Letters from the SAL

Getting started with eCogniton webinar recording

2011-12-01T17:24:00.002-05:00

A recording of the AmericaView sponsored webinar, Getting Started with eCognition, that I gave today is now available for viewing via this link. You may also be interested in another video, Segmentation Algorithms in eCognition. The segmentation video uses the same data and contains a link to download the data from the eCognition Community site.

Leveraging Geospatial Knowledge to solve Geographic Secrets

2011-11-05T19:49:00.000-04:00

This post on leveraging geospatial knowledge to solve geographic secrets comes from one of our interped GIS techs, Dan Koopman.

--

Urban Exploration is a hobby that has skyrocketed in popularity over the past 20 years in the United States. It is loosely defined as the discovery, infiltration, and exploration of areas traditionally deemed "off limits."

It is probably no mistake that the increase in popularity of Urban Exploration parallels the public's access to new and ever-improving geospatial data in the form of free services such as Google Maps. What started as a technical tool aimed at getting people from Point A to Point B, has raised awareness within the public of what comes between - in other words, the value of public geospatial data beyond driving directions. The discovery and exploration of abandoned places is, at its heart, a geospatial question that begs to be answered and addressed by datasets ranging from high-resolution imagery to census information.

These leaps in awareness and usership, however, have come at a price to Urban Exploration. Increased access to information about delicate abandoned buildings and the resulting traffic in and out has led to many negative consequences in the hobby. These consequences range from increases in security aimed at keeping explorers out of sites, to outright abuse and degradation of sacred historic buildings by vandals. To counteract this abuse, two unspoken rules obeyed by the community for sites not yet heavily trafficked have been established:
a) The exact location of the site must never be revealed
b) However, Clues leading serious and respectful explorers to the site can be given upon request, so long as some amount of work must be undertaken in order to find the site.

This being said, I hope that whoever chooses to read this article takes from it not a free shot at a really cool place, but an example of how using geographic knowledge in conjunction with free geospatial technology can help you find a virtual needle in a haystack. I also feel comfortable in sharing this since the location discussed is so remote that the effort it would take to reach would not be worthwhile for your average vandal.

Jake Reinig is a talented young photographer and explorer from San Diego, California. I was visiting his website when I came across some beautiful pictures of some double-decker passenger train cars left abandoned in a desert. Reinig, being a responsible explorer, spared only two descriptions in his explanation of the site: in the "Desert east of San Diego." I became so interested in the idea of several large passenger train cars being stranded in the middle of the desert that I decided to try and find them using Google Earth.

I encourage the reader to check out Jake's post that contains the images used to find the trains:
http://www.jakereinig.com/blog/?p=1901

In an initially naïve move, I figured that I could just "fly around" in Google Earth in the desert east of San Diego, and find the cars in the imagery. I soon realized however that this was a ridiculous and ineffective method.

In the last of the pictures on Reinig's post, there is a very distinct conical mountain in the background behind the railroad tracks. I opened up the Google Earth preferences and set the vertical exaggeration to 2/3 - this allowed me to try and match the feature that I was seeing with the digital elevation model that is loaded in Google Earth. Unfortunately, as I found out, there are many conical mountains in the desert east of San Diego. It only took about an hour of fruitless navigation to figure that out...

After the realization that trying to find a train car in 2500 mi2 of desert was literally like trying to find a needle in a haystack, I started to assemble more contextual clues and leverage geospatial information.

I started working under these assumptions:

Specified in Description
* Desert
* East of San Diego
Implied
* It's a railroad
Visually Interpreted Clues
* The trains are on a spur
* The spur likely ends after the train cars
* There is a smallish trestle over a river

It occurred to me that there may not be too many railroad lines running through the desert in that part of the world, so I googled "California Railroads." I cherry picked a very simple map of the all historical railroads in California courtesy of www.djcooley.com and zoomed in on Southern California:

The map reveals that there is only one railroad line that connects San Diego to the desert, and it cuts across the mountains east of San Diego and across the southern part of the Anza-Borrego Desert eastward to El Centro. I went on the assumption that the trains had to be located on a spur off of this railroad line, narrowing down the potential targets.

At that point, I began to interpret the pictures on Reinig's site a little more thoroughly.
Perhaps the single most important photograph to solving the secret is the second to last one, for two reasons:
a) It shows the fact that the trains are on a spur
b) More importantly, it reveals that the "main line" of tracks runs in a SE to NW fashion. I arrived at this crucial relationship when I noticed that the sun in that photograph is setting. Because the sun always sets in the West, and the main line is closer in the left of the frame than the right, it must have been constructed on a SE to NW trajectory.

Drawing this relationship out on paper gave me a map to compare with the imagery for the area, greatly improving the likelihood of finding the trains. But that's not all - the first picture clearly shows the spur ending after the trains. It also shows a bit of topography to the East of the trains as well as to the North.

Another crucial piece of evidence is revealed in the final picture, which shows the conical mountain. Because of the newfound contextual knowledge that the railroad tracks run SE to NW, it can be assumed that there is a considerably unique conical mountain to the South and East of the trains.

I added these new geographic clues to my list:

Geographic Clues
* Because the sun sets in the West, and this is clearly a sunset, the main rail line must run from the SE to the NW, with the spur roughly N/NE
* There is a very distinct conical mountain to the South of the Spur
* There is a small railroad trestle going over a stream
* One of the photos reveals that there is a hill to the SE of the trains as the photo is looking down on the trains
* There is a distinctive conical mountain to the SE of the site

I also began to look at Geologic clues. In the second-to-last picture, there is considerable vegetation - sagebrush and low-lying desert bushes. The density of cover is relatively high. The first picture shows there to be large sandstone outcroppings and boulders, uncharacteristic of a low desert. I also added these clues to my list:

Geologic Clues
* There are large smoothed sandstone outcrops in close proximity to the trains
* There is sand that has been blown over the tracks
* There is considerable vegetation and brush
* The end of the spur has a large rock outcrop

These new clues brought me back to El Centro. I suspect that it is called El Centro for a reason... To the east there is nothing but sandy flat areas without any distinct topography... This description definitely does not fit in with the land cover seen in the pictures, and is missing that crucial conical mountain feature to the southeast of the trains.

With these clues compiled, a significantly reduced search area began to take shape:

a) EAST of San Diego, in (or near) the desert
b) WEST of El Centro
c) NORTH of the Mexican Border (it's assumed to be in the United States)
d) SOUTH OF or DIRECTLY ON the only railroad that connects San Diego to El Centro.

This graphic shows the only possible places the site could be. The red line represents the railroad line drawn in from the California Railroads Map. The black box shows the West, South and East borders.

I also drew out the trajectory of the main line and spur that I figured out from the sunset picture:

Armed with the Hand Drawn Map showing the clues, I began following the railroad line west out of El Centro. The drawn map was invaluable as you could eliminate any section that didn't fit the 2 dimensional vectors of the SE-NW line. That is to say, there are an awful lot of sections that don't meet those criteria. Still, I was hung up on the location being on the Eastern edge of the mountain range...

I had isolated an area that met the description well - the tracks were SE-NW, there was a conical mountain in the SE, there was sand over the tracks, a small trestle over a stream and even an abandoned water tank - unfortunately there was no spur. I turned on the "Historical Imagery" toolbar and turned on previously acquired imagery but there was no evidence of a spur.

After that failure, I turned on the "Photos" layer in Google Earth and was surprised by how many images came up. Basically, users upload and "tag" the locations of their pictures to Google Earth's servers. These images are then added to a central database and made available through the "Photos" layer. In the area that I was focusing on, I found one tagged as "Carrizo Gorge Railway."

Without divulging any more information, within several minutes I was tracking the railroad line further west, hand drawn map in hand. When I finally found the site, I must admit that it was not as far from civilization as I had expected. It is actually the quiet neighbor to an, er, interesting type of ranch...

Also unexpected was the degree to which my hand drawn map matched the geometry of the railroad spur when I held it up to my computer screen - it was uncanny how accurate it was.

It only took three hours of intensive research, but through some historical mapping, geologic identification, geographic knowledge, and contextual interpretation it was possible to gather enough clues to resolve a train in the middle of a large desert. Of course, none of the individual parts can lead you there alone; it takes a common platform to make sense of all the data. In this case, all that it took was leveraging a free geospatial platform to find five train cars in a roughly 2500 mi2 area - a pretty decent feat considering that GIS was just a twinkle in someone's eye when the cars were placed there.

Storm surge maps - is bad data worse than no data?

2011-08-28T12:49:00.000-04:00

Due to Hurricane Irene, my esteemed university cancelled the first day of classes on Monday, thus postponing the first session of the course I am teaching, the GIS Practicum, until Wednesday. I thought I would use the time to generate a few slides showing the power of GIS during a crisis situation. I have quite a few family and friends in NYC and thus I was interested to see if they would be affected by the storm surge. What I found, after browsing three sources of storm surge data, were some massive inconsistencies. In one case in particular I felt the presence of poor storm surge data on the map was worse than having no data at all.

Let's start with Esri's US Tropical Storm Map. I like it, because it includes precipitation data with the storm map and it worked fairly well on my Xoom. The storm surge data, however, just doesn't pass the snuff test. A single grid cell covers many square kilometers leading one to believe that their area, nearly 20km inland, could be subjected to a 9ft storm surge. From the Esri map you can basically tell that areas near the ocean are subject to storm surge. Not in the least bit helpful.

Compare this to the work Steven Romalewski and his OASIS team put up. Clearly Steve and his team had access to localized, more authoritative data through the NY State Emergency Management Office. Unlike the data Esri is displaying, the surge models actually makes sense when you compare them to topographic and hydrographic data sets. I could zoom into areas where my friends and family live and find out if they were likely to be affected. Would I trust the Esri map for that? No way!

Finally, the National Hurricane Center has their map. The Google Crisis Response Center map using the same underlying data (the graphic below is from Google). The data appear to be more detailed that what Esri is serving up, but the coarseness of the grid once gain results in some unrealistic patterns in which small section of the shoreline are excluded.

First, all of these organizations should be commended for taking the time to make this data available, but I do wonder, if in this case, Esri in particular is undermining their credibility by serving up data that is at its best, uninformative and confusing, and at its worst, wrong. I would be grateful to hear your thoughts on this matter, particularly if you happen to be a storm surge modeling guru.

Unmanned glider imaging - GoForce One

2011-07-24T16:12:00.001-04:00

Below is a write-up from one of our recently graduated students, Bobby Sudekum. Bobby has worked for us in the Spatial Analysis Lab for over a year, doing everything from heads-up digitizing to LiDAR processing, to project management. With GoForce One Bobby combined his geopspatial expertise with his passions for planes, skiing, and photography. His work on GoForce One ended up winning him this year's ASPRS Central New York region student award. You can find out more about Bobby on his web site. He will likely be moving on from the SAL in the fall so if you are looking to hire a highly talented and motivated individual for your organization, Bobby might be the perfect fit.

--

Go Force One
Bobby Sudekum

When I was younger, I was always interested in aviation. I loved flying the wind up rubber band-powered balsa wood planes. I dreamt about owning an r/c plane but for some reason, but Santa never came though. Fast-forward 15 years to my introductory Remote Sensing class. Little did I know that when I signed up for this class, I would be participating in a service learning project centered around flying r/c planes equipped with cameras to learn more about Remote Sensing. I was ecstatic.

Fast-forward again to the fall of 2011. I am sitting in Economics of Public Policy, day dreaming about a mountain bike film in which they use a camera mounted on a zip line to achieve some stunning angles of mountain bikers. I began to wonder “How can I achieve the same effect with a skier, but with out the zip line?” I’m thinking, scratching my head, tapping my pencil...and then it dawns on me. Airplanes. I remember back to my Remote Sensing class. The planes we used were by no means powerful nor big, and the cameras also fared on the heavier side of things. After class, I immediately called my roommate who was an engineer. “Mike, think about this. Glider plus a GoPro (a small hd video camera) plus deep powder plus a skier.” There was a long pause followed by, “Bob, it’s a great idea but I don’t think it will fly, its just too heavy.” I was not convinced.

I began to look around the Internet for the largest foam glider I could find. I found one with a 54” wingspan for $9.95. I jumped on it, and bought two. The planes arrived and both of my engineering roommates shook their heads. I didn’t care. I knew, from the experience I had in my Remote Sensing class that it was possible. I chopped the nose off one of the planes and put a piece of Velcro on the camera and the nose. On a cold January morning, I threw the glider-camera combo off the balcony of a friend’s house. Not only did it fly, it flew better than anyone thought it would.

My roommates and I thought I had come up with a million dollar idea. We had this idea that we could capitalize on this new technology. For the next couple of months, we tested Go Force One (GoPro + Air Force One) in secrecy. Some of the footage we recovered from the plane was unbelievable, some of it was less than optimal. Halfway through the ski season, I sought for a mechanism to share my work with others. So I made a video with some of the footage and a little information about how it works. I thought that if people could build off of my idea, my plane could end up soaring longer and higher. Open source if you will. I have since then received a lot of good feedback from the video, and my eyes have been opened to many new ideas. I am glad I shared my idea with people.

