Letters from the SAL

The role of morphology algorithms in land cover mapping

noreply@blogger.com (Jarlath O'Neil-Dunne) — Tue, 14 Mar 2017 18:26:00 +0000

Back in the old days, maps always had to look good. Accuracy was sometimes a bit of an afterthought. With the advent of digital remote sensing in the 1970s and 1980s, people began to question the validity of the land cover data that were being generated from automated digital remote sensing processes. Russ Congalton led the accuracy assessment charge with his seminal work on the subject (pro tip: read his book). Accuracy assessment is now standard practice in the remote sensing community. If you are doing land cover mapping and want to get it published, you have little chance of doing so without some sort of an accuracy assessment.

Along the way we sort of forgot that the output from any land cover mapping effort had to look good, as I have pointed out in a prior blog post. This was not much of an issue when most land cover data were derived from 30-meter Landsat imagery as the resolution meant that the data were inherently blocky in appearance. Focusing strictly on accuracy made sense. In the last decade, high-resolution land cover has moved into the mainstream. Whereas it is often hard to tell if a Landsat-based land cover dataset has errors because of the mixed-pixel effect, it is easy to do so with high-resolution land cover. For high-resolution land cover mapping accuracy is one measure, but I would argue that cartographic representation might even be more important if you expect people to trust your data products.

Object-based approaches to land cover mapping, in which pixels are grouped together through the application of segmentation algorithms, reduce the salt and pepper effect that often plagued some of the early pixel-based feature extraction work. Objects are not perfect, though, the output any segmentation algorithm is cartographically flawed when compared with the work of a human. Morphology algorithms play an important role in high-resolution land cover mapping, not in terms of accuracy, but in terms of improving the cartographic representation.

Let's take a look at workflow in which morphology algorithms are integrated into an automated feature extraction project. This example, which draws upon our work mapping land cover in support of San Diego's tree canopy assessment, focuses on buildings. When we look at maps, we expect the edges of buildings to have straight lines. Unfortunately, the output from the segmentation process does not result in straight lines. To improve the appearance of the buildings we applied a series of morphology steps that included smoothing, generalization, and finally orthogonalization. Does it improve accuracy? Nope. But anyone looking at the squared-off buildings would likely say the output is superior.

It is important to note that the processing trade-offs for morphological algorithms can be substantial. In our work the final step of "making this look good" often takes more computing resources than all of the segmentation and classification steps combined.

Figure 1. Color infrared aerial imagery

Figure 2. Lidar Normalized Digital Surface Model (nDSM), representing the height above ground.

Figure 3. Building objects (blue lines) overlaid on the lidar nDSM.

Figure 4. Original building objects (blue) compared to orthogonalized buildings (yellow) overlaid on the lidar nDSM.

Figure 5. Close-up view of Figure 3.

Figure 5. Orthogonalized buildings.

A better elevation model by integrating UAS and lidar

noreply@blogger.com (Jarlath O'Neil-Dunne) — Thu, 22 Dec 2016 02:05:00 +0000

A lot of times people end of comparing one data type to another as if it is some sort of contest. Recently, we have seen such comparisons being made for elevation data derived from lidar collected using traditional manned aircraft to that derived from Unmanned Aircraft Systems (UAS) imagery. I think the real benefit is not having to choose one data type of another, but rather combining data to create the best possibly product. Such a process maximizes the strengths of each type of data and minimizes the weaknesses. Airborne lidar is unmatched in its ability to provide high-resolution data over fairly broad areas along with the capacity to measure the terrain underneath tree canopy. Unfortunately, airborne lidar data are costly, processing times are slow, and new collections are rare. UAS provide a cost-effective means for acquiring accurate high-resolution elevation data at a low cost with an exceptionally quick turnaround. The photogrammetric processing techniques to extract terrain data have advanced substantially in the past two years, but there is no way to acquire elevation data for what you cannot see, and thus forest understory is virtually impossible to map.

This is a short summary of a project we worked on to create an improved Digital Elevation Model (DEM) using a combination UAS photogrammetric and airborne lidar elevation models. I define a DEM as a surface model in which above-ground features (e.g. buildings and trees are removed).

UAS imagery

The best existing elevation data for the area came from airborne lidar data that had ~1-meter resolution. The lidar data are great, but at over 6 years old the lidar DEM no longer accurately reflected the terrain due to recent construction, nor did the lidar have the detail needed for to meet the planning requirements.

Lidar DEM

We acquired imagery using our senseFly eBee RTK. The eBee RTK is a fantastic system for acquiring precision imagery without the need for ground control. Using Pix4D and some in-house routines we processed the imagery to create a bare-earth DEM. As you can see from the graphic below the DEM shows excellent detail in the open areas but falls apart in the forest due (top of graphic) to the inability to match points between the images. Further into the forest, we did not get any elevation data.

UAS DEM

Using Quick Terrain Modeler we extracted the area from the UAS DEM were we had the best detail and combined it with the lidar DEM to fill in the areas of gaps creating a composite DEM. Quick Terrain Modeler's intuitive tools made for a smooth workflow and the visualization environment is unmatched. I was amazed at how well the two datasets aligned, both horizontally and vertically.

Composite DEM created from a combination of UAS photogrammetric and lidar elevation models.

It was really encouraging to see that traditional airborne lidar and photogrammetrically derived UAS data can be integrated to create a seamless, accurate product.

