Methodology

Using satellite imagery and digital image analysis-interpretation techniques, the Land Cover Remote Sensing project examines and maps annual land use distribution during the main agricultural crops growing season from May to August. The study area covers the entire state of North Dakota.

Satellite imagery acquisition:

Satellite imagery covering the entire state of North Dakota are provided by USDA's National Agricultural Statistics Service (NASS), Foreign Agriculture Service (FAS) and Farm Service Agency (FSA). The project is currently using medium-resolution (30m spatial resolution) LANDSAT multispectral (TM, ETM) satellite imagery. Cloud free (or minimum cloud percent) imagery captured during the agricultural crops growing season (May-August) are selected. To improve the separability of individual land use (crops in particular) a multitemporal approach is used. This means, for each location two images are utilized in the classification process, one captured early in the season (May-June) and another captured mid or later in the season (July-August). Due to the wide-area coverage analysis the study area is divided into analysis districts based on satellite imagery observation dates, i.e. regions with corresponding image observation dates are grouped together.

Ground truth data:

Every year during the crop growing season (May-August) extensive ground truth data is collected for the development of land use spectral signatures. The data is collected during the NDASS's June Agricultural Survey (JAS). Additional data is also collected during JAS follow-up visits in July-August. The ground truth exercise involves field visits to approximately 420 sample sites (segments) across the state of North Dakota for the purpose of identifying and delineating fields/training sites. The segments/sample sites are approximately one square mile in size.

Image processing and analysis:

Image processing and analysis functions are performed using a land use classification methodology developed by NASS's Research and Development Division, Spatial Analysis Section. The processing is performed using an imagery processing software, PEDITOR. This program was developed by NASS and is a modular system of Delphi programs which runs on Windows 95 and NT platforms. A Visual FoxPro graphical user interface, named RSP, is used for file management and PEDITOR module initiation. For more information about the software and the methodology visit http://www.nass.usda.gov/research/Cropland/SARS1a.htm.

Product distribution:

The image files are distributed on CD-ROM in ERDAS Imagine .gis and .img format. ESRI's ArcExplorer, a freeware image GIs browser, is distributed as the display tool. All of the final categorized PEDITOR images from the state or project are exported into ERDAS Imagine as .lan binary files. But first, the original raw images from from the Landsat TM satellite, are co-registered to an orthorectified base using automated block correlation techniques within PEDITOR on imagery provided by EarthSat Inc., GeoCover Project. The registration coefficients are then applied to the categorized scenes for mosaicing. Before mosaicing, the scenes are prioritized and cut by scene or political boundary. The categories from individual scenes are reassigned, so only one unique set of categories exist across the entire mosaiced scene. Cloud masking is performed during mosaicing to minimize cloud extent. No attempt is made to smooth or feather the image to improve appearances across scene boundaries. Once mosaicing is finished, the image is exported to ERDAS Imagine's LAN format. Further processing in ERDAS Imagine consists of colorizing, labeling , and building image attributes. The image is then ready for export to GIs/.img format for ArcExplorer distribution. The images are projected into the UTM projection (1983 datum). Since the state of North Dakota has two UTM zones (13, 14), the data files are distributed in UTM zone 14.