Osgb Accuracy?

[moleskinner]

Looking for a little help.

 

I have been locating caches on my PC Memory Map and downloading to Garmin 12 GPSr. Have noticed that WSG84 downloads replicate the Memory Map co-ords, butwhen switching to OSGB system on both products, downloads on the garmin are consistently 3m "out on the Eastings and 9m "out" on the Northings.

 

Is this just conversion discrepencies? but which one is more accurate ?

[+wildlifewriter]

(etc) ...Is this just conversion discrepencies? but which one is more accurate ?

 

Both are accurate but (to oversimplify) the Earth is round - unless you are in Arkansas - whereas maps are flat. (Except when you let the kids fold them.)

 

To represent a curved surface on a flat map, needs a positional distortion known as the "map projection". There are many ways of doing this, according to the use being made of the map, the area covered by it, and so on.

 

A whole lot more about this, here... GIS mapping

 

Every map -digital or paper - that you might have, should if it's any good, specify the "datum". That's the setting you were altering on your GPS'r. If the datum selected is different from that under which the map was drawn, you will see the 'discrepancies' you mention. However, this is not a fault in the GPSr - it's only doing what you told it to do.

 

Now... I assume that the MM maps you mention are OSGB ones. And you would think that the correct applicable datum for these would be "OSGB", right?

 

Wrong.

 

The MM maps have been adapted for use in GPS applications, and seem to use WGS84, so that's what they use when displaying a Lat/Long entry.

 

Hth,

 

Paul

[+wildlifewriter]

I inserted the wrong link in that last post. (Well, it's late and I am an old man and will probably die, soon.)

 

The link should have been: National GPS network

 

Note the bad-news comment about accuracy levels in the transformation calculations.

 

Paul

[moleskinner]

Wildlifewriter, thanks for taking the time to post reply.

 

I have some (basic) knowledge of the geodesic models used for the globe and I had previously visited the site you refer to, but don't think it adresses the issue I am experiencing.

 

MM OS maps are based on WGS84 and must use a conversion to display BNG co-ords - I think. The dicrepencies I refer to are not those encountered when using BNG co-ords on GPS with WGS84 datum selected (not to be done!) as this can lead to 100s of metres out compared to the 10m or so I was referring to.

 

MM's conversion to display BNG co-ords must be different to that of the GPS as the results are not equal and I was wondering how to ascertain which is more accurate (as they are not the same they cannot be equally accurate - can they??)- it would help to know this when using OS maps and BNG co-ords as an aid to locating caches which I trust other cachers sometimes do?

 

thanks again,

Moleskinner

[+Naefearjustbeer]

I have noticed this in the past, Highland Nick had given me some co-ordinates to look at and I used Memory map to convert, I got different results from him when he used his etrex to convert.

I can only asume that memory map and garmin use different calculations to convert between different formats, But given the accuracy of the gps I think the discrepancies will make no difference in the real world when you are searching for caches. You will be taken to the general area and you can then use your geocachers eye to find the exact spot. I didnt loose any sleep over the differences maybe Nick took it and investigated further but I dont know, It is all rocket science to me

 

Cheers

 

Donnie

[+The Forester]

The reason for the discrepancy is that the box uses a highly simplified set of datum shift parameters. The problem is not in converting between the two spheroids, only in the datum shift.

 

The Ordnance Survey grid co-ordinates are based upon a legacy set of triangulations and network adjustments which have imperfections of many metres in places.

 

Picking a suitable dX;dYdZ data set is necessarily arbitrary and can only be optimised for one area. If you perform the conversion for somewhere else a hundred miles away with the same parameters you will find a different error.

 

When these little handheld single frequency GPSs came onto the market, these differences were academic because the SA errors were several times greater. Also, until just a few years ago the OS used to charge lots of money to convert just a single point from WGS84 to OSGB36. They refused to divulge the parameters they used, though in any individual case it was quite easy to back-calculate their results by reducing both sets of co-ords to their Cartesian co-ords and subtractively deduce the correction parameters to use.

 

They had divided the country up into numerous blocks. I can't remember whether it was 9 or 19. They used the adopted 7 parameters for that particular block for each calculation they did.

 

Because the OS didn't divulge their magic numbers, the GPS manufacturers couldn't replicate them. Anyway it would have been very expensive in software terms to do all this work for one country when that algorithm is not applicable anywhere else in the world.