Vigilant Guard 2016

noreply@blogger.com (Jarlath O'Neil-Dunne) — Sat, 30 Jul 2016 13:33:00 +0000

On July 29th our UAS Team participated in Vigilant Guard 2016, a statewide catastrophic event exercise. The UAS Team collected imagery of "damaged" bridges and roads as part of a simulated flood. A broad range of products including videos, oblique photos, orthorectified imagery, and 3D models were provided to emergency managers within hours of the team being activated. For our summer UAS intern, Matthew Mitchell, this marked the cumulation of weeks of hard work developing expertise in a wide-range of UAS technologies.

The UAS Team in action

Orthorectified UAS imagery

Bridge inspection UAS imagery

3D model derived from UAS data

Chicago Regional Land Cover Dataset

noreply@blogger.com (Jarlath O'Neil-Dunne) — Thu, 16 Jun 2016 00:18:00 +0000

We are happy to report that you can download the final version of the high-resolution land cover data set for the Chicago region. This data set supersedes the Chicago region tree canopy data set we published back in February as we made some improvements to the tree canopy class. We mapped a total of 10 counties: Cook County, DuPage County, Kane County, Kendall County, Lake County (IL), McHenry County, Will County, Lake County (IN), LaPorte County and Porter County (Figure 1).

Figure 1: Counties included in this data set.

We used a combination of imagery, LiDAR, and ancillary vector data sets (e.g. road center lines), ran it through our award-winning object-based image analysis system, then followed it up with tens of thousands of manual corrections because even our award-winning automated approaches are not perfect. We mapped seven land cover classes: 1) tree canopy, 2) grass/shrub, 3) bare soil, 4) water, 5) buildings, 6) roads/railroads, and 7) other paved surfaces. Looking for more information - we took the time to write up detailed metadata, which is contained in the XML file in the download.

Figure 2: Leaf-on imagery

Figure 3: High-resolution land cover

We are grateful to all the counties listed above who were kind enough to share their data with us and to all of our project partners: the USDA Forest Service, American Forests, the Bank of America Charitable Foundation, the Morton Arboretum, the Chicago Metropolitan Agency for Planning, and the Field Museum.

UVM Humanitarian OpenStreetMap Team, Spring 2016

noreply@blogger.com (Anonymous) — Tue, 07 Jun 2016 12:07:00 +0000

What is HOT?

In crisis situations like an earthquake, volcanic eruption, avalanche or epidemic, first responders need up-to-date digital maps to navigate the affected area. However, many remote regions around the world simply do not have the data necessary for aid groups to safely traverse city streets and locate rural villages post-disaster. In the event of a disaster, the Humanitarian OpenStreetMap Team (HOT) mobilizes volunteer mappers around the world to focus their efforts on data deficient regions.

What we do

OpenStreetMap volunteers create digital maps in a web browser using satellite imagery to identify road networks, buildings, and natural resources. Volunteers may also add names of features or directional flow of rivers and roads when knowledge or image quality allows. The end product is similar to any navigation app but includes features of high importance to responders such as hospitals, school fields that serve as helicopter landing pads, and pedestrian-only dirt paths. Responders on site in the disaster zone can download these data at any time and use it to inform their disaster response decisions in the field.

Who we are

The University of Vermont Humanitarian OpenStreetMap Team is a fluid group of volunteers made up of students, community members and seasoned GIS professionals. The team hosts mapping parties to teach newcomers how to contribute in the Aiken Center on UVM campus. In its past three events the team has mapped flooding in Sri Lanka and two areas in Ecuador affected by earthquakes and volcanic activity, involving over 70 volunteers.

Ecuador, March 26th

In August 2015, the Cotopaxi volcano in Ecuador began spewing ash and smoke that threatened nearby towns and prompted the president of Ecuador to declare a state of emergency. Due to its precarious location a mere 30 miles from the capital city of Quito, the USGS Volcano Disaster Assistance Program preemptively activated Humanitarian OpenStreetMap to help create much-needed data for the area in preparation for a major volcanic eruption. On March 26th, the UVM team held a mapathon to support this effort and it proved to be the most well-attended event yet. With 40 people attending, the team was able to create and identify over 6,000 buildings in the area. In the event of an eruption, VDAP provides support for the country’s response team through geospatial data, monitoring equipment, and guidance.

Tri Delta brought 30/40 volunteers to map for VDAP in Ecuador

Ecuador, April 20th

Less than a month later, Ecuador suffered another deadly natural disaster. On April 16th, a 7.8 magnitude earthquake shook the northern coast of Ecuador. It caused hundreds of deaths, thousands of injuries, and destroyed roads, buildings and other infrastructure. Meanwhile back on UVM campus, a group of around 28 people began to assist mapping of building footprints and road networks in an effort to create a “before” and “after” earthquake map. This assisted government relief workers identify routes that were not damaged or destroyed.

VPR coverage of the event

UVM employees using satellite imagery to digitize roads and buildings

Sri Lanka, May 28th

On May 14th, a low-pressure system in the Indian Ocean caused torrential rains across Sri Lanka. At its peak, 19 of the country’s 25 districts were affected by widespread flooding and landslides. On May 28th, the UVM HOT once again held a mapathon to create building footprints to support relief efforts. Fueled by hummus and pretzel snacks, the team contributed to a global effort that created over 125,000 buildings in the area. As of June 6th, 87 people had been killed and over 5,000 buildings had been damaged or partially damaged by the storm. Sri Lankan officials blame illegal landfills and construction in unstable areas for the damage wreaked by the storm. The Finance Minister recently announced new building code policies that would require businesses to obtain environmental approval prior to construction.

