Video
Modeling of Archeological Phenomena with Satellites
Transcript
Karen: Hello, and welcome to the NPS Archeology Program Speaker Series for fall of 2013. My name is Karen Mudar and I'm an Archeologist in the Washington office. This webinar series is devoted to geophysical technologies in archeology. I'm sorry to stop, I'm having a little bit of trouble with our ... Let’s try this. Here we go. On December 19th, NPS Archeologists Anne Vawser and Amanda Davey are going to speak to us about GIS applications. Anne is the leader of the NPS Midwest Archeologyl Center’s Archeological Information Management team. The center is the repository for over 5 decade’s worth of archeological research at almost 60 parks in the Midwest Region. The team is tasked with building and maintaining spatial attributes of these data in a geographic information system.
Anne and Amanda will talk about ways that GIS has become an integral part of archeology conducted at the center. I'm looking forward to that. I’ve heard them talk before, and they’ve had very interesting research to report. Last week, Drew LaBounty at Voyageurs National Park talked to us about a GIS project that he’s carrying out at the park. Until I heard his lecture I didn’t realize how closely Drew had worked with retired NPS archeologist Jeff Richner. Richner, as you remember, was awarded the 2012 Cotter Award for Project Excellence, for a project that resulted in an archeological and ethnohistorical synthesis of the historic use of the area occupied by the park, by the Bois Forte Band of Minnesota Chippewa Indians. Drew’s GIS project provides a visual dimension to Richner’s and others research, and it enriches our ability to use these data.
It was a great example of some of the many ways that GIS can integrate archeological, historic, and ethnographic research. Before I introduce today’s speaker, I want to remind people that the presentation will be recorded, so please set your phone to mute unless you’re speaking. Also if you’re having difficulty viewing the images for the presentations, you might check your computer’s operating system. We had some trouble with our speaker for today. Windows 7 and Internet Explorer seems to be a winning combo for PC computers. We want to apologize for our tardiness in posting the webcasts from the Fall webinar series. Our web designer has been very busy, posting a chapter on cultural resources and fire to the archeology research manual, and he’s finished building the web pages – Yay! - and the working group is reviewing the product.
We hope to make the module available soon, and I will announce the URL in the E-gram when we do. I first learned about today’s speaker through an article in Park Science. Alan Sullivan is Professor and head of the Department of Anthropology at the University of Cincinnati. His research is focused on the ways that the dynamic interplay between anthropogenic fire and intensive wild plant production affects economic decision making and regional abandonment patterns. Today he will discuss the possibilities as well as the challenges that arise from satellite remote sensing as extended to direct predictive modelling of small scale archeological phenomena at the regional level of analysis. Of special interest to you, I think, he will be using research that he’s been conducting on the South Rim of the Grand Canyon as an example.
Alan suggests that you ask questions during his talk rather than waiting until the end. The cultural resource folks that use some of these trained me to monitor the question team - thanks guys! - during the last webinar so I’ll be checking that as well in case it’s hard for you to get to a phone. I’ve been looking forward to this presentation with great interest. I'm curious to see the progress that’s been made on using satellite photography for archeological research. I realize that I’ve forgotten to advance my screen here. Let’s see if we can make this work. Here we go. Thanks for speaking with us today, Alan, and we look forward to your talk.
Alan: Terrific. I'm happy to be here with everyone in this virtual world, and my strategy today is to introduce everybody to some of the capabilities of satellite remote sensing, and importantly some of the interesting applications that we’ve been exploring with certain classes of satellites. I think I’ll be very candid here, I come at satellite remote sensing as an archeologist. I know a little bit about some of the technical details, but that shouldn’t dissuade you from asking me questions. If I can’t answer them, my collaborator Kevin Magee will certainly be able to do that. Karen, can you hear me okay?
Karen: Yes, we can.
Alan: Okay, terrific. I'm going to start with the main body of the presentation now, okay?
Karen: Yes, please do!
Alan: You’ve got our banner page, and I'm going to quickly move to what I consider to be some essential reading for those who would like to get both their feet and their brains wet in satellite remote sensing. These are 3 books that I would highly recommend to people who get inspired by our webinar today. The first one I highly recommend by Sarah Parcak, that’s how you pronounce her name. She’s at the University of Alabama in Birmingham, and this book provides a wonderful overview of the capabilities of satellites going back several decades, as well as some examples of her work in Egypt, where she has been using satellite remote sensing to understand - to discover and understand - the spatial patterning of large sites in Egypt. The next 2 items are fairly recent. The first one is an edited volume that illustrates, I think, on a global basis, many applications of satellite remote sensing.
I like this book a lot because it provides some opportunities, particularly for American archeologists to gain a sense of how European archeologists, in this case, Italians, have been using remote sensing in environments that in many cases have attributes that are similar to the ones that we work with. The last volume is the more recent book, I think it just came out this year, and it represents the wide range of applications of satellite remote sensing, I think there’s studies in there with lit up LIDAR as well, and it gives a very distinctive American perspective on applications of satellite remote sensing.
I will say that, as far as I can determine, there are probably nearly 100 archeologists who actively engage in applications with satellite remote sensing to archeological problems. The number of archeologists who are involved in this type of research has accelerated dramatically, particularly during the last decade. Here we have just the basic outline of many of the topics that I want to talk about today, and like most people I get really irritated when people read PowerPoint slides, so I’ll just go over these very quickly. First of all, I'm going to introduce you to what it is, what is the basic phenomenon that we focus on, satellite remote sensing. I’ve got here some things, some high points really that I want to discuss with you about some satellites and their properties.
It’s important to pay attention to 2 topics here, one is spatial resolution, which is actually the size of the pixel on the ground surface. The other is spectral resolution which refers to different regions of the electromagnetic spectrum, we’ll talk a little bit about that. There’s a joke, of course, this is satellite remote sensing humor but indeed the basic products of satellite remote sensing begin with 256 shades of gray. That is, indeed, very exciting. Then we’ll go into some basic examples of what you do with the images, and how you could process them and manipulate them with software. I’ll very briefly talk about the range of indices that have been used to illustrate differences between archeological phenomena and environmental phenomena. Then we’ll get into some archeological applications here focusing on differentiating anthropogenic phenomena from the natural world.
