Notifications
Clear all

Residuals plot

9 Posts
5 Users
0 Reactions
51 Views
(@geodesist)
Posts: 49
Member
Topic starter
 

Hello, Dear colleagues

Please I need your help,

Could you tell me, what's the interest and why we use the Residuals plot (for statistical study).

Thank you

Cordially?ÿ

 
Posted : January 8, 2021 9:20 am
dgm-pls
(@dgm-pls)
Posts: 274
Member
 

It tells you the amount and direction that your control is weakest in.?ÿ You can use this info to tighten up your control network if the residuals are too high.

 
Posted : January 8, 2021 10:11 am
rover83
(@rover83)
Posts: 2346
Member
 

Plots of the residuals primarily assist in analysis of a least squares solution and blunder detection. Histograms will show skew and clusters of residuals. Ideally, systematic errors and blunders will not be present in the dataset, leaving you with random error, which in theory should plot as a normal curve. There are some methods to calculate a theoretical cut-off for maximum acceptable size of the residuals, which you can use to decide whether to accept an observation, or remove it from the dataset and re-run the adjustment.

However, standardized residuals (rather than the raw residuals) are generally more useful for determining whether or not an observation is likely a blunder. There are a couple of methods available, such as data snooping and the tau criterion, for computing the cut-off point.

XyHT had an excellent 4-part series on the subject. Here's the first one (has links to the rest of the series at the bottom):

https://www.xyht.com/surveying/methods-analyzing-data-part-1/

 
Posted : January 8, 2021 12:04 pm
geeoddmike
(@geeoddmike)
Posts: 1556
Member
 

I could not find the post by Jim Frame but a good example of the utility of plotting data is here:?ÿ

453AD771 6782 4579 8AF8 A586C01CBA4D

Outlier detection is greatly simplified via the use of graphs. In GPS data processing high-end tools include residual plots to analyze data and detect outliers. While detecting and fixing cycle slips are highly automated nowadays, in the ƒ??old daysƒ? we had to manually (creating an edit file) fix cycle slips.

As for the the topic of the statistics of residuals, I recommend:

https://geodesy.noaa.gov/library/pdfs/NOAA_TR_NOS_0065_NGS_0001.pdf - Allen J Pope paper.

The source of the graphic (that also includes a good discussion of precision v accuracy) is here:

https://wou.edu/chemistry/courses/online-chemistry-textbooks/foundations-general-organic-biological-chemistry/chapter-1-measurements-chemistry/

HTH,

DMM

?ÿ

?ÿ

 
Posted : January 8, 2021 12:30 pm
geeoddmike
(@geeoddmike)
Posts: 1556
Member
 
3D0961C1 9508 439E B5A6 A6C30B37DD87

Another example of the utility of plotting from XKCD...

 
Posted : January 8, 2021 12:55 pm

(@geodesist)
Posts: 49
Member
Topic starter
 

@geeoddmike Thank you for your valuable help. Really much appreciated

 
Posted : January 9, 2021 4:42 am
(@geodesist)
Posts: 49
Member
Topic starter
 

@dgm-pls Thank you for your valuable help. Really much appreciated

 
Posted : January 9, 2021 4:43 am
(@geodesist)
Posts: 49
Member
Topic starter
 

@rover83 Thank you for your valuable help. Really much appreciated

 
Posted : January 9, 2021 4:47 am
(@mike-marks)
Posts: 1125
Member
 
Posted by: @rover83

Plots of the residuals primarily assist in analysis of a least squares solution and blunder detection. Histograms will show skew and clusters of residuals. Ideally, systematic errors and blunders will not be present in the dataset, leaving you with random error, which in theory should plot as a normal curve. There are some methods to calculate a theoretical cut-off for maximum acceptable size of the residuals, which you can use to decide whether to accept an observation, or remove it from the dataset and re-run the adjustment.

However, standardized residuals (rather than the raw residuals) are generally more useful for determining whether or not an observation is likely a blunder. There are a couple of methods available, such as data snooping and the tau criterion, for computing the cut-off point.

XyHT had an excellent 4-part series on the subject. Here's the first one (has links to the rest of the series at the bottom):

https://www.xyht.com/surveying/methods-analyzing-data-part-1/

That graphic example works great if you have one angle/distance blunder and can isolate the problem to a single observation & reobserve, been there done that.?ÿ Not so much if several stations are suspect in a traverse.?ÿ The great advance that network observations (thanks to EDMs) provide is an adjustment where one station (or several) sticks out as a stinker and needs re-observations.?ÿ Or, perish the thought a first order control station is BS and if you ignore it your network is golden.

It's a new world where distant GPS translocation can distort local accurate transit & tape surveys to where they're a few feet off and your neighbor owns 3 feet of your driveway.?ÿ I still hold?ÿ record monuments but where there are none it appears the GPS guys win out because of superior accuracy from section corner a half mile away.

 
Posted : January 17, 2021 8:15 pm