Before HOT activation in an area of Sri Lanka affected by the May floods

After HOT activation focusing on building and road creation in areas affected by the floods

Example of satellite imagery (Bing) from Sri Lanka's western coast used to digitize roads/buildings

Delaware High-Resolution Land Cover

noreply@blogger.com (Jarlath O'Neil-Dunne) — Mon, 14 Mar 2016 15:47:00 +0000

The second version of the high-resolution land cover dataset we developed for Delaware is out. This is the most detailed, highest accuracy land cover dataset ever developed for the state. It was generated by processing massive amounts of high-resolution imagery, LiDAR, and existing vector datasets.

Download a zipped version of the dataset from this link. Metadata are included. The land cover should be considered current as of 2014, which is the date of the LiDAR. It should be noted that the best available imagery came from 2013 (leaf-on) and 2012 (leaf-off) respectively.

This mapping effort is part of a larger initiative to map land cover within the Delaware River Basin and the Chesapeake Bay Water. The Delaware River Basin mapping is funded by the William Penn Foundation as is being carried out in collaboration with Shippensburg University. You can find out more about that project on the Delaware River Basin Land Use Dynamics project site. The Chesapeake Bay Watershed mapping is funded by the EPA and is being done in collaboration with the Chesapeake Conservancy. For more information on the work being done in the Chesapeake Bay Watershed visit the Chesapeake Bay Land Use Workgroup.

High-Resolution Land Cover

High-Resolution Imagery

High-Resolution LiDAR Hillshade.

Classes
0 - Background
1 - Water
2 - Emergent Wetlands
3 - Tree Canopy
4 - Scrub/Shrub
5 - Low Vegetation
6 - Barren
7 - Structures
8 - Other Paved Surfaces
9 - Roads/Railroads
10 - Tree Canopy over Structures
11 - Tree Canopy over Other Paved Surfaces
12 - Tree Canopy over Roads

UVM Humanitarian OpenStreetMap Team - A Year in Review

noreply@blogger.com (Anonymous) — Wed, 02 Mar 2016 17:14:00 +0000

Humanitarian OpenStreetMap Team (HOT) is an international community of volunteers (www.hotosm.org) that uses satellite imagery to create detailed digital maps in response to humanitarian crises. These maps can then be accessed by aid workers on the ground which increases the speed and efficiency of relief efforts. The data is also a potent tool to promote economic development and entrepreneurial innovation in developing countries.

Over the past year, we have seen some of the worst humanitarian crises in recent times, ranging from the Ebola Outbreak in West Africa to the 7.8 magnitude earthquake which devastated Nepal. In response to these crises, we often organize mapping parties or “mapathons” to train students and members of the community how to contribute to OpenStreetMap.

Since the UVM Humanitarian OpenStreetMap Team started mapping a little over a year ago we have hosted 10 mapathons working on projects in Guinea, Sierra Leone, Mali, Nepal, Mozambique, Afghanistan, Chile, Nigeria, Saint Lucia, Fiji and Mexico. Over 200 people have attended these mapathons and are now equipped to contribute to OpenStreetMap independently in response to future crises.

We’d like to thank the Rubenstein School of Environment and Natural Resources, as well as our community partners, Trader Joe’s, Myers Bagels, Sherpa Foods, Brueggers, August First, Great Harvest, Manhattan Pizza, Mr. Mikes Pizza and Shaws for making our work possible. Also, thank you Tri Delta and Delta Upsilon for adding so many crisis mappers to the mix!

Overview of last 3 mapathons in as many months

HIV Medication Distribution, Doctors Without Borders, Mozambique - December 20th

Digitized 347 roads, 173 residential areas and 107 waterways to improve understanding of where patients are located and the fastest route to get to them for Doctors Without Borders campaign to distribute HIV medication.

Before and after the recent OpenStreetMap activation in Mozambique.

Polio Vaccine Campaign, Nigeria - February 14th

Digitized 1,587 roads to help eHealth Africa to track vaccination teams, delivery of vaccines, internally displaced people (due to Boko Haram), distance matrix analysis and planning new vaccination campaigns.

Community members from Addison county, VT and New York mapping in Nigeria.

Economic Development and Disaster Preparedness, Saint Lucia / Disaster Response, Fiji - February 28th

Digitized 2,440 buildings and 246 roads in Saint Lucia to support urban planning, entrepreneurial innovation and disaster preparedness.

Digitized 761 buildings and 147 roads in Fiji in response to Cyclone Winston, the strongest tropical storm to make landfall there on record, which displaced tens of thousands of people and devastated infrastructure.

Dozens of UVM students contributing to the mapping effort in Saint Lucia and Fiji.

Lake Arrowhead Ice Jam

noreply@blogger.com (Jarlath O'Neil-Dunne) — Sat, 27 Feb 2016 17:01:00 +0000

The point at which the Lamoille River flows into Lake Arrowhead in Georgia, Vermont is known for ice jams. We spent quite a bit of time in the spring of 2015 trying to capture an ice jam, but the exceptionally cold winter followed by a dry and slow spring thaw kept the ice jams away. When we received notification late yesterday that an ice jam had formed, causing the closure of Lake Arrowhead Road we sent the UAS Team up to capture some imagery.

With UAS we can gather imagery when we want, at a very high resolution, and then orthorectify it to make it suitable for mapping. This gives us products that are unmatched for documenting emergency situations, such as the road closure, and analyzing the contributing factors, the ice in this case.

UAS imagery captured on February 26, 2016 over the point at which the Lamoille River flows into Lake Arrowhead.

Let's start with our existing state 50cm imagery of the area, acquired under leaf-off conditions. We have great imagery in VT, but the resolution simply is not good enough for this type of emergency response and analysis, never mind that it was taken at the wrong time of the year.

VT 50cm orthophoto acquired under leaf-off conditions, 1:250 scale

Moving on to the UAS imagery we can see not only key transportation features, such as the road edge and road markings, but also the ice conditions. The ice that has been pushed over the road, causing the road to be temporarily closed, can be seen in the in February 26th imagery.

