I passed Chi Squared!
I changed my centering error from .02 to .05.
I passed Chi Squared!
> I changed my centering error from .02 to .05.
I'm pleased for you. But even the 0.02 number is enormous. Perhaps you could post the residuals section of the listing report.
More thoughts
One reason for cross-ties is to help isolate blunders (assuming the cross-ties don't include blunders). The other reason is to strengthen the network by short-circuiting the accumulation of error.
It seems like your new distances from points 10 and 11 work pretty well with the older data, but the new angles don't fit so well. I haven't found any explanation.
What Star*Net is telling you with the ellipses is true - you don't know where points 6,7,8 etc are very accurately relative to 1-2, but put that in perspective: the large ellipses don't relate to the relative accuracy of 6-7, 7-8, etc. Look at the relative ellipses, too. They tell you how the nearby points relate to each other.
A lot of the reason for large ellipses at the lower left part of the figure is depending on a short leg 1-2 to measure angles from. The centering error on a short leg makes that angle sloppy, and then that slop is leveraged all the way out to point 6.
If you remeasure the angles 1-2-9 and 1-2-3 with fresh setups ( turn and re-center your tribrachs to randomize the errors) and enter several measurements (and they aren't in significant disagreement) it will take an average, it will give that average more weight, and thus reduce the ellipse sizes somewhat.
Just for education, try doing away with point 1 and holding the bearing of 8-9 to see what that does to the ellipses.
Or see what the ellipses look like if you free point 1 and hold point 6 (and use the unconstrained initialization value for point 1).
---
If you calculated an average of angle 7-8-9, go over your arithmetic again. If not, remeasure it now that you are getting better at it. The last version I ran showed a likelihood that a lot of your error was there, corresponding to about 10 minutes of angle or 0.12 ft centering.
Your recent measurements should have better centering than 0.05, no? If you can't pass with 0.02 or 0.03, that tends to suggest there are still some poor measurements in there.
"tried changing an angle value": You can't just change measured numbers to something you think would fit better. You can omit that measurement (use a * after it, and see what size it "wants" to be) or figure out how to give that measurement a larger std err if there is some knowledge of bad centering or such.
Looking At Your Point 10 And 11 Observations
2-10-3 92°17'57"
3-10-9 133°30'53"
9-10-2 134°12'07"
-----------------
Sum 360°00'57"
Totally unacceptable.
4-11-6 114°24'20"
6-11-8 59°17'47"
-----------------
Sum 173°42'07"
4-11-8 173°48'28"
Even Worse
Also, you observed 7-8-6, which gives a 0.014' error from traverse point 6 to re-observed point 6.
Unacceptable.
Paul in PA
Professors: The Dog Ate My Homework
> 2-10-3 92°17'57"
> 3-10-9 133°30'53"
> 9-10-2 134°12'07"
> -----------------
> Sum 360°00'57"
>
> Totally unacceptable.
I found out what happened here...33" was in my field book; I put 53" into star net. Still, if 57" is unacceptable, then 37" is probably as well. I'm taking three readings each time, and averaging. I'm pretty sure instrument error is not the cause, because both Face Left and Right came out to the same angle, (to the nearest 10", which is all the instrument will do.
What would be the greatest deviation expected for three sets of readings with an instrument with 10" accuracy? I'm not sure the minimum reading is the same as accuracy, because it could be off 10" either way. It's also got only single axis compensation. The objective of course, is to eliminate any observation errors at all so that the only thing left is the capability of the machine. If I can do that, I'm ok with it.
I also increased the distance from 1-2 to what it really is (230.8'). That helped tighten things up some. Here's the Input:
And the plot:
And the new output with residuals:
Now, for this:
> 4-11-6 114°24'20"
> 6-11-8 59°17'47"
> -----------------
> Sum 173°42'07"
>
> 4-11-8 173°48'28"
> Even Worse
I shot myself in the foot again. 4 was directly behind a power pole, from where I set up on 11. There just weren't many options. I had to hack my way through the woods, and as it was, had only a crack between all the trees to the edge of the pole. So I read up on how to do an Offset shot with the instrument, but blew it in the field, so just had my wife lean the prism out enough (southeast) from 4 to get the shot, measured the offset with a plumb bob, then tried to do it with trig (tangent of 6" /202.65') and add the 6'48" to the angle to get me to where 4 really was (or where I thought it was, lol).
The problem is, I only did it for 4-11-8, and not the other two. Argh. So I just corrected the table in star net with the angles I measured, knowing that the spot I observed was somewhere around 170.695' from 3. Not sure what to do with that. I can't very well "move" my stations half way through this mess. That makes the difference 27", which might be within the limits of my instrument.
And, finally:
> Also, you observed 7-8-6, which gives a 0.014' error from traverse point 6 to re-observed point 6.
>
> Unacceptable.
I'm just not sure I understand this statement. Still looking at these readings.
In the mean time, I'm on to looking at and trying some of Professor Bill93's suggestions.