For those of you who joined the webinar hosted by Ken Kvamme, many of the conceptual principles that he talked about apply equally as well to satellite remote sensing. We’re trying to find anomalies in many cases that the human eye cannot detect in natural environment. There are some basics here that I’ll go over as well about the geo-referencing of images, and how you do that. This topic here called Deriving Spectral Signatures is a basic operation when you’re trying to not only identify the unique spectral characteristics of archeological phenomena, but to apply them in a predictive manner to unknown areas. Then we’ll talk a little bit about Ground-Truthing which is an absolute necessity when you’re dealing with predictive models.
We’ll talk a little bit about evaluating the results of these predictive models and then at the end, I want to make a plea for using older satellite images such as Landsat images to actually enhance our abilities to understand what humans today are doing to the environment particularly on public land. Here is, and I'm sure everybody will recognize the electromagnetic spectrum from their high school physics book, and what I'd like to do here, however, as you can see on the left, there are certain wavelengths that are most commonly used in satellite remote sensing, and you can see there’s only a bit of the entire electromagnetic spectrum that we actually use. We actually can see with the human eye, which is of course itself a remote sensor, it’s a very narrow range along the entire spectrum.
Now, what you see here is an example of the extent to which variation and light can be used to detect differences in some common kinds of artefacts, made from or are symptomatic of different kinds of materials. Just let me give you a little bit of background here, because what you’re seeing here is illustrative of the way that sensors on satellites work. What the spectrometer, that is calibrated through absolute brilliant light, you can then image as a satellite would do different kinds of archeological remains. You can see here that there are clusters or concentrations of different classes of artefacts based on how they reflect light.
The wavelength bar at the bottom is - I think - should be calibrated to 10 to the minus 3 very short wavelengths, but you can see for example that, with respect to natural materials, fire cracked rock, and fire hardened daub, they are differentiated from ceramics, which themselves are differentiated from rocks, which themselves are differentiated from obsidian. For those of you who are familiar with obsidian, particularly grey or black obsidian, you know very well that color of obsidian does not reflect light at all.
These kinds of examples illustrate how we can take advantage of the electromagnetic spectrum, when the sensors that are on a portable spectrometer are actually installed on satellites that orbit the earth, like these two. These are some of the most common kinds of ... These 2 are the satellites that produce the most common kinds of images that archeologists have employed in different kinds of remote sensing studies of archeological phenomena. You can see that the satellites differ in 2 respects from the ones I mentioned earlier. One is spatial resolution, that is the basic size of the image, pixel-wise, that is the unit of observation of the satellite. The other is the spectral band and the spatial resolution of those different bands. You can see that if you just compare those variables between these 2 satellites, that they differ dramatically.
Each is useful for different problems, as I’ll illustrate in just a while. I thought I would use this wonderful slide here that I got from one of the books that I mentioned just a while ago to illustrate the process involved in acquiring satellite images, and then using them for archeological purposes. The entire system is very different than that of, say, GPS, where we just basically acquire signals and then differentially correct them. As you can see here this whole process starts with our friend the sun. He sends electromagnetic energy at different wave lengths to us. Some of that energy can be corrupted or affected by the passage of the energy through the atmosphere.
The energy then gets reflected back up in space to be captured by one of the several satellites that you just saw. It is then processed and stored and sent back, transmitted to a receiver, and that receiver then stores the images. These private companies, they have their way of disseminating the images. Sometimes the images are free through government websites, but ultimately the user is the one who actually then begins to process the satellite images with various software. Here you can see some of the principle differences between satellites, in this case our friend IKONOS and Quickbird with respect to 2 of the principal features of the images, spectral resolution and spatial resolution. I could draw your attention to a couple of things here, one is these band numbers.
I think my little pointer is working okay, those are the band numbers, and you can see that the spatial resolution over here for the spectral band is less, that is, the size of pixel is greater than for basically black and white or panchromatic band. There’s also variation between the satellites and again I’ll use my little pointer here to show that IKONOS images have higher resolution because the size of the pixel itself is smaller and, depending on your problem, you can choose which of the satellites and their images are more appropriate for your problem. Now if we take a look at Landsat, we can see, that when we look at the same band here that the spatial resolution is far greater than we saw for IKONOS and Quickbird.
The point of this figure here is to illustrate what these bands actually do, and I think that it’s important, therefore, to look at the lower table to show that these bands have associated wavelengths, various colors that are processed electronically, and that they’re useful as the title of the column indicates, for different kinds of applications. In archeological applications, as you’ll see, we actually do focus on these first 4 bands, particularly in the area where I work, which is the semi-arid area in the American Southwest. This is one image of several that you’ll see this afternoon that illustrate the effect of different spatial resolution on the same archeological phenomena.
Here this is the famous site, of course, of Tikal National park in Guatemala, and you can see by comparing these images side to side, that in fact the IKONOS image picks up extremely small phenomena including little stelae down here in the courtyards and plazas of Tikal. You can see actual steps on some of the pyramids. Whereas if you look over here to the Landsat image, many of those distinctions cannot be differentiated, and that doesn’t mean that a Landsat image is bad or inferior, it simply means that it’s more appropriate for certain problems than others, given the nature of the resolution. We’re faced with these 2 different aspects of satellite imagery, spectral resolution and spatial resolution.
One of the virtues of some of the advanced software that’s been developed over the past several decades is that there is a way to take advantages of the spectral resolution, the range of spectral resolution and the range of spatial resolution in many of the satellites that we visited so far. Here is one process, this is an image manipulation process called pan-sharpening, and what it does is through various types of statistical manipulation of the digital data that are encoded in the various images. You can actually take a high resolution panchromatic or black and white image that has very high spatial resolution because the pixel image is small, in contrast to the low resolution or large pixel size of the multi-spectral image, and using either one of these transformation processes, just as an aside, we use this principle component analysis type of spectral sharpening to produce what in many cases turns out to be high resolution multi-spectral images.