UAS 4cm imagery acquired on March 19, 2015, 1:250 scale

UAS 5cm imagery acquired on February 26, 2016. 1:250 scale

Hundreds of overlapping photos were collected during the February 26th flight. We then processed those overlapping photos in Pix4D to create a 3D point cloud. Using Applied Imagery's Quick Terrain Modeler software we analyzed the height of the ice, finding that it was above the road in some areas (~17cm in the example below).

3D profile showing the ice to be higher than the road.

In addition to being useful for emergency response and transportation planning the UAS imagery gives us an unprecedented method for comparing ice conditions between years. In the image below snowmobile tracks are clearing visible on March 19, 2015 whereas the same area is open water in February 26, 2016. The large ice chunks pushed on top of each other in the 2016 imagery are the result of the recent warm temperatures and rains that caused flooding throughout much of the state.

You can download the imagery in JPEG2000 format and the 3D point cloud in LAZ format. For viewing the point cloud I recommend Quick Terrain Reader - it's free!

DISCLAIMER: The views, opinions, findings and conclusions reflected in this presentation are the responsibility of the authors only and do not represent the official policy or position of the USDOT/OST-R, or any State or other entity.

UAS Mapping of Winooski River Flooding - 25Feb2016

noreply@blogger.com (Jarlath O'Neil-Dunne) — Fri, 26 Feb 2016 01:42:00 +0000

We got hit with some freakish early season flooding in Vermont, and in the immortal words of Milli Vanilli "Gotta blame it on something. Blame it on the rain." To be fair, it wasn't just the rain, but the combination of rain, warm temperatures, and ice in the rivers.

Road closure in Middlesex

Our Unmanned Aircraft Systems (UAS) Team got the early call to assist in mapping the flooding along the Winooski River. From the perspective of UAS operations the conditions were terrible: a mix of fog along the river (cold river, warm air), rain, low clouds and winds. Nevertheless we were able to capture hundreds of pictures along the Winooski River for key railway sites in addition to the flooded roads in Middlesex. You can download a KMZ (Google Earth) file of the four site locations from this link.

Sites mapped

For each location we produced an orthorectified image along with a 3D point cloud. Note that the absolute elevations in the point cloud could be off by a few meters, so please only use them to make relative measurements. You can download the imagery, in JPEG2000 format, and the point clouds, in LAZ format, from the links below. We also have links to online map services.

Flight 307 - Richmond

Imagery Point Cloud Web Map

Flight 308 - Richmond

Imagery Point Cloud Web Map

Flight 309 - Jonesville

Imagery Point Cloud Web Map

Flight 310 - Middlesex

Imagery Point Cloud Web Map

A short video showcasing the Middlesex mapping.

DISCLAIMER: The views, opinions, findings and conclusions reflected in this presentation are the responsibility of the authors only and do not represent the official policy or position of the USDOT/OST-R, or any State or other entity.

Chicago Regional Tree Canopy

noreply@blogger.com (Jarlath O'Neil-Dunne) — Fri, 05 Feb 2016 14:25:00 +0000

This data set has been superseded by the Chicago region land cover data set.

We are excited to announce the release of the Chicago Regional tree canopy dataset (scroll down for download links). The dataset was developed by the University of Vermont Spatial Analysis Laboratory with support from the USDA Forest Service, Morton Arboretum, American Forests and the Bank of America Charitable Foundation. Collaborators includes the Chicago Metropolitan Agency for Planning, the Field Museum, Cook County, DuPage County, Kane County, Kendall County, Lake County (IL), McHenry County, Will County, Lake County (IN), LaPorte County and Porter County.

Counties included in the dataset

This is the most detailed, accurate, and highest resolution tree canopy dataset ever developed for the region, and one of the largest tree canopy datasets in the United States. We developed a rule-based expert system using the eCognition platform to automatically extract tree canopy, then followed it up with a thorough manual review. The key input datasets were leaf-on imagery from the National Agricultural Imagery Program (NAIP) and LiDAR obtained from a variety of state and local sources. Without the imagery and LiDAR the project would simply not have been feasible.

Leaf-on NAIP imagery

Leaf-off LiDAR

Tree Canopy

We have three versions available for download:
1) Chicago regional tree canopy in the Albers Equal Area coordinate system
2) IL portion only in IL state plane
3) IN portion only in IN state plane.

Note that the full regional dataset contains nearly 30 billion pixels and is 3.46GB when uncompressed. All the datasets are in IMAGINE raster format (.img), which is supported by a broad number of open source and proprietary geospatial software packages. Full metadata is present for each dataset. Note that the default zip utility used by some versions of Windows has a bug that gives an error message that saying that you need petabytes of free space to unzip the file. If this happens simply use another utility such as 7-zip or Winzip.

Humanitarian OpenStreetMap Team - Nepal, Mexico and Afghanistan

noreply@blogger.com (Anonymous) — Wed, 02 Dec 2015 16:19:00 +0000

Nepal/Mexico Mapathon

The last week in October was exceptionally busy for members of the Humanitarian OpenStreetMap Team at UVM. On October 25th the Spatial Analysis Lab hosted a mapathon to create detailed digital maps in Nepal in response to a massive landslide that blocked the Kali Gandaki River. Coincidentally, hurricane Patricia had just hit Mexico’s western coast and we were able to delegate a team of volunteers from Tri Delta to work in Puerto Vallarta. Thankfully, Mexico was well prepared and there was a minimal loss of life, although there was severe damage to infrastructure as a result of flooding and winds. Despite breaking records for highest sustained wind speeds globally and barometric pressure in the Western Hemisphere, Patricia quickly dissipated once it made landfall after hitting a steep gradient of elevation into a mountainous region.