 

OS policy has fundamentally changed and they have embraced GPS wholehertedly and spent a lot of money on devising and improving a very sophisticated mathematical model with which to do the conversion to a very high level of accuracy. At first you had to pay about £135 to buy the program, but now it is avilable free online. It's called Grid InQuest and you can probably find it with a google search. It produces what mount to the definitive conversion between OSGB36 and WGS84.

 

Using this correct conversion for a particular locality, we can back-calculate what the correct dX; dY and dZ parameters should be. If your devices allow you to input user-defined set of conversion prameters, you can then use those ones to quasi-replicate what the full and hugely complicated official polynomial model would produce -- certainly to sub-metric accuracy.

 

Trying to explain the maths in easily understandable terms without a worked example is hopelessly difficult, so please give me a set of example co-ords to play with and I'll show you how the trick is done. Try to use a dataset that both of your machines have already given you results for.

 

Cheers, The Forester

[moleskinner]

Phew, thanks The Forester. Think I understand what you are saying. But perhaps Naefearjustbeer had a point(thanks NFJB), that difference is neglible cos we do not need the accuracy anyway - (cannot plot this level of accuracy even on OS 1:25,000 and can use Lat/Lon co-ords anyway for cache locating). So perhaps not worth pursuing further.

 

Still would be nice to know if Memory Map (MM) or GPS was the more accurate conversion for my area i.e. Scotland.

 

So if you can be bothered!

 

I tried converting the following Lon/Lat co-ords (WGS84) with both machines and got the results below:

 

Lon,Lat (deg mm.mmm) - 55°58.305'N, 4°17.496'W (Downloaded from MM to GPS which then displays this identically)

 

Memory Map conversion to OSGB - NS 57082 77791

Garmin 12 GPS conv. to OSGB - NS 57084 77783

 

Thanks again

Moleskinner

[+wildlifewriter]

Still would be nice to know if Memory Map (MM) or GPS was the more accurate conversion for my area i.e. Scotland

My 2d worth...

 

Working this out from the OSTN02 transformation, I get the following:-

 

From: N 55 58.305 W 04 17.496

Convert to Deg.Min/Sec: N 55 58 18.3 W 04 17 29.76

Convert to OSGB(36): 57081.2 , 77788.9

 

You pays yer money and you takes yer choice

 

Incidentally, I also tried to see what Garmin's MapSource software would do with these figures. It gave the same result (exactly) as your Garmin GPSr. (No surprises there, then.)

 

So... why the discrepancy? At a guess, perhaps GPS software designers were worried (quite rightly) that their wee machines did not have enough memory/processor power to use the horribly cumbersome OS tables and equations. They probably employ a simpler (but slightly less accurate) formula which can be implemented at some reasonable speed.

 

You and I would be the first to complain if our Garmin receivers paused every now and again, with a screen display which read...

 

----------------------

FiGURING OUT

WHERE WE ARE

 

(Please wait...)

 

----------------------

 

 

Paul

[+The Forester]

The rigorous algorithm converts your WGS84 co-ords to OS grid co-ords of:

257,081.23E 677,788.92N

 

Thus we can see that the MM calc is 2.22m 020° from true position

And the G12 is 6.54m 155° from the true position.

 

That resolves which of the two boxes has the lesser error in West Central Scotland.

 

I quite agree with Naefearjustbeer that chasing very fine accuracy of OS co-ords is not very useful in the context of geocaching. The geocaching datum is WGS84 and the resolution of the third decimal minute is 2.15m at your Latitude.

 

Anyway, for we in Scotland who are not currently using the WAAS/EGNOS signal, we can’t get GPS accuracy of that sort without a lot of averaging.

 

So, the simple answer to your original question is that your MM gives better OSGB36 conversions than your G12.

 

For very special situations, such as using the OS 1:1,250 scale sheet, we would need to calculate the local dX;dY;dZ and I’d be quite happy to do so, but for general purposes you can just use your MM for such conversions, perhaps applying a correction for the couple of metres shown above if you want to be picky.

 

Actually, your G12 has a little known undocumented feature which allows you to extract phase difference data which can be postprocessed to produce a static fix accuracy of well under 10cm if you have the patience and a bit of expertise, but that’s beyond the scope of this discussion.

 

 

Cheers, The Forester

[moleskinner]

Thanks to both Wildlifewriter and The Forester.

 

I am extremely impressed by the knowledge both of you have on this subject which is backed up with the conversions, especially as they are pretty much identical!

 

So GPS is slightly less accurate for OSGB36 co-ords, but not to lose sleep over it.

 

Got to be said this is a pretty useful site as I have asked Memory Map the same question without response!

 

Case closed I think, and thanks once again.

 

Moleskinner