It’s a way of this pan-sharpening or often called pan-merging processes is one of the ways to take advantage of the attributes of high spectral resolution, and high spatial resolution. Here’s an example from our home town here in Cincinnati. This is as the little legend indicates, an IKONOS pan-sharpened image, and what that means is that the original image was in, the panchromatic image was in one meter pixels and the spectral image was in 4 and by taking advantage of the procedure that we just discussed we have this nice pan-sharpened image of Cincinnati. For those of you who are not really familiar with Cincinnati, this is the old Riverfront stadium here that’s now been replaced by the Great American Ball Park, this is Paul Brown Stadium where the Bengals are going to try to secure some play off.
This is their practice field, which I wish they would use more frequently. At any rate, if we go to the next image, and this is northern Kentucky here, these are all the various bridges across the Ohio River. As you see here, this is the Licking River which you can see here bringing in lots of sediment and agricultural runoff from our neighbors in Kentucky. Nonetheless, to go to the next image, this is a false color infrared image of the one we just looked at. Now the Bengals’ practice field is red, if I got that right, I hope so. My students know I'm red-green color blind. It’s red, right? All right, thanks. You can see that there are other features of this landscape which pop out more, given the nature of that transformation.
Now, one of the ways that archeologists in particular can take advantage of much of the processing software, is to use various kinds of what we call vegetation indices, where you can manipulate the actual images from the satellite digitally, with software to produce different effects. One of the most common one is this one here called NDVI, right here. It’s simply calculated by taking the nearest red band, subtracting the red band, dividing it by the nearest red band. The idea here is that really healthy vegetation, the values of NDVI is going to be near one. On the left here you can see a Landsat true color image of Cincinnati, again this is the downtown area here. We are actually coming to you from this building right here, you can see my little dot, I'm speaking to you right from there, this is the university.
The idea here is that, given the nature of your problems you may want to focus on certain areas of vegetation, or vegetation health, and these vegetation indices help you do that. Here, for instance, you can very clearly see the Ohio River, the downtown area, kind of the forested area up here which is a residential neighborhood. With an NDVI transformation, both the river and the downtown area which is basically a concrete jungle, they all look the same to the human eye, and you can pick out, as well areas that are still relatively heavily vegetated in the Cincinnati area. We’ll come back to this particular issue in a moment. Here we go, okay. This image here is simply a sampling of the different kinds of manipulations that can be conducted with satellite images.
We just talked about the NDVI vegetational transformation, SRI is another one that’s very simple in calculation compared to NDVI, and all the rest of them here that you can see involve various kinds of transformations of the bands on satellite images. As you can see over here, under Description, they allow you to do different things again depending on your problem, whether you want to accentuate edges or you want to accentuate the interiors or some other aspect of vegetational communities. Just to give you a sense of the extent to which many of the transformations that can be conducted on digital images by software, how those transformations can affect our basic understanding of what we’re looking at,
here on the left in this panel you can see a rock hard image, and this is a small panel in our project area just out of the Grand Canyon, and with the naked eye you can make out just a couple of little legs or something. When you take this image and manipulate it with software as you can see on the right here, you get a completely different sense of what humans had done to that particular rock art panel. This little figure here is known by my students as Mr. Tomato Head among other names that he’s been given, but without the processing capabilities, this image, among others, you never would have been able to see clearly with the naked human eye.
The idea here and the reason I show you this is not only is it interesting in, in of itself, but it’s an exemplification of the processing capabilities that archeologists can use when they want to try to find information about archeological phenomena at the regional level. My principle focus today, as Karen mentioned earlier, is I want to illustrate both some possibilities and problems of using satellite remote sensing, satellite archeology at the regional level when we’re dealing with fairly large bodies of land. Right here what we’re looking at, we’re looking out from Grand Canyon National Park. This is the south rim of the Grand Canyon here. This is the famous desert view area over here. You have this enormous territory that is managed today by both the National Park Service Grand Canyon National Park, and Kaibab National Forest.
In this several hundred square kilometer area, we have indications of fairly extensive human occupation. What we don’t have a really good sense of is the density and the distribution of various kinds of archeological phenomena across these huge areas, which, of course, has implications for heritage management, as well as lots of interesting questions that archeologists are inclined to ask. We were just looking at an area, I’ll run my little pointer, we were just kind of looking in this direction from the park out this way.
What you’re looking at here in all these areas here, and up here - and this is down in the forest, up here in the park - the number of survey areas that we’ve been inspecting over the past couple of years. In this whole area are several thousand different kinds of archeological sites as you can see here by the legend on the left. What we would like to know is, do these patterns continue? Are they uniform, are they clustered? Are they evenly distribute across the area? It’s taken us about 2 decades to get to the point where we know that where these sites are, these are all differentially corrected GPS points on here these different phenomena,
we’d like to know whether there are different areas in this huge territory that were reserved for different kinds of activities, or whether the archeological phenomena co-exists over large areas, a perfect opportunity to employ satellite remote sensing. What I’m going to do now is take you through the process that we’ve been using over the years to try to determine the extent to which satellite remote sensing can be used to inform us about the regional archeological record and its properties. This is just a true color Landsat image of northern Arizona for those of you who are familiar with this part of Arizona.
Flagstaff is way down here, Wupatki National Monument over in this area here. This is very clearly the south Rim of the Grand Canyon here. I think Grand Canyon Village is in this area here, and this is the area where we’ve been focusing on, it’s called the Upper Basin, it’s a downfaulted block of the Coconino Plateau. It’s administered, as I mentioned, this area in here, this green ecosystem, is administered by both the park and the forest. This, as the legend indicates, is a false color infrared image, similar to the one that we looked at when we visited Cincinnati not too long ago.