Students and community members using OpenStreetMap

Sherpa Foods generously donated 80 momos and Trader Joe’s came through with snacks and drinks to keep our mappers energized. The combination of delicious food and listening to Alabama Shakes (a mapathon crowd favorite) provided for a good Sunday’s worth of mapping. Together we were able to digitize 5,329 buildings and 549 roads in Nepal and Mexico with a record number of 37 volunteers!

Afghanistan Mapathon

On Monday October 26th, the day after our mapathon, a 7.5 magnitude earthquake hit remote areas of Eastern Afghanistan and Pakistan damaging and destroying thousands of homes. OpenStreetMap was activated on Wednesday the 28th to provide road networks and residential area polygons to humanitarian organizations and on Friday the 30th the Spatial Analysis Lab hosted another mapathon with support from Trader Joe’s to work in Afghanistan. The dataset was used to identify remote villages and determine the fastest route to distribute food and other essential supplies to those affected. This task was especially urgent as snowy and rainy conditions left people stranded and desperate for assistance.

Below are examples of OpenStreetMap data before/after recent activation and imagery of Nepal, Mexico and Afghanistan:

Kathmandu, Nepal before OpenStreetMap’s recent activation

Kathmandu, Nepal after OpenStreetMap’s recent activation

Imagery of Kathmandu, Nepal provided by Bing

Puerto Vallarta, Mexico before OpenStreetMap’s recent activation

Puerto Vallarta, Mexicol after OpenStreetMap’s recent activation

Imagery of Puerto Vallarta, Mexico provided by Mapbox

Takhar-Badakhshan Province, Afghanistan before OpenStreetMap’s recent activation

Takhar-Badakhshan Province, Afghanistan after OpenStreetMap’s recent activation

Imagery of Takhar-Badakhshan Province, Afghanistan provided by Bing

Learn More: http://learnosm.org/en/

Task Manager: http://tasks.hotosm.org/

IRC chat: http://irc.openstreetmap.org/ (for any quick questions)

How did you contribute?: http://hdyc.neis-one.org/

Feel free to shoot me an email if you want to get involved or have any questions at noah.ahles@gmail.com

Tree Canopy Assessment Tools

noreply@blogger.com (Jarlath O'Neil-Dunne) — Thu, 19 Nov 2015 00:15:00 +0000

We have published the latest version of Tree Canopy Assessment Tool. This is a simple ArcGIS model that allows you to compute tree canopy and land cover metrics for any polygon feature class according to the US Forest Service's Tree Canopy Assessment protocols.

The model requires two input layers, a raster land cover and a polygon geographies layer to summarized the land cover by. The land cover raster must contain seven classes: 1) tree canopy, 2) grass/shrub, 3) bare soil, 4) water, 5) buildings, 6) roads/railroads, and 7) other paved surfaces. Outputs consist of tree canopy metrics in one table and land cover metrics in another table.

OpenStreetMap Post Earthquake Mapping, Coquimbo, Chile

noreply@blogger.com (Anonymous) — Wed, 14 Oct 2015 18:50:00 +0000

On September 27th, the Spatial Analysis Lab held its first Mapathon of the fall 2015 semester. The focus of this Mapathon was to create a detailed digital dataset of roads, buildings and open areas, in Coquimbo, Chile in response to the 8.3 magnitude earthquake that hit off of Chile’s coast on September 16th. Coquimbo experienced widespread damage to infrastructure and flooding due to the earthquake and subsequent Tsunami that swept through parts of the city. In fact, in parts of Chile the ground shifted 4.6 feet (1.4 meters) as a result of the earthquake.

Students and community members using OpenStreetMap

Coquimbo already had some data before this recent OpenStreetMap activation in its road network. However, most of the roads were unclassified and to plan an effective humanitarian response the road type has to be determined. For the most part, we were creating building polygons, classifying roads and connecting roads with smaller paths to increase the efficacy of routing systems. We were also creating polygons delineating open areas which can then be used for potential helicopter landing pads and/or temporary camps for internally displaced people.

Coquimbo, Chile before and after OpenStreetMap’s recent activation

During this Mapathon, we were fortunate enough to have Steve Bower present on the history of OpenStreetMap and how data is being created and utilized for humanitarian, commercial and public purposes on a global scale. Steve started contributing to OpenStreetMap following the earthquake in Nepal and has quickly integrated himself into the Humanitarian OpenStreetMap Training Team over the past several months.

An infographic created by Steve Bower to show the complex nature of the OpenStreetMap system

Learn More: http://learnosm.org/en/

Task Manager: http://tasks.hotosm.org/

IRC chat: http://irc.openstreetmap.org/ (for any quick questions)

How did you contribute?: http://hdyc.neis-one.org/

Results/Stats Breakdown: http://resultmaps.neis-one.org/

Map your edits: http://overpass-turbo.eu/ - Hover over the area you were working, press the Wizard button and enter into the query box user:username e.g. user:nahles

UAS team deployed to aid in train derailment

noreply@blogger.com (Jarlath O'Neil-Dunne) — Mon, 05 Oct 2015 23:40:00 +0000

This morning the Amtrak Vermonter train derailed in Northfield, VT. Responding to a request from the Vermont Agency of Transportation, we immediately put together a volunteer Unmanned Aircraft Systems (UAS) team and deployed to the area. We promptly set up a launch site, coordinated the acquisition area with disaster management officials, and then conducted UAS flight operations.

Train derailment site.

