North American Landscape Characterization (NALC) Project/Campaign Document

The NALC project area includes the conterminous United States and Mexico.

Processing: Generating triplicates involves a multi-stream approach. The 1980's image is precision corrected and registered to a map base using a three-step approach and is used as the cartographic base to which the 1970's and 1990's data are registered. A relational data base has been developed and maintained, which contains the corner coordinates and quadrangle names for United States 1:24,000-scale topographic maps. This data base was augmented with similar information for the 1:250,000- and 1:50,000-scale topographic maps for Mexico acquired from the Instituto Nacional de Estadistica, Geografia e Informatica (INEGI) in Aguascalientes, Mexico. The latitude and longitude coordinates of the 1980's scene to be used for a given triplicate are intersected with the map data base to identify the topographic maps which fall within that particular Landsat Worldwide Reference System-2 (i.e., Landsats 4 and 5) path/row.

The 1980's image is precision corrected and registered to a map base. In rare situations, multiple 1980's images are used to produce a cloud-reduced composite. The registration entails the selection of image and planimetric source control points for use in developing the model for precision correction. Control point sources used include 1:100,000-scale USGS digital line graph (DLG) data and 1:24,000-scale USGS topographic maps for triplicates within the United States, and 1:50,000-scale maps for triplicates outside of the conterminous United States. The DLGs are components of the National Digital Cartographic Data Base (NDCDB) and are comprised of the various thematic layers (e.g., transportation, hydrography, hypsography, political boundaries) depicted on the 1:100,000-scale topographic map series (USGS, 1989). For triplicates within the United States, the DLG data are the primary source for control point selection. The DLGs are interactively overlain onto the imagery to facilitate visual correlation of area features between the imagery and the DLG data. Once a feature match has been identified, the image (line, sample) and DLG (latitude, longitude) coordinates for a specific point along the feature are extracted and compiled in a control point file. Approximately 30 to 35 points are extracted.

The next step in triplicate generation involves mosaicking the DEM data and transforming the DEM data's original geographic projection to a Universal Transverse Mercator (UTM) projection. The DEM data used in the NALC project are derived from Defense Mapping Agency (DMA) Level-1 Digital Terrain Elevation Data (DTED) (USGS, 1987), which were digitized from the standard National Topographic Map Series 1:250,000-scale maps. These maps provide complete United States coverage. The DEM data, often referred to as the 3-arc-second DTED data, are available as gridded files corresponding to 1-degree-latitude by 1-degree-longitude blocks. These blocks are mosaicked by path/row, then projected and resampled to 60- by 60-meter pixels in a UTM projection.

Elevation values for the ground control points are retrieved from the DEM data. The ground control points containing X, Y, and elevation values are corrected for relief displacement. The relief-corrected control points are then used to compute the coefficients for a second order polynomial model that is used to geometrically correct and reproject the 1980's MSS image to a UTM ground control coordinate system. Using cubic convolution, the image is rectified and resampled to a UTM-projected output image comprised of 60- by 60-meter pixels. As of July 28, 1994, a full terrain correction is also applied to the 1980's image by correcting for the effects of relief displacement on a pixel-by-pixel basis using the previously created DEM image.

A verification of registration quality is performed using control points selected from either 1:24,000-scale USGS maps or 1:100,000-scale DLG source material. Selection of 12 to 15 verification control points proceeds in a manner similar to the selection of registration points. Scenes must meet quality objectives of total Root Mean Square Errors (RMSEs) of less than 1.0 pixel for United States scenes, and total RMSEs of less than 1.5 pixels for Mexican scenes. Registration control points are reselected for scenes which fail to meet quality objectives.

The final geocoded 1980's component for each triplicate is a six-band image. Bands 1 through 4 are comprised of MSS bands 1 through 4, band 5 (an Normalized Difference Vegetation Index (NDVI) computed from MSS bands 2 and 4), and band 6 (a pixel identity band in which pixel values greater than or equal to 1 correspond to valid image data and pixel values of 0 correspond to non-image fill). The non-zero pixel identity values can be referenced to the alphanumeric attribute labels in the metadata records. The pixel identity images can be used to disaggregate a composite image into its constituent source images in order to accommodate scene-specific processing requirements (e.g., atmospheric correction, normalization of solar illumination angles).