Through trial and error we found what best worked for our UAV. We learned that it is designed to break, so we bought a decent amount of epoxy to glue it back together. We learned that you must throw it a lot to get the results we were looking for, because it flies where it wants to fly. Many people have asked me why I have not tried putting a motor and some controls on it. My answer is simple. It is a simple machine. It is something I can bring to the top of the mountain, through behind a skier, and not worry about it crashing into a tree. For now it is a simple machine, but in the near future I plan to move to more technical devises more specifically r/c planes.

This past semester, I enrolled in a special topics class, the focus of which was to apply object-based image analysis (OBIA) techniques to tackle real-world issues identified by non-governmental organizations through the Planet Action network. For my project I generated a 6 class land cover data set for the Sierra de las Cruces mountain range to the west of Mexico City, Mexico using Trimble’s eCognition Developer. The data will be used to monitor the depletion of forest cover over time.

At the end of the semester it occurred to me that OBIA techniques might be applicable to the images captured via Go Force One. Using a single frame from one of the flights as a test case I set out to see if it would be possible to classify the image like I did for my project in Mexico. Figure 1 shows the original frame used to classify the image.

The process began through the application of a multiresolution segmentation algorithm to aggregate pixels into image objects as seen in Figure 2.

Using a combination of spectral and spatial properties, the objects were then classified as either sky, tree, or snow. The results, which are rather impressive, can be seen in Figure 3.

The implications of this simple experiment are in my opinion outstanding and exciting. It is exciting because I think that small UAV like Go Force One, represent the future. Technology such as Go Force One, combined with OBIA techniques offers an end-to-end solution that is robust, transferable, and most importantly, affordable. Although we already have high-resolution spatial imagery of much of the earth, much of it is outdated and it is sometimes it is difficult to stay up to date. With a UAV, imagery could be gathered for a small city more often, or perhaps imagery could be gathered moments after a natural disaster to monitor damages.

Looking to future, I plan to continue to fly Go Force One, refine its design, and also move onto other aerial imagery projects. My next project is Go Force 2.0. Go Force 2.0 will be an r/c plane with a stronger camera capable of capturing sub-meter spatial imagery. I hope to team up with a seasoned pilot; he or she will fly it, I’ll do the camera and analysis work. My biggest plans though are to one day fly a UAV over a small town, gather imagery, and apply OBIA techniques to map land cover change over much more rapid time intervals than is typically possible. To find out more about Go Force One, and to see some of the shots, please go www.vimeo.com/bobbys/goforceone.

Digitizing the Wacom Way

2011-05-19T21:16:00.000-04:00

A key component of our land cover mapping is manual quality control. We accomplish this though the application of "heads-up" digitizing techniques within ArcGIS. A single project can often require tens of thousands of manual corrections made for a given project. With so many edits even a minor improvement in per-polygon digitizing speed can yield huge gains.

We were intrigued by the possibility of using a pen-driven interactive display to perform manual corrections because most people have extensive experience drawing with pens, and are familiar and precise with their control. This allows the end user to draw their edits in a natural fashion directly on a digital map. We contacted the folks at Wacom who were kind enough to ship us their DTU-2231 to test in our production environment.

The DTU-2231 has 22” screen, which is bright and crisp. There is a power button that switches the display on and off, and a display menu similar to what you would see on a standard monitor (brightness, contrast, screen position, etc.). Also included is a stylus pen that can be stored behind the monitor in a recessed holster when not in use. The stylus is the size of a standard pen, and has three buttons: two buttons near the point and one button on the end. The end button is similar to an eraser on a pencil. The display ships with software that allows for customization of the buttons, sensitivity of the screen, and different profiles for different software installed on the computer.

In terms of the Wacom Tablet’s application in GIS Editing, several key advantages surfaced:

The tablet allows for a standard mouse to be connected, allowing the user to switch between the fast pen for editing and more “computer friendly” mouse for doing other things.
The software allows the user to enter custom “keystrokes” for the buttons. In the “customize mode” window of ArcGIS, the user is able to enter “shortcuts” for the editing tools. In our case, we have assigned “e” to switch to the feature selector tool, “c” to switch to the pan tool, and “{alt} + s” to save edits. After assigning these shortcuts in ArcGIS, we assigned the stylus buttons to match: The “eraser” button is assigned to “save edits,” the primary button is assigned to “feature select,” and the secondary button is assigned to “pan.” In assigning these shortcuts to the different buttons on the pen, we can easily create a feature, select a feature (to delete or reshape), pan the map and save our edits without ever having to move the mouse around the screen – all of these functions take place right on the stylus.
Edits are fast and precise

There are however some drawbacks:

When editing attribute changes need to be made, the pen can actually slow things down. A mouse is the most effective way to change a feature’s attributes since there is inevitably a lot of clicking that goes on.
The editing templates in ArcGIS 10 can “confuse” the pen and the custom buttons will no longer work. In ArcGIS 10’s editing templates, the user “preselects” the attributes for the polygon before going to create it. This creates a problem as the user must constantly be predefining the attributes before drawing. This problem can be fixed however by turning off the “create features using templates” in the “ArcMap Advanced Settings Utility.”
Another potential problem that we ran into occurs when individual users change the customized settings for their own purposes each session. Deciding on customized buttons is something that should be done as a group, as having to redefine the buttons every session can take 15 minutes to complete. That is of course a human resource problem as opposed to a Wacom problem.

For our needs at the SAL, the Wacom Tablet represents a time effective tool for editing single data type vector datasets (i.e. editing a vector layer solely composed of polygons representing tree canopy). The editing process is fast, precise and natural. We ended up purchasing the DTU-2231 and will likely purchase another one this summer.

GeoNet - Channel Extraction from LiDAR

2011-05-08T19:33:00.003-04:00

This post is a summary of one of our graduate students, Terry Barrett, who did some excellent work getting GeoNet to run on a sizable (billions of points) LiDAR dataset.

---

GeoNet is an open-source Matlab code for stream-channel extraction of high resolution digital elevation data, developed by Passalacqua et al. at the National Center for Earth-Surface Dynamics at the University of Minnesota. With the release of version 1.0.1 last September, the code has matured enough to be capable of adaptation to production use. One of our projects had need of stream-channel extraction for its one billion point, one meter resolution LiDAR DEM, and I was assigned the task of scaling up GeoNet and producing the stream-channel network for the study watershed.

GeoNet is designed to extract the channel network for a single watershed of maximum extent of about 2000x2000 elevation points, which it can do in a reasonable time (a few minutes) if the code is compiled. When I began using GeoNet v.1.0.1, there were no compiled versions offered for Windows operating systems, but I was able to compile it for both 32 and 64-bit using the C++ compiler included in Visual Studio 2010.

Development of the production code was done using a 3.75x3.75km subset of the watershed DEM, which produced a code that can: tile a geotiff raster into buffered tiles, sized for GeoNet; process those tiles through GeoNet to determine the channel network; trim the resulting channel-network shapefile tiles to the original tile dimensions; and connect the tiled-channel shapefiles with channels that conform to stream-flow logic.

Here is the output on the subset DEM (runtime on a 32-bit machine was about two hours; the blue dots are the connecting channel locations):

Here are some representative channel connection locations from that run that show the robustness of the connection logic (green lines are trimmed-tile channels, red dots are identified channel ends, and blue lines are the channels added to provide logical connections; tile boundaries are shown as dashed lines):

The logic correctly connected the nearest independent channel ends for all locations except one, and the code was fast enough for the available time on the production machine, and so it was then used to run the entire watershed DEM.

Here is the output for the entire watershed (run in two halves, in parallel, on a 64-bit machine; runtime ~ 36 hours):

The connector channels:

There are three reasons why the channel network has a few holes and doesn’t go to the edge of the DEM:

1) GeoNet’s curvature calculation fails on tiles with more than about 20% missing data, which affected both the edge of the DEM and some interior areas with missing data.

2) A few tiles crashed the Dinf mexfile (which calculates contributing area); I examined the tiles, and could find no significant differences between them and nearby tiles that did not crash.

3) A few channel shapefile tiles (all lowland areas, possibly covered by water) crashed the connecting code; they were about 3X the size of the next largest files, and so generated overly-large connection matrices.

The project was successful, insofar as it produced an automatically generated shapefile network over most of the watershed, but there are areas for further improvement and development:

1) Prevent holes in the network – by debugging/improving the areas of code mentioned above; using a watershed DEM with a larger buffer (perhaps 2km instead of the 10m that the watershed DEM currently has); and/or performing the tiling with another program, such as ERDAS IMAGINE (I modified some share-code, geotiffwrite, to write the geotiff tiles; may still have some bugs).

2) Validate and calibrate the channel network through comparison to networks generated by other methods (manual and automatic) and to ortho-photos of the watershed. Both GeoNet and my production code have a number of input parameters (all kept at default settings in the above analysis), and my preliminary investigations into the sensitivity of the channel network to changes in these parameters suggest that there is a lot of room for optimization.

3) Further improve the connecting-channel logic – the subset DEM used for development still had one location where the logic produced an unnecessary connection channel (this location has overlapping independent channels on the stitching boundary), and there are areas where unnatural stream loops are created.

4) The final step required to create a channel network most useful for further geoprocessing requires that the connection channels and the tile channels be merged into a single network of poly lines that properly conforms to sub-watershed regions. This may require “smart” tiling, with considerably more overlap (and processing time) than used by the current production method.

ASPRS 2011 Roundup

2011-05-05T14:51:00.002-04:00

If there was one theme from this year's ASRPS 2011 conference it was death by PowerPoint. I'm no Steve Jobs when it comes to presentations, but I know enough to not make it miserable for the audience. The graduate students in my GIS Practicum class deliver far superior presentations than many of the seasoned faculty at ASPRS 2011. Slides with endless bullet points, more formulas on a single slide than a person could read in 20 minutes, tiny graphics, the list goes on. This would be bad enough, but for a geospatial society you would think members would realize the value of a good picture. Compared to the quality of the presentations at the urban forestry conferences I have been going to, the ASPRS sessions were, for the most part, a massive disappointment.

That being said there were a number of top-notch presentations. I particularly enjoyed the session I moderated on "using multiple data sources." Tommy Jordan and his team at CRMS are doing fantastic work training the next generation of geospatial professionals by integrating undergraduate students into the production workflow for a recent LiDAR and imagery collect. Although the sensors and techniques we used are wildly different, I learned a lot from Ken Stumpf's work on land cover mapping in Alaska. Ken is with Geographic Resource Solutions. I am glad I stuck around for the final LiDAR session as two grad students, Colin Gleason from SUNY ESF, and David Kelbe from RIT, gave presentations that were filled with useful information and fantastic graphics. I attended the AmericaView session, and although I had seen much of the work before it was still inspiring.

Paul Ramsey gave a brilliant opening keynote on open source software. He did not pull any punches in going after Esri et al. I think all of us left his presentation questioning why we ever purchased commercial software. The problem with open source - they don't have the money to sponsor conferences like ASPRS.

Episode 4 of the Geospatial Revolution project premiered on day 1. Impressive is an understatement. It left me proud to be part of this profession and yet humbled by the great things others are doing. The theme of the conference was "Ride of the Geospatial Revolution," but I cannot help if ASPRS might soon find itself drowning in the location-based geospatial tidal wave. The largest booth in the exhibitor was Microsoft. That's right, the same people that sell Office were showing off their remote sensing capabilities. Who would have predicted that five years ago? I can't help but wonder what will happen to all of these companies whose business model is to sell imagery. Why, in these difficult economic times would an organization pay vast sums of money when Microsoft or Google will provide them with the data for free or at a much lower cost. No doubt there will always be the need for mapping companies, but the Bing Maps Global Ortho Project is going to change the remote sensing landscape.

Although the presentations were largely a let down I thoroughly enjoyed the Hot Topics session on the challenges involved in breaking the 85% accuracy barrier. Joe Knight from UMN was an amazing moderator. Chuck Olson, a man whose work on the elements of image interpretation serves as the basis for everything we do was in attendance. He offers a fantastic perspective from decades of experience, but it was disappointing to hear him say that all automated techniques rely solely on spectral information. WRONG, WRONG, WRONG. More than a few folks, us included, are using object-based approaches to incorporate spectral and spatial information into an automated process. Just check out Infoterra's work for proof.

Although Paul Ramsey's talk made me want to check out PostGIS, the proprietary vendors are doing some awesome work. I got a preview of the enhancements coming in the next release of eCognition. Amazing! The transition to Trimble has not slowed things down one bit and eCognition continues to redefine what is possible in terms of image analysis. Had some great conversations with teams from MDA, Trimble Geospatial, and ERDAS.

Keith Pelletier and I ran an Object-Based Image Analysis workshop on Monday. Over 30 people attended, and it was a lot of fun. I learned a lot from their questions and I hope they learned a lot too. Thanks to all of you who attended. I hope you will stay in touch.

You can check out all of the Tweets from ASPRS 2011 at #asprs11a. A lot of the tweets are me complaining about the presentation, sorry, but someone has to speak up.

Will I go next year - maybe. ASPRS really needs to post some guidelines to the presenters or better yet, make them all watch Death by PowerPoint the night before. Scientific presentations can be enjoyable.

