Soil spectral effects on 4-space vegetation discrimination

REMOTE SENSING OF ENVIRONMENT 15:155-165 (1984)

155

Soil Spectral Effects on 4-Space Vegetation Discrimination A. R. HUETE and D. F. POST Department of Soils, Water and Engineering, University of Arizona, Tucson, Arizona 85721

R. D. JACKSON USDA-ARS, U. S. Water Conservation Laboratory, Phoenix, Arizona 85040

The influence of soil background on vegetation discrimination in four-band reflectance space was examined. Dry and wet reflectance data were obtained for 20 soils covering a wide range in spectral properties with a hmad-held radiometer. Principal components analysis was used to study the distribution of soil spectra in 4-space and to define a mean soil line. Soil-specific background lines were similarly derived and used to examine the overall cloud of soil spectra in individual soil form. Reflectance data from a rid.l-canopy wheat plot were used to compute unit vector coefficients in the greenness direction from the mean soil line and from the individual soil lines. Analysis of the mean soil line showed that it was not possible to discriminate bare soil from low vegetation densities. Greenness measurements were shown to be sensitive to both soil type and soil moisture condition. In contrast, the use of individual soil lines as a base to measure greenness minimized soil background influence and improved vegetation assessment, particularly at low green plant canopy covers.

Introduction The study of green vegetation from satellite multispectral scanner observations is hampered by the influence and variability of soil background. The problem is particularly noticeable in arid and semiarid environments where there has been little success in extracting vegetation information from areas with less than 30% green canopy cover (Hutchinson, 1982). Canopy spectral responses can vary with the amount of soil exposed, soil color, soil moisture and cultural practices (Colwell, 1974; Rao et al., 1979; Jackson et al., 1979; Kollenkark et al., 1982). As a result, much effort has been devoted to the development of multispectral band indices that can isolate green vegetation from soil background. Kauth and Thomas (1976) and Kauth et al. (1979), in a study of the distribution of soil spectra in four-dimensional LANDSAT MSS signal space, found most of the variability of bare soil signals to be attributed to brightness as nearly all spec~JElsevier Science Publishing Co., Inc., 1983 52 Vanderbilt Ave., New York, IVY 10017

tral data fell along a line extending from the origin. Soft spectra have been shown to vary with surface moisture content, organic matter content, particle size distribution, soft mineralogy, soil structure, surface roughness, crusting and presence of shadow (Angstrom, 1925; Bowers and Hanks, 1965; Stoner and Baumgardner, 1981). The 4-space diagonal line, describing maximum variations in soft spectral response, has been named the "soil line" or soft brightness vector (Kauth and Thomas, 1976; Richardson and Wiegand,

1977). The emergence of green vegetation over a soil causes composite red radiance to decrease because of chlorophyll absorption and overall infrared response to increase as a result of leaf mesophyll structure. Thus, deviations of spectral data from the bare soft line, in an appropriate direction, may be attributed to the presence of green biomass. Richardson and Wiegand (1977), using infrared and red band combinations, developed a two-dimensional perpendicular 0034-4257/83/$3.00

156

HUETE, POST AND JACKSON

vegetation index (PVI) with the orthogo- the soft line does not match that derived nal distance of a spectral point to the soft by Kauth and Thomas (1976), greenness line as the measure of vegetation density. may become sensitive to soft background Kauth and Thomas (1976) developed a and vegetation analysis becomes less religreen vegetation index (GVI) as the four- able. Furthermore, bare soft spectral dimensional perpendicular distance of a deviations from the soft line in the soil spectral point to the soft line. They com- brightness-plant greenness plane may inputed linear combinations of the four MSS terfere in the discrimination of green bands (Tasseled Cap Transformation) to vegetation from bare soft. Soft noise may enable projection of 4-space spectral data be of little concern to the current onto a plane defined by a soft brightness LANDSAT System; however, in future and a plant canopy "greenness" axis. In sensor systems, especially those that may such a plane, soft and plant spectral be- be more agriculhtre-intensive, soft spectral havior are least correlated to one another, behavior must be better understood if we and plant greenness measurements are are to improve our knowledge of cominsensitive to soft background. Kanth and posite soft-plant canopy systems. Thomas (1976) transformed Condit's soil The purpose of this research was to reflectance data (Condit, 1970) into demonstrate the nonconstant behavior of LANDSAT MSS signal space and dis- soft background when a single soft line, covered a second component of soft re- composed of softs with widely ranging flectance attributable to red softs. Since spectral properties, is used as a base to this component was orthogonal to both measure greenness. Improvements in the soft line and greenness, they sug- vegetation discrimination with the use of gested the concept of a "plane of softs" soft-specific background lines are diswith vegetation growing outward from cussed. the plane. The soft line concept has become widely Experiment and Calculations accepted in the interpretation and analyTwenty softs, with a wide range in sis of multispectral data (Fukuhara et al., physical and chemical properties, were 1979; Jackson et al., 1980; Thompson and Wehmanen, 1980; Wiegand and used (Table 1). The softs ranged from the Richardson, 1982). Although the soil line black, organic-rich Cloversprings loam and may shift with atmospheric conditions and dark brown Comoro gravelly loamy sand, LANDSAT calibration factors, it has gen- to the dark red, high iron Whitehouse (B) erally been assumed that there exists one sandy clay loam and the highly reflective unique soft line encompassing a wide bright silica sand and yellowish brown range of soft types and soft surface condi- Superstition sand. tions. Jackson et al. (1980), however, reThe softs were air-dried, passed through ported deviations from linearity when a a 19-mm sieve, and placed inside 50 cm × wide spectral range of softs were plotted. 50 cm × 25 cm wood boxes. Two boxes Such deviations among soil types could were used for each soft, and all boxes shift the slope of the soft line and alter the were positioned outdoors on a support orthogonal distance of a vegetation spec- frame at the Campbell Avenue Experitra to the soil line. Thus, in areas where mental Farm, Tucson, Arizona. Between