It shows some very distinctive features, vegetation features of this area. This is the very famous Tusayan Ruin here, located right here, one of the earliest excavated sites in the entire park service system. As you can see this is one ecosystem, the upper basin this whole area here, and this is the Coconino rim territory here. Over here this is a wilderness area in the forest system. This little area right here is what a forest fire looks like or other byproducts, really, of a forest fire this was a controlled burn actually, in a Landsat false color infrared image.
This, as you can see, is an IKONOS image of a portion of the upper basin and this is basically what you would get to begin your remote sensing project, something like this that you clip and shape to your project area. I need to make a couple of caveats here, when you begin to select your images … Here, just let me situate you. Again this is the South Rim of the Grand Canyon; this is our upper basin area here. When you select your images, it’s really important to preview them, this is one image that we looked at when we were beginning our studies, and it has blemishes, which are represented by these clouds here and here, as well as the shadows of the clouds here and here.
This image, because our project area is right underneath the clouds, was rejected because there’s no way that we can process away essentially the obscuring effects of clouds and their shadows. In this image we will not want to … You can see here we have an inversion of clouds over the Grand Canyon and eastern Grand Canyon and, actually, the Colorado River gorge kind of runs this way. You have this really beautiful image or situation in the Grand Canyon that, if you were using a satellite image, very clearly this would not be one that you would select - beautiful but useless for our purposes.
Here’s a image, an IKONOS image, a panchromatic IKONOS image that has been clipped and stretched to fit our project area, it is remarkably sharp in terms of the contrast between different features. There are millions of one meter by one meter pixels in this image that cover … You can see there’s a major highway here, there was a quarry, there’s a quarry here. This is the major entry way SR state route 64, the eastern part of the Grand Canyon. It’ a beautiful thing, it’s not a photograph it’s simply registering variation in 256 bands of grey, very exciting.
However, let’s look at the close up, so if we go back one second. We’re going to focus on this area here, so I can give you a sense of what you can actually see and determine is useful for your problem. That area is this area basically in here, and as the caption says on the bottom this is an IKONOS 1 meter pan-merged image. Which means that we’ve taken the high resolution of the panchromatic image and merged it, sharpened it to include some color. Let me point out a couple of features here. You can very clearly see now you can distinguish roads, black top roads from Forest Service roads and other roads. This is a little dam here, a cow tank, this is a car … It’s an interpretation that little white speck there, of course is an interpretation that we make of variation in the imagery.
This is our old friend, the Tusayan Ruin. This is a creative painting, beautiful painting of what life would have been like at the Tusayan Ruin at the end of the 13th century. This is one of the largest sites in the area. There’s not a lot of evidence for all the activities you see here, but that’s a matter for another discussion. We want to use the Tusayan Ruin to illustrate some aspects of image processing and the usefulness of satellite imagery and archeological research. That was a painting, a creative rendition of the Tusayan Ruin, here’s what it looks like on the ground. These ruins were excavated in 1930, this point here is the same point as this one here, and we’re going to be looking at these different features in a satellite image now of the Tusayan Ruin.
This is it. Again you can see the different effects of spatial resolution. The Tusayan Ruin is under here somewhere in these large 30 meter pixels of a Landsat image. On this IKONOS image here, you can actually see the individual ruins that we were just looking at. They go this way, they come back this way and this little feature here is an excavated kiva. I want to emphasize these are not photographs, these are process satellite images and 256 shades of grey. What we’ve got here is our basic strategy for trying to not only detect but predict small scale anthropogenic anomalies in this vast landscape. I’m focusing specifically on small scale phenomena because, in my experience and the experience of many of my colleagues, they are abundant but not well understood.
Our strategy here is to derive what we call spectral signatures, from known archeological sites like the Tusayan Ruin as well as hundreds of others. Then to develop a predictive model, we project those spectral signatures about which I’ll say more in a second, to unknown areas, thereby creating maps that are basically predictions of the locations of archeological sites. The first step, as Sarah Parcak reminds us,is that we have to geo-reference our satellite image over known ground control points, this is an absolute essential, particularly in areas of the west and the southwest that have a lot of topographic relief. You need a dozen or more - minimally - ground control points, so that you can warp and drape your satellite image to ensure that the pixel that you’re analyzing with your software, is the same pixel, or refers to the same pixel on the ground surface.
I want to show you a couple of different ways that we’ve done this. Here is where the Tusayan Ruin is actually located on the ground surface right here. Then what we do is we find a very distinctive feature on the ground surface like right here. In this IKONOS image again we can see the Tusayan Ruin itself running here, here and here’s that 1 Excavated Kiva right here. Our ground control point is actually down here which is a drain in the parking lot of the Tusayan Ruin which is of course administered by Grand Canyon National Park. When you establish a system of ground control points you have to ensure that they’re spatially distributed across your landscape. If we go over here to the image on the right, we are going to establish another ground control point miles away from the Tusayan Ruin.
We are going to establish a point here, this is what it looks like closer up, and here’s what it looks like when we expand the satellite image. Let me draw your attention to the fact that these images are so highly resolved that you can actually see the fence line between Kaibab National Forest here and Grand Canyon National Park here, this is a physical fence that separates those 2 entities of different agencies. What you’re looking at here, is actually a post for a cattle guard in the fence that separates the park from the forest. Here are some examples of more ways that we found useful to establish ground control points.
This area here, up here is the new parking lot at the visitor center of Desert View and they have these little tongue creek islands with little posts right in the middle of them, the perfect place to establish a ground control point. Here’s another one that illustrates as well how sensitive satellite images are. This is at a pullout near Lipan Point right here. This is another drain or another sewer in the parking lot. Right here you can see a concrete berm that is highly differentiated from the natural terrain here and, of course, this is the actual Grand Canyon itself. The image was clearly taken somewhat later in the day given the attitude of the shadows and the fact that the Canyon is already dark even though it’s good visibility outside.