>
> Paul in PA
Explain Your Instrument
I am unsure what kind of 10" instruments are still kicking around.
When setting up a single axis instrument for precise work, it needs to be leveled through a full circle before accepting that it is level. Also the vertical circle has to be checked and adjusted. I am assuming the probability that your distances were not shot direct and reversed and that you did mot shoot/record backsight distances.
2-10-3 92°17'57"
3-10-9 133°30'53"
9-10-2 134°12'07"
-----------------
Sum 360°00'57"
Totally unacceptable. Maybe the 57" was 37" or 27"?
Full circle should be within 360° +/- 15" range.
Also, explain your 7-8-6 angle.
Your line of site may have been too close to the pole which can bend the line. At a minimum you need a full prism site with the center 0.2' clear.
To get better results measure 2 temp points 1.00' from 6, one nearer and one farther away, do a D&R to each and calculate the Distance-Distance intersect point. If it is tough to set the point at exactly 1.00' record your offsets clearly and the distance between them (measure check), geometry can be your friend. I sometimes set temp nails and sometimes just set up 2 bipods clear of each other rather than walk back to move it.
I most often do this to sight corner markers I find after my traverse is set.
Paul in PA
Explain Your Instrument
> I am unsure what kind of 10" instruments are still kicking around.
>
> When setting up a single axis instrument for precise work, it needs to be leveled through a full circle before accepting that it is level. Also the vertical circle has to be checked and adjusted. I am assuming the probability that your distances were not shot direct and reversed and that you did mot shoot/record backsight distances.
>
I shot a few backsight distances, but when I have, they're very close if not exactly the same.
I'm using a Topcon GTS-203.
I have been leveling all three ways around the tribrach (I usually go around twice).
That said, I have noticed this when plunging the scope:
From 10 to 9, for example, Face Left Vertical readings averaged 02' 3.33", but when I flip and reverse, they average 03' 43.3", a full minute and 30" off. That shouldn't be, should it? But even so, why would it impact the horizontal angles?
Can't say when the last time the unit was calibrated. Should I have it done?
As for 6,7, and 8, They add up to 179 58'27.5", a full minute, 32 seconds shy. Looks like I'm going back to the corner from hell to try again, but here's the question:
At very short distance (like 40'), a centering error of even .02 feet could result in an error of almost the minute and a half I'm off.
Points 6 and 8 are spikes in hard pack; But point 7 is a 3 foot wooden stake, that I'm sure probably wafts around at least .05' if not more.
If I think that's what's causing the error, can I just eliminate it from the Starnet table? I mean bad data is bad data, correct?
If You Eliminate 7, You Need
5-6-8 and 6-8-9.
Do you mean you are shooting the top of a 3' high stake?
That Topcon should be giving you 5-6" precision on 2D&R. What is your least count settings for angles? I have seen much better results on 5" instruments with 5" readings.
That vertical angle spread affects your angular readings when you turn away from the direction that you leveled the gun in.
Set your gun up and level the gun looking at the building, then zero on a vertical line such as a building. Check your sight line and angular reading as you go up and down that vertical line.
Then relevel the gun turned 90° from the building, then zero on building. Check your sight line and angular reading as you go up and down that vertical line.
Paul in PA
More thoughts
> It seems like your new distances from points 10 and 11 work pretty well with the older data, but the new angles don't fit so well. I haven't found any explanation.
I think the problem was my failed Offset Shot...Learning those is on my to do list. The alternative would have been taking a chain saw to the power pole...a No-no around here.
>
>
> Just for education, try doing away with point 1 and holding the bearing of 8-9 to see what that does to the ellipses.
I think I'm getting a feel for how powerful Starnet can be, isolating errors.
Here's what I tried:
I fixed point one and the angle 1-2-3 (which I'm pretty confident of. Here's what happened:
>
>
> Your recent measurements should have better centering than 0.05, no? If you can't pass with 0.02 or 0.03, that tends to suggest there are still some poor measurements in there.
I definitely need to work on centering. I'm building a laser plummet I can use for my prism setups. Stay tuned.
>
> "tried changing an angle value": You can't just change measured numbers to something you think would fit better. You can omit that measurement (use a * after it, and see what size it "wants" to be) or figure out how to give that measurement a larger std err if there is some knowledge of bad centering or such.
I tried that with the 44 degree angle at 8 (between 7 and 6, but it didn't change. Not sure what's wrong there but I'm still working on it.
If You Eliminate 7, You Need
> 5-6-8 and 6-8-9.
>
> Do you mean you are shooting the top of a 3' high stake?
>
Uhhh....Guilty Your Honor! I get it now. Won't do it again!
Here's what happens if I completely eliminate 7:
Doing this I find I can reduce the centering error setting down to .03 without busting the Chi Squared limit. (.02 does it), which might be telling me that the centering error on 7 could have been the culprit for a lot of my problems.
> That Topcon should be giving you 5-6" precision on 2D&R. What is your least count settings for angles? I have seen much better results on 5" instruments with 5" readings.