The following systematic, radiometric, and geometric corrections are applied to 1970's MSS CCT-X data to generate the EROS Digital Image Processing System's EDIPS-P product. The CCT-X formatted data are preprocessed to correct for line length adjustments, variable detector response, band registration, nonlinear mirror-scan velocity, Earth rotational skew, and detector-to-detector offsets. The images are destriped to compensate for variations in the radiometric response of the individual detectors prior to geometric registration, because the noise is scan-line dependent. Using the satellite ephemeris data and platform navigation model, an interim systematic correction is then applied to generate a UTM-projected output image with a north-up orientation. Automated cross-correlation procedures (Bernstein, 1983; Scambos and others, 1992) are then implemented to extract control points from the 1970's and 1980's images to compute coefficients for image-to-image registration. This involves the use of a single band from each of the 1970's input images. Once an accurate transformation is developed, the grid for the interim systematic correction is convolved with the image-to-image transformation grid to produce coefficients, which facilitate a single-step registration of the 1970's P-product with the map-registered 1980's image. The net result is an image registration procedure that only involves one step of resampling. As of July 28, 1994, a full terrain correction is also applied to the 1970's image by correcting for the effects of relief displacement on a pixel-by-pixel basis using the previously created DEM image.

Cross-correlation procedures are also used to extract over 100 control points that are used for verification of image-to-image registration quality (1970's to 1980's) with target total RMSEs of less than 1.0 pixel. Previous studies have shown that the use of polynomial transformations alone on CCT-X format data yield only a 1.5-pixel to 2.0-pixel internal image accuracy once map projected, but the use of a satellite model overcomes this problem. Should the scene fail to meet quality objectives, image-to-image cross-correlation parameters may be altered and new registration control extracted or hand-selected image-to-image control may be used.

The last step involves the creation of a pixel identity band to accompany each of the 1970's images. Each pixel identity value is indexed to the specific 1970's scene used to fill-in the WRS-2 path/row tile with pixel values of 0 representing non-image fill. Typically, two or more 1970's images are required to provide complete coverage of a WRS-2 path/row tile.

The procedures for the 1990's image registration are similar to those for the 1970's data except that the 1990's data are acquired as P-level products. An interim systematic correction is applied to generate a north-up, UTM-projected image. Automated cross-correlation procedures are used to select control points and a transformation grid is computed for the image-to-image registration. The interim and precision transformation grids are convolved into a single grid which is then applied to the 1990's input to create its 1980's coregistered equivalent. Similar to the 1970's processing, this involves only one step of resampling. As of July 28, 1994, a full terrain correction is also applied to the 1990's image by correcting for the effects of relief displacement on a pixel-by-pixel basis using the previously corrected DEM image.

The verification of image-to-image registration quality is performed using cross-correlation procedures as was performed on the 1970's image. The target RMSE for this registration is less than 1.0 pixel. Should the scene fail to meet quality objectives, image-to-image cross-correlation parameters may be altered and new registration control extracted or hand-selected image-to-image control may be used.

Similar to the 1980's image, the 1990's component of each triplicate is a six-band image comprised of MSS bands 1 through 4, an NDVI image, and a pixel-identity image.

Prior to March 20, 1995, cloud-reduced compositing was performed after all the image data were coregistered. This step was performed only in cases where 1990's scenes with 30 percent or less cloud cover were not available. To minimize the amount of cloud cover in the 1980's or 1990's triplicate component, the EROS Data Center adapted Advanced Very High Resolution Radiometer (AVHRR) cloud compositing procedures for use in NALC triplicate generation (Eidenshink, 1992; Holben, 1986). The compositing process operates on image pairs and is based on the MSS NDVI:

This index is sensitive to variations in surface characteristics, such as biomass, and is sensitive to clouds (Justice and others, 1985). The NDVI is computed for each of the images to be used for compositing purposes. The maximum NDVI value determines which input image pixel brightness values will be used to constitute the output image. This maximum NDVI decision rule is computationally efficient and yields consistent results. A 1980's and/or 1990's triplicate component which has been composited will have a maximum NDVI image and a pixel identity image.