Tree Canopy Assessment Reports

2011-04-22T00:32:00.002-04:00

It's been great to see the growing interest from decision makers in our tree canopy assessment products. Although our tree canopy assessments originally started out as "urban" tree canopy assessments, and were thus known by the acronym "UTC," more and more we are now doing the assessments for entire counties and thus use the more generic term "tree canopy assessment." Of course "UTC" sounds better and is certainly great description for our urban work. Our reports eventually end up on the Forest Service's UTC web site, but due to the time lag in the posting process and the general interest in the reports I thought I would post a collection here. The report is only one of the outputs from an assessment, but arguably the most visible one. We learn a great deal from our collaborators and continually update the format to reflect many of the great suggestions we have received. As always, don't hesitate to contact me if you have any questions about (urban) tree canopy assessments.
Cumberland, MD
Baltimore, MD
Frederick, MD
Howard County, MD
Montgomery County, MD
Montpelier, VT
Philadelphia, PA
Rutland,VT
Virginia Beach, VA
Washington DC

OBIA Spectral Gradient Tutorial

2011-04-20T11:42:00.004-04:00

One of the challenges in getting started with object-based image analysis (OBIA) is the complexity inherent in high-resolution remotely sensed data sets. Inspired by one of Bruce Gorham's OBIA workshops we started developing some examples using pseudo data. I recently posted a tutorial rule set to the eCognition Community site that I am calling "Spectral Gradient." This post provides an overview of that rule set. Please note that the rule set is designed to be a self contained tutorial in that the description of each process step is present within the rule set. To access the documentation click on the appropriate parent process then scroll to the lower right of the Process Tree window. The show/hide comment icon will appear, click on it and the documentation will be displayed in a sub-window.

The tutorial starts with this image.

The objective is to classify "Red Circles." Red Circles must meet the following criteria:
- Have a pixel that is at least 90% pure red
- Not contain any pixels that are pure white
- Have the geometric attributes consistent with a circle.

In the image above, the circle on the right along with the rounded rectangle and the triangle have pixels that are at least 90% pure red. The circle in the top left does not.

A quadtree segmentation is used to crate image objects. While it is not the most robust segmentation algorithm, it is fast.

Pure red objects are classified using the "ratio of red" feature. This feature measure the amount of that the red band contributes to the total brightness of the object. Setting a ratio of red threshold of 0.9 is equivalent to saying "classify objects that are 90% or more red."

Red objects are grown into other objects providing the brightness of those objects is less than or equal to 254 (e.g. not white). An infinite loop is used, meaning that the grow region algorithm will run until the criteria can no longer be fulfilled.

The merge region algorithm is first used to merge objects that are classified as Red then those objects that are unclassified. This is a necessary prerequisite if geometric attributes are to be used to classify circles.

Circles have low asymmetry and roundness values, thus both criteria are used to evaluate Red objects as circles.

The end result is that we have successfully extracted the only circle containing "pure red" pixels.

I'm on Twitter

2011-04-17T22:31:00.001-04:00

After seeing all of the good remote sensing info Tyler Erickson was putting out and not wanting to miss Mike Galvin's urban forestry tweets, I finally got my act together and joined Twitter. You can find me @jarlathond.

ASPRS 2011 OBIA Workshop

2011-04-16T17:19:00.000-04:00

My colleague Keith Pelletier and I are offering an object-based image analysis (OBIA) workshop at the 2011 ASPRS conference in Milwaukee. If you are interested in attending the conference and the workshop there is still time to register. Our workshop is #9 and it is an all-day workshop running on Monday the 2nd. I will also be giving a presentation, Incorporating Contextual Information Into Object-Based Image Analysis Workflows, during the Using Multiple Data Sources session on Wednesday the 4th. A description of the workshop is below.

This full-day, advanced workshop is designed to help participants harness the true power of object-based image analysis (OBIA). It is recommended that participants have a strong foundation in remote sensing and GIS, and at least some exposure to OBIA. This workshop is particularly well suited to individuals who are finding it difficult to extract information from the latest generation of high-resolution imaging and LiDAR sensors using OBIA techniques. Specific emphasis in this workshop will be paid to moving beyond the standard “segment and classify” approach that is typically employed in most OBIA projects, to an iterative workflow that better mimics the type of mapping carried out by human analysts by fully incorporating the spectral, geometric, and contextual information present in an image. Through a series of lectures, demonstrations, and hands-on exercises, participants will be exposed to the methods that will enable them to build effective and efficient OBIA routines.

The workshop will be divided into four parts. In the first part, the theoretical foundation for the effective application of OBIA technology will be laid out by drawing from the remote sensing, neurobiology, and cognitive sciences literature. This will be followed by a review of the current approaches to OBIA, with particular attention to some of the pitfalls that often prevent OBIA technology from being applied to its full potential. The second part will focus on effective approaches to and best practices for object-based feature extraction, including a thorough review of segmentation algorithms. The third part will cover more advanced topics, including: 1) image object fusion, 2) pattern recognition, 3) morphological routines, and 4) context-based classification. The workshop will conclude with recommendations on how to design and deploy enterprise OBIA systems capable of processing of datasets containing billions of pixels.

Demonstrations and exercises will make use of a broad range of remotely sensed (e.g. imagery and LiDAR) datasets and a particular focus in the exercises will be integrating remotely sensed and thematic datasets in an OBIA context. Participants are encouraged to bring their own computers to use during the hands-on exercises. OBIA software will be provided (requires Windows XP, Vista, or 7).

New LiDAR new article on NYC tree canopy mapping

2011-04-13T16:45:00.000-04:00

Head on over to LiDAR News to check out the latest article published by my colleague Sean MacFaden on LiDAR-based Tree-canopy Mapping for New York City.

Streaming NAIP into ArcGIS

2011-04-13T14:03:00.002-04:00

With all of the tree canopy mapping we do here in the SAL you can imagine that we are big fans of the National Agricultural Imagery Program (NAIP). Each state is typically imaged on a 3 year cycle with the current standard being 1m resolution products. Best part for us tree canopy mappers is that the images arer typically leaf on! While you can download NAIP products from the NRCS Geospatial Data Gateway you can stream NAIP into ArcGIS by establishing a new ArcGIS Server connection. Please note that this service only displays true color NAIP. Four band NAIP has to be procured from the APFO.

Step 1
Double-click on Add ArcGIS Server in ArcCatalog.

Step 2
Click Next once until you get to the General window where you can choose the type of ArcGIS Server connection. Make sure the radio button for Internet is selected and enter the following URL in a Server URL field.
http://gis.apfo.usda.gov/arcgis/services
Click Finish.

Step 3
Within ArcCatalog under GIS Servers navigate to arcgis on gis.apfo.usda.gov and go into the NAIP folder. Select the state image service you wish and load it into ArcMap.

Fun with NAIP

2011-04-07T20:03:00.000-04:00

My colleague, Keith Pelletier, found this in the 2010 National Agricultural Imagery Program (NAIP) for Lancaster County, PA. I imagine the aliens are impressed.

Planet Action - The Projects

2011-03-31T19:21:00.000-04:00

This semester I am supervising a small, but dedicated group of students who are applying their expertise in remote sensing to work on three projects supported by Planet Action. Thanks to the eCognition software provided by Trimble Geospatial and satellite imagery provided Spot Image the students get to work with cutting edge technology.

We are working with collaborators from France, Mexico, and Columbia on the following projects:

Mountainous land use and land cover mapping in the French Pyrenees to support the monitoring of human-nature interactions in the Haut-Vicdessos mountain area. This project is in collaboration with Thomas Houet from the GEODE Research Laboratory, University of Toulouse.
Mapping mangrove forests in the Bocas del Atrato Region of Colombia to support REDD policies.
Forest quality assessment in the mountains west of Mexico City to support ecological restoration.

Toronto High-Resolution Land Cover

2011-03-12T19:30:00.000-05:00

The City of Toronto has posted the high-resolution land cover data set we generated as part of the urban tree canopy assessment to their Open Data Portal. It's great to see more and more cities opening up their GIS data holdings.

Tsunami Planning: The Need for LiDAR

2011-03-11T10:39:00.003-05:00

As I write this post the first waves from the tsunami that originated from the devastating 8.9 magnitude have reached Hawaii. The Honolulu metropolitan region is home to over 900,000 people and contains numerous businesses, and key pieces infrastructure, including the airport. By all accounts from the media the excellent alert system that our government has in place will prevent the loss of life. Predicting the significance of the tsunami's impact is difficult, but Honolulu is fortunate in that LiDAR was acquired back in 2009. Unlike traditional digital elevation models (DEM) that only represent the bare earth, LiDAR digital surface models (DSM) are a 3D representation of all features. When I came into work this morning I loaded in the LiDAR DSM for Honolulu into Quick Terrain Modeler, draped on top of it the existing aerial imagery, and then ran some flood simulations for the Honolulu International Airport (HNL). HNL is an interesting airport. It is literally on the Pacific Ocean, with its "Reef Runway" being the world's first major runway to be constructed entirely off shore.

The graphics below show what HNL would look like if the ocean levels were to increase by 5 meters. Please note that these graphics are done for demonstration purposes only. There is no indication that the sea level will rise by 5 meters, nor has this analysis been subjected to detailed quality assurance and quality control procedures. Inundation mapping is far more complex than just raising the sea level. If you are looking for the official inundation maps please visit the the Honolulu government web site. Here is a nice article on how the updated inundation maps were generated using LiDAR. There is also an interesting report from Spatial Informatics Group on the The Protective Role of Natural and Engineered Defense Systems in Coastal Hazards that was prepared for HI. I hope that those communities affected by this tragedy are able to recover quickly.

Landsat and ArcGIS

2011-03-08T02:56:00.000-05:00

Over on the AmericaView blog is a video I created that shows how to download Landsat imagery from USGS GLOVIS site and then stack the layers, change the symbology, and generate an NDVI layer in ArcGIS. My hope is that this video will make Landsat imagery more accessible to regional and municipal GIS professionals. Thanks to AmericaView for making the funding to generate this video available.

ArcGIS 10 on a MacBook Pro

2011-03-05T15:48:00.000-05:00

Although I have been running ArcGIS 10 on my hefty Dell T7500 workstation for many months I just got around to installing ArcGIS 10 on my 17 inch MacBook Pro. I am running Windows 7 64-bit SP1 on my Mac via VMware Fusion. While I've had the occasional program crash within VMware Fusion (mostly the Java updater) I have been really happy with the performance overall. I use the MacBook Pro for teaching and from an ArcGIS geoprocessing standpoint it's faster the aging Dell Precision 390 workstations we have in our teaching lab.

At ArcGIS 10 ESRI introduced a new license check out capability. I was worried that the license check out would not work though VMware Fusion, but I logged in from home via VPN, started up the AcGIS Administrator, checked out the licenses, quit VPN, fired up ArcGIS 10, and everything worked like a charm. For large organizations like our university this new license check out capability will be a big time saver. The two of us responsible for generating license files spent untold hours in previous years generating license files and trouble shooting ArcGIS license manger issues for those users looking to go on the road with ArcGIS. I think this will save us a lot of time.

ERDAS IMAGINE 2011.0.2

2011-03-05T15:33:00.000-05:00

Although there is no mention of it on the main ERDAS web site or in the ERDAS Community there is a service pack out for IMAGINE 2011. Installing it will upgrade you to 2011.0.2. It's a bit confusing checking for service packs for IMAGINE as their old support site is still up and running, but it appears that for IMAGINE 2011 all service packs are being posted to the the new erdas.com web page under PRODUCTS > [MAIN PRODUCTS > [SUB PRODUCT] > DOWNLOADS. You can find the download for 2011.0.2 here.

USB 3.0

2011-03-04T16:32:00.000-05:00

When it comes to transferring large raster or point cloud data sets that are in the terabyte range, mailing hard drives around is always faster than FTP. Of course copying data to a USB hard drive can be sloooow. If you have not done so already consider in investing in USB 3.0. We've been getting a 10x boost in transfer speed with USB 3.0 as compared to USB 2.0.

USB 3.0 requires different ports than USB 2.0, but all you need to upgrade is a USB 3.0 PCI card, which will run you less than $50. You will also need as USB 3.0 compatible hard drive. We have been using the Seagate GoFlex drives, which support both USB 2.0 and USB 3.0 through an interchangeable base plate. This is very handy in that we can use USB 3.0 to transfer the data on our end and then switch out the base plate to the USB 2.0 model prior to sending the drive to our collaborators, most of whom are still using USB 2.0.

Prioritizing Preferable Locations for Increasing Urban Tree Canopy in New York City

2011-03-02T22:59:00.000-05:00

Our article on Prioritizing Preferable Locations for Increasing Urban Tree Canopy in New York City is out in the online journal Cities And The Environment (CATE). Abstract is below. CATE offers free access to all of its peer-reviewed articles. The latest issue contains the proceedings from the 2010 MillionTrees Research Symposium. GIS played a role in the majority of the research published in the proceedings.