Within 60 minutes of us arriving on site we had transferred a KML file to the incident commanders that contained over 280 photos of the derailment site and the surrounding area, each tied to a location on the ground.

Train derailment imagery displayed in Google Earth. Each one of the white push-pins shows the location of a photo taken by the UAS. Clicking on one of the push-pins opens up the corresponding photo. Follow the link above to download the KML file.

The UAS we use, the senseFly eBee, is specifically designed to acquire mapping-grade data. Each photo is geotagged during the flight, enabling us to generate a KML file containing all the images upon landing. This allows us to provide the response teams with imagery that is comprehensive, tied to a location on the ground, and easy to display using the freely available Google Earth. Upon returning to our lab we processed the imagery to create a seamless orthophotomosaic (JPEG 2000 format), which can be loaded into GIS software for mapping, measuring, and integration with other mapped data. You can view the image mosaic online using this link. Although other assets were able to acquire overhead imagery and video, our UAS team provided a unique capability in that we were the only ones able to produce GIS-ready products suitable for mapping.

Image mosaic of the acquisition area displayed using GIS software.

Thanks to funding from the US Department of Transportation, the Vermont Agency of Transportation, the USDA Forest Service, and AmericaView we have been able to transform what began as a research project into an operational asset, capable of rapidly responding to incidents across the state. This marks the third time our UAS team has deployed for disaster response and came right after we had participated in a joint federal disaster training exercise. In the past year we put a tremendous amount of effort into building the operational capacity. This has included everything from developing detailed flight checklists to the extensive training of our personnel. The effort clearly paid off.

The UAS team after delivering the imagery to the incident commanders.

Timeline of the events:

11:50 - UAS assistance request received

12:15 - airspace coordination completed

12:25 - mission preparation completed

12:35 - UAS team departs

13:30 - UAS team arrives in Northfield

13:40 - launch site established

13:50 - flight operations commence

14:15 - flight operations complete

14:20 - KML file produced

14:35 - KML file delivered to incident commanders
17:00 - UAS team returns to lab
19:35 - orthophoto mosaic produced

Prior to releasing UAS imagery we carefully evaluate it for privacy concerns. The images we collected for this response efforts were done after all injured personnel were removed from the site and the angle of the imagery makes it impossible to identify individual persons. Furthermore the imagery is comparable to that acquired by the state and media outlets that have been publicly released.

We are grateful to both our funding sources and our commercial partners, Spatial Informatics Group, Applied Imagery, senseFly, and Spatial Networks.

DISCLAIMER: The views, opinions, findings and conclusions reflected in this presentation are the responsibility of the authors only and do not represent the official policy or position of the USDOT/OST-R, or any State or other entity.

Pennsylvania Statewide High-Resolution Tree Canopy

noreply@blogger.com (Jarlath O'Neil-Dunne) — Sat, 19 Sep 2015 01:36:00 +0000

We are thrilled to announce the first public release of the Pennsylvania statewide high-resolution tree canopy dataset (data download link - 2.3GB!). (Technically this is the second version as the first one was for internal review). This dataset maps tree canopy for the entirety of Pennsylvania at a resolution of 1m, making it 900 times more detailed than the National Land Cover Dataset (NLCD)! With our landscapes becoming increasingly fragmented and heterogeneous high-resolution datasets add precision and accuracy to any analysis.

We primarily relied upon Pennslyvania's publicly available LiDAR data, which was acquired in 2006, 2007, and 2008. We supplemented the LiDAR data with leaf-on imagery acquired in 2010 through the National Agricultural Imagery Program (NAIP).

Needless to say, the project was a massive undertaking, with the raster dataset containing close to 200 billion pixels. To extract tree canopy from the LiDAR and imagery we used Object-Based Image Analysis (OBIA) techniques. A rule-based expert system was developed that made use of segmentation, classification, and morphology algorithms to extract tree canopy features based on their height, spectral, and spatial properties. This process allowed us to resolve some rather tricky issues with the data, the chief of which was the spotty LiDAR returns in the deciduous treed areas due to the leaf-off nature of the LiDAR (see graphic below). The data then underwent a manual review. If you are interested in more about our methodology we recommend this paper from the 2014 ASPRS conference.

Sample from the Pennslyvania statewide tree canopy dataset. On the left is a LiDAR Normalized Digital Surface Model (nDSM), which represents the height above ground. In the center is the NAIP color infrared imagery, and on the right, the mapped tree canopy. The large forest patch in the example, despite showing up as closed canopy in the imagery, appears spotty in the LiDAR nDSM due to the leaf-off acquisition. By using the imagery in conjunction with the LiDAR we were able to avoid under representing the tree canopy.

There is no charge for this dataset, but we would be grateful if you would give us credit in any publications, reports, derivative datasets, etc. We will be making additional revisions to the data to improve its quality, and thus any feedback is certainly welcome. Please note that this dataset is independent from those areas we have mapped for Urban Tree Canopy (UTC) assessments. Future releases will integrate these two projects.

If you are a state with LiDAR and are interested in a similar product please let us know. Our techniques are efficient, accurate and effective, enabling high-resolution tree canopy to be mapped statewide at a very reasonable cost. We will be releasing statewide tree canopy for Delaware and Maryland in the coming months. In addition we are working with multiple partners to develop comprehensive high-resolution land cover for the Delaware River Basin and the Chesapeake Bay Watershed.

Funding for this project came from the NASA Carbon Monitoring System program and it was developed in collaboration with our outstanding colleagues from the University of Maryland. The techniques used stem from collaborative research carried out with the USDA Forest Service.

