Thematic Mapper Studies of Semiarid Shrub Communities

Susan L. Ustin, John B. Adams, Christopher D. Elvidge, Marcel Rejmanek, Barrett N. Rock, Milton O. Smith, Randall W. Thomas, and Roy A. Woodward Susan L. Ustin and Roy A. Woodward are research botanists in the Department of Botany, University of California, Davis 95616, and Space Sciences Laboratory, University of California, Berkeley 94720
John B. Adams is a professor and Milton O. Smith, a research scientist, Department of Geological Sciences, University of Washington, Seattle 98195
Barrett N. Rock is group leader and Christopher D. Elvidge is research associate in the Geobotany Remote Sensing Group, Jet Propulsion Laboratory, Pasadena, CA 91109
Marcel Rejmanek is assistant professor, Department of Botany, UC Davis
Randall W. Thomas is senior fellow, Space Sciences Laboratory, UC Berkeley
They are all involved in a multidisciplinary study of remote sensing of arid lands.

Abstract

Although remote sensing has been successfully used for many geological purposes, it has been less successful when applied to understanding vegetation and community ecology (Goetz et al. 1983, Siegal and Gillespie 1980). Our ability to interpret spectral information remains especially limited in arid environments, where the vegetation is sparse and shrub communities with similar physiognomies often differ markedly in species composition and ecology.

Despite these difficulties, remote sensing has become a useful tool in arid zone ecology and range management programs. Use of these techniques is expected to increase substantially because of the improved spatial and spectral resolution of advanced sensors and imaging spectrometers (Goetz et al. 1985). These developments should eventually allow investigators to identify surface features directly from unique spectral properties and should have a significant impact on plant ecological research, ranging from the analysis of ecosystems to environmental physiology of dominant species.

In this paper we illustrate how thematic mapper (TM) satellite data may help analyze semiarid vegetation, and we discuss how spectral data can be transformed to yield an array of ecological information. This scanner detects solar energy reflected from the earth's surface in six visible and infrared bands and surface-emitted thermal energy in one infrared band. We will discuss four techniques that can be used individually or together to draw inferences regarding edaphic variation, community composition, canopy architecture, and physiological or phenological activity. Although the underlying biophysical mechanisms determining surface reflectance properties are only partially understood, more quantitative relationships are expected as analytic techniques improve.

Study area

Our study area is located on a Sierra Nevada bajada (alluvial fan complex) west of the town of Independence, in the Owens Valley of southeastern California (Figure 1). Elevations range from 1250 m on the valley floor to 1900 m on the upper fans. This region is part of a broad ecotone between typical Mojave and Great Basin desert shrub communities to the south and north. Tueller (1980) has reported that these communities are extremely difficult to distinguish with Landsat multispectral scanner data. Vegetation patterns are complicated by grazing and wildfires on the bajadas, abandoned farmland on the valley floor, and recent desertification resulting from diversion and removal of water, making satellite data even more difficult to interpret (Carneggie et al. 1983).

Although most of the bajada supports a Great Basin sagebrush community, elevational trends are evident. Figure 2 illustrates the general distribution and abundance of the most common species along the elevational gradient. At the junction between the Sierran foothills and the bajada, typical Sierran species are common, but typical Great Basin desert species dominate at all lower elevations. The upper fans support a Purshia-Artemisia (bitterbrush-sagebrush) scrub, which grades into a Coleogyne (blackbrush) scrub at middle elevations (BLM 1979). The lower fans support a mixed desert scrub, including some Mojave Desert elements. The valley floor is dominated by alkaline (saltbush) scrub.

To understand how vegetation contributes to TM imagery, we analyzed five areas in the Artemisia-Coleogyne zone, where past wildfires had removed the vegetation, and compared these with adjacent unburned areas for percent cover, soil surface type, and shrub size. Fires on these slopes virtually eliminate shrub cover, producing large areas of exposed bare soil. Cover by annual or weedy vegetation is low in the initial stages of community recovery. The decrease in bare soil surface as the community develops is nearly linear with time; it takes an estimated 70 years to reach the vegetation cover found on unburned areas at similar elevations. Although the five most abundant species in the mature community (93-97% of vegetation cover) reestablish within the first ten years after removal, changes in species proportion occur (Ustin et al. 1986).