This paper presents a set of Geographic Information System (GIS) methods for identifying and prioritizing tree planting sites in urban environments. It uses an analytical approach created by a University of Vermont service-learning class called “GIS Analysis of New York City's Ecology” that was designed to provide research support to the MillionTreesNYC tree planting campaign. These methods prioritize tree planting sites based on need (whether or not trees can help address specific issues in the community) and suitability (biophysical constraints and planting partners’ existing programmatic goals). Criteria for suitability and need were based on input from three New York City tree-planting organizations. Customized spatial analysis tools and maps were created to show where each organization may contribute to increasing urban tree canopy (UTC) while also achieving their own programmatic goals. These methods and associated custom tools can help decision-makers optimize urban forestry investments with respect to biophysical and socioeconomic outcomes in a clear and accountable manner. Additionally, the framework described here may be used in other cities, can track spatial characteristics of urban ecosystems over time, and may enable further tool development for collaborative decision- making in urban natural resource management.

Using LiDAR to Map the Green Infrastructure

2011-03-02T22:47:00.001-05:00

My latest article has been published in LiDAR News. It gives a brief overview of the importance that LiDAR plays in accurately mapping trees in urban areas. We are making excellent progress in mapping New York City's urban tree canopy using the recent LiDAR collect. I hope to have a post out on that project next month. For more on how New York City's LiDAR is being used for solar mapping check out Matt Ball's article on Spatial Sustain.

NLCD 2006

2011-02-16T16:57:00.001-05:00

The 2006 version of the National Land Cover Database is out. You can download the Provisional Products and Supplementary Layers. Please note that you must use NLCD 2001 version 2.0 for any 2006 to 2001 comparisons.

Hiring: Deputy Director of GIS and Information Management

2011-02-14T08:57:00.002-05:00

The City of New York just posted a great job opening for a Deputy Director and GIS and Information Management in the Parks & Recreation's Forestry, Horticulture, and Natural Resources division. With NYC's MillionTrees initiative moving forward and a 15 billion point LiDAR data set, this is a great chance to be on the cutting edged both in terms of policy and technology. Just make sure to get your application materials in by March 4th.

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

City of New York Parks & Recreation

Citywide Job Vacancy Notice

Posting № 83342

Office Title: Deputy Director of GIS and Information Management

Division: Forestry, Horticulture & Natural Resources

Civil Service Title: Computer Operations Manager Level MI

Salary: $65,000 - $80,000

Location: Olmsted Ctr., Queens

The divisions of Forestry, Horticulture & Natural Resources are central to executing the greening components of PlaNYC and MillionTreesNYC. Central Forestry & Horticulture oversee and coordinate forestry operations and green infrastructure projects. The Natural Resources Group’s mission is to conserve and restore New York City’s natural resources and habitats for biodiversity and ecosystem services. These divisions include the New York City Urban Field Station, Parks’ research partnership with the USDA Forest Service.

MAJOR RESPONSIBILITIES

• Under general supervision of the Director of Research & Analysis, manage a wide variety of data and information resources to support operations and analytics for Forestry, Horticulture and Natural Resources. Work with managers and cooperating researchers on analyses and data dissemination/coordination. Provide prompt information and GIS assistance to agency partners and stakeholders. Work with other city/state/federal agencies on specific projects like mapping green infrastructure priorities in roadbeds, watershed and natural areas.

• Supervise division’s GIS and information management staff who execute spatial and non-spatial analyses. Develop cartographic, statistical and narrative materials to support operations and research projects. Coordinate the preparation of analytical/presentation quality data summaries, analyses and maps for stakeholders and the general public.

• Supervise the collection of spatial and non-spatial data from a variety of sources including field observations, imagery, GPS, and existing databases and maps. Coordinate the storage, organization, and maintenance of data relevant to natural resource conservation, ecological monitoring and restoration, urban forestry and green infrastructure. Maintain and publish a data catalog of the divisions’ data.

• Work with Parks’ Information Technology (IT) division to define and implement an information model that starts with the collecting of data and ends with delivery of information to all users. Develop data standards that promote information relationships and discourage data redundancy; develop and enforce quality standards. Ensure proper documentation of methodologies and creation of metadata.

• Work to ensure the quality and integrity of the forest inventory and ecological assessment geodatabases. Oversee cartographic support and conduct spatial analyses of inventory and associated service request, inspection and work records, and to track ecological restoration work and support restoration planning.

• Research, recommend and implement new methodologies and solutions through the use of technology. Foster the use of mobile computing, Global Positioning Systems (GPS) and remote sensing data to support urban forestry and natural resource initiatives.

QUALIFICATION REQUIREMENTS

1. Six (6) years of progressively responsible full-time paid experience supervising or administering computer operations involving a large-scale third generation computer at least 18 months of which shall have been in a managerial capacity.

2. Education at an accredited college or university may be substituted for the general experience described above (but not for the 18 months of managerial experience described above) at the rate of one year of college for 6 months of experience up to a maximum of 4 years college for 2 years of experience. In addition a Master of Business Administration, Master of Public Administration or any other Master’s Degree in Management of Administration may be substituted for an additional year of general work experience. However, all candidates must possess the 18 months of administrative or managerial experience described above.

This position is exempt from NYC residency requirements.

PREFERRED SKILLS/QUALIFICATIONS

1. A degree with major study in Geography, Geographic Information Systems, environmental sciences, ecology, forestry and other related disciplines such as horticulture, landscape architecture or city planning.

2. Extensive experience using GIS software specifically ArcGIS, familiarity with ArcMap extensions, and GIS-based applications.

3. Experience managing and analyzing very large data sets and keeping detailed project logs.

4. Strong computer software experience with all Microsoft Office applications; familiarity with data visualization software such as Tableau.

5. Strong communication skills, especially with analytical products, as well as process documentation and sharing knowledge.

If you have any questions regarding this vacancy, please call the Personnel Division at 212-830-7851.

To apply, please submit cover letter & resume with JVN # to:

David Terhune, Director of Personnel

City of New York/Parks & Recreation

24 West 61st Street, 2nd floor

New York, NY 10023

Email: grow@parks.nyc.gov

Fax: (917) 849-6490

Attention: JVN# 846-11-84657

Getting 'N Sync with Google Earth

2011-01-06T11:06:00.002-05:00

First, apologies to those of you who expected to this post to be about America's greatest boy band ever, but I had to do something to increase traffic to the blog.

The latest releases of both Quick Terrain Modeler (7.1.2) and ERDAS IMAGINE (2011) offer the ability to synchronize the viewer with Google Earth. This is extremely handy for general orientation and given the fact that practically every geospatial professional has Google Earth installed, it just makes sense. Hats off to both of these companies for making our lives easier. Sometimes it's the little things that make a difference. There is at least one script I am aware of to make this work in ArcGIS, but it would be nice if this functionality was built-in.

ERDAS IMAGINE 2011 view synchronized with Google Earth

Quick Terrain Modeler view synchronized with Google Earth

Capitalizing on Local Government’s Investment in LiDAR

2011-01-04T15:00:00.000-05:00

My article, Capitalizing on Local Government’s Investment in LiDAR, has be published in the latest issue of LiDAR News. You can read a related take on this topic on Paul Beaty's blog The Field Guide. I disagree with Paul on paying for a point cloud classification that goes beyond differentiating ground classes in most cases. This is because classified points (e.g buildings, vegetation) don't equate to information (2D or 3D features) that can be readily integrated into most GIS software packages used by local decision makers. I do agree with him that rasterized LiDAR products are the way to go for most GIS users, but I highly encourage people to first view the point cloud and not blindly run a raster interpolation of the data.

Renaming .img Files in ERDAS IMAGINE

2011-01-04T10:47:00.000-05:00

In a previous post I covered some of the issues associated with large raster display in ArcGIS 10. It appears that all of these issues are result of using ArcGIS 10 to rename raster files that are in IMAGINE (.img) but contain an associated .ige file. The .ige file is a spillover file for "large" raster datasets in IMAGINE format. For more on .ige files please read Paul Beaty's detailed post on his blog The Field Guide. The short of it is that for 8-bit data a .ige file is generated when the .img file exceeds ~2.1GB. When this occurs the .img file becomes less than a 1MB and essentially serves as a pointer to the .ige file. The problem when renaming these larger rasters in ArcGIS 10 is that the .img file is not updated to point to the new .ige file name.

If you have renamed one of these larger rasters in ArcGIS 10 ESRI has posted a fix. Please note that this fix does not actually allow you to rename large rasters in IMAGINE format, just correct the problem if you have. The best way I have found to rename large rasters in IMAGINE format is to use ArcGIS 9.x (I run it concurrently with ArcGIS 10 using VMware Workstation) or use ERDAS IMAGINE. The screen capture below shows how to rename a raster using ERDAS IMAGINE 2011.

ESRI Hiring "Senior Raster Product Engineer" - Hope This Will Fix Bugs

2010-12-16T06:38:00.000-05:00

I noticed ESRI is hiring a Senior Raster Product Engineer. I really hope this person will be able to smooth out the raster issues with ArcGIS 10. ArcGIS 10 has some major improvements, particularly with respect to geoprocessing and the Raster Calculator. Unfortunately when it comes to basic raster display, ArcGIS 10 is one step forward and two steps back.

We have logged three bugs so far: 1) certain large floating point rasters do not display after zooming in past a certain scale, 2) certain 8-bit rasters do not display after zooming in past a certain scale, 3) renaming large rasters in ArcCatalog results in all cells being converted to NODATA. We found workarounds for all of these, but they require the use of ERDAS IMAGINE. Ironically, having image processing software in addition to ArcGIS 10 is perhaps now more important than ever. The key is to stay away from large raster datasets in ArcGIS 10. If you do have a large raster dataset and access to ERDAS IMAGINE, use the DICE tool in ERDAS to cut up the raster into smaller sections then build a raster catalog in ArcGIS. This should alleviate the three bugs I mentioned above.

Webinar: Raster Processing in ArcGIS with Map Algebra and the Raster Calculator

2010-12-09T20:57:00.000-05:00

I will be presenting a webinar on Raster Processing in ArcGIS with Map Algebra and the Raster Calculator December 20th from 2:30PM - 3:30PM EST. The webinar is hosted by the Vermont Center for Geographic Information (VCGI). To register click here. Below is a brief description of the workshop. Hope to see you there.

Map Algebra is the underlying language for Spatial Analyst tools and operators. This workshop will demonstrate simple and effective techniques for processing raster datasets in ArcGIS within the Raster Calculator using Map Algebra expressions. This workshop is designed for ArcGIS users who are familiar with raster data, but don’t have much experience with either Map Algebra or the Raster Calculator. Demonstrations will be done using both ArcGIS 9.3 and ArcGIS 10, and key changes to Map Algebra and the Raster Calculator at version 10 will be covered.

Live 3D textures in Google Earth

2010-12-05T17:29:00.000-05:00

Robert Pless and his group from Washington University in St. Louis do some fascinating work using images from web cams to update the 3D textures within Google Earth. I was in St. Louis and got an impressive demo in which his team showed billboards in New York City's Times Square displaying within Google Earth in real time. Make sure you check out the Project Live 3D website where they have an excellent overview video.

RFP for Vermont Orthophotography

2010-10-22T11:08:00.001-04:00

The State of Vermont as issued a request for proposals (RFP) for Vermont Statewide Orthophoto Acquisition Services. Previous orthophotos acquired by the state were panchromatic and acquired by a film-based system (see below). It will be nice to have true digital, multispectral imagery.

Clipping Raster Data in ERDAS IMAGINE

2010-10-18T23:42:00.000-04:00

Clipping a raster dataset using a polygon layer in ERDAS IMAGINE is not rocket science, but knowing the work flow certainly makes things easier if you have never done it before.

Here are the steps:

Load your raster dataset into the viewer
Load your polygon layer
Select the polygon layer and then click on Copy followed by Paste to create an AOI layer
Select the AOI layer and run the subset tool with the AOI > Viewer option enabled.

This video provides a quick demonstration of the process.

Thematic Raster Display in ArcGIS

2010-09-24T17:31:00.000-04:00

We do a lot a high-resolution land cover mapping and due to the number of features we map have really no choice but to store the data in a raster format. When working with these large rasters in ArcMap we found that artifacts can appear at certain map scales when pyramids are present. Deleting the pyramids will solve the problem, but this would result in unacceptable display speed for very large raster layers such as the 10GB land cover dataset I use in the example below. If you click on the image to enlarge it you will notice black pixels (roads) appear to surround the buildings (red).

We found that changing the Display settings to use a Majority resample method noticeably improves the display of thematic raster datasets at larger map scales (we review our 1 meter land cover layers at a scale of 1:2500). To change the resample setting go into the properties of the layer > Display tab > Resample during display using.

Once we adjust the resample method our land cover layer no longer has the strange artifacts.

Build pyramids and compute statistics in batch

2010-09-21T09:27:00.000-04:00

As we all know its much easier display raster datasets if they have both pyramids and statistics. When working with hundreds or thousands of raster datasets the process of building pyramids and computing statistics can be painfully slow. In version 2010 of IMAGIINE ERDAS introduced parallel processing. For every IMAGINE Advantage license four processes can be initiated. My Dell T7500 supports running 16 simultaneous processes and thus thanks to this and the number of IMAGINE licenses we have I can compute statistics and pyramids on 16 rasters at a time. It's easy to set up a batch process using the IMAGINE batch processing wizard. This is a real time saver for large tiled datasets.