With our ground control points then we can take our satellite image, warp it over those ground control points, trying to ensure that the pixels that we’re going to analyze are ones that are actually referred to on the ground. One of the most common ways that software does those kinds of transformations, is what we call an MNF transfer, and that is where you try to remove noise from the satellite image that’s created by electromagnetic energy passing through the atmosphere, energy that’s added to the image of the consequence of it being both stored and translated from the satellite to receivers on the ground.
This little figure here is just intended to show how MNF noise reduction works. Over here, on the left, is your basic satellite image that’s been warped over the ground control points. You can see roads and other things. The image on the right is actually just pure noise, just random chaotic pixel noise that has been added to the image. With this MNF transfer, where you conduct various statistical procedures that remove the noise here from the image here, yielding what is a very useful image for various types of analyses, including the derivation of spectral signatures of different kinds of archeological phenomena that may be lurking here in the image.
This is where satellite remote sensing gets a bit technical, so I’ve chosen some examples to illustrate how our knowledge of bands and of processing of different kinds of archeological phenomena can actually be taken advantage of to create different kinds of predictive models. This is the second step, step 2 in developing direct predictive models of small scale phenomena from satellite imagery. Right here we have your basic unmodified Quickbird image for instance, one of our satellites. You can see that there are 4 different bands here red, green, blue, near-infrared and by virtue of software processing we can develop a signature for one type of archeological phenomena called a masonry structure, just using 4 bands. That signature looks like this.
There’s not a lot of information there, not very useful because this particular signature can refer as well to other kinds of archeological phenomena, kinds other than masonry structures. What you want to do then, is augment that Quickbird image by adding or incorporating different transformations, like the MNF transformation that we just talked about, to remove chaotic or random noise that’s confusing the development of the signature. We can also use our old friend NDVI to create yet even more distinctive aspects that will yield a very classic signature. This transformation here is a tasseled cap transformation that’s supposed to respond to differences in wetness and dryness.
Then with software you can also create ratios between the different bands blue, green, blue, red and so on and so forth. Here you see a much refined and highly specific masonry structure spectral signature that actually includes the original one here, plus all of these different aspects of how the bands have reacted to the transformation to yield a very distinctive masonry structure spectral signature. The next part of this process is that once we have what we call a desired or a spectral signature that refers to a specific type of archeological phenomenon, we’re interested in taking that spectral signature and projecting it to unknown areas, knowing full well that the nature of the pixels in these unknown areas might vary, might not match up point by point with our spectral signature.
In fact, we - when we’ve done some experiments - we found that there could be a great deal of variation between the spectral signatures of phenomenon that look like this, or look like this. There’s another transformation that you can run called a match filter transformation which you see here, again it’s a statistical process involving software where the user, in this case the archeologist, says, “Well, ok, I know I’m not going to get spectral signatures that look exactly like this, but if I can get some that are within 90% of or very close to the desired spectral signature, I’ll consider them essentially the same, or having a high probability that they’re going to match the reference or previously determined spectral signature. This is unacceptable, so I am not going to consider this, in view of my thresholding procedure, a match. In fact, it’s a low probability of being a match.”
This is a very useful way, this match filter operation. It’s a very useful way to take data that may not absolutely correspond to your desired spectral signature and establish probabilities that you’re comfortable with. In many cases, your comfort level for establishing these probability levels is based on your experience with the range of known variation of the archeological phenomena you’re trying to model. The end product of these procedures is to create kind of a library, if you will, of spectral signatures of different kinds of archeological phenomena.
In this heuristic slide you can see that we have about 5 different kinds of common archeological phenomena in the Grand Canyon area. You could see if we focus over here, around band 3 and these are reconfigured bands by the way, but you can see that masonry structures have a very different reading and signature by the time the signature approaches band 3 that helps to differentiate masonry structures from sherd and lithic scatters here, and brush structures here. In other words these archeological phenomena, once their spectral signatures are extracted from and processed with software have optical properties that clearly allow the software to distinguish them, creating thereby finger prints of different kinds of archeological phenomena that we anticipate can be then projected to unknown areas to create predictive models.
Here’s what some of these things look like, you can get appreciation for how the actual phenomena themselves do have properties that register differently on the satellites. This is a pile of fire cracked rock which you can see here, and it has a lot of variation in its reflectance from some very light colored rocks here that have been heavily fired to some darker material. This pile of rock also varies texturally, there’ some big rocks here and over here and some fine, very fine grain rocks here. Notice as well that this feature, this big old feature here contrast dramatically with a surrounding ground surface here in terms of its color.
All these differences are going to be registered on the satellites in different shades of gray and then processed with software. In contrast, what you see here is a brush structure, that is composed just of dead wood that has been concentrated and built up by humans, maybe about 100 years ago. You can clearly see that this type of archeological phenomenon, a brush structure, contrast in every way with the fire crack rock pile we just examined. Both of those types of archeological phenomena contrast dramatically with this mound here, which is the remains of a fairly large masonry structure that itself is composed of rocks of relatively similar sizes and shapes. Not a lot of contrast among the rocks, they’re all relatively same size, shape and color.
Step 3 - we take all of our spectral signatures from these different archeological phenomena. Those signatures have been developed from what we call training data and the training data are here on the left, these are known sites locations of which are known through GPS. In this case to simplify my presentation, I focus just on masonry structures. There’s approximately 350 dots in these 3 different areas where we have been working. These two here in a park, this big area here is in the forest - Kaibab National Forest. Nonetheless when you take all of these structures, extract their spectral signatures, you can then project them as we’ve done here on the right portion to unknown areas. Survey areas that haven’t … Or areas that have not been surveyed, and the idea is that if you navigate to these different spots, you will find the remains of a masonry structure. Let’s see how good we do.
Step 4, this is essential. You will find when you go to the literature, and there are just dozens and dozens of remote sensing studies that have been published in archeological prospection. The International Journal of Remote Sensing, the journal called Remote Sensing, the journal of Archaeological Science, where there are a lot of feasibility studies that have been done and a few predictive modelling studies that have been done, but not too many are accompanied by this very important Step 4 where you actually go out and ground-truth your prediction, and what we found is that our pixel-based predictions of archeological phenomena, in this case masonry structures, were not very good.