Vegetation indices

A number of methods have been developed to separate vegetation from soil information in satellite data (Jackson 1983, Kauth and Thomas 1976, Kauth et al. 1978, Richardson and Wiegand 1977). The most common for agricultural purposes are band ratios, which are chosen to maximize the contrast between vegetation and soils. Unfortunately, these ratios are sensitive to the variations in soil brightness characteristic of arid lands (Elvidge and Lyon 1985, Huete et al. 1985) and predict vegetation less well than other methods at our site (Ustin et al. 1986). A potentially more reliable method is derived from scatterplots of TM band pairs. For a wide range of rocks and soils, red and near-infrared reflectances (TM band 3 and TM band 4) are highly correlated, forming a well-developed baseline diagonal to the axes. Vegetation spectra depart from the rock-soil baseline because photosynthetic pigments reflect little red light, whereas leaf and canopy structure reflect much of the incident near-infrared energy. This spectral relationship is the basis for the perpendicular vegetation index (PVI), which measures the orthogonal distance of a pixel from the baseline (Richardson and Wiegand 1977). Since PVl depends on chlorophyll absorbance, it should yield information about the physiological status of the canopy. The counterpart to PVI is the soil brightness index (SBI), which measures the distance along the rocksoil baseline from the scatterplot origin to the pixel. SBI responds to several factors, including rock and soil brightness, shadowing, and changes in vegetation density.

PVI and SBI images were developed for the bajada to examine the spatial distribution and relative density of vegetation (Figure 3). As expected, the PVI image clearly distinguishes irrigated fields, valley lands with shallow groundwater, the Sierran foothills, and the riparian vegetation along the perennial streams as having high vegetation density (Figure 3a). Gradients in vegetation density can be seen on the fans, with lower PVI values near the foot of the bajada and in the central portion of the image. In areas where cover is already sparse, such as near the foot of the bajada, areas of low vegetation (< 10%) stand out as having lower PVI values than the surrounding surfaces, demonstrating the ability to discriminate the presence of vegetation under these conditions. But changes in shrub cover at middle elevations on the bajada are more complex, and in some cases PVI values are inconsistent with vegetation cover. For instance, the youngest burn (a) is not detectable on Figure 3a, when in fact projected canopy cover is only one-half that in the surrounding area, and differences are even greater when canopy volumes are compared. Although PVI is reported to be highly correlated with green leaf area index and biomass in crop systems (Tucker 1979, Wiegand 1984), leaf area in our communities is very low (LAI < 1), and PVI may be responding to additional vegetation features, such as canopy architecture. Recently, Pinter et al. (1985) showed that canopy structure had a significant effect on reflectance characteristics in visible and near-infrared wavebands, end Jackson and Pinter (1986) found that erectophile and planophile canopies of equal biomass and LAI could differ by 30% in PVI.

Ground measurements show that the spectral properties of the dominant species of this community change dramatically over the growing season as a result of physiological and phenological changes (Ustin et al. 1985, 1986). Gradients in PVI values across the fans are most apparent in May and markedly less distinct by July and October, consistent with declining physiological activity and lower leaf area after the summer drought begins. Thus, examining TM scenes at different times during the year may be useful for monitoring changes in physiological activity and phenology of the vegetation even under such sparse canopy conditions.

Most of the variation seen in the SBI image (Figure 3b) is due to differences in the amount of bare rock-soil surface on the bajada rather than soil characteristics per se. All of the burns are clearly visible; their brightness correlates inversely with relative vegetation cover. The oldest burn (b) is still visible, however, even though its cover is now as high as that of surrounding surfaces. This may be because the SBI is sensitive to shading from vegetation and boulders; the canopy height on the burn is much lower than in adjacent areas. Both indices largely conform to expectations about surface conditions, but both conflict somewhat with ground data, indicating that they are affected by interactions that are not yet understood.

Correlating vegetation and spectra

From areas where detailed vegetation analyses were available, we used multivariate analyses to establish relationships between the vegetation and spectral characteristics. A stepwise discriminant function analysis (Afifi and Clark 1984, Jennrich and Sampson 1983) showed that either ground-based vegetation transects or TM spectral data could be used to distinguish all burned and unburned localities. Significant variables (P < 0.05) included DN values (relative brightness) of all seven TM bands (but not ratios or vegetation indices), percent cover of the five dominant species, and the proportions of total shrub, litter, and soil cover. Thus, despite the general homogeneity of the vegetation, samples from each locality clustered uniquely.

