the reports. Far too few of our fellow professionals make that same effort.
But I have to admit that I a bit dismayed at some of the responses. No matter how you massage the statistics, you can't take data that is not as good as it needs to be and turn it into good data.
You need more redundancy. End of story. Manipulate the expected errors any way you want, define the reference bearing any way you want, but at the end of the day what you have is a data set that needs more redundancy. None of the other things address that basic weakness.
Larry P
> But I have to admit that I a bit dismayed at some of the responses. No matter how you massage the statistics, you can't take data that is not as good as it needs to be and turn it into good data.
That would be true only insofar as there wasn't enough information presented to have reliable estimates of the standard errors. Given the apriori estimates of standard errors that Bowtie provided, there wasn't particularly any problem with his observations. The problem lay in the survey design in how Pt. 103 was connected to the network.
> You need more redundancy. End of story.
No, for a simple survey like that more redundancy definitely is not absolutely needed. Sure, the closing angle on the quadrilateral would have been nice, but it wasn't necessary. The missing part wasn't redundancy, but was Bowtie Surveyor actually testing his total station and accessories to have realistic a priori values of the standard errors that were used in weighting of observations.
Without having quite good estimates of those values of standard errors, he would have needed to make a very big project out of a simple survey in order to assess them from the adjustment statistics. That is an extremely inefficient way of getting at an answer that should have been known when he rolled up to the job.
> Manipulate the expected errors any way you want, define the reference bearing any way you want, but at the end of the day what you have is a data set that needs more redundancy.
That overlooks the importance in how the reference bearing is chosen when a record bearing basis is used. The point should be obvious that a survey oriented to a monumented line 15 ft. long has much greater azimuth uncertainty throughout than one oriented to a monumented line 1000 ft. long. It would be dumb to overlook that fact.
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.
Kent, Close The Traverse To Give StarNet More To Work With
Much more time has been wasted talking it up, than actually making the observation.
StarNet was not designed to support shortcut methodology, but to adjust proper methodology.
Paul in PA
Kent, Close The Traverse To Give StarNet More To Work With
> Much more time has been wasted talking it up, than actually making the observation.
>
> StarNet was not designed to support shortcut methodology, but to adjust proper methodology.
As I demonstrate in a thread above ( http://beerleg.com/index.php?mode=thread&id=282085 ), the closing angle 178-102-101 does virtually nothing to the station coordinate error ellipses. Adding 103-101-109, however, is a significant improvement in the relative positional uncertainty of Pt. 103.