We found these two true positives in these areas here, but we were a little bit disappointed because many of our predictions were off, they were too few true positives. The number of false positives was way too high, false positives of course the way you make a prediction that something’s going to be there and there’s not, and then there were false negatives, where in fact we have predicted something was not going to be there and it was. We had a problem, then, that we thought we could rectify as follows. As a consequence of ground-truthing your predictive model, one of the things we discovered and this is one of our true positives here, this is a substantial mound of masonry,
you’ll notice that you could see the entire extent of the room block here, so there’s no trees growing over this structure canopy. There’s no vegetational canopy obscuring this structure. In contrast, this one down here which can be another masonry structure not quite as large, right in through here is actually obscured by trees. We were beginning to think, “Wait a minute! There is some chaos being introduced into our predictive model because in one case masonry structures can be detected by the software under thick canopy as in this case and another case it can’t be.”
We thought, “There ought to be one way to resolve this.“ We also were suspicious to that one major source of the false positives we were getting were vegetational processes, which are very common in areas of the American west where you have large trees growing up through bedrock creating concentrations of material such as this that can fool the software. Doesn’t fool the sensors on the satellite because they’re just registering variation and light, but these kinds of differences can be misinterpreted by the software as a spectral signature of human activity when it’s not.
Back to the drawing board, we decided to undertake two additional steps as you can very clearly see here. One what’s the source of the problem? We predicted that it’s a combination of the effects of canopy cover and other non-anthropogenic processes like the one I just illustrated. We decided to take a different approach, so this approach is called the segmented object-oriented approach, where the unit of observation is not the individual pixel, but segments or clusters of pixels that the remote sensing people call objects. This approach takes advantage of some of the … some experience as well as some of the verification results that we did, basically our knowledge of the archeological record of this area which is ... The archeological record of this area we’re working is fairly common throughout the American Southest and the West.
Where you have, in this case, a masonry structure as you can see, I’m highlighting with my pointer and it’s obscure. It’s got a lot of vegetation growing on it. In contrast, the area out in front of the structure is full of artefacts, it’s an artefact scatter that’s associated with the structure, a lot of people call it midden out here. You can see there’s hundreds of artefacts out here as indicated by these red flags. We thought, “Let’s take advantage of the optical characteristics of this mound that has vegetation growing on it, the mound itself as well as all of the pixels that are out here in the sun obscured area.” You see here in the bottom figure that yeah, this is an IKONOS image of this site and you can see that here is the obscured masonry, looks a lot like the surrounding vegetation, but this area out here is different. What we do is, in the software, we create what we call objects which are areas of varying size that are unobscured by vegetation that we are going to use to predict the locations of the obscured masonry structures.
Back to our processing software, again. All of this should now be familiar to everybody. Now we’ve got an unmodified QuickBird image and an augmented QuickBird image. We know exactly which bands are used here in the unmodified one. Here is the spectral signature of the ground surface of one of those objects that refers only to the artefact scatter. You’ll notice that it looks very different than the 4-band spectral signature we saw from masonry structures. To make the argument and the spectral signature as accurate as possible, we did, over here, a MNF transfer, and then we use different combinations of other transformation like an NDVI and a Tasseled Cap to create a really dramatically different type of signature for the artifact scatters that are associated with the masonry structures.
What we did was then say, “Okay, lets adopt this approach and see what the variation is in these areas now that have been defined in terms of their spectral properties of artefacts on the ground surface. You can still see here that the areas of interest, here is one here, 225 square meters, these are still relatively small scale phenomena, because the size of the square really that would contain an object about 225 square meters and area is about 15 meters on the side, or roughly the size of the pixel of a Landsat panchromatic image pixel.
These are still small scale phenomena we just picked, because that’s what we’re trying to model. We also then went back and resampled our training data to show how successful this segmented object- oriented analysis might be and you can see here, here is Kaibab National Forest and an obscured canopy where you had basically chaotic results in comparison to the nearby Grand Canyon National Park, same type of vegetational situation, unobscured canopy, you’re having just unsystematic results which I think is indicative of the expanse at which we had poor predictive success earlier on.
Now here, look at how consistent and accurate our predictions are of known areas, objects based on the training data. We think now, as the consequence of the second method that we’re getting close to actually refining a predictive method, if you will, for small scale phenomena. Just, let me briefly go over the slide that shows our results so far of the application of this new method and I know you can read it for yourself, but I just want to go over some high points here that we were looking for to take advantage of the spectral characteristics of the satellite images, some method that it would focus on take advantage of the actual differences between the archeological phenomena and the environmental phenomena and use distinctive combinations of those special differences to create these objects.
I think that this last bulleted item here is the one I would like to draw your attention to and that is that archeological sites themselves have environmental effects and, in some cases, are independent from the natural environment that doesn’t have archeological phenomena. All of those aspects of the regional record then can be taken into account with this particular style of analysis where you’re actually looking at how vegetation that grows in archeological sites can be used in combination with the archeological materials to create very distinctive types of archeological signatures. Our next step, of course, is to project these new types of predictive units to the regional record and ground truth them and that’s what we’re doing right now. Our next step is to actually do the projections and the ground-truthing.
That said, I would like to also introduce you to another way that archeologists have been using variation in vegetation to predict archeological sites and this is from work that’s been done in the Mayan area, it’s in jungle, so very heavily vegetated and archeologists working there for decades like Tom Sever and Tom Garrison and others, have been focusing on the vegetational signatures of archeological phenomena. This approach is a little bit different than ours because they’re trying to identify actual signatures of vegetation that would register underlying archeological phenomena and to give you an extent to how our methods differ, this is a famous site of San Bartolo in Guatemala.