To test the predictive capacity of these discriminant functions, we analyzed an independent data set and determined the proportion of correctly classified transects and pixels. The classifications appeared robust; erroneous decisions involved placement in the next most similar class (such as from one unburned area to another). Significantly, the proportion of correctly classified pixels (> 93%) was much higher than the proportion of correctly classified ground transects (> 50%). This agrees with the results of several other authors who have found that satellite data are more accurate than the ground data used to verify community membership (Curran and Williamson 1985, Justice and Townshend 1981, Smedes 1975), although interpreting the integrated spectral response is difficult for heterogeneous scenes (Batista et al. 1985, Duggin and Philipson 1985, Kaufman 1985, Toll 1985). The value of the spectral data is still more evident when one considers that at least 15 times as many personnel-hours were required to collect and analyze the ground data.

We used canonical correlation analysis to further evaluate the predictive capability of spectral and ground variables. The spectral distinctiveness of these sites was compared by plotting samples on the first two canonical axes, representing 95% of the spectral variation. As expected, this analysis revealed that the unburned sites were very similar to each other, whereas the burned sites were all distinct, and that the oldest burns were most similar to the unburned sites. However, we were not able to relate specific vegetation variables consistently to DN values in each of the spectral bands or to the multispectral patterns. Strongest correlations were found between percent rock cover (r² = -0.76 to -0.89) and reflective TM bands (excluding the thermal band) and between percent bare soil exposed (r² = 0.74 to 0.75) and TM reflective bands, again indicating that most of the spectral variation is determined by the relative proportions of vegetation and bare rock-soil surfaces. Unfortunately, transect data could not be precisely paired with specific pixels, so that only site means were compared in our analysis, substantially reducing the number of independent observations and limiting statistical confidence. A larger set of vegetation measurements might produce a significant correlation between species composition, soil type, and reflectance characteristics, especially during peak spring growth.

Multispectral clustering models

We tried a different approach, delineating ecologically meaningful spatial units on the bajada by interpreting spectral variation with a clustering algorithm (developed after Ball and Hall 1967) on data from all seven TM bands. Twenty classes were identified that exceeded a minimum separation distance (defined by a Swain-Fu test), nine of which occurred on the bajada. Each class was characterized by a unique spectral signature, and although albedo (brightness) generally increased downslope, it did not account for class differences. Each pixel was assigned to a class using a maximum likelihood method, and each class was represented by a color for display purposes (Figure 4).

The first six classes, represented by greens and grays, are associated with irrigated, riparian, or upper Sierran foothill areas. In the lower Sierran foothills (left), predominantly shown as oranges, tans, and bright blues, spatial patterns indicate albedo differences caused by topography. Bajada classes show little topographical effect and are primarily related to exposures of different alluvial surfaces, but they also reflect plant cover, community composition, and land use patterns (Table 1). Dark purple, lavender, and cream-colored units on the fans correspond to older granitic alluvial surfaces with a high cover of large boulders, and these are overlain by younger surfaces (red and gold) on the central fans (Vaughn 1983). Gray, olive, and tan units near the valley floor are finer-grained and more alkaline than the higher elevation terrace units. However, if classes corresponded only to substrate characteristics, we would not observe the burned areas on the bajada. In all cases, the burned areas belong to classes seen at lower elevations where vegetation cover is also lower, suggesting that canopy cover is modulating the effect of soil composition. Canopy cover (including litter) declines from about 40% to 15% from the top to the bottom of the bajada. The two most recently burned areas (a, b) have less vegetation cover and correspond to much lower elevation fan classes; older burns (c, d) resemble adjacent classes or are only partially distinct from surrounding surfaces. Linear features in the olive unit (e) are disturbed areas of low vegetation cover (< 15 %), mostly associated with water-spreading levees and gravel ditches.

The classes represented on the lower part of the bajada appear to be more consistent in delineating vegetation types than the classes at higher elevations. Dark green irrigated fields (f) are distinct from riparian communities. Tan units (g) underwent a decline in shrub cover after the 1977-78 drought and now support a low cover of weedy annual vegetation (Griepentrog and Groeneveld 1981). Gray units (h) correspond to an alkaline saltbush scrub community, and cream units to a mixed Mojave-Great Basin desert community. On the other hand, all midslope communities are dominated by Coleogyne, regardless of classification. Although the purple and lavender units closest to the Sierran foothills correspond to the Purshia-Artemisia scrub, they also correspond to the Coleogyne-Artemisia scrub somewhat lower.

