height and image data mis-aligned in some cases

[unregistered]

I have noticed that in some cases the height data and the landsat images are not aligned.

I downloaded the 'add-ons' 04-australia.zip and 13-machu_picchu.zip.

If I zoom into Sydney Australia and 'tilt' the camera I can see great lumps in the ocean right on the coast and some of the headlands are flat. It looks like the height and image are a kilometre or so out of alignment - I know Sydney so I can see that the 'topography looks right but the images are out of register.

Similar issue at machu picchu, if you go there and tilt the camera you can see that the river that in reality goes around the 'spur' of mountain is displayed as going up the mountain and down the other side.

Do you think it could be some config on my end?
Is there some way from me to play with some offsets somewhere?
Is the data being handed out to everyone flawed?

[cloud]

i saw this too. see this missaligned river. i guess it's about 100 m missalignment. how can this happen?

http://n.ethz.ch/student/apfister/river.png

PS: try to find where it is ;)

[Beansprout]

There's a few bugs concerning place names being misaligned too. Seems like something's wrong in the algorithm WW uses.

[TomServo]

Originally posted by TheBeansprout@Nov 4 2004, 07:16 AM
There's a few bugs concerning place names being misaligned too. Seems like something's wrong in the algorithm WW uses.
Quoted post


Chris mentioned a few times he thinks that applying a curve to the data (to map it to the globe) is one of the causes for miss-located items on the maps.

[cmaxwell]

Actually, this error is caused by the Landsat imagery. Unfortunately, the JPL Landsat imagery is very poorly aligned and stitched together, causing severe problems with geo-accuracy of the imagery in places south of the equator. If you look at a more overall picture of the australia and new zealand area, you'll notice the Landsat "swaths" do not match up well and are off by tens of kilometers in some cases. Lucian is aware of this and he'll be re-making the JPL Landsat mosaic in the future.

In the meantime, we're searching for other sources of data that will hopefully solve this problem of geo-accuracy in Landsat imagery.

Stay tuned. :)

[ear1grey]

Originally posted by cmaxwell@Nov 9 2004, 05:46 AM
Lucian is aware of this and he'll be re-making the JPL Landsat mosaic in the future.
Quoted post


Hopefully such a resample might resolve the ghosting in southern New Zealand too.

/me attaches example, crosses fingers.

[Jonatan Kelu]

A funny one to demonstrate this problem is if you have a look at Uluru (Ayers Rock) in Australia. You can see the image of it flat on the ground, and then a few kilometres from the image, the land rises up in the shape of the rock. So the rock is there, but its image is a few kilometres off.

[Lucian]

Originally posted by Jonatan Kelu@Nov 19 2004, 12:07 PM
A funny one to demonstrate this problem is if you have a look at Uluru (Ayers Rock) in Australia. You can see the image of it flat on the ground, and then a few kilometres from the image, the land rises up in the shape of the rock. So the rock is there, but its image is a few kilometres off.
Quoted post


That is one bad thing, isn't it. Turns out the precision in the UTM2Geographic transformation I was
using was not that great, I will redo that code for the next version of the mosaic. To make it worse,
about 50% of the scenes were re-processed since I built the mosaic, I assume some georeferencing
errors were caught. It will not be perfect however.
Both the Landsat and the SRTM are georeferenced based on tie-points, which are not very dense
(or well located) in remote areas such as the Australian outback, or most of the south South America.
The SRTM itself is patched in places, so if an area looks funny, it is because data is not available and
has been filled in. The Gibraltar Rock is a good example of that, there are some strange spikes in the
water right next to it!


Lucian

[Lucian]

Oh yeah, the US is aligned right in most places, because it uses the National Elevation Dataset (NED),
not SRTM, and tie-points are very precise and dense.

Lucian

[Bill S]

Originally posted by TomServo@Nov 4 2004, 12:33 PM
Chris mentioned a few times he thinks that applying a curve to the data (to map it to the globe) is one of the causes for miss-located items on the maps.
Quoted post


There also seems to be an issue with altitude - placenames are placed at sea level so their position depends upon which side of the screen they are on: select any area in the Andes, zoom down, then move it across the screen - towns will move from one side of a lake to the other.
Just read of the new enhancements - placenames heightmapped - please ignore this post

Having great fun with this software!

[Beansprout]

Bump.

Been a few threads about elevation etc recently...so hopefully people will see this one B)

[SwingKid]

Since other people are mentioning this I thought I'd add some examples from my region of interest in south-eastern Europe. The monument under the cross-hair here:

http://www.waves.utoronto.ca/~pavacic/loma...ter/kerch03.jpg (http://www.waves.utoronto.ca/~pavacic/lomac/theater/kerch03.jpg)

Should be about 0.00200 degrees south, on the hilltop here:

http://www.waves.utoronto.ca/~pavacic/loma...ter/kerch04.jpg (http://www.waves.utoronto.ca/~pavacic/lomac/theater/kerch04.jpg)

(sorry, using v1.2c here, coordinates shown in the images)

You can see this better far away to the south-east in he next image. The dam, the bays around the reservoir and the river to the east all look like they should be about 0.00200 degrees Lat. south of where they are with respect to the terrain. So it looks like it may be a consistent mis-alignment over a large area of tiles.

Thanks for a great system! Hope this helps,

-SK

[SwingKid]

Oops, forgot to include the "next image":

http://www.waves.utoronto.ca/~pavacic/loma...ter/jvari04.jpg (http://www.waves.utoronto.ca/~pavacic/lomac/theater/jvari04.jpg)

-SK

[DaveD]

Here's my 'find' for misalignment: Queen Mary's Peak--2160m [37d06m S; 12d17m W]--a tiny inhabited South-Atlantic volcanic island. The LandSat image is perfectly flat; due north a shadowy gray cone rises sharply out of the ocean (with an attached placename tag). The northern coastal edge of the island rises about 800m up the side of the 'ghost' cone. Exciting topography! Dave

[DaveD]

This is the *same* place; the peak is one of 3 islands in this group.
[You can easily convert the coordinates: 37deg + .(6/60); south is (-); lat= -37.1;
12deg + .(17/60); west is (-) long= -12.283]. My coordinates were taken from Encarta, so they're close but not exactly the same!

Oddly, some other islands in the same region are spot-on. Dave

[Guest]

My entry for misalignment is posted on the worldwind web site:
http://worldwind.arc.nasa.gov/forums/index.php?showtopic=465
Certainly someone has noticed this before?