Montgomery County MD Tree Canopy Explorer

noreply@blogger.com (Jarlath O'Neil-Dunne) — Thu, 10 Sep 2015 20:50:00 +0000

In late spring we completed an updated mapping of Montgomery County, Maryland's land cover. The driving factor for this analysis was to understand how Montgomery County's tree canopy is changing over time. Thanks to the County's investments in high-resolution imagery and LiDAR we were able to map change with such detail that even newly planted trees and portions of tree canopy that were removed were detected. With this information the County now has a handle on tree canopy change, and can summarize the data at a range of scales from the property parcel to the watershed.

The good folks at Montgomery County developed a web-based application that allows anyone to compute the amount of area in a user-defined polygon, giving the general public access to the data we developed.

UAS Mapping of July 2015 Storm Damage in Barre, VT

noreply@blogger.com (Jarlath O'Neil-Dunne) — Fri, 24 Jul 2015 23:39:00 +0000

UAS collection areas

The severe storm that dumped close to half a foot of rain on Central Vermont in a matter of hours on July 19, 2015 caused Gunners Brook to overflow its banks, severely impacting the City of Barre. Homes were ruined, roads next to Gunners Brook collapsed due to severe bank erosion, and sediment covered much of the downtown area.

The University of Vermont's Unmanned Aircraft Systems (UAS) team responded to the disaster, acquiring detailed imagery for two key areas, providing the most comprehensive inventory of the event. This video shows how UAS imagery was used to assess the damage.

You can download the both south section image mosaic and the north section image mosaic in JPEG2000 format. JPEG2000 files can be viewed using most CAD/GIS software. Looking for a free tool? Try the ERDAS ER Viewer.

Great Brook July 2015 Storm Damage - UAS Mapping

noreply@blogger.com (Jarlath O'Neil-Dunne) — Wed, 22 Jul 2015 23:58:00 +0000

On July 19th a series of storms moved through Central Vermont unleashing nearly 6” of rain. The flooding resulted in Great Brook, which flows into Plainfield, VT spilling over its banks, damaging and destroying homes and roads. The State of Vermont responded with rescue teams and the towns immediately set out to protect their communities. As the sun rose the next day it became clear that this area, so used to these disasters, had yet again found itself knocked down and starting the process of recovery all over again. The University of Vermont immediately activated the Unmanned Aircraft Systems (UAS) team. After coordinating with local officials and the Vermont Agency of Transportation, the UAS team set out to Plainfield to map the extent of the damage.

Fortunately, the UAS team had been gathering data for the Great Brook/Plainfield area since December of 2014 as part of a project to map woody debris, which has long been a source of concern due to logs clogging culverts and bridges resulting in flooded roads. The extensive database of imagery collected prior to this recent flood event enabled rapid change detection mapping. Within 24 hours we completed flight operations and conducted a preliminary assessment. This video highlights some of the dramatic changes that have taken place and the utility of using UAS data for disaster response, assessment, and long-term monitoring.

The imagery can be viewed in this web map viewer.

We have posted the imagery (in JPEG 2000 format) and the 3D point cloud (in LAZ format) for download.

UAS Flight Operations Highlight Reel

noreply@blogger.com (Jarlath O'Neil-Dunne) — Thu, 16 Jul 2015 22:09:00 +0000

Our video production guru Gavin Zeitz put together this highlight reel of our UAS flight operations over the last several months.

An object-based approach to LiDAR QA/QC

noreply@blogger.com (Jarlath O'Neil-Dunne) — Tue, 26 May 2015 19:11:00 +0000

LiDAR data is a bit complicated to QA/QC. First of all the datasets are huge, often comprising billions of points. Then there is the issue of confirming that everything from the number of returns to the classification of the points meets the specifications. Birds, haze, classification errors - the list of things that can influence the quality of a LiDAR dataset is long.

The most common method of conducting a quick review of LiDAR data is to simply summarize the information in a grid cell. Figure 1 shows and example of this. The number of returns is computed for each 20m grid cell. The problem is that grid cells don't confirm to the features the LiDAR data are interacting with. For comparison, Figure 2 shows color infrared aerial imagery for the same area. The low point density over the water is to be expected given that water typically absorbs the LiDAR signal, but the low point density over other areas may be of concern. Unfortunately the grid cells don't conform to the boundary of the lake making the identification of such issues less than intuitive.

Figure 1. Point density summarized using a 20m grid.

Figure 2. Color infrared aerial imagery.

To overcome the limitation of grid-based summaries we developed an object-based approach to LiDAR QA/QC using eCognition. The gist of the approach is that we automatically generate polygons through the application of segmentation routines from a combination of the LiDAR intensity data and the best available imagery. We then summarize key LiDAR and imagery metrics for each polygon. As the polygons generally conform to land cover features, metrics, such as point density, are more meaningful. Figure 3 shows each polygon symbolized by the number of returns per unit area (square meters in this case). We can also bring in attributes of the imagery to identify areas that are not of concern. Figure 4 shows the mean near-infrared (NIR) values from the imagery, which help to highlight the lake.

Figure 3. Number of returns per unit area.

Figure 4. Mean near infrared value derived from imagery.

The other advantage of having a polygon layer is that we can run some simple queries. Figure 5 shows an example of a query used to highlight polygons that have fewer than 5 points per unit area but an NDVI of greater than 0.2. With higher NDVI values indicative of vegetation, the fact that we have a lower than average number of returns may be cause for concern.

Figure 5. Polygons with a low number of returns yet relatively high NDVI.

We could get fancy and actually relate the LiDAR metrics to land cover, but this would require more effort. The routine, as it stands now, ran over a 1.9 billion point dataset in 35 minutes.