These results demonstrate the limitations of this technique: since the classes are based on spectral mixtures of soil and vegetation in varying proportions, neither can be uniquely identified unless the vegetation type is edaphically restricted. Because classes are site dependent, vegetation density and community structure cannot be directly inferred without ground verification. Yet once ground conditions are associated with classification patterns for a given area, soil types, habitat features, and vegetation characteristics can be delineated. Despite its limitations, this was the only analytical method in which subtle community changes in species composition could be recognized, suggesting that it may have wide practical application for assessment of land use patterns, range quality, desertification, and other habitat changes.

Multispectral mixing model

A fundamentally different approach for separating the spectral effects of substrate and vegetation has been developed using a multispectral mixing model (Adams and Adams 1984, Adams et al. 1986). In addition to distinguishing vegetation and soil, this model attempts to separate the height and cover components of vegetation using the shade-shadow concentration in the image. By dissecting components of canopy architecture, this approach may provide an improved method for estimating biomass, and, through comparison of TM scenes from different seasons, for detecting annual primary productivity. Like the two-dimensional vegetation indices, the mixing model relies on prior measurements of in situ spectral responses of soils and vegetation, and statistical techniques for separating soil, vegetation, and topographic brightness variations in the pixel data. The mixing process first identifies the most extreme spectral signatures, four for this TM scene, and additively mixes them until the best fit to the TM data on a pixel-by-pixel analysis is found. In the present case, two "end-members" are representative light and dark soils, the third is a vegetation component typical of dormant Artemisia-Coleogyne scrub, and the fourth is a shade component, accounting for topographic and other shade-shadow variations. Dormant Artemisia-Coleogyne was used for this analysis because it best models the spectral response in this December image.

Figure 5 illustrates the shade and vegetation components for the model generated from the winter TM scene. This figure extends from the snow-covered Sierra Nevada to the White-Inyo Mountains and to the Alabama Hills near the lower right corner. The shade portion of the image is expressed with tones from light to dark, corresponding to increasing concentration of shade and shadow-casting features. This effect can be seen on north-facing slopes, which are dark, and south-facing slopes, which are light, since the illumination is from the southeast. In the absence of topographic effects or roughness caused by boulder fields, the shade concentration increases with vegetation cover and canopy height, as confirmed by site inspection. The surprising congruence between this model and SBI (see Figure 3b) supports the conclusion that most albedo variations on these fans are due to shadow-casting features. The total range of mean canopy heights on the image, except for scattered trees, is about one meter. Three levels of spectrally determined vegetation-cover concentration are indicated by the color groups, gray (lowest), brown (intermediate), and green (highest). Overlaying the shade image with the vegetation concentration image allows the vegetation cover to be modulated by the relative concentration of shade. This produces a range in brightness for each of the three colors: darker grays and browns predict higher shade concentration than lighter grays and yellows.

To illustrate the sensitivity of this method in detecting subtle differences in vegetation concentration at very low densities, we have decreased the lowest vegetation concentration included in the green category in two-step increments. In Figure 5a only the highest vegetation concentration is green, corresponding to greater than 65% canopy cover, whereas the green in 5b and 5c corresponds to > 55% and > 45% canopy cover, respectively. For simplicity, the brown and gray classes include larger cover ranges, but they too could be subdivided. In 5c, the brown category represents 45% to 25% vegetation cover, dark gray 25% to 20%, and light gray less than 20% vegetation cover.

Ground measurements support the conclusion that the spectral mixing model can accurately predict vegetation cover to within 5%. For example, the transition area between the brown and green categories at (a) in Figure 5c has a mean cover of 44.2%, while the oldest burn (b) has a cover of 41.9% and is brown. The golden brown at (c) is at the lower limit of the brown class and has 24.3% cover. Although data suggest that the model responds linearly to changing cover, we emphasize that this conclusion must be regarded as tentative pending further evaluation of the model.