If you're looking at the numbers for the 201 and 202, I think the 203 is not as good. In Parameters, "Minimun Angle", the choices are only 10" and 20". I have it set on 10".
>
> That vertical angle spread affects your angular readings when you turn away from the direction that you leveled the gun in.
>
> Set your gun up and level the gun looking at the building, then zero on a vertical line such as a building. Check your sight line and angular reading as you go up and down that vertical line.
>
> Then relevel the gun turned 90° from the building, then zero on building. Check your sight line and angular reading as you go up and down that vertical line.
I'll definitely try what you're suggesting, but can't do it on my test site. The only buildings are a ways away, and, except for one with a pretty tall chimney, I wouldn't trust them at all...100 year old barns aren't exactly the straightest things in the book around here.
>
> Paul in PA
Next Steps?
Here's what I know so far that I need to work on:
1. More precision leveling the instrument and centering on stations. Rebar or spikes in the dirt only; no stakes!
2. Lot more work understanding how to use Starnet to its full potential.
3. How to do proper offset shots, to get around trees, etc.
4. More care recording, transferring, calculating angles and distances!
5. Anything else?
All that said, if we step back and look at this traverse, would it approach the precision required for a "rural survey"? Early on, I calculated the ratio to be something on the order of 1:5900, but a lot has changed.
The Starnet Adjusted Traverse closes completely of course (that's the point, right?). So to with the Compass Rule. But the question is, once outright blunders are removed, would the normal course of action (if this were a real survey) to be plot the Starnet-ed or Compass Ruled locations, and call it a day? I mean, I'm not setting the foundation for the World Trade Center.
How does a surveyor truly know when "Good" is "Good Enough" for a given application? Is it all just about how the numbers come up?
If You Eliminate 7, You Need
You got a tighter fit by holding too many fixed values and telling the program things you haven't measured. That's ok as a diagnostic tool, but not to claim as results.
Holding point 2 is a good move; if you really want to use it as your reference point it will give tighter results than basing on point 1.
However, after picking point 2 as the fixed one and a bearing to hold, you should NOT hold angle 1-2-3 fixed. Use EITHER (A 1-2-3 !) OR (B 1-2 !). You only get to arbitrarily pick three (3) things (the ! marks). Two coordinate values for point 2 and either one bearing or one angle uses up your free choices. Imagine sliding a map around. You get to freely pick an X position, a Y position, and a rotation for the map without distorting the map. To keep the statistics accurate you can't fix more things.
Also, after picking point 2 as the fixed one, you have NOT measured coordinates for point 1 or point 6. Those are calculated from your other measurements. To keep the statistics accurate you can't tell it you measured coordinates (unless you are using GPS and know your standard error).
My run of your data, with these corrections, is a lot better than prior results. The residuals are getting closer to what we might expect.
That leaves the problem of leveraging angle 1-2-3 out to point 6, which magnifies into more uncertainty on that end of the network than you probably would want to accept. Adding a measurement of 1-2-9 and repeats of those two angles should tighten things up.
You could hold an arbitrary bearing on one of the long lines and also tighten things, but I under you wanted your results referred to B 1-2, so a different line probably isn't an option.
If You Eliminate 7, You Need
>
> That leaves the problem of leveraging angle 1-2-3 out to point 6, which magnifies into more uncertainty on that end of the network than you probably would want to accept. Adding a measurement of 1-2-9 and repeats of those two angles should tighten things up.
>
I really appreciate your help! I think I'm starting to get a feel for what's driving the errors.
My next step is to re-do 1-2-9 AND 1-2-3 completely. I've never back sighted to 1 with a prism on a tribrach with an optical plummet (only to a survey stake 3 feet tall...argh!). I'll take extraordinary care to center on the top of the rebar (which extends from the earth at both 1 and 2 about 8 inches, and feel pretty solid). I'll do Left and Right x 3 both ways, to both points.
This'd go a lot quicker if I quit my day job.
Next Steps?
> How does a surveyor truly know when "Good" is "Good Enough" for a given application? Is it all just about how the numbers come up?
That is what the LS gets the big money for, grasshopper.
If The Choices Are 10"And 20", You Have A 20" Instrument
Typically you can set an instrument to about 1/2 it's precision. A lot of 5" instruments can be set to 1". The info I Googled says a Topcon 203 can be set to 5" or 10". What are the letters associated with the numbers 203.
Paul in PA
If The Choices Are 10"And 20", You Have A 20" Instrument
It's a GTS, not a CTS.
So, if I have a 20" instrument, what should I put into Starnet Options for the Angle accuracy? And, more important (to me right now): How close to the capability of the instrument am I coming on this exercise?
Big Money
> > How does a surveyor truly know when "Good" is "Good Enough" for a given application? Is it all just about how the numbers come up?
> That is what the LS gets the big money for, grasshopper.
You don't fool me. The guys around here do this for the love of the Art. I can tell.
😀