ArcGIS 10 Large Floating Point Raster Display Bug

2010-09-16T07:17:00.000-04:00

For those of you, like me, that work with very large raster datasets, ArcGIS 10 has a nasty bug that causes floating point rasters to not display at certain scales. ESRI Support lists this bug as # NIM061128. The problem does not exist in ArcGIS 9.x releases so this is obviously due to some of the underlying modifications made to ArcGIS raster handling at version 10. It's a bit difficult to work with your data if you cannot see it, so we will be holding off on making the full transition to version 10 until this bug is fixed. Right now I have version 10 running on a virtual machine I set up using VMware Workstation. This allows me to run ArcGIS 10 and ArcGIS 9.3 on the same computer.

LiDAR: An End User's Perspective

2010-08-20T23:20:00.000-04:00

This video, LiDAR: An End User's Perspective, is the second presentation I gave at the New York City LiDAR workshop on August 2, 2010. I posted a recording of my first presentation, LiDAR 101, a few days ago. In this presentation I share some advice based on the experience my colleagues and I in the Spatial Analysis Lab have gained over the last six years working with LiDAR data. Certainly not all encompassing, but I do hope that it provides some useful information.

The LiDAR point cloud visualizations, surface models, along with the various LiDAR-based analyses were performed using Quick Terrain (QT) Modeler. At the end I demonstrate a workflow for automatically mapping land cover by fusing imagery, LiDAR products from QT Modeler, and vector datasets using eCognition.

Is the Peer Reviewed Literature the Best Place to Publish?

2010-08-18T17:12:00.003-04:00

This New York Times article resurrected a question that pops up in my head every time I read a peer reviewed article in a remote sensing journal that involves some type of land cover mapping: is this really the best medium for publishing this work?
The typical journal article with have an introduction, methods, results, and discussion. There will be 1-2 screen captures of the final product, most likely in black and white, and then an accuracy assessment. The screen captures are typically of a relatively small area, and often at a scale not suitable for assessing the quality of the data. The accuracy assessment is of course done by the very people who did the analysis, which is hardly an independent assessment. The problem is further exacerbated by the fact that the slow pace of peer review means that an article published today is likely the result of techniques applied 2, 3, or more years ago. In the rapidly evolving remote sensing field this is a lifetime. There are strong arguments for the peer review process, but if the reviewer does not have access to the end product they are simply reviewing a story about the data.

I am of the opinion that if some sort of final product, such as a land cover map, is the basis for the article, that dataset should be made available. With all the advances in mash-ups and GIS servers it is not that difficult to do. What of instead of publishing those papers to peer reviewed journals they were just posted to a blog, with an accompanying link to the map? The article could be published as soon as the project is complete, as opposed to waiting years, and readers could view the final product and provide feedback. This would allow others to replicate the techniques in a timely manner, which is particularly valuable given the ways in which a software package can change in a 1-2 year period. It would also allow persons with expert knowledge of the area, particularly decision makers who might make use of the data, to assess its quality.

You will notice that I did not use the term "accuracy." "Quality," while subjective, is a superior measure for assessing land cover maps, particularly those derived from high-resolution land cover data. Quality implies not only accuracy, but a realistic representation of the landscape. The distinction is illustrated quite well in the graphic presented below. This datasets (not mine) passed peer review where it was purported to have a near 90% overall accuracy. The screen capture shows three land cover classes (tree canopy, impervious, and grass/shrub) with the legend purposely omitted and the colors designed to be confusing. The land cover is from a medium density urban area at a scale of 1:1,000. While the data might meet some peer reviewed threshold for accuracy, I think it is safe to say it falls short on the quality front. The simple fact is that for a residential area, if you have a map depicting tree canopy, grass, and impervious and the end user cannot make out what those classes are without a legend, you have a problem. It is relatively easy to hide these shortcomings in peer reviewed article. If one does not have to display the end result, there is an incentive to focus instead on a "fancy" methodology, throwing around terms like "neural networks."

Below is an example from one of our projects displayed at a scale of 1:1000, although from a different medium density residential area. Once again, I am not displaying the legend, but the same colors are used to represent tree canopy, grass, and impervious. I would be interested to hear feedback, but I bet you can make out the land cover classes based on the arrangement of the features. Interestingly, the accuracy of this land cover data set is only slightly better than that of the one above. The difference is more subjective, the difference is in the quality.

Did the Land Cover Change?

2010-08-18T11:39:00.003-04:00

The City of Roanoke, Virginia released the following press release on their web site.

In 2002, an Urban Ecosystem Analysis found that Roanoke’s tree canopy was 32 percent, but since then the city has planted almost 3,000 new trees, with a goal of achieving 40 percent tree canopy in 10 years. In early 2010, a Virginia Department of Forestry report showed the city’s tree canopy to be 48 percent of total land area. The report also allows city forestry staff to target specific neighborhoods, blocks, or public properties that do not yet meet the 40 percent goal.

On the surface it would lead one to think that planting 3000 trees resulted in the overall tree canopy increasing from 32% to 48% (over 17 sq km) in a relatively short period of time. This equates to 1.4 acres of tree canopy for each one of the 3000 trees planted. Even if natural growth is factored in this rapid increase in canopy is impossible to achieve in a temperate ecosystem, particularly as some tree canopy would likely have been lost during the same time period. The 2002 Urban Ecosystem Analysis was done by American Forests, and appears to be a pixel-based classification of high-resolution satellite imagery. The 2010 report from the Virginia Department of Forestry was based on the 2008 1m NAIP, with object-based techniques used extract land cover. The bottom line is that using land cover datasets derived from different sensors using different methodologies to report change yields misleading information.

Change detection, particularly high-resolution land cover change detection in heterogeneous urban areas, is extremely challenging, but it is something we as the remote sensing community will need to master if we are to provide decision makers with accurate information regarding the impact of policies and initiatives. In the meantime the City of Roanoke could estimate the change in canopy from 2002 to 2008 by taking the two sets of images, dropping a few thousand sample points, and categorizing them and canopy or not canopy based on manual interpretation. This technique tends to be quite accurate for producing city-wide estimates of change. Unfortunately it would not help in determining where the change occurred. This would require a more costly approach, generating consistent land cover datasets from both the 2002 and 2008 time periods.

LiDAR 101: NYC LiDAR Workshop

2010-08-16T23:13:00.000-04:00

The New York City LiDAR acquisition this past spring received quite a bit of press. In early August the New York City Urban Field Station and the Mayor's Office of Long Term Planning and Sustainability organized a workshop to introduce various city agencies to LiDAR in hopes of finding the best way to capitalize on this investment and to educate the end user community. This video is a rerecording of a talk I gave, entitled "LiDAR 101." I would like to thank Sean Ahearn from CARSI for providing me with one of the LiDAR tiles.

Turning Data Into Information

2010-05-13T06:20:00.008-04:00

High-resolution land cover is one of the most useful datasets a planner can have in his/her GIS. Whether it's setting urban tree canopy goals or implementing a storm water utility based on impervious surface area, you need high-resolution land cover information. Unfortunately, the millions of dollars spent each year on high-resolution remotely sensed datasets, such as imagery and LiDAR, don't always result in corresponding high-resolution land cover information. There are two reasons for this: 1) traditional "pixel-based" approaches yield unacceptable low accuracies when applied to high-resolution data and 2) manual interpretation, while accurate, can be prohibitively expensive.

Beginning in 2005 we embarked on a series of collaborative projects with the US Forest Service that focused on assessing tree canopy in cities. That collaboration has now expanded to include more than 30 communities in the US and Canada. High-resolution land cover was a key input to the assessment process. The decision makers we collaborated with were demanding, they wanted not just high-resolution data, but accurate high-resolution data. Furthermore, developing the land cover dataset had to be cost effective. After all, there is no point in assessing the tree canopy if the assessment does not leave you with any funds to plant trees.

We came up with some automated, cost-effective methods of mapping land cover that leverage existing investments in remotely-sensed and GIS datasets from a variety of federal, state, and local sources. Although the primary driver for the land cover datasets was the urban tree canopy assessments we realized that the land cover datasets we were generating had a much wider range of uses. In collaboration with the Forest Service we put together this fact sheet titled, Turning Data Into Information. With all this money spent on data, it makes sense to budget a bit more to turn it into useful information.

Although doing this type of work is still somewhat of a niche field there are at least a few organizations that have this capability both in academia and private industry. When it comes to contracting the work out it's important to consider the overall accuracy of the end product. We are of the opinion that 90% should be the absolute minimum, with 95% the desired accuracy; particular if the dataset is going to be used as a basis to map change over time. I recommend asking for a sample of previous work prior to entering into a contract, and not just a screen capture, get the actual data for a reasonably sized area.

Segmentation algorithms in eCognition

2010-04-13T09:52:00.006-04:00

In this video I provide an overview of the most common segmentation algorithms available in eCognition including:

Incorporate thematic (vector) datasets by turning them into image objects
How to reduce processing time by stringing together multiple segmentation algorithms, using sub-levels, and maps
Alternatives to multiresolution segmentation
Compare segmentation results using maps.

If you are an eCognition user, you can download the project containing the data and the rule set from the eCognition Community web site.

Webinar: OBIA and LiDAR

2010-03-24T13:53:00.002-04:00

Definiens is hosting a webinar on the topic of OBIA and LiDAR on April 7th. You can register at the link below.
https://www2.gotomeeting.com/register/992960218

Seeing the trees through the city

2010-03-16T19:56:00.010-04:00

From a high-resolution land cover mapping perspective the urban environment is a worst case scenario. 2D and 3D heterogeneity make detecting features of interest extraordinarily difficult. Active sensor technology, particularly LiDAR can help to compliment, and in some cases, take the place of, data from passive sensors.

Below is an example of some graphics I generated for the Million Trees Research Symposium, held March 5th and 6th in New York City. As New York City lacks wall-to-wall LiDAR coverage, I pulled some examples from our work in Baltimore to illustrate the utility of using LiDAR to overcome some of the limitations of high-resolution satellite and aerial imagery. LiDAR generates its own electromagnetic energy, and thus it generally produces consistent height measurements regardless of the natural lighting condition.

As cities such as New York, Boston, and Los Angeles seek to track the impact of their tree planting initiatives over time, they would be wise to consider investing in LiDAR collections every 5-10 years. There is no doubt that imagery has an important roll to play, but in the "urban canyons" it's LiDAR that allows one to see the trees through the city.

Point cloud from downtown Baltimore. The arrows point to street trees in the "urban canyon."

2007 color infrared (CIR) 1m resolution National Agricultural Imagery Program (NAIP) data of the same area. The street trees are obscured by the building shadows.

Normalized digital surface model (nDSM) derived from the LiDAR data. The street trees are clearly visible.

While there has been a lot of talk recently about LiDAR-imagery fusion for automated feature extraction it has been my experience that extracting information from a combination of LiDAR and imagery requires more than a query of height information from the LiDAR and spectral information from the imagery. In the urban canyons it may be more advantageous to use only the LiDAR. Detecting these areas, by the presence of shadows, is of course best done using the imagery. This approach goes beyond simply populating the point cloud with spectral information from the imagery, towards an intelligent system; one that maximizes the capabilities and minimizes limitations.

I would like to thank the City of Baltimore for giving us access to their LiDAR data, particularly our friends in Baltimore City Recreation and Parks.

Large Dataset Processing in eCognition: Overlapping Tile Method

2010-02-25T22:46:00.001-05:00

This video is part of the AmericaView eCognition software tutorial series. It explains a work flow known as "Overlapping Tile Processing." Overlapping tile processing is an efficient method for processing very large datasets using eCognition Server. It is one way of overcoming the memory limitations that are inherent to object-based image analysis.

This presentation is recommended for intermediate to advanced users of eCognition.

NAIP 2010: Proposed States

2010-02-04T19:36:00.004-05:00

I found the list of states slated for acquisition as part of the 2010 National Agricultural Imagery Program (NAIP). I don't exactly know what the 3 tiers refer to, but assume it's something to do with funding, with Tier 1 states falling into the "definite" category and Tier 3 states into the "don't bet on it" category. NAIP consists of 1 meter resolution aerial orthophotos. By default, only 3 visible bands (red, green, blue) are acquired, but states have the option to "partner" and contribute funds towards a 4-band product. You can find out the cost of partnering via this link. The 4-band product includes a near infrared band, which is very useful for land cover mapping.

Here is a little-known fact about the fourth band. The contractors flying the data are in most cases acquiring the fourth band, regardless if it is paid for up front. If you are working in a state and NAIP was acquired in 2009 or later and the only deliverable was a 3-band product, try and contact the contractor for that state. You will likely find that they can provide you with a 4-band product, for a price. I received once such price quote from a contractor and it was very reasonable.

What makes NAIP unique is that, unlike most orthophotos, it is acquired during the summer months. It does not make sense to me why a product designed for agricultural mapping does not have the near infrared band included by default, but there you have it.

OBIA and LiDAR article in "Geo: International"

2010-02-01T18:53:00.004-05:00

Some of our work using automated feature extraction to map tree canopy is featured in the most recent issue of Geo: International.