In striking contrast to the Sierran fans, Figure 5 indicates that most vegetation cover on the fans and foothills of the White-Inyo range is less than 25%. The only areas of higher cover occur along stream canyons and on north-facing slopes. On the other hand, the valley floor exhibits a wider range of shade and vegetation conditions than do the fans. The large light area at (d) coincides with a pristine locality of highly alkaline soils (Groeneveld et al. 1986), whereas (e) is bounded by the 1872 earthquake fault line, which blocks water movement. The fault itself can be seen because a surface springline permits higher cover (Groeneveld et al. 1986). In some cases, however, very dark areas (f) do not indicate tall vegetation but rather sites of unusually dark soil and surface moisture, which mimic the spectral signature of shade. Such "errors" illustrate the iterative nature of this model, since substituting the dark soil type of the valley bottom for the dark fan soil eliminates this confusion. Because of the need to limit computational demands, only some of the heterogeneity in a scene of this size can be modeled at any one time; here we have chosen to illustrate the shrubs and soils of the bajada.

Inspection of the three vegetation density images reveals that in some cases shading is not correlated with apparent cover. For instance, alkali grasslands on the valley floor (g) are yellow-golden in Figure 5a (implying 25% cover) but become green in Figure 5c (implying 45% cover, in agreement with ground measurements). This suggests that the low shading of grass-sedge canopies, in some cases less than 10 cm tall, may underestimate cover relative to shrub-dominated fans. Comparisons between communities must be made cautiously because the model is so sensitive to different vegetation types, especially those with differing canopy architectures. However, by changing the vegetation type selected as the spectral end-member, the model can be repeatedly tested and evaluated for such effects. Thus, inferences can be made regarding the physiognomy of the dominant vegetation.

Conclusion

These examples illustrate the range of vegetation information that can be obtained from the thematic mapper in a semiarid region. By using a variety of analytical methods on a common data set, considerable qualitative and quantitative information emerges about environmental conditions. Using multitemporal imagery, such comparisons should yield information regarding phenological cycles and physiological conditions of vegetation for short time intervals or habitat changes and successional processes for longer periods. Other sources of habitat information may be linked in an interactive computer database, including topographic, geologic, and soils maps. Because of the spectral characteristics of different sensors, a multisensor approach itself can increase the sophistication of analysis. The variety of new sensor systems, particularly the imaging spectrometers currently available or under development, will greatly enhance the specificity of remotely sensed data (Goetz et al. 1985). As both instruments and methods of analysis develop, remote sensing will become an increasingly important tool for ecological research.

References cited

Proceedings of the International Symposium on Remote Sensing of Environment Third Thematic Conference: Remote Sensing for Exploration Geology

Adams, J. B., M. O. Smith, and P. E. Johnson. 1986. Spectral mixture modeling: a new analysis of rock and soil types at the Viking Lander I site. J. Geophys. Res., in press.

Afifi, A. A., and V. Clark. 1984. Computer Aided Multivariate Analysis. Lifetime Learning Publ., Belmont, CA.

Ball, G. H., and D. J. Hall. 1967. A clustering technique for summarizing multivariate data. Behav. Sci. 12: 153-155.

Batista, G. T., M. M. Hixson, and M. E. Bauer. 1985. Landsat MSS crop classification performance as a function of scene characteristics. Int. J. Remote Sens. 6: 1521-1533.

Bureau of Land Management (BLM). 1979. Owens Valley Unit Resources Analysis I, II, III. Bishop, CA.

Carneggie, D. M., B. J. Schrumpf, and D. A. Mouat. 1983. Rangeland applications. Pages 2325-2384 in R. N. Colwell, ed. Manual of Remote Sensing, vol 11. American Society of Photogrammetry, Falls Church, VA.

Curran, P. J., and H. D. Williamson. 1985. The accuracy of ground data used in remote sensing investigations. Int. J. Remote Sens. 6: 1637-1651.

Duggin, J. J., and W. R. Philipson. 1985. Relating ground, aircraft and satellite radiance measurements: spectral and spatial considerations. Int. J. Remote Sens. 6: 1665-1670.

Elvidge C. D., and R. J. P Lyon. 1985. Influence of rock and soil spectral variation on the assessment of green biomass. Remote Sens. Environ. 17: 265-279.

Goetz, A. E H., B. N. Rock, and L. C. Rowan. 1983. Remote sensing for exploration: an overview. Econ. Geol. 78: 573-590.

Goetz, A. E H., G. Vane, J. E. Solomon, and B. N. Rock. 1985. Imaging spectrometry for earth remote sensing. Science 228: 1147-1153.

Griepentrog, T. E., and D. P. Groeneveld. 1981. The Owens Valley Water Management Report. County of Inyo, Bishop, CA.

