Technique Validation

The ability of the resolution enhancement algorithm to recover surface detail from the low resolution measurements was demonstrated with simulated measurements in LHW. For actual measurements, the lack of ground truth makes it difficult to objectively evaluate the effectiveness of the algorithm. However, we have observed a remarkable correlation between the observed Ku-band radar backscatter characteristics ( and ) and vegetation formations. This topic is explored in depth in the following section. Noting this, we have compared the backscatter from the enhanced resolution images to a map of vegetation to evaluate the capability of the radar images to discern vegetation features smaller than those resolvable in the low resolution ``binned'' images. For simplicity only the values were used to make this comparison.

The values from both the enhanced resolution and low resolution ``binned'' images were extracted along a line from W S to S S. This line was arbitrarily chosen to cross a variety of vegetation types and conditions. Figure 4 shows the location of this line in an expanded view of both the enhanced and low resolution images. (An image of the full extended Amazon Basin at enhanced resolution is shown in Fig. 6.) Comparison of the images in Fig. 4 reveals that the enhanced resolution image contains more detail than the low resolution ``binned'' image.

Figure 5 shows plots of the values from the enhanced and low resolution images along the study line (note that is the value in dB at a incidence angle). Referring to this figure we note that the value from the enhanced image (solid line) appears to contain more detail than the value from the low resolution ``binned'' image (dashed line). Further, it is clear that the low resolution line is not merely an averaged version of the enhanced resolution line. Using a small-scale land cover map published in 1978 by UNESCO [15] for reference, the vegetation types along this line were determined. The extent of each type along the line is indicated by the vertical dashed lines. As discussed further in a later section, the highest response, in general, occurs for moist tropical forest with reduced values for woodlands. Grasslands typically have the lowest response.

While there is a good comparison between the map and the radar imagery, the land cover map has some limitations which must be kept in mind when using it for comparison. For example, the map contains unavoidable generalizations because the amount of detail is limited by the map scale. In addition, there is some uncertainty in the placement of vegetation boundaries due to inaccurate source material and because of the frequent existence of broad transition zones (ecotones) between vegetation classes. Classification schemes used to generate the map also demand that grouping of vegetation species be performed. Within a given class, there may be variations in the relative species densities within the class. Thus, spatial variations in the radar response within each nominal vegetation class can be expected. Further, the radar response seems to be more dependent on the canopy density and vigor rather than just the physiognomic type.

In Fig. 5, type ``a'' corresponds to moist formations which include ombrophilous forest (which is the dominant formation along the line), and related degraded woodland and grassland mosaics. This has the highest response. Type ``b'' consists of variable subhumid medium tall grasslands with woody broadleafed evergreen woodlands. Being an ecotone, this can be expected to have a higher woodland density with a higher response than similar formations further from type ``a''. Type ``c'' is a subhumid degraded formation disturbed by cultivation. As discussed later, this type can be expected to exhibit a wide variation in vegetation due to various stages of clearing and regrowth. This will result in a variable response. Type ``d'' is subhumid grassland mixed with shrubs. This small region has the smallest response which is well-defined in the enhanced resolution image. Type ``e'' corresponds to subhumid medium-tall grasslands nearly devoid of shrubs. We note a close correspondence in the location of this type on the image and on the map though the observed response is somewhat higher than generally observed for this vegetation type. Type ``f'' is a subhumid tropical evergreen seasonal lowland forest. Type ``g'' is edaphic tropical grassland with palms while type ``h'' is edaphic tropical tall grassland with palms. The dip in the enhanced resolution plot indicated by ``i'' corresponds to the location of a transition within type ``a'' between a grassland and tropical forest.

Unfortunately, the precise response for each vegetation type along the study line is not known. However, we note a close correspondence with the details of the response from the enhanced resolution image and the vegetation formations along the study line. These are only weakly evident in the low resolution image. Although the correspondence is not perfect, and noting the cartographic uncertainties in the map previously mentioned, we conclude with some confidence that the resolution enhancement algorithm has, indeed, provided resolution enhancement.