Definien’s Christian Weise and Juan José Cáliz Rodríguez describe how a new generation of image analysis software supports the analysis of large volumes of high resolution image data from different sources.

Thanks to the UTC blog from Casey Trees for pointing this out.

Jefferson County, WV High-Resolution Land Cover

2010-01-29T21:32:00.005-05:00

Several months ago the Jefferson County Commission (West Virginia) asked us to develop a high-resolution land cover dataset for them. Like many counties, when it came to remotely sensed data they were data rich, but information poor. In 2005 NRCS had flown high-resolution LiDAR, and the USDA acquired 1m color infrared aerial imagery as part of the 2007 NAIP acquisition. A wonderful remote sensing database by any standards, but Jefferson County found they could not answer very basic questions such as, "How much tree canopy do we have?"

To assist Jefferson County we built an object-based image analysis (OBIA) system using the Cognition Network Language (CNL) found in eCognition to map 7 land cover classes from the imagery and LiDAR. This was followed by a detailed manual review of the entire dataset. The project was not without challenges, the chief one being the slight (0.5-4m) horizontal offset between the imagery and LiDAR. Fortunately, CNL provides a robust set of algorithms, allowing us to build an expert system that yielded, accurate and cartographically pleasing results.

The end result was a high-quality land cover dataset developed with minimal costs to Jefferson County. As this project used readily, freely available datasets it demonstrated the excellent return on investment that can be had from federal remote sensing programs.

I set up a simple viewer, overlaying the land cover data in Google Maps, with a few clicks of the mouse using the ArcGIS extension Arc2Earth.

Note that the imagery present in Google Maps is more recent that what was used for the project and thus differences between the imagery and the land cover are the result of changes in the landscape. Jefferson County no has an excellent base layer to which they incorporate these changes and track them over time.

Image Processing to Improve Image Objects

2010-01-25T01:44:00.011-05:00

A few years ago when object-based image analysis (OBIA) software was gaining popularity a number of us in the remote sensing community (myself included) went through a healthy “pixel bashing” phase. After all, applying some of the standard image processing techniques to extract features from high-resolution multispectral imagery with just a few bands just seemed silly given the importance of shape, size, and contextual information compared spectral information. Image processing techniques no longer seemed all that valuable.

My thinking has changed a lot in the past year and I now often employ image processing techniques to aid in my OBIA work, but do so at a different stage in the process than most of the recent articles I have seen in the literature. In those articles, image processing techniques are typically used to generate layers that provide additional attributes for the image objects (e.g. NDVI), which are in turn used to assign the image objects to a particular class. In this post I am making the case for using image processing routines to create more meaningful objects by employing the image processing routines prior to segmentation.

I worked on this example with Dr. Carl Diegert from Sandia National Labs. Dr. Diegert introduced Gaussian noise into an image of bright rectangles arranged in an arc. Despite the noise one can clearly see the rectangles when zoomed out to the extent of the image.

When we zoom in to point at which pixels become visible it is difficult to determine where the edges of the rectangles are.
As image segmentation algorithms work at a “local” scale they yield highly unsatisfactory image objects in a case such as this. As the image objects do not correspond to the rectangles, it is impossible to classify the rectangles by their geometric (e.g. shape, size) properties.

If we apply a median filter, the contrast between the rectangles and the background improves.

We can then use the median filter layer as an additional input layer for the segmentation algorithm. This produces image objects that better represent the rectangles.

Finally, we can extract the rectangles based on their geometric properties alone. In this example I created a function that computed the likelihood of belonging to the “bright rectangles” class based on the area (number of pixels), shape index (border length divided by 4 times the square root of the area), and asymmetry (ratio of lengths of the minor and major axes).

The advantage of using geometric properties instead of brightness is that they are likely to be more stable under varying conditions.

This was a simplified example, but the theory holds regardless of the data you are working with. Consider using edge detection layers when working with aerial or satellite imagery and slope layers when incorporating LiDAR digital surface models.
For those of you using eCognition version 8 you can download the project, data, and rule set from the eCognition Community site. There is no bouncing around between software packages - image processing, segmentation, and classification are all done within the eCognition environment.

FEMA, flood mapping, and LiDAR

2010-01-24T16:40:00.006-05:00

A recent news article on the dim view local officials have of FEMA's flood maps reminded my own less than rewarding experience with the FEMA flood mapping program. It should be no surprise that in the last several years FEMA began using LiDAR to update flood maps. It's hard to beat LiDAR when it comes to generating high-resolution digital elevation data over broad areas. My main criticism with the flood mapping process relates not to the quality of the maps themselves, but to the contracting out of the flood maps, specifically the deliverables that the contractor is required to provide. I have never been successfull in obtaining the point cloud data from any FEMA flood mapping project. It appears the main problem is that the contractor hired to perform the flood mapping is the sole keeper of the LiDAR data. These contractors have no monetary incentive to either preserve or share this data (and I don't blame them if that's not what they were paid to do). The LiDAR point cloud is a rich source of information, something that should be preserved for long-term analysis, not thrown away by FEMA's contractors. Preserving the point cloud would also provide a mechanism for independent verification of the flood maps themselves.

What should FEMA do? In my opinion they should turn over all LiDAR contracting to the USGS. Once the data is acquired FEMA can contract out the actual flood mapping work. The USGS has an excellent track record of not only sharing data, but in getting interested parties to invest in "buy-ups" to either expand the area of coverage or improve the specs. FEMA has done quite a bit of LiDAR collection (through their contractors) in my state (Vermont), yet no one I have spoken with (VT has a very tight-knit geospatial community) has ever heard FEMA discuss collaborative opportunities. Our state USGS representative on the other hand, is constantly trying to build support for new LiDAR initiatives.

If the news reports about the quality of FEMA's flood maps are true and my suspicions about the point cloud data being tossed are well founded, FEMA's flood mapping money has not been well spent. I would encourage anyone from FEMA or anyone who has had luck getting FEMA LiDAR data to post a comment. I hope I am wrong.

IMAGINE 2010 first looks: batch processing

2010-01-08T09:16:00.004-05:00

I installed ERDAS IMAGINE 2010 before the holidays, but only just got around to testing out the new batch processing capabilities. One of IMAGINE 2010's strong points is the ability to use multiple threads for certain batch tasks, including image compression. Most would agree that there is lackluster support for multicore processing in the geospatial software arena, so it's nice to see things are coming around.

The screen capture below is from the new IMAGINE 2010 process list. I needed to compress several hundred GeoTIFF files to JPEG 2000. My new Dell T7500 workstation has 16 cores, allowing me to compress 16 tiles simultaneously. Needless to say this is a huge time saver.

Mapping the Green Infrastructure

2009-11-22T22:47:00.006-05:00

My presentation on Mapping the Green Infrastructure from the 2009 eCognition User Summit in Munich Germany. I discuss object-based image analysis (OBIA) systems and how we leveraged eCognition software to automatically generate an accurate high-resolution land cover dataset for the entirety of Jefferson County, West Virginia (550 km2) using freely available imagery (USDA NAIP), LiDAR (NRCS), and vector datasets (Jefferson County Commission). Neither the imagery nor the LiDAR were acquired for this project, making this a wonderful example of how automated feature extraction can generate a return on investment from existing datasets. Imagery and LiDAR can provide a gold mine of information, but this type of work requires a robust OBIA system, capable of processing billions of pixels worth of data. While much of the geospatial software industry has been lagging behind in the enterprise category the good folks at Definiens have been offering thier enterprise ready eCognition Server product for many years. An interesting side note - the parcel-based summaries (summarizing land cover for hundreds of thousands of parcels) I show early on in the presentation took just over 8 minutes in eCognition Server compared to 36 hours in ArcGIS (before it crashed). I will cover this is another post.

Getting Started with ArcGIS Model Builder

2009-10-24T20:49:00.003-04:00

Model Builder provides an intuitive environment for non-programmer types to string together geoprocessing tasks. In this video I demonstrate how to create a basic model and make the model suitable for use by those with no Model Builder experience by exposing parameters and generating help. The end result is that a model that, to the end user, looks just like any other tool in ArcGIS.

Going from LAS to Raster in QT Modeler

2009-10-13T04:55:00.007-04:00

QT Modeler continues to top my list of software packages designed for viewing, editing, querying, and exporting LiDAR data. In this video I demonstrate a work flow we employ for taking LiDAR data in LAS format and exporting it to a raster for use in ArcGIS. The best part about QT Modeler is that there is 64-bit version so you can load up on memory and process very large LiDAR datasets that quickly exceed the capabilities of ArcGIS.

Meaningful Objects

2009-10-08T21:58:00.009-04:00

It has been said before, and it is certainly worth repeating, that object-based image analysis (OBIA) represents a paradigm shift in the analysis of remotely sensed data. As the name implies, OBIA centers on the application of segmentation algorithms to group pixels into objects. Objects, particularly when it comes to high-resolution imagery, are far more meaningful than pixels as they have spectral, spatial, and topological information.

In diving through the peer reviewed literature over the past several years there seems to be very little written about insuring that the objects created from segmentation algorithms actually represent the features of interest. Unfortunately, most of the literature has focused on comparisons of image segmentation algorithms and discussions relating to “optimal” segmentation parameters. While intellectually stimulating, these writings ignore the fundamental issue in land cover mapping applications of OBIA - image objects need to approximate polygons that a human would digitize. I would argue that “optimal” has nothing to do with the particular algorithm or the parameters used, and everything with the quality of the output. It is only when objects represent features of interest that we can begin to apply some of the key elements of image interpretation, such as size and shape, to assign those objects to meaningful classes.

So what is the best way to create meaningful objects? We need to recognize that any segmentation algorithm is just an algorithm. That algorithm has no way of knowing the objects you are interested in, nor can we expect it to replicate human cognition is a single run. Over the past few months the SAL has been putting a lot of thought into this process of creating meaningful objects. We’ve come to the conclusion that the biggest problem with the vast majority of papers presented in the peer reviewed literature is that they follow a simple linear process in which image objects are created through the use of a segmentation algorithm, and subsequently assigned to a land cover class.

We believe that the best way to create meaningful objects, and in turn, accurate land cover maps, is to take an iterative approach. In this iterative approach there is still an initial segmentation, but once we have image objects they are subject to additional functions beyond classification such as morphology, fusion (with other image objects), and re-segmentation. In many cases we find it takes a combination of these processes, sometimes via loops, to create meaningful objects.
Below is a video recap from a presentation I gave at the AmericaView Fall Technical Meeting held at the USGS EROS Data Center on this topic. If you are a Definiens user and would like to take a look at an example rule set and project that employs this methodology you can download the one I demoed at the AmericaView Fall Technical Meeting from the eCognition Community. The video below is a short recap of the talk I gave on this subject.

Metadata templates in ArcGIS

2009-09-23T05:39:00.004-04:00

Metadata. Few people bother creating it and no one enjoys doing it, but it's a crucial step if you want to make geospatial data useful to others. Metadata standards are cumbersome, with many fields containing repetitive information. In this video I created for the GIS Practicum course I teach, I show how to create a metadata template in ArcCatalog and apply that template to an existing dataset. Thanks to Keith Pelletier who pointed out that you can safely use the find an replace tools in Microsoft Word 2007 to make changes to your metadata in XML format.

The only element of image interpretation successfully automated?

2009-09-21T21:02:00.009-04:00

In reviewing the list of workshops for ASPRS/MAPPS 2009 Specialty Conference I noticed that the eminent Dr. Charles Olson is providing a workshop of the fundamentals of image interpretation. From previous posts you can tell that I am a strong believer in the elements of image interpretation. Olson does a fantastic job of discussing how the remote sensing community sacrificed accuracy by focusing on automating only one element of image interpretation (tone) using statistical algorithms in this 2008 paper. I was nevertheless surprised to see that the workshop description cites tone as the only element successfully automated.

I will be the last to argue that the remote sensing community has completely come around and is now successfully using the all of the elements of image interpretation to automatically extract features. One only has to review a recent syllabus from an upper level digital image processing course (I won't post links to the offenders) to see that pixel-based unsupervised and supervised methods, which ignore the elements of image interpretation and all the advances in neuroscience over the past several decades, are still widely taught as the de facto methods of automated feature extraction. (I even found one syllabus that grouped supervised and unsupervised classifiers under the heading of "pattern recognition," when nothing could be farther from the truth). That being said, I think Olson is entirely incorrect when he states that tone is the only element successfully automated. A team from The University of Queensland Brisbane has a very nice article in the latest issue of PE&RS, demonstrating how mapping banana plantations was automated using tone, texture, and context. Microsoft Research has some other interesting examples that clearly show that many of the elements of image interpretation can be incorporated into an automated process.

Now if you put me on a desert island and asked me to bring only one reference to guide my image interpretation it would still be Olson's 196o classic, Elements of Photographic Interpretation Common to Several Sensors. Nothing beats a human when it comes to accuracy, but tone is no longer the only element to be successfully automated.