Groeneveld, D. R. C. D. Elvidge, and D. A. Mouat. 1986. Hydrologic alteration and associated vegetation changes in the Owens Valley, California. In Proceedings of the Arid Lands Today and Tomorrow International Arid Lands Research and Development Conference. University of Arizona Press, Tucson, in press.

Huete, A. R., R. D. Jackson, and D. E Post. 1985. Spectral response of a plant canopy with different soil backgrounds. Remote Sens. Environ. 17: 37-53.

Jackson, R. D. 1983. Spectral indices in n-space. Remote Sens. Environ. 13: 409-421.

Jackson, R. D., and P J. Pinter. 1986. Spectral response of architecturally different wheat canopies. Remote Sens. Environ., in press.

Jennrich, R., and P Sampson. 1983. Stepwise discriminant analyses. Pages 519-537 in W. J. Dixon, ed. BMDP Statistical Software. University of California Press, Berkeley.

Justice, C. O., and J. R. G. Townshend. 1981. Integrating ground data with remote sensing. Pages 38-58 in J. R. G. Townshend, ed. Terrain Analysis and Remote Sensing. Allen & Unwin, London.

Kaufman, Y.J. 1985. The atmospheric effect on the separability of field classes measured from satellites. Remote Sens. Environ. 18: 21-34.

Kauth, R.J., P E Lambeck, W. Richardson, G. S. Thomas, and A. P Pentland. 1978. Feature extraction applied to agricultural crops as seen by Landsat, vol. ll. Pages 701-721 in Proceedings of Technical Sessions, LACIE Symposium. NASA Johnson Space Center, Houston, TX.

Kauth, R. J., and G. S. Thomas. 1976. The tasselled cap—a graphic description of the spectral-temporal development of agricultural crops as seen by Landsat. Pages 4B.41-51 in Proceedings of the 2nd International Symposium on Machine Processing of Remotely Sensed Data. Purdue University, West Lafayette, IN.

Pinter, PJ., R. D. Jackson, C. E. Ezra, and H. W. Gausman. 1985. Sun-angle and canopy architecture effects on the spectral reflectance of six wheat cultivars. Int. J. Remote Sens. Environ. 6: 1813-1825.

Richardson, A. J., and C. L. Wiegand. 1977. Distinguishing vegetation from soil background information. Photogramm. Eng. Remote Sens. 43: 1541-1552.

Siegal, B. S., and A. R. Gillespie. 1980. Remote Sensing in Geology. John Wiley & Sons, New York.

Smedes, H. W. 1975. The truth about ground truth. Pages 821-823 in Proceedings of the 10th International Symposium on Remote Sensing of Environment. Environmental Research Institute of Michigan, Ann Arbor.

Teensma, P D. A. 1981. Sagebrush (Artemisia tridentata Nutt.) and fire in Owens Valley, California. M.A. thesis. University of Oregon, Eugene.

Toll, D. L. 1985. Effect of Landsat thematic mapper sensor parameters on land cover classification. Remote Sens. Environ. 17: 129-140.

Tucker, C. J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8: 127-150.

Tueller, P T. 1980. Remote sensing for range management. Paper presented at the Remote Sensing for Resource Management Conference, Kansas City, MO, 23-30 October. US Soil Conservation Service and NASA.

Ustin, S. L., B. N. Rock, and R. A. Woodward. 1985. Analysis of substrate and plant spectral features of semi-arid shrub communities in the Owens Valley, California. Pages 347-359 in Proceedings of the International Symposium on Remote Sensing of Environment Fourth Thematic Conference: Remote Sensing for Exploration Geology. Environmental Research Institute of Michigan, Ann Arbor.

. 1986. Use of remote sensing techniques in the analysis of semi-arid shrub communities. In Proceedings of the White Mountain Research Station High-Altitude Symposium. White Mountain Research Station, University of California, Los Angeles, in press.

Vaughn, D. E. 1983. Soil inventory of the Benton-Owens Valley area: parts of Inyo and Mono counties, California. BLM CA 83009-7111. Bureau of Land Management, Bishop, CA.

Wiegand, C. L. 1984. The value of direct observation of crop canopies for indicating growing conditions and yield. Pages 1551-1555 in Proceedings of the 18th International Symposium on Remote Sensing of Environment. Environmental Research Institute of Michigan, Ann Arbor.

1998, Center for Spatial Technologies and Remote Sensing (CSTARS)
University of California, Davis