This report describes an initial assessment of the absolute geolocation accuracy of the NASDA JERS-1 Amazon mosaic imaged during low-water season, September-December 1995. The analysis is based on ground control points located on 1:100,000 scale maps; future work will augment this assessment using additional map points plus other sources including GPS-coded aerial video, ground-based GPS readings, and published control point locations for Brazilian airstrips.
Currently, basinwide digital elevation data are not available at sufficient resolution to permit correction of the terrain effects resulting from SAR imaging geometry. In May 2000, the Shuttle Radar Topography Mission is scheduled to acquire interferometric data from which a high-resolution digital dataset of 80% of Earth's land surface will be produced. In the meantime, mosaic users can carry out their own terrain corrections for smaller areas by digitizing elevation contours from high-resolution maps. The control points used for this analysis are all located on flat terrain at elevations < 500 m above sea level. Most of the Amazon basin is < 500 m in elevation, although hilly relief causing minor terrain distortions is common. Basin topography can be studied at roughly 1-km resolution using the USGS Global 30 Arc Second Elevation Data Set. The area covered by the JERS-1 Amazon mosaic extends beyond the boundaries of the Amazon basin into areas including the western slope of the Andes and the Orinoco watershed; for the Amazon basin boundary, see the hydrology section of the EOS Amazon Project home page.
Forty-five control points were located on Brazilian and Peruvian 1:100,000 scale maps using features including road intersections or sharp bends, bridges, power lines, airstrips, and stream confluences. Because larger stream confluences and lakes are subject to locational shifts depending on water levels, only smaller stream confluences were used. Map coordinates were determined manually using a ruler. Accuracy of map coordinate location (visually estimated) ranged from 0.2-2.0 mm of map distance. Ninety-five percent of the points were estimated to be locatable on the maps to within 1 mm or less; this corresponds to a ground uncertainty of 100 m, the approximate mosaic pixel size. JPL's DGX software was used to display mosaic tiles and obtain sample and line numbers and equivalent longitude and latitude for pixels corresponding to control points. All points were judged to be locatable to within 2 lines or samples, 82% to within 1.5 lines/samples, and 60% to within 1 line/sample; 16% were judged to be locatable with sub-pixel certainty.
The WGS 84 geodetic datum was selected as the reference datum for the Amazon mosaics; it is also the datum used by the Global Positioning System. Three datums were used by the makers of the map sheets on which control points were located: South American 1969 (Brazil), Provisional South American 1956 (Peru), and WGS 84. Non-WGS-84 coordinates were converted to WGS 84 using the standard Molodensky transformation. PC software to perform datum conversions between WGS 84 and other datums is available from the National Imagery and Mapping Agency. Control point locations are shown in Fig. 1.
Table 1 shows longitude and latitude (in WGS 84) for the 45 ground control points, as located on the maps and on the NASDA mosaic. Differences in longitude and latitude between the mosaic and the maps in decimal degrees (mosaic coordinates minus map coordinates) are shown in Table 2. The differences (converted to meters) are plotted as functions of longitude and latitude in Fig. 2. Two points (19 and 23) with unusually high errors in latitude are omitted from the plot to better show overall trends. Fig. 2 shows a linear relationship between longitude and longitude error, as well as a weaker variation of latitude error as a function of longitude; errors increase moving from east to west. At the east end of the basin (-48°), mosaic and maps are in close agreement, differing by less than 1 km in longitude and about 2.5 km in latitude. Moving west, the difference increases to about 8 km (longitude) and 4.5 km (latitude) at -76°. No trends as a function of latitude are seen. Fig. 3 shows the errors plotted as vectors, scaled by a factor of 50.
The differences shown in the preceding figures are a combination of uncertainty in locating control points on maps and mosaics, possible systematic distortions introduced during SAR processing or during mosaicking, and error introduced in pegging the mosaic to the WGS 84 datum. In order to minimize the latter error, the mosaic was shifted by amounts equal to the mean mosaic-map differences in longitude and latitude. Control points with unusually large errors (points 19 and 23) were excluded from the mean calculation. Errors for these points may be due to factors unrelated to processing or mosaicking, and the points are being re-evaluated (though they are still included in subsequent analysis other than the mean calculation). The resulting mean adjustments are -0.030744° (longitude) and 0.030923° (latitude) for the NASDA mosaic. These adjustments were subtracted from the original mosaic locations to shift the mosaic as close as possible to the map locations. Reference latitude and longitude given in the table of mosaic tile specifications on the NASDA CD-ROM already include these adjustments. The equivalents for these shifts in terms of meters at the equator are -3422 m in longitude and 3442 m in latitude.
The adjusted control point coordinates, along with the original map coordinates, are shown in Table 3, and Table 4 gives the NASDA-map differences (converted to meters) for the adjusted coordinates. The rms error in meters relative to the maps is 1856 m (longitude) and 605 m (latitude). Table 5 shows what percentage of points are located within a given error for the adjusted mosaic.
| Error | NASDA lon | NASDA lat |
|---|---|---|
| <= 100 m | 2% | 16% |
| <= 500 m | 27% | 56% |
| <= 1000 m | 44% | 87% |
| <= 2000 m | 51% | 96% |
| <= 5000 m | 100% | 100% |
Vector plots of the NASDA-map differences for the adjusted mosaics are shown in Fig. 4, and Fig. 5 shows a histogram of the mosaic-map differences (with means equal to 0) after adjusting the mosaic. The obvious geographic variability in the error vectors (Fig. 4) indicates that users can improve the geolocation accuracy of the mosaic by systematic corrections such as rotation, if accuracies given in Table 5 are insufficient for their applications.
For the five mosaic tiles that had five or more control points, the mosaic adjustment described above was repeated on a tile-by-tile basis to examine within-tile error: each tile was optimally shifted, and the resulting errors relative to maps calculated. Plots of the errors in terms of meters are shown in Fig. 6. The arrangement of the tiles in these plots is consistent with their relative geographic positions (cf. Fig. 1). It appears that tile 113, located near the center of the basin, has smaller errors than the others, although the number of samples is small. For longitude, 97% of the 33 points were within 1 km of the map location, 82% within 500 m, and 42% within 200 m. For latitude, 94% were within 1 km of the map location, 88% within 500 m, and 61% within 200 m.