Sweat the small stuff

2009-09-04T23:16:00.009-04:00

Several years ago we performed our first urban tree canopy (UTC) assessment in Baltimore City, which lead to Baltimore City establishing one of the first UTC goals in the nation (40%). The land cover data used to determine Baltimore's existing tree canopy percentage of 20% came from the Strategic Urban Forest Assessment (SUFA) dataset. SUFA relied on pixel-based classifiers to extract land cover information from 2001 IKONOS satellite imagery.

A few months ago, with funding from the Baltimore Ecosystem Study (BES) and in collaboration with the US Forest Service and Tree Baltimore, we decide to repeat the land cover mapping to see if we could improve upon the SUFA dataset. We took a data fusion approach, integrating LiDAR, color infrared NAIP imagery, and vector datasets within an object-based image analysis (OBIA) environment. While we knew that SUFA underestimated tree canopy, due largely to the fact that the pixel-based approach failed to detect individual trees and smaller forest patches, it was rather surprising to find that it underestimated tree canopy by a full 7 percentage points. Now one might argue that tree canopy increased from 2001-2007, but after reviewing the data we discovered that it was more likely that tree canopy decreased during that time period due to development. Remember that book Dont' Sweat the Small Stuff? Turns out that when it comes to tree canopy in urban areas, the small stuff adds up and the source data and methods can have a big impact on the final percentages. As Baltimore seeks to increase its tree canopy though individual tree plantings, the assessment of whether these plantings are impacting overall tree canopy will hinge on accurate land cover mapping. A data fusion approach using OBIA technology delivers far superior results.

Some examples of the comparison between 2001 SUFA and the new 2007 UTC assessment data are below. You can download the 2007 land cover data here.

2007 color infrared NAIP

2001 SUFA 3-class land cover
(Tree canopy is dark green)

2007 UTC assessment 7-class land cover
(Tree canopy is dark green)

I love you NAIP 2008.

2009-07-23T06:29:00.002-04:00

A lot of love for the 2008 NAIP in Virginia Beach.

UVM becomes 7th Definiens Center of Excellence

2009-07-08T13:25:00.003-04:00

Definiens selected the University of Vermont's (UVM) Spatial Analysis Laboratory (SAL) as their 7th academic Center of Excellence. The SAL also administers UVM's ESRI Development Center (EDC).

ArcGIS Advanced PDF Export

2009-06-13T22:57:00.006-04:00

ArcGIS 9.3.1 provides the capability to export PDF documents in which the end user can interactively query the attributes of vector features. I provide some tips and tricks on using the advanced PDF export function within ArcMap in this video.

ERDAS IMAGINE GLT

2009-06-10T13:02:00.004-04:00

Here is a video I put together that provides an overview of the IMAGINE Geospatial Light Table (GLT). Those raised in the image intelligence community know the power of using an electronic light table (ELT) to exploit imagery. When manually examing data from sensors such as Quickbird and GeoEye, that acquire data greater than 8-bits, the ELT interface is a must-have as it provides easy access to tools such as dynamic range adjust (DRA). The folks at ERDAS have done a very nice job of taking the traditional ELT interface and building in additional capabilities. If you use are an IMAGINE user and still use the classic viewer it's time to make the jump to the GLT. I guarantee you won't ever go back.

OBIA, A Data Fusion Approach

2009-05-21T22:15:00.004-04:00

Below is a rerecording of a presentation I gave at the 6th annual Pennsylvania Remote Sensing Conference on May 18th. I discuss the object-based image analysis (OBIA) data fusion approach we take to land cover mapping. Please click on the graphic to view the video.

Urban Tree Canopy Assessment - Rockville, MD

2009-05-20T23:42:00.005-04:00

The video below presents the results of the urban tree canopy (UTC) assessment we just completed for Rockville, MD. High resolution land cover is the basis for the UTC assessment. Rockville is covered by both leaf-on color infrared imagery and LiDAR data, but mapping land cover proved to considerably more challenging than we expected. Some filtering was performed on the LiDAR data, resulting in a very inconsistent first return dataset (we did not have access to the point cloud). By taking a data fusion approch, combining the LiDAR and imagery in Definiens we were able to get around this, but it was not easy. I discuss these issues a bit in the video, but hope to cover the process in more detail in a future post. Imagery was prepped using ERDAS IMAGIN E. LiDAR data was prepped using QT Modeler. The UTC metrics are generated using a geoprocessing model within ArcGIS. Please click on the image below to view the video.

LiDAR intensity

2009-04-27T15:19:00.008-04:00

For those of us interested in LiDAR the in the scan blog is one of the best resources out there. The recent post on intensity was fantastic, but I was very surprised to see that conclusion of the post was that intensity is not a particularly useful LiDAr attribute. I was surprised because we have made rather extensive use of LiDAR intensity data over the past few months for our high resolution land cover mapping work. I should note that I am in no way a LiDAR expert, just an end user. We've found LiDAR intensity data to be particularly effective for impervious surface mapping. We are just wrapping up a land cover project in Rockville, MD that made use of both imagery and LiDAR intensity data.

Color infrared (CIR) aerial imagery acquired in 2007 (leaf on):
LiDAR intensity data acquired in 2008 (leaf off):
The paved paths in the golf course, while visible in the aerial imagery, are much more easily distinguished in the LiDAR intensity image.

We take a data fusion approach to automated feature extraction in which LiDAR and imagery data are integrated. Using the Definiens/eCogniton software suite we build an expert system to automatically extract features based on the properties of both the LiDAR and the imagery.
The end result is quite good (keep in mind that at a scale of ~1:300 we are zoomed in far beyond the scale at which either the imagery or the LiDAR data is accurate to).
LiDAR data was prepped using QT Modeler.

Vermont joins AmericaView

2009-03-29T09:16:00.002-04:00

Vermont was recently accepted into AmericaView. The VermontView consortium is lead by the University of Vermont' s Spatial Analysis Lab and includes the following partners: VCGI, USGS, NRCS, USDA Forest Service, VACC, ANR, and VAPDA.

Should NLCD 2001 be used to estimate tree canopy in urban areas?

2009-03-28T22:34:00.005-04:00

The simple answer is - no. We generated land cover data for three areas: Baltimore County, MD, Burlington, VT, and Des Moines, IA. Object-based image analysis techniques were employed to derive land cover from high resolution imagery and LiDAR. We then compared the estimates from our land cover data (>95% overall accuracy) to that from the NLCD 2001 tree canopy layer. The differences were striking, with NLCD underestimating tree canopy by 27% to 48%.
This isn't to say that NLCD is a bad product, it's not. NLCD tree canopy estimates are derived from 30m Landsat data. Urban areas are highly heterogeneous, with much of the tree canopy existing either individually or in small clumps. It's unreasonable to expect a sensor, such as Landsat, with only 7 bands (6 of which are used to derived the NLCD tree canopy layer) to detect such features. The graphic below shows that NLCD is clearly biased towards large clumps of trees, just as one would expect when performing land cover mapping using 30m pixels.
Accurate tree canopy estimates in urban area require the use of scale appropriate remotely sensed data. Landsat, for all its greatness and longevity, is not that dataset.

ElderSmile

2009-03-12T23:50:00.004-04:00

Our article on the ElderSmile program is in the April edition of the American Journal of Public Health.

Societal changes, including the aging of the US population andthe lack of routine dental service coverage under Medicare,have left many seniors unable to afford any dental care whatsoever,let alone the most advanced treatments. In 2004, the ColumbiaUniversity College of Dental Medicine and its partners institutedthe ElderSmile program in the largely impoverished communitiesof Harlem and Washington Heights/Inwood in New York City. Thelong-term goal of this program is to improve access to and deliveryof oral health care for seniors; the short-term goal is to establishand operate a network of prevention centers surrounding a limitednumber of treatment centers. Preliminary results indicate substantialunmet dental needs in this largely Hispanic and Black elderlypopulation.

The Elements of Image Interpretation

2009-03-06T06:20:00.009-05:00

There are generally five accepted elements of image interpretation: color, texture, pattern, shape, size, and location. From my readings, it appears that these principals were first documented the 1940s. Makes perfect sense as World War II and the subsequent Cold War resulted in the rapid development of military remote sensing capabilities. Coupled with this increase in capabilities was the need to thoroughly understand what principals should guide a photointerpreter. If remote sensing is considered to be both an art and a science, the trade craft employed by the military photointerpreters during these early days was certainly an art form.
I would like to put forth the argument that the advent of digital image processing techniques caused some sectors of the remote sensing community to focus too much on science of remote sensing, and less on the art form. In short, the trade craft was lost. No where was this more evident than the widespread use of pixel-based classifiers, such as the unsupervised and supervised routines that were commonly employed to extract land cover information from digital imagery.

When one considers the elements of image interpretation, it is clear that techniques that rely solely on the spectral (color) values associated with individual pixels violate the very principals of image interpretation. Perhaps much of this was due to the limitations of the technology at the time. After all, it was not until Definiens introduced their object-oriented software a decade ago that the shift towards object-based image analysis (OBIA) techniques that better replicated the human cognitive process, began. Nevertheless, when examining the literature over the past decades in which pixel-based classifiers where used, rarely do I see a disclaimer along the lines of "these techniques violate the very principals of image interpretation, but they are the best tools we have."

OBIA requires one to understand the elements of image interpretation. Perhaps the "art" is back in remote sensing.

Rock River LiDAR, First Looks

2009-03-05T10:25:00.006-05:00

I received the initial delivery of the LiDAR acquired in 2008 for the Rock River watershed (St. Albans area) from this USGS yesterday. Here are some screen shots of the point cloud displayed in QT Modeler. Once all of the deliverables are in, the good folks at VCGI will distribute the data. We will be using the LiDAR in combination with the 2008 NAIP to derive high resolution land cover. LiDAR will play a key role in helping map and model critical source areas.

NAIP 2008: states with 4-band imagery

2009-02-20T06:17:00.002-05:00

2008 NAIP is making its way to the public domain. Several states participated in the "upgrade" to 4-band (blue, green, red, NIR) imagery. The map below shows 2008 NAIP products by state.

NAIP acquisition cycle - partner up!

2009-02-08T13:52:00.003-05:00

The USDA has release the NAIP acquisition cycle for 2009-2011. NAIP is acquired as a true color (3-band) product by default. It is relatively affordable to partner-up with the USDA to upgrade to a 4-band true color/color infrared product. For example, it cost us in VT only $20k to upgrade from 3-band to 4-band in 2008. 4-band imagery is far superior for everything from impervious surface mapping to change detection. With state orthophoto programs feeling the pinch, NAIP provides an affordable way to insure consistent imagery exists for your state. As NAIP is public domain anyone can use the data.

The migration to 64-bit - has it started?

2009-02-05T22:24:00.002-05:00

If you are like me, that heavy duty dual quad-core 64-bit workstation sitting under your desk that keeps your toes warm during the winter doesn’t get much of a workout from most of your desktop geospatial applications. The vast majority, and certainly the most prevalent applications, are written for 32-bit operating systems. Thus, they can make use of 2GB of memory at the most. With the size of geospatial datasets getting larger each year, you, like me, have likely received your share of error messages relating to memory limitations when working with these large datasets.

Needless to say I was pretty excited when I heard from the folks at Applied Imagery about the 64-bit version of their flagship QT Modeler LiDAR software. Although I've had the 64-bit version of QT Modeler kicking around for a few weeks it was only this afternoon that I finally got around to increasing the memory on my Dell T7400 running Vista-64 to 32GB (at a cost of a little over $1000). I fired up the 64-bit version of QT Modeler, loaded in the entire Baltimore City LiDAR dataset (122 LAS tiles), and was happily flying through billions of LiDAR points like I was in Google Earth.

Will more companies follow Applied Imagery’s lead, I don’t know, but let’s hope so.

Scared of the Geodatabase?

2009-01-08T05:49:00.005-05:00

Geodatabases have been around for quite some time but a good number of folks I talk to in Vermont are still managing their GIS data in legacy formats (shapefiles/coverages). The presentation below is a recap of my talk at the VSDP January 2009 roundtable in which I provide a gentle introduction to geodatabases for those individuals who are considering migrating their data to a geodatabase, but have been a bit scared off by all this talk of mulituser editing, replication and what not. In the presentation I discuss feature datasets, topologies, and domains.

4-band NAIP

2009-01-06T21:19:00.008-05:00

Displaying 4-band imagery in ArcGIS can be confusing at first. The video below is a quick recap of the presentation I gave on displaying Vermont's 4-band NAIP in ArcGIS at the January 2009 VSDP roundtable.

Baltimore County high resolution land cover

2008-12-23T10:28:00.007-05:00

Some initial results from our land cover mapping in Baltimore County. Over 175GB of imagery, LiDAR, and vector datasets were incorporated into the land cover classification process. We used Definiens Developer to build the expert system and eCognition Server to apply the rule sets. An example, at a scale of 1:2,000, is posted below. For comparison, some previous work done using a pixel-based classifier is also shown. The object-based image analysis (OBIA) method we employed vastly improves the accuracy, but there are some problems with confusing street lamps and fences with trees due to the use of LiDAR.
National Agricultural Imagery Program (NAIP) orthophoto.
Our work using OBIA techniques.
Original pixel-based classification.

