The state of Maine is a picture of natural beauty, from its rocky shorelines to its vast northern forests. Maintaining the state’s natural resources is a cooperative balancing act between private land–owners, conservation organizations and government. One of the key tools used to manage natural resources and planning for Maine’s future is remotely sensed landcover data, which have been widely used for modeling urban and forest growth, estimating imperviousness, determining cumulative impacts of landcover change, and predicting wildlife habitats (Figure 1).

Meeting New Data Requirements
In early 2004, geographic information system (GIS) staff from several Maine state agencies identified the need for new state landcover data. Users knew the landcover data were crucial to their work and were concerned about the age of available data, all of which were based on 1992 imagery at 30-meter-pixel resolution—far too coarse for many applications.

The staff began putting requirements into a request for proposal (RFP), and several state agencies contributed funds to reach the necessary goal of $300,000. The group also began collaborating with the U.S. Geological Survey (USGS) and the National Oceanic and Atmospheric Administration (NOAA), which already were working together on 30-meter landcover data for a new National Land Cover Dataset (NLCD).

It was clear that Maine’s project and the federal projects would benefit from collaboration. USGS and NOAA agreed to reprioritize their schedule and put Maine at the top of the 2005 mapping list. An RFP was released, and Sanborn (www.sanborn.com) was selected to prepare a new landcover dataset and a separate, but related, imperviousness dataset for Maine based on 5-meter resolution 2004 imagery. The result was the Maine Landcover Dataset 2004 (MELCD 2004).


A Collaborative Effort
The winning proposal provided a thorough way to collaborate and integrate Maine’s effort with the federal 30-meter project. The proposal included a combined ground data collection process, as well as a classification based on the USGS/NOAA NLCD classification, with slight modifications to meet Maine’s needs. Finally, both projects used the same base imagery, Landsat Thematic Mapper, for the initial classifications; Maine’s imagery was enhanced later with higher-resolution data. Collaboration was key to the project’s success, because it allowed Maine to share costs for some of the project’s most expensive phases: satellite imagery collection and processing, training data, and accuracy assessment.

Landsat Thematic Mapper imagery collection, registration and mosaicking costs were covered by USGS and NOAA, which needed them for classifying the NLCD map. The imagery data and the classification, with some additional modifications, were the basis for Maine’s final product. These processes would have been costly for Maine to absorb (Figure 2). Expenses related to collecting training and accuracy assessment data were shared between USGS, NOAA and Maine, with Maine providing 1,000 hours of staff time in the field, including associated travel costs. The value of the USGS/NOAA contribution is estimated at $300,000.

Maine’s project differed from the federal projects by requiring a slightly modified classification system, a 2004 date for satellite imagery and 5-meter resolution. The classification needs were met by modifying the USGS/NOAA classification just enough to meet the needs of Maine users (Figure 3). Forestry classes were expanded to indicate cut types, all nonforested wetlands were collapsed into a single wetlands class, and three other classes were added (blueberry fields, roads/runways and alpine vegetation). These changes to the federal classification were based directly on Maine user needs and the state’s budget.

In 2004, Spot Image Corp. (www.spot.com) collected panchromatic 5-meter imagery with its SPOT-5 satellite to meet the state’s vintage and finer spatial resolution requirements. The original SPOT collect was supposed to occur during the leaf-on period, but weather conditions kept this from happening. A few scenes extended into the fall and early
winter season.

Sanborn developed a  5-meter impervious coverage over the urbanized portion of the state from the 5-meter imagery. The coverage was edited and quality controlled to produce a 90 percent accurate data layer that was used for the MELCD.

A Two-Stage Approach
Landcover within Maine was developed in two distinct stages. The first stage was to develop a statewide landcover dataset consistent with the USGS/NOAA landcover map. The second stage entailed updating existing landcover data to 2004 conditions; refining the classification system to Maine-specific classes; and refining the spatial boundaries to create a polygon map based on 5-meter imagery.

Image analysis techniques used to produce the map combined supervised classification using Classification and Regression Tree (CART) algorithms and spatial modeling. Three Landsat image dates provided the ability to discriminate specific landscape elements. For example, spring imagery was useful for classifying wetlands and separating conifers and broadleaf species; fall imagery was useful for discriminating broadleaf species.

After creating NOAA’s Coastal Change Analysis Program (CCAP) base map, Sanborn used image segmentation to refine the spatial boundaries of the landcover classes. The company fused 30- and 5-meter imagery to create simulated color 5-meter imagery. These data were segmented using eCognition image-classification software from Definiens Imaging (www.definiens-imaging.com). The process groups areas with similar pixels and labels them as a unit (Figure 4), producing segments that were labeled using automated methods to build the final MELCD.

After classification was completed, Sanborn analysts reviewed the map and modeled and/or edited specific classes by hand to remove class confusion. The final product was subjected to a statistically valid accuracy assessment that indicated an overall accuracy of 75 percent, with individual class accuracies in most classes exceeding 70 percent.

User Benefits

The first users of the new landcover data were thrilled to see the difference in pixel resolution and the scale at which the data could be used (Figure 5). The data break new ground for users, allowing them to conduct analyses at levels previously impossible by using new tools developed specifically to work with MELCD data (Figure 6). Additionally, integrating MELCD data with federal data puts Maine’s data into a much larger regional context if desired (Figure 7).

The final data were delivered in May 2006 and are being distributed via the Maine GIS Data Catalog (http://megis.maine.gov/catalog). As a result of the product’s collaborative nature, Maine users are able to get a wide variety of landcover and imperviousness data for their needs, including:

  See  Featured Images
 
  Subscribe to Earth Imaging Journal

  Go to Home Page