This is the work of William Saturno and many of his colleagues and what they’ve done is that they’ve taken the spectral signatures of vegetation with these very distinctive yellowish areas and overlaid them on known sites like this giant complex here in the Mayan area to predict the locations of smaller outlying settlements and they’ve been very successful in that regard. I would like to point out that these are large scale phenomena and some of our largest sites might sit within this little one here that I’m circling with my pointer. That’s another approach that they found very successful. Back to Landsat now. Landsat, of course, has in combination to IKONOS and QuickBirds special resolution is fairly large. Landsat, however, because it goes back so far in time, that is, the record of Landsat images goes back several decades to earlier satellite like Landsat 5, Landsat 4.
You actually have a record of vegetational change that you can employ to determine how environments and ecosystems have changed. In the application I’m going to share with you now, how heritage resources might be threatened by changes to the environment. Here we have a Landsat image, taken in 2007. As you can see, it’s a 2-color image and I will point out some features here. This is a quarry that we’ve used for road building activities. I also want to draw your attention to this little feature here. These are archeological sites, every one in this little red circle and they’re located in an area of really healthy vegetation. This is pinyon- juniper woodland, that whole area here and these are other archeological site.
I want to focus on this area right inside this little symbol here. There we have a healthy forest in 2007, same area in 2010. I want to draw your attention to this phenomenon here. You can see that a significant portion of the forest is gone, it’s unhealthy. All you’re looking at here is bare ground, as I move my little pointer back and forth. This is an area, this area in here, is an official designated woodcutting area that was established by the Forest Service and we have archeological sites in and around it and those sites, of course, were not only threatened, but in this case destroyed. We can go to … this is where not only Landsat, older Landsat images. are useful, but some of the transformations that we’ve talked about before, here is an NDVI image of the same area that we just looked at, we’ve reversed what constitutes the vegetated areas when we examined NDVI in Cincinnati and you can see the extent to which vegetation has been removed from the forest and here is the rest of the forest in good shape.
Go down here, notice in 2012, the extent to which the wood cutting area has extended to include double the number of archeological sites. We think that this is one way that we, as archeologists, can help the Forest Service plan further what they call forest management activities to avoid damaging or threatening different kinds of archeological sites. This is the aftermath of … This is what some of what these activities look like on the ground, here is one earlier woodcutting area. These are actually on-the-ground images of, or an on-the-ground image of what we just saw. Here is the newer one, this is the large lithic scattering here, this is a sign that was intruded right through the site, you can see different kinds of would be aftermath of woodcutting here.
We’re trying to encourage the Forest Service, in this case, with our satellite imagery that we can assist in identifying areas that are less likely to have archeological sites that might be affected by official woodcutting area. In contrast, we can use satellites to actually predict or estimate the effects of woodcutting on ecosystems and all of the heritage resources that might be in them. In this image here on the left, this is an image of pixels that had standing trees on them in 1990 and then through the years up until 1997, all of these pixels, then all the black you see here were pixels on the ground that had trees removed from them by illegal woodcutting activity.
This is another exemplification of the extent to which Landsat 7 imageries looked at year by year can create and absolutely accurate view of forest health and the extent to which forests are being degraded by illegal woodcutting activity, each one of these pixels here has had trees removed from it illegally. You can then determine rates at which pixels are being changed from healthy to unhealthy, measure that, those changes by NDVI and create predictive models as we have here on the right, that show how many resources, known resources as you can see here under what I call mapping unit types over here are going to be affected by the expansion of illegal woodcutting activity. You can see that basically, this part of our national forest is being deforested by illegal woodcutting activity.
I can tell you that since this model was produced in 2007, the situation here was actually under-predicted and that the situation up here in the park was over predicted, which is okay, but it does illustrate the extent to which we can use libraries of older satellite images to help inform practices about how to manage both the heritage resources and the natural resources on public land. Back to the electromagnetic spectrum - beautiful rainbow over the South rIM of Grand Canyon, and with that, Karen, I’m more than happy to take questions.
Karen: Thanks so much Alan, what a great lecture, do people have questions? Well, I hate to expose my ignorance yet again, but I’ll jump in with a couple of questions of my own. First of all, I have to say how ironic it is that you are a blue-green color blind as you’ve undertaken this research.
Alan: Well, I could comment on that! I only employ graduate students who were certifiably not color challenged as the same as I am and I’m forbidden from ever using a Munsell soil color book. I always get that yanked out of my hands, but go ahead yeah, it is kind of ironic, I agree.
Karen: Can you talk a little bit about the commercial availability of the satellite images? Do you buy them? Where do they come from?
Alan: That’s a really good question and we, the University of Cincinnati, because we’re a public institution, some of the images we can acquire just because we have licenses, some images we have to purchase specifically for our analysis. I think in the Park Science article that you referred to earlier, Karen, we had a little table of the prices of some of these images and some can get very pricey depending on whether the satellite has to be targeted, there’s a private corporation satellite, in some cases they have to be targeted specifically to acquire the scene that you’re interested in and that really jacks the price up thousands of dollars.
We’ve been trying to employ legacy images, when we’ve had to acquire an image for our own purposes, we’ve had to spend several thousand dollars. My assumption is that those in the federal system should be able to take advantage of satellite images that have been acquired by your neighbors perhaps and other agencies have way reduced costs than what we have to pay.
Karen: Interesting. Is there anybody out in our audience who can comment on that? One of the reasons why I was interested in having you speak today Alan was because I don’t think we have very many people in the Park Service who are working in this sort of research and it’s great to have somebody who can demonstrate the applicability of this methodology.
Alan: If I can elaborate on your observation Karen, I think there’re great opportunities here, I could talk a little bit about the spool-up cost and perhaps some of the hidden costs. We at the university have a licensing agreement with the company that developed and markets some of the software that I was mentioning earlier, it’s called Envi, and that license is a little bit pricey, but we get it because we are a research one university at some discount. I think that the federal government would be in a position to negotiate the software cost down. Second thing is that you do have to have some instruction, some training into how to migrate your GPS data and your GIS data and use them in combination with the satellite processing software, like Envi.
