A Classified Success

A 2.491Mb PDF of this article as it appeared in the magazine—complete with images—is available by clicking HERE

Energized by the launch of the International Polar Year in 2007, two companies came together that same year to initiate their own commitment to the planet: Planet Action (PA). Co-founded by satellite data provider Spot Image (now Astrium GEO-Information Services) and GIS provider Esri, Planet Action provides geospatial technology and expertise to organizations working on climate change causes, impacts and solutions.

To understand the significantly challenging mission of Planet Action’s projects, consider the small sampling of some of the land issues before them.

• Mexico loses on average 350,000 hectares of forested land every year, an area about the size of Rhode Island.

• Colombia loses nearly 200,000 ha of forests every year–though the true figure may be higher since an estimated 100,000 ha of native forest are illegally cleared annually.

• France has a mountainous region in the Pyrenees with an uncertain environmental future. The abrupt abandonment of this formerly human-dominated land in the Pyrenees in the 20th century, coupled with natural afforestation, has produced an organically grown land-use conundrum–how to understand the impacts of this severe land-cover change and create a balanced ecosystem that supports both biodiversity and the return of human interest in the land.

These three examples are only a razor-thin slice of the environmental issues layering the globe. However, they do represent three successful beginnings on that elusive road to environmental sustainability, the ultimate goal for the hundreds of projects supported by Planet Action. And on a personal note for five students at the University of Vermont (UVM), these three projects seeded a unique work-study class and helped to launch five budding careers in the geospatial industry.

A Work-Study Debut
Since its launch in 2007, Planet Action has supported more than 600 nonprofit projects worldwide in various domains such as biodiversity, forestry and water resources. Through the donations of satellite images, GIS solutions and image processing systems by program partners, Planet Action supports both small, local agencies such as NGOs, universities and research centers as well as large agencies such as UNESCO, WWF, the Green Belt Movement and IMAZON. Planet Action donated more than 1,000 satellite images alone in 2011 and program partners such as Trimble have donated their eCognition® image analysis software licenses to aid users in their image processing tasks–to date, more than 100 eCognition Developer licenses have been donated to more than 50 projects in 43 countries.

Though Planet Action started small with a roster of 13 projects, by 2010 it had increased to 500, and has added more than 150 since. Three of the 2010 projects were those from Mexico, Colombia and France, all of which similarly required satellite imagery to classify and map the land cover, but for notably different environment types–mangroves, dense forests and an alpine ecosystem. As each of these environments present unique classification challenges, the project managers required an image-analysis system that could delve beyond traditional pixel colors and recognize contextual elements to clearly distinguish different vegetative types. To achieve that, each respective project manager wanted to produce their data layers with Trimble’s eCognition, an object-based image analysis (OBIA) software, but they didn’t have the in-house resources to take on the task.

Enter the UVM’s Spatial Analysis Laboratory (SAL). SAL’s personnel use eCognition extensively for their varied work that includes biodiversity analysis, land-cover mapping and urban tree-canopy assessments. Their work has been so successful, SAL’s Director Jarlath O’Neil-Dunne–who has been an eCognition user since 2002–was able to establish the university’s eCognition Center of Excellence in 2009 (now Trimble Imaging Innovation Program), which has helped to further augment SAL’s prestige and revenue.

With such a prominent profile, it did not take long for Trimble to connect O’Neil-Dunne with these Planet Action projects in need of image-analysis help–a connection that came at about the same time a group of five students involved with the lab were keenly interested in becoming more proficient in OBIA.

"This was an ideal need-need pairing," says O’Neil-Dunne. "Not only were the projects particularly well-suited for object-based image analysis, they would enable these students and the project managers to learn the capabilities of the software through hands-on, real-world projects with deadlines, specific deliverables and end users greatly invested in the outcomes."

And with that, a work-study class was born.

