Well, I can probably count on one finger the number of people
who both read this page and are interested in LIDAR base
maps, but Eddie sent me this note in response to my whining
about the LIDAR at Rosaryville. I want my minutes back for
the time lost on #9!.
Hey dude,
I just read your log regarding rosaryville. I've been
playing with the original lidar data that they used - Greg Lennon
sent me a piece of it. I have to tell you that this is rather poor
lidar data, and its likely that Price Georges county got screwed by
the contractor. The reason Jon couldn't map the gullies from the
lidar data, and the reason you (correctly) felt that the contours were
soft is because the lidar data was apparently collected leaf-on.
Greg had to smooth the resulting bare-earth data to make the contours
because there was so much leftover vegetation corruption of the bare
earth. That is, because of the leaves there were very few ground
returns even recorded. Have a look at some images I made on this web
page:
http://www.catchingfeatures.com/eddie/rosary/page1.html
Have a look at the top two links. First one is the "bare-earth" lidar data
that the contours were made from. This has been filtered to remove the slowly
varying elevation changes, leaving behind only sharp features - somthing
called unsharp-masking. The upper right corner of this image is the NW corner
of the O map. You can see the flooded lake that we went around and the forest
to the N of it. Notice all the crinkly looking things in the forested area.
This is 1m posting lidar data - normaly very good, but in this case you can't
make out much of any detail on the ground. Now check out the lower left of
this image. There are some cornfields down there and you can actually see
places where deer have walked through and bedded down. This is a hint that
the data was collected early in the season. As it turns out the data was
collected in October. The second link there is the google maps image
of the same area, which in this case is leaf-off and there is nothing
standing in any of those fields.
Now have a look at the last link on that page (fr_lr_compare.jpg).
The top two panels show a piece of the lidar from near my apt. Left is
first return and right is last return. In this case, the
last returns are indeed mostly reflections off the ground, as the
laser pulse first reflects off of high branches, and then hits the
ground below. Note that the powerlines and most of the tree canopy is
gone in the last returns image. The only thing remaining are the large
tree trunks and some thicker understory. All of this remaining stuff
can be easily filtered to leave behind very clean bare-earth.
Now look at the bottom two panels. This is the same first-returns left,
last returns right, but from the Rosaryville data. Almost no difference
between the two. The laser is simply not penetrating to the ground.
So now when you try to filter the remaining vegetation, there is almost no
ground information to work with. You end up throwing out nearly every
point, and thats why the bare ground in the Rosaryville unsharp masked
image looks so ratty. Greg had to smooth pretty heavily to keep all those
jaggies from corrupting the contours, which made them softer. Also
might explain the higher than expected #9.
The third link on that page shows some bare-earth unsharp maked data from the
Mckeldin map that Ted is working on. Left is 2m res lidar, right is 1m res.
Both are under full forest canopy. This image covers about 700m from top to
bottom. Note the level of detail in both. Vegetation filtering was excellent
in both cases (was leaf off). The 1m data is just exquisite, with pits,
faint trails, ditches etc easily identifiable. The Rosaryville data was also
1m, so should have looked very much like this if it had been taken leaf-off,
and Jon would have had not trouble at all mapping the gullies right
off the lidar.
Anyhow, this is something I've always been worried about. People using lidar
blindly without minding the limitations of how the data was collected. There
is good lidar and bad lidar, just like any other remotely sensed data. THis
is about the worst I've seen. So please don't lose faith in it :) Ted's new
map is excellent (his first one), and the new Prince William map used for the
US Champs also had very good contours. Greg is going after PG county now to
see what gives and hopefuly we can get Jon some better data to fix the
Rosaryville map. It seems like a pretty nice area.
Oh, and here is a nice dataset of Great Falls in D.C. Just found this
last week. Pull out an old map and see what you can identify:
http://www.catchingfeatures.com/eddie/gf/page1.html