Flooding in Cambridge, VT

noreply@blogger.com (Jarlath O'Neil-Dunne) — Thu, 16 Apr 2015 13:30:00 +0000

As noted in a previous post we have been monitoring ice and water in the Lamoille River where is passes through Cambridge, VT since mid-March as part of a US Department of Transportation funded project. The Lamoille causes flooding in Cambridge nearly every year, often impacting travel on Route 15, a crucial link in the transportation network in this part of Vermont. Yesterday our team captured imagery of the flooding using our senseFly eBee. We used a color infrared (CIR) camera because water absorbs the majority of near infrared light, making it ideal for mapping the extent of the flooding. At this time the flooding has only impacted agricultural fields, but the water is right up to edge of Route 15. The imagery produced from the eBee is orthorectified, allowing it to be readily integrated into GIS software, where measurements can be made to compute the flooded area. Acquisition time for the imagery (largely area than shown below) was 75 minutes. It took 3.5 hours to generate the digital orthophotos.

Color infrared imagery collected on April 15, 2015 showing evidence of flooding.

Color infrared imagery collected on April 9, 2015 prior to any flooding.

In March we collected data at a slightly higher resolution and with additional overlap, allowing us to produce a 3D point cloud that we then turned into a detailed terrain model. We ran some quick flood simulations using Quick Terrain Modeler and found we were able to simulate the April 15th conditions quite well. By our calculations a rise of about 1 foot will push the water onto Route 15. With rain forecast in the coming days that could very well happen.

3D point cloud from the March 20, 2015 flight.

3D terrain model generated from the point cloud.

Simulated flood conditions.

Mapping spring changes to the Lamoille River in Cambridge, VT

noreply@blogger.com (Jarlath O'Neil-Dunne) — Fri, 10 Apr 2015 18:16:00 +0000

This is a follow-up on two previous posts (Montpelier and Arrowhead Mountain Lake) containing more examples of how we are using our eBee to map spring ice and flood conditions on Vermont rivers that have historically caused disruptions to transportation networks through a US Department of Transportation grant. The images below are 4 centimeter resolution color-infrared data acquired at three time periods this spring where the Lamoille River flows through Cambridge, VT. This area has been a hotbed of flooding in previous years. It marks the first time remotely sensed data has been used to monitor the river and evaluate how these changes might impact the transportation network.

March 20th, 2015

April 2, 2015

April 9, 2015

Mapping Winooski River ice conditions in Montpelier

noreply@blogger.com (Jarlath O'Neil-Dunne) — Wed, 08 Apr 2015 12:17:00 +0000

This is a follow-up to our previous post on using high-spatial, high-temporal resolution imagery from our senseFly eBee to map changing ice conditions on Vermont Rivers. The Winooski River in Montpelier has a history of ice jams and flooding. As part of our Department of Transportation grant we have been gathering color-infrared imagery of the Winooski River to evaluate the use of such imagery to map changes in ice conditions.

Below are examples of that imagery for the Winooski River in Montpelier from March 30th and April 6th of this year, in addition to the best available existing aerial imagery, which was taken a few years back. With the eBee we are able to get imagery of the area of concern when we needed it. The color-infrared qualities make it possible to make out relative ice thickness along with the presence of open water. You will notice some white dots in the April 6th imagery - it was snowing at the time of the flight!

We were also fortunate to have computer science students from Montpelier High School join us for the flights and participate in launch operations.

Existing true color aerial imagery of the Winooski River.

eBee color-infrared aerial imagery acquired on March 30, 2015. The presence of ice, ice thickness, and open water can be easily distinguished.

eBee color-infrared aerial imagery acquired on April 6, 2015. The river is now free of ice. The blurry white dots in the image are snowflakes (it was snowing during the flight).

Threshold Classification in eCognition

noreply@blogger.com (Jarlath O'Neil-Dunne) — Mon, 06 Apr 2015 04:40:00 +0000

This video demonstrates effective techniques to simple threshold-based classifications within eCognition using the new interface that was introduced in version 9.0.2.

Monitoring ice conditions with high-spatial and temporal resolution data

noreply@blogger.com (Jarlath O'Neil-Dunne) — Sun, 05 Apr 2015 00:32:00 +0000

Technically it is spring in Vermont, but with the exception of yesterday it feels more like mid-December. I woke this morning to an inch of snow covering the ground. Nevertheless, the birds are chirping and the snow is melting, which means that the ice jams are coming. Ice jams, can cause flooding both due to the backup of water or the release of water when they finally breakup. The flooding is associated with damage and disruptions to our transportation networks. Remote sensing technology should be ideally for tracking and mapping ice jams, but it is not widely used for a number of reasons. Freely satellite imagery (e.g. Landsat) lacks both the temporal and spatial resolution. High-resolution commercial satellite imagery tends to be a bit on the pricy side and is often confounded by cloud cover. Manned aerial systems do offer a solution, but it is difficult to justify the cost for small area mapping. The result is that we are left with an incomplete picture of current ice conditions in our streams and rivers.

As part of a project funded by the US Department of Transportation we are looking into remote sensing technologies that can inform managers on threats to the transportation network, such as ice jams. This spring we have been collecting high-resolution color infrared aerial imagery using our senseFly eBee to better document ice conditions. It typically takes us 30-90 minutes to acquire the imagery, then another 4-8 hours to process it into high-resolution orthorectified GIS-ready imagery. Below are some examples from Northwestern Vermont where the Lamoille River enters Arrowhead Mountain Lake. It is an area that has experienced a high number of ice jams in the past. The changes in the ice conditions are clearly evident, with areas of open water and thinner ice appearing in the April 2nd imagery. We are continuing to gather data for this location and others. Be on the lookout for updates on future blog posts.

March 19, 2015

April 2, 2015