"A" for Effort
Starting in January 2011, O’NeilDunne assigned the class of five their respective projects–two students for Mexico, two for Colombia and one for France. After a two-week tutorial on the eCognition software, the students had four months to acquire the available geospatial imagery, process it and deliver the required maps to the end users. Although O’Neil-Dunne recognized this class would be a unique opportunity for the students, he also knew it would come with risks.

"Being able to apply their knowledge and skills to real-world projects would be an incredibly effective way to learn, but it would come against the backdrop of a tight deadline, and the expectations of the project managers who they’ve never met and for whom English is not their native tongue and who are in three different time zones," he says. "It was a risky proposition, but it all worked out very well."

And to help ease communication and support technology knowledge transfer between the Planet Action project managers and the UVM students, Trimble offered the use of its dedicated online platform called eCognition Community where they could discuss issues, exchange ideas and blog results.

For all three projects, the students were provided Spot images, and any other available supplemental data such as digital elevation models (DEMs). Using eCognition, they incorporated the imagery into the software and created customized "rule sets," a workflow of if-then scenarios to automatically classify specific vegetative types and environs, which were then mapped according to user specifications.

For Mexico, they produced digital thematic maps of the Sierra de la Cruces region, a 200,000-ha forested area west of Mexico City; the maps definitively show urban areas, water resources, grasslands and specific tree types including pine, fir and oak. The land-cover classification is serving as the basis for further ecological field data being collected and the ongoing analysis of landscape patterns to help develop strategies to curb rampant deforestation in the area.

In Columbia, the objective was to map the present extent of the mangrove forest, as well as other wetlands and flooded forest wetlands within the Darien region, which lies near the border with Panama and has been heavily impacted by logging, mining, and clearing for agriculture and cattle ranching. Based on the OBIA processing, the students were able to deliver detailed maps of the mangrove forest and associated wetlands; these maps are aiding the project team in their goal to develop community-driven proposals to halt the deforestation and degradation of the forest.

In addition to the detailed classification maps, O’Neil-Dunne also required the students to provide the processing knowledge used to generate the data products–the cu
stomized eCognition rule sets.

"The ability to clearly distinguish a wetland, for example, which is quite hard for traditional pixel-based image processing solutions to adequately do, and to automatically map it is a powerful tool for these projects," says O’Neil-Dunne. "Even more valuable is to have the `process tree,’ the step-by-step procedure of how the software was instructed to identify vegetative types like mangroves, for example. That provides the project managers with a master knowledge base for today and for the future."

A Right-Brain Approach
For the French project, that "future" has already arrived. Indeed, early classification work on the Pyrenees project not only showed Thomas Houet, the project manager and associate researcher at the French National Research Center (CNRS), that they would be able to meet their mapping needs, it triggered a different exploratory path for the project and led to a successful grant to continue the project.

When UVM student Daniel Koopman was initially assigned the Pyrenees project, his primary objective was to produce a land-cover classification map of a 245-square-kilometer area in the Vicdessos valley in the Ariège region of southwestern France.

Designated a long-term "Human-Nature Observation" site and supported by the Institute of Ecology and Environment of the CNRS (InEE– CNRS), this particular area of interest (AOI) in the Midi-Pyrenees region has suffered significant land-use changes over centuries of upheavals. For 6,000 years the region was dominated by the agropastoral industry–both growing crops and raising livestock–coupled with decades of industrial activity; then, in the late 19th, early 20th century, rapid depopulation–largely due to two World Wars and the economic and social attractiveness of urban areas–precipitated the end of transhumance (leading livestock to higher regions for summer months), severely shrinking the size of cultivatable land. This chain of events has allowed the region to organically transition back to a natural ecosystem, leaving the region with an uncertain role for modern-day land-use possibilities. And that uncertainty has been of particular concern for the InEE-CNRS, which has been striving to understand these landcover changes and their impacts over the last 50 years in order to help authorities better shape environmental planning and development for the region.