My experience and that of my students is that you can get this training in about a semester or so, and so I would think that given the expensive use of GIS by the park service, that that should not be a major problem, the that I’m referring to is to have people migrate their GIS skills to the manipulation of satellite images with Envi software, for instance. Beyond that, there’s not a lot of … I’m not trying to market this method or anything obviously, but I think that in the private sector, some of the initial startup costs could be substantial if you were just trying to start up a company that analyze satellite software images, but I think those costs have been driven down dramatically because of the availability of the images and of the software. The available training itself, I think, positions a lot of people who have GIS training to take advantage of all of these opportunities.
I think the park service is perfectly positioned to take the next step from GIS and I know there’s a lot of activity in that area to experimenting with and designing predictive models because, given the nature of planning in the park service, at least the nature of planning that I’m familiar with, if you had some idea of the likelihood of the presence of archeological sites in different area, you could really plan effectively as to whether or not you want to put a camp ground or a trail or create yet another visitor experience.
Karen: I think you’re right. Do you do the work with the Grand Canyon through a CESU or are you an outside entity that’s taking … that’s just doing your research on inside the park boundaries?
Alan: Well, right now it’s the latter. There is an entity called the Colorado Plateau CESU, it’s out of Flagstaff, and we’re in the process of preparing our application for that, so I’m optimistic that we’ll be able to look even closer with a number of different agencies particularly in northern part of Arizona.
Karen: Good, I see that you’ve got money from NCPTT.
Alan: Yeah, that was wonderful, they actually supported some of our earliest efforts in this regard about 7 or 8 years ago and we were very grateful for that funding because it allowed us to do some pilot studies and to experiment with many of the different method that I’ve discussed today. I like to tell people that NCPTT is way more than learning about how to preserve grave stones in old buildings. They were visionary in the extent to which at least they investigated in us early on.
Karen: Do people out in our audience have any question?
Anne: Hey, Karen, this is Ann Vawser. You had asked earlier whether people had access to satellite data. , I thought I would share after all about captures access to some of the satellite data and Park Service does have some free and some reviewed top to access to satellites, to various satellites through a variety of treatments and your best bet there is to work with the inventory and monitoring folks, because they are the ones that that are primarily accessing that satellite information for vegetation monitoring, but certainly, if they are getting the data then it’s there and it would be available, too. They may already be doing some analysis or have access to the software that was mentioned and if parks were interested in looking at archeological applications they might be able to work with them on that.
Karen: Thanks for that, Anne, that’s good information. Do people have any other comments or questions they wanted to ask?
Michael: I have a question, Karen.
Karen: Yes sir.
Michael: This is Michael Peterson, archeologist from Redwood National Park.
Karen: Hi, Michael!
Michael: I just had a question for Dr. Sullivan. We have a lot of vegetation out here and so that’s one of my first step is to use satellite imagery and various other forms of imagery with GIS, before I even do the survey. My question is I don’t know if you have any thoughts on using these little drones like these helicopters with different cameras like a stabilized go pro or even I was thinking about the UV camera for really a high resolution flights to investigate areas?
Alan: I do have some thoughts in that regard. We have thought about using drones and balloons and other things in the Grand Canyon area, and of course when you do that, when you consider that there are all these noise, abatement rules and regulations that were brought to our attention, so we didn’t really pursue that. As you may know, in the Grand Canyon area, there are these no fly zones essentially to enhance the visitor experience. My thinking is that what you’re talking about there, I’m wondering what kind of image we would actually get from drones or other low flying aircraft that we could … If you want to look at a large area, I personally would be a little bit concerned about stitching together the images into a coherent scene, because I’m just not sure how regular and systematic the pixel sizes would be.
Here, I’m expressing my ignorance completely, but knowing I’m very comfortable with the standardized image attributes that we get from either archived Landsat images or the commercially available ones. They’re highly standardized and systematic and they come with all kinds of locational information that essentially makes our job easier than having to figure out exactly where each image is. That would be my only reservation right now is actually geo-referencing the images you might acquire from the various kinds of airborne vehicles you mentioned into the software.
Michael: Yes, thank you, excellent points.
Karen: Does anyone else have questions? I wanted to ask you about your masonry signature that you had interpreted, developed or recognized in your research. I wanted to ask you, does a rock have to be all the same kind of rock for the signature to be the same? I can see for certain areas, that won’t be a problem. If they are all sand stone or they are all the same type of rock derived from the underlying geology, but if you’re working in an area where you might have a lot of glacially deposited rock, would you expect to see different signatures based on the mineral composition?
Alan: Absolutely, and the reason I say that is our friends in geology have been using spectrometers and other objects, or rather instruments, like that to build up what we call spectral libraries of the signatures of different kinds of rocks and I think that … Then you can access the spectral libraries online and try to match up your spectral signature of saying a known rock with spectral library that’s already online. I think that that’s a good model for archeologist to try to pursue, in other words and getting back to your question, Karen, in the area like we have in the Grand Canyon where everything is based on limestone, if we encounter some structures that are made of compositions that are of rocks that have different compositions than limestone or some other object, yeah, it would be great to be able to refer that association to a spectral library.
For instance down in Wupatki, which is not far away, only about 60 miles away, they do have structures that are made out of different kinds of rocks, so I would expect even though you might have a 2-room masonry structure, if it should work differently, it should have a different spectra’ signature if they were made out of basalt and limestone than if the structures were made out of limestone exclusively or basalt exclusively.
Karen: Okay, that makes sense. Does anyone have any more questions? Alan, thank you very much for a very interesting talk.
Alan: You very welcomed and if anyone has any questions of a technical nature or of a substandard nature, my email is just my name Alan.sullivan@uc.edu, send me your questions and I’ll get right on them.
Karen: You can also get in touch with Alan through me.
Alan: Terrific.
Karen: Thanks, again.
Alan: Thank you for the opportunity, Karen.
Karen: You’re welcome!
Alan: Okay, bye.
Description
Alan P. Sullivan, 12/5/2013, ArcheoThursday
Duration
1 hour, 24 minutes, 13 seconds
Date Created
12/05/2013
Copyright and Usage Info