Classified polygons over Horry County SC
Horry County, South Carolina, is among the fastest growing areas in the nation. With miles of connected beaches, the county faces significant development demands resulting from millions of annual visitors and an influx of retirees moving to its coast. To help manage these impacts, the county has embarked on a land-use/land-cover mapping project that utilizes an innovative methodology and existing geospatial datasets.
Maximizing Source Imagery
"Land-use/land-cover data is the critical piece of information that our stormwater group has been missing for its hydrologic watershed modeling," said Tim Oliver, Horry County's assistant director of information technology and geographic information systems. "As a rapidly developing coastal county, a comprehensive dataset is necessary to understanding the impact from one development project to the next."
Fugro EarthData was already working with Horry County to update the government's orthoimagery and planimetric mapping when the need for land-use/land-cover data was identified. Bill Shinar and John Knowlton, who serve Fugro EarthData as regional manager and project manager, respectively, recommended a new approach to image classification that would draw on the county's recently acquired airborne imagery and a newly developed, semi-automated approach to image classification.
The new methodology was proven by Fugro EarthData in 2007 during a project for the National Oceanic and Atmospheric Administration's Coastal Services Center (NOAA CSC) to map seagrass habitats along the Texas Gulf Coast. Like the Horry County project, the Texas effort relied on multispectral four-band digital imagery collected from a Leica ADS40 sensor.
Automating Image Classification
"While there are some differences between shallow-water versus land-based image classification, the techniques used for either project type are basically the same," explained Chad Lopez, senior digital imaging analyst for Fugro EarthData.
Traditional land-use/land-cover mapping is accomplished using manual photointerpretation methods, which can be extremely time consuming and lacking in detail. Automated pixel-based classifiers exist, but these can have high error rates, especially when used on high resolution imagery, causing labor-intensive manual error correction.
In contrast, Fugro EarthData's semi-automated approach streamlines the process through use of object-oriented feature extraction software and classification and regression tree (CART) analysis. Managing inputs from a wide variety of ancillary data sources, the software classifies groups of pixels, rather than individual pixels, and incorporates both shape and context into the classification process. The result is more efficient map production with less operator bias and greater overall consistency.
For the Horry County project, Fugro EarthData resampled the original 6-inch resolution orthoimagery to 48-inch resolution and mosaicked the data into nine processing regions. An image segmentation process created polygons (groups of pixels) from the imagery that adequately captured low, medium, and high density built-up urban areas as well as vegetation and mapped them using a classification scheme derived from the Anderson Land Cover/Land Use Classification.
Overcoming Obstacles
Even with this highly advanced image classification approach, there were still a few production obstacles to overcome. Take the classification of trees versus shrubs. "It can be tough to distinguish from nadir imagery whether vegetation is 3 meters tall or 5 meters tall," Lopez said. To address this issue, Fugro EarthData utilized the county's 2005 lidar data to derive vegetation height. The information, while a few years dated, proved an important processing input that enabled a more efficient and accurate classification.
Creating polygons in urban areas posed another challenge. Though trees are a common characteristic within built-up neighborhoods, the classification software tends to identify trees and houses as different feature classes. As a result, trees were being split out from their neighborhoods with separate polygons. As a solution, Fugro EarthData utilized the near-infrared spectral band to derive an adaptive texture image, which was used for the segmentation process. Since a mix of trees and houses provides greater texture than a forest alone, polygons could accurately be built around the entire neighborhood—trees and all.
Expanding Data Applications
Final deliverables for the land-use/land-cover mapping project are slated for later this month. Accuracies for individual feature classes are estimated at 80 percent while overall accuracies are estimated at 90 percent. In addition to stormwater management applications, the county is also considering other potential uses of the land-use/land-cover data, including future land-use planning, fire fuels modeling, and hurricane debris estimating.
That these existing and potential applications all derived from the same source data is not lost on Tim Oliver. "Let's face it," he said, "without the ADS40 imagery, the land-use/land-cover project wouldn't have been possible." By maximizing the value of the imagery beyond its original conception, Horry County is ensuring taxpayers a high rate of return on their initial investment; proof that geospatial technology really can help governments do more with less.
www.fugroearthdata.com
|