Although historical changes in the region have been well-studied,¹ Houet wanted to capitalize on technological advancements to more efficiently and effectively map recent changes over the last few years. But first, he needed to have Koopman–who is experienced in classifying alpine ecosystems–create an accurate, present-day, land-classification map for the AOI, which is comprised of a lower valley, an intermediary zone favored by agricultural activity, and a high-mountain zone dominated by dramatic changes in vegetation.

"Producing land-cover maps for mountainous areas such as the Midi-Pyrenees has been problematic for traditional land-classification technologies because slopes, sun and satellite angles strongly affect spectral measurements," says Houet. "The advanced OBIA technology now enables us to overcome these unique challenges and map these regions with accuracy we couldn’t achieve before."

Koopman was given two 5-meter-resolution Spot scenes acquired in 2010 and a 30-m-resolution DEM produced from ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) data. He also had ancillary datasets including a vector layer, buildings layer and water layer, to help with the classification–all of which, coupled with the satellite data, were integrated into eCognition to build the customized rule set. Using his alpine-studies experience, Koopman first defined the landscape characteristics of each zone–typical tree cover, slope angles, proximities to urban areas–and then began writing rules to train the software to mimic the process the human brain uses to distinguish specific environmental features.

"Teaching the software to perform a sophisticated process that the brain does superiorly, in microseconds, is one of the main challenges of OBIA, but it is also what makes it special," says Koopman. "The `left brain’ of the software is typical of traditional image-processing tools in that it matches the spectral properties of a pixel with its typical land class, but it also has a `right brain,’ which allows you go beyond spectral properties and more into contextual and relational areas. For example, you can not only classify a tree, you can specify the exact type of tree based on its geographic position– certain tree types only grow at certain elevations. The ability to consider an object’s surroundings gives you the power to create much more accurate classifications–datasets that I can create in one single software package, using a replicable methodology, in minutes."

In four months, Koopman had developed the needed process tree to distinguish 11 different class types, including valley, intermediate and mixed alpine grasslands, snow, coniferous forest, deciduous trees and wetlands, and map them. Although the rule set took months to build, it only took 10 minutes to run the workflow and create the classification map. Koopman delivered the map and associated eCognition rule set–the classification gold, says Koopman–to Houet in May 2011.

Based on the map detail and his own eCognition experience gleaned from collaborating on the project, Houet secured funding from the CNRS to continue to collaborate with Koopman on the project, but with a slightly more challenging objective. In September 2012, Koopman travelled to France to help conduct fieldwork to determine if they can apply eCognition to the existing land-cover map in order to deduce land-use change.

"Land-use change adds more complexity because it requires you to make sophisticated assumptions and add human components to the analysis," says Koopman. "The goal is to see if we can develop a flexible enough rule set to create land-use maps independent of time, so a rule set that can be applied both to historical remote sensing data and future datasets acquired."

A Work-Study Sequel?
Though a risky prospect at the outset, O’Neil-Dunne’s debut Planet Action work-study class can be classified as a success. Not only did his five students meet each project’s requirements in the span of one semester, he also took pride in knowing that their eCognition class experience helped them secure full-time jobs in the geospatial industry.

Koopman was hired by Geostellar, a solar energy company, in January 2012, and uses eCognition software daily in his work. And according to O’Neil-Dunne, the other former students are all working in the GIS arena. "I think these projects were big selling points on their resumes," says O’Neil-Dunne.

With the students’ careers well underway, O’Neil-Dunne hopes to play matchmaker again with the right Planet Action projects and the right group of students. With Planet Action’s unabated mission, chances are high that he will succeed.

Mary Jo Wagner is a Vancouver-based freelance writer with 20 years experience covering geospatial technology.

¹ Houet T., Ribière O., Vacquié L., Vidal F. et Galop D. (2012) Analyse spatiale de l’évolution des modes d’occupation et d’usages des sols sur le Vicdessos de 1942 à nos jours, Sud-Ouest Européen. n°33, pp. 41­56

A 2.491Mb PDF of this article as it appeared in the magazine—complete with images—is available by clicking HERE