Vermont Spatial Data Partnership January roundtable

2008-12-23T10:08:00.003-05:00

The Vermont Spatial Data Partnership (VSDP) January roundtable is just around the corner - January 7th in Randolph. I will be giving a presentation discussing some of the new remote sensing data that will be appearing soon, specifically the 2008 4-band NAIP. My other session will provide a gentle introduction to those folks who are looking to migrate their data to either personal or file geodatabases using ArcGIS. Video summaries of both presentations will be posted to this blog.

Remote sensing industry outlook.

2008-12-17T11:34:00.007-05:00

The latest issue of Geoworld has the geospatial industry outlook for this coming year. Five industry experts from the remote sensing sector gave their take on the remote sensing industry. I was a bit disappointed in the interviews as most just gave the usual spiel about all the satellites that were operational or soon to be operational. I think we are starting to reach the point at which high resolution imagery is losing the "wow" factor. Statements like "That's great that I can see my neighbors car, but how is this imagery helping me make better decisions?" are already starting to surface, and will continue as budgets tighten. Purchasers of remotely sensed data will want to see a return on their investment, and not spend 6 figures just so the maps their GIS shop produces have nice background imagery. They will want answers to their key questions. For example, "What is the distribution of impervious surfaces by land use in our city?" or "How much room do we have to plant trees on park land?" In short, they will want actionable intelligence, not just imagery.

It was nice to see at least one of the experts interviewed raise this issue. Ron Lake, the CEO of Galdos Systems said:

Increased volumes of remotely sensed data without concomitant increases in the ability to rapidly generate and integrate derived information products will greatly limit the utility of the new remotely sensed data.

Pixels may not be the answer, but they are part of the solution

2008-12-12T19:53:00.004-05:00

This post expands upon my previous post regarding object-based image analysis (OBIA). While pixel-based classifiers simply don't work with high-resolution remotely sensed data this does not mean that pixels don't have a valuable role in extracting information from high-resolution remotely sensed data. For one, image objects are created from pixels, so they constitute the building blocks of image objects. Moreover, the application of the appropriate pixel-based processing algorithms can greatly improve OBIA by improving the way in which image objects are generated (e.g. edge detection) or adding features that make image objects more interpretable (e.g. NDVI). In the end, we need a way to turn the data we have into information. OBIA provides the most effective way to replicate manual image analysis by moving the classification process from meaningless pixels to meaningful objects. When image obects are present we are no longer stuck with using the spectral properties of the pixel to classify our data, we have texture, context, shape, size, etc. Pixels are not the answer, but they are part of the solution.

What's happening in remote sensing today?

2008-12-11T11:59:00.005-05:00

I make an appearance in EarthTV's what's happening in remote sensing today video. Kass Green starts things off making a key point about exploiting today's high resolution imagery - you can't make maps the old way through pixel-based classification.

Helping NYC plant one million trees

2008-12-11T00:19:00.010-05:00

With all the wonderful things going on in geoweb land it sometimes gets lost that the true power of GIS is that it can help people make better decisions. Doing the analysis to provide actionable intelligence requires a tremendous amount of work. As I mentioned in a previous post I am co-teaching the GIS Analysis of NYC Ecology class this fall in which students are helping the New York City Dept. of Parks figure out where to plant those one million trees. Students gave their final presentations this past Monday. Each group selected a neighborhood in a borough (based on their previous neighborhood prioritization) and worked to prioritize the parcels within that neighborhood for tree plantings. Fantastic presentations all around; the students clearly invested some time in slogging through hundreds of datasets, and thought long and hard about what matters when it comes to planting trees. The presentations were innovative, inspiring, and serve to remind all of us that GIS can help us make this planet a better place to live. It was great to see the urban tree canopy assessment work we did back in 2006 (provided some of the justification for the million trees initiative) reach to the next phase.

I’ve posted some screen captures of the work the students did below, and links to the PowerPoint presentations. In a nod to the geoweb – yes, it would be much better if the data in the presentations were served up in an online map. I hope to get some of them up on the web in early 2009 now that we have a central ArcGIS Server running on campus.

You can read more about what the class did in this article. This work this class did also played a key role in helping the NYC-Forest Service-UVM team win the 2008 New York State GIS Partnership award.

Building shading calculated to identify areas where trees would have difficulty growing due to the lack of sunlight.
Impervious surfaces derived from high resolution aerial imagery summarized at the parcel level to help prioritize parcels in need of greening.
Parcel prioritization for the Brownsville neighborhood in Brooklyn; factors evaluated included stewardship potential and land suitability.

Calculation of how many trees could be planted in the Brownsville and Ocean Hill neighborhoods based on available land and Forest Service crown diameter data.Links to PowerPoint presentations:
Bronx parcel prioritization
Brooklyn parcel prioritization
Queens parcel prioritization

Definiens Developer customized import

2008-12-09T00:44:00.007-05:00

The customized import tool is an efficient mechanism for batch project creation in Definiens Developer. Setting up a customized import is a bit confusing the first time you do it so I created this video outlining the process. In this particular example I set up a customized import process to create projects for series of imagery and LiDAR tiles along with a thematic layer (water polygons). I tiled the imagery and LiDAR in ERDAS IMAGINE using the Dice tool with following settings:

If you don't want to employ tiling and stitching in Definiens, or you can't because you don't have eCognition Server, loading tiled images with overlapping extents can help dramatically reduce unwanted edge effects. You can process each of the tiles individually then mosaic the output from Definiens using your GIS or image processing software. When you mosaic the output it's important to select a software package that supports the generation of seamlines so that you can generate a cut line down the middle of the overlapping area. That being said, processing the data through eCognition Server is much more efficient.

Extracting tree canopy from LiDAR

2008-12-06T21:46:00.009-05:00

This video demonstrates some techniques for extracting tree canopy from a LiDAR dataset using Definiens Developer. The sample rule set I generated employs the contrast split segmentation algorithm to extract tall image objects from the LiDAR and assign those objects to a tree class. The contrast split segmentation algorithm does a fantastic job of highlighting trees, but the image objects resulting from the contrast split algorithm are jagged and thus don't look like trees all that much. To amend this issue I employ some reshaping operations consisting of growing tree objects into surrounding pixels, followed by a circular morphological filter to smooth out the rough edges. For those of you unsure what this business of object based image analysis (OBIA) is all about I highly recommend watching the Definiens OBIA made easy video starring Stickman.

Leveraging the geoweb in desktop GIS

2008-12-06T07:25:00.007-05:00

Despite all the great things the geoweb has brought us, when it comes to geoprocessing most of us are still using ArcGIS Desktop. The geoweb, for the most part, is an afterthought. It's something we take advantage of when the analysis is complete, as a way to share our results. Most of the data we work with is stored if not locally, at least on a nearby server. For students just learning GIS it can be difficult to understand how the geoweb can influence their analytical workflow.

I was asked to give a presenation on the geoweb yesterday to a geography introductory GIS course. As the students had been working in a desktop GIS environment all semester I wanted to avoid the mistake I believe made by some presentations on the web/neography that imply desktop GIS is dead. Desktop GIS, much like desktop word prcessing is still going to be valuable 5 years from now. In addition to discussing the geoweb and presenting some wonderful examples of neogeography I came up with a little demo to show the class how the geoweb can be leveraged in a desktop GIS environment. You can click on the graphic below to view the presentation. I stream in data from a web coverage service (WCS) to ArcMap then run a geoprocessing model using local data to extract values from the WCS raster, symbolize the resuling point data, and export the data to KML for use in Google Maps.

Definiens 2008 user meeting

2008-12-06T06:20:00.009-05:00

I attend this year's Definiens user meeting in Denver on November 21st. As a long time user of Definiens/eCognition it was great to meet some of the brains behind the software. I presented an overview of our urban tree canopy (UTC) assessment work in which Definiens technology plays a key role, allowing us to extract actionable intelligence from very large geospatial datasets to inform the decision making process. You can view the portion of my presentation where I demonstrate one our rule sets by clicking on the graphic below.

There were a lot of great presentations but one that particularly intrigued me was a demonstration that Peter Becker (ESRI) and Gregor Willhauck (Definiens) put on, showing how Definiens Developer can ingest data from ArcGIS Image Server. While this capability is still in the beta phase it points to the larger trend we see within the geospatial industry of traditional desktop operations moving out to the cloud.

The cloud has certainly changed things in terms of web mapping applications with services such as Amazon S3 allowing anyone to serve up massive amounts of data, but I have not seen any examples, until this presentation, on how the cloud could be leveraged to help extract information from remotely sensed datasets. There is an unprecedented amount of remotely sensed data being gathered. Unfortunately, the capacity to analyze it not there and most of it end up serving as nothing more than a pretty backdrop in Google Earth, ArcGIS, etc. This is to be understood given the challenges of extracting information from high resolution remotely sensed data. The datasets are so large that desktop processing is not feasible. The cost of moving to the enterprise is steep - purchasing the servers and the software (eCognition Server) to run on those servers starts in the tens of thousands of dollars. Making such a purchase can be difficult to justify if you are only going to use the service a few times a month. What Peter and Gregor presented gives me hope that in a few years we will be able to take leverage the knowledge of a community of experts that is scattered across the globe and the capabilities of the cloud with a pay as you go scheme. For example, GeoEye-1 acquires an image, you access a portion of the image via ArcGIS Image Server to build your rule set, then pass the rule set off to someone with a server farm running eCognition Server to handle the processing of GB of data. You pay for the imagery and the server time. No mailing DVDs, no buying and maintaining expensive hardware and software.

Neighborhood tree planting prioritization

2008-11-17T23:52:00.005-05:00

The students in the GIS Analysis of NYC Ecology course that am I co-teaching this semester presented the results of their tree planting prioritization analysis done at the neighborhood level to our collaborators at the NYC Parks Department today. Good stuff all around. As you can imagine a tremendous amount of work goes in to simply deciding what variables should be analyzed. Each team analyzed a minimum of six variables, computed Z scores, and generated an overall prioritization ranking. The graphic below shows what the Brooklyn team (Dan Erickson, Michele Romolini, and Jiaxin Yucame) came up with based on:

Asthma Hospitalization Rate
TRI Facilities
Household Poverty Rate
Race Distribution
Home Ownership
Vacant Land
Possible Urban Tree Canopy
Stewardship Activity Level
Number of Schools
Number of Community Gardens

Linking vegetation and health: the importance of resolution

2008-11-16T15:08:00.004-05:00

In this month's medical journal Lancet a team of researchers from the University of Glasgow (my Granny's alma matter) and the University of St. Andrews published results from a study examining health inequities in relation to green space. The found that - Populations that are exposed to the greenest environments also have lowest levels of health inequality related to income deprivation. Physical environments that promote good health might be important to reduce socioeconomic health inequalities.

For our Urban Tree Canopy Assessment work we have compared tree canopy estimates derived from our high resolution land cover data (we map every tree) to those obtained using the 2001 National Land Cover Database (NLCD) tree canopy layer. As you can see from the below example from Cumberland, MD, the results are striking. NLCD underestimates tree canopy for the city by a full 20 percentage points (48% vs 28%).

Naturally I was interested in the source data used in the Lancet article. Turns out there is a land use layer for England that maps out open spaces down 5m2. I was encouraged to see high resolution data being employed as otherwise small open spaces in urban areas would have undoubtedly been excluded, yielding misleading results. Both Landsat and NLCD are valuable resources, but it is important to realize that when it comes to analyzing vegetation in urban areas there are numerous small patches (community gardens, small parks, backyards) that have tremendous social and ecological value. These patches cannot be mapped from 30m pixels. Fortunately the widespread availability of high resolution remote sensing combined with advances in automated feature extraction make it possible to generate high resolution land use/cover datasets.

Join field tool in ArcGIS 9.3

2008-11-16T11:42:00.006-05:00

Remember the cumbersome process in Model Builder of passing values from one table to another - add a field, join the tables, calculate the values, remove the join? In 9.3 ArcGIS added the Join Field tool. This allows you to pass an entire attribute field from one table to another in one fell swoop. Now if only they would come up with a Rename Field tool.

IMAGINE 9.3 intial take

2008-11-16T11:07:00.004-05:00

The good folks at ERDAS have released IMAGINE 9.3 and we got it up and running in the SAL this past week. The theme of the 9.3 release is speed - application startup is much faster as is the display of vector data within the viewer. Nice work ERDAS! For more info on the 9.3 release check out Paul Beaty's blog.

UVM becomes an ESRI Development Center

2008-11-16T10:43:00.000-05:00

Thanks to the good folks at ESRI we (the University of Vermont) have been granted status as an ESRI Development Center (EDC). The benefits to our students will be tremendous and will really allow us to better equip the next generation of geospatial professionals. Interested about GIS @ UVM? Visit the UVM GIS page. Of particular intersest to undergraduates is our new minor in geospatial technology.

Welcome

2008-11-16T10:35:00.000-05:00

Welcome to my blog Letters from the SAL. The SAL, or Spatial Analysis Laboratory, has been my home since 2000 when I enrolled in graduate school at the University of Vermont. After finishing my graduate degree (it took a bit longer than expected thanks to a taxpayer funded vacation to the big sandbox) I refused to leave my desk and was eventually hired on as a researcher. I hope you enjoy my postings, which will be an unorganized collection of ramblings on geospatial technology.