Return to Dr. Matt Nolan's 2011 Photography Page
Of course the purpose of our air photos is to do something scientific with them. Just looking at the images individual yields a lot of useful and interesting information usually, but there is a tremendous amount of additional information that can be extracted from them. One of the way is by using them to make topographic maps. The photos have to be acquired in a certain way to ensure this can be done reliably, but usually we are taking the time to do so. Below are a few snapshots and some data files (wth a viewer so you can fly yourself around them) of a somewhat random selection of projects. These can be thought of as back-of-the-envelope projects just to assess data quality or visualization; none of these data should be used quantitatively in the form presented here.
Pilgrim Hot Springs
In fall 2010, we acquired several thousand images around Pilgrim Hot Springs. One of the goals was to make an orthorectified mosaic of these images, using coarse terrain data from USGS. That's easy enough to do, but I wanted to see whether we could use terrain data made from the images to orthorectify the sameimages, so that we could get a much higher resolution and more interesting project. Because making a topographic map was never a mission objective, the lens used was long and matched the field of view of a FLIR camera flown simultaneously, and thus the base-to-height ratio is not optimum for topographic mapping and there is insufficient overlap. But I thought that by processing pairs from adjacent flight-lines, I could make a complete map. The block below way my only idle test of this for an hour's effort. By finding the right pairs, I was able to fill the holes, but this did add some noise. Ultimately I think we could make a nice map at 50cm spatial resolution and about that in vertical accuracy, though no doubt there few holes and slivers would remain unmapped. One of the issues was the air was really turbulent during this mission, so roll's were large.
Here you can see the overlapping sections of many photo-pairs. I focussed on filling the central section, where the buildings associated with the hot springs are located.
Here is the same map with colors representing height. You can see individual trees in some places.
A 3D oblique view.
I was surprised how well the trees worked out. One could use this for canopy height measurements fairly accurately.
During our Jago and Hulahula River missions, I usually try to swing over McCall Glacier to check things out. Here is the result from one pair of images from June 2010. This map compares exceptionally well with recent lidar we contracted out, and at a fraction of the cost.
Processing these for topography seemed like an interesting challenge, as these were shot with only making a controlled mosaic in mind and used a long lens to try to look straight down into the water without getting reflections. Despite the poor geometry, I was able to extract a lot of quantitative information from this test pair.
I thought this might make a nice Christmas Card for local fish tank owners. You can actually see elodia blooms in the bathymetry, especially in the image below.
The oblique images look like something from a model train set.
Thermokarst slumps near the Noatak River
One of the primary mission objectives for the Kangilipak and Kavuchurak blocks we flew this summer was to measure volume change caused by thermal erosion of the landscape. Much of the ground here is underlain by glacier ice that is likely more than 10,000 years old. As it gets exposed for whatever reason, it melts and causes slumps.
Here are two overlapping pairs over a lake with a large slump on its lower left side.
The results we are getting yield spatial resolution on the order of 30-50cm and a noise level on about the order. We can clearly resolve the topography of ice wedges between polygons with these maps, despite the signal being on the same order as the noise, because the noise is somewhat random and the signal is somewhat ordered.
The primary objective of the Noatak block was the measure the topography of several dozen pingos there. I really like the ways these came out, I had never really spent much time looking at pingos before. They have clearly different vegetation communities on their north and south sides, and all kinds of interesting topography. We can make useful maps of these at 15-25cm spatial resolution, with similar noise. I think it's definitely worth downloading the viewer so you can see this one in 3D.
The blobs surrounding the pingo are clusters of willows and spruce trees. We can easily measure canopy heights and spatial distribution with these maps, since the stands are so isolated.
One of my main airborne projects is mapping change in the Jago and Hulahula Rivers, as part of our ecological studies. Below are some example products from our flights this summer.
Here are about 100 images stitched together into a DEM and associated orthoimage. It was quite challenging flying at a constant height above ground while staying directly over the river, but we captured everything. If you look close you can see how at the turns the images are sucked to the outer edge as we banked the plane to make the turn, yet we still got the river beneath. This section goes from McCall creek almost to Bitty.
There are four image seams in this image -- can you find them? I cant. I was amazed at how well these stitched together. Remember, this is also an orthoimage with underlying topography.
Here I made the seams more visible by zooming in on the point cloud. The middle section of overlap has higher point density so doesnt pixelate as quickly. I'm pretty sensitive to detecting stitching error based on my panoramic photography experience, but I couldnt find any.
An oblique on the point cloud, with some vertical exagerration.
An oblique on a DEM. You can see some shading differences at the seams. The outliers are now visible as spikes. These seem mostly associated with water bodies, like the river and puddles, where refraction messes with the geometry.
Here's a close up of the DEM surface at one of the seams. The vertical offset here is about one meter. The tussock field has an amplitude of about 50-75cm, and is likely caused as an artifact of merging point clouds that are slightly offset and have different spacing. This model is just the first pass attempt, so even though it's a bit ugly, it can only get better from here with more refined processing as is it's really not bad in terms of accuracy. Floodplains like these are perhaps the most challenging mapping environments, as the lack of topography not only brings the signal closer to the noise, but also reduces the strength of the 3D solution itself, causing the errors to get larger. Having only a single strip of images, rather than a block, also weakens the solutions. But even given all this, I think this is an amazing first-pass result. Within single pairs, the noise level is about the same as in previous examples, about 25 cm.
Another view of some of the artifacts, looking at the underside of the surface. Most of the spikes are associated with water, and the largest vertical offset I found at a seam was this one at about 1.5 meters.
Please remember that these examples are meant for visualization only, the scales and rotations are not necessarily accurate.
To interactively fly around some of this terrain, get the *reader* here (not the trial version of the full software).
Slump2 -- 53MB
Slump3 -- 10MB