SOIL SPECTRAL EFFECTS ON 4-SPACE VEGETATION DISCRIMINATION

157

TABLE 1 Physicaland Chemical Characteristics of the 20 Test Soils PA.ltTICLESIZE DISTRIBUTION

SoIL SV.xatv.s~ Cloversprings Comoro Holtville Whitehouse (B) Whitehouse (A) Avondale Grabe Pinaleno Vint Agua Pima Brazito Contine Laveen Pimer Hayhook Mohave Gila Superstition Silica sand

TZXTOa.CL CLASS loam loamy coarse sand silty clay sandy clay loam loamy coarse sand loam loam sandy loam loamy fine sand fine sandy loam loam sandy loam sandy clay loam loam silt loam sandy loam sandy clay loam very fine sandy loam sand coarse sand

% % % % % OIIGANIC EXT. % GaAVEL CLAY SAND M^TrFm IRON CaCOa

0 28 0 13 13 0 0 0 0 11 0 12 2 5 0 12 4 6 0 0

21.1 4.5 40.6 33.3 7.2 22.1 20.3 7.1 3.9 11.5 26.2 9.6 25.6 19.2 19.8 4.8 22.9 9.9

40.3 84.2 8.7 48.6 79.3 40.1 41.2 71.4 82.4 59.5 35.5 68.2 52.2 46.3 20.9 69.9 59.4 53.7

5.7 2.2 1.7 1.5 1.5 0.9 1.6 1.3 0.5 1.3 1.1 1.3 0.6 1.O 1.0 0.4 0.7 1.5

1.8 1.5 0.9 2.5 1.5 0.8 0.7 0.9 0.8 0.5 0.7 0.4 0.7 0.8 0.6 0.9 0.6 0.5

0 0 10 0 0 5 5 0 8 5 4 4 3 9 15 0 2 6

1.6 95.9 0.0 100.0

0.2 0.0

0.2 0.0

3 0

MLrNSELLCOLOR DRY

MOIST

10YR 3/2 10YR2/1 10YR 4/2 10YR2/2 10YR 5/3 10YR3/3 2.5YR 4/6 2.5YR3/6 5YR 5/6 5YR 4/4 7.5YR 5/4 7.5YR4/4 10YR 5/2 10YR3/2 7.5YR 5/4 7.5YR4/4 10YR 5/4 1OYR4/4 10YR 5/3 10YR4/3 10YR 5/3 1OYR3/3 10YR 5/3 10YR4/3 7.5YR 5/4 7.5YR4/4 7.5YR 5/4 7.5YR4/4 10YR 6/4 1OYR4/4 7.5YR 4/4 7.5YR4/4 7.5YR 5/4 10YR4/4 10YR 6/3 10YR4/3 10YR 7/4 10YR 8/1

10YR5/4 10YR7/1

aSoil series are arranged in order of their "brightness." All laboratory analyses were completed by the Soil Conservation Service National Soil Survey Laboratory, Lincoln, Nebraska. The laboratory methods and procedures are presented in Soil Survey Investigations Report No. 1, prepared by the soil survey staff.

experiments, the boxes were covered with polyethylene bags to minimize dust contamination. Reflected solar radiation was measured with a portable Exotech Model 100A 1 radiometer with spectral bandpass intervals similar to LANDSAT 4-MSS bands 1, 2, 3, and 4 (0.5-0.6 #m, 0.6-0.7 #m, 0.7-0.8 #m, and 0.8-1.1 #m, respectively). Measurements were taken over all soils in a dry state and at numerous drying stages after 2.5 cm water was added. The radiometer was hand-held so as to obtain a nadir view with a 15 o field-of-view lens from a height of 1.5 m above the soil 1Trade names and company names are included for the benefit of the reader and do not constitute an endorsement by the University of Arizona or the U. S. Department of Agriculture.

surface. The experiment was conducted 16-18 June and 19-23 July 1982 between the hours 0930-1130 and 13301530 MST. Only data from cloud-free days are reported here. Reflectance was calculated by ratioing the average of eight target radiance measurements to the irradiance. Irradiance measurements were taken from a standard BaSO4 plate approximately every 10 min. Measurements were first taken over all soils in a dry condition throughout the day during the hours mentioned. In subsequent days, one set of replicate soils were wetted in the morning, and readings were taken over all soils, dry and wet, throughout the day. The wetting and drying cycles of the replicate soils continued throughout the experiment while the other

158

HUETE, POST AND JACKSON

set of softs remained dry at all times. Readings were made on all softs during each measurement period which lasted approximately 20 min. Thirty-five sets of spectral measurements were collected per soft in each of the four bands. A principal components analysis was performed to determine the overall distribution of soft spectral data in four dimensions, and the unit eigenvectots of the first component were used to define the direction of the mean soft line. Unit vector coefficients defining greenness were calculated orthogonal to the mean soft line following an algebraic procedure described by Jackson (1983). Green wheat spectra taken over an entire season with the same Exotech radiometer (data reported by Jackson et al., 1983) were used to establish a relationship between greenness and percent green wheat canopy cover. However, only full canopy wheat spectra (no soft showing) was used to compute the unit greenness vector. Thus, the soil brightness-plant canopy greenness plane was defined by the first principal component of bare soil spectra and a green wheat spectral point. Principal components analyses were similarly used on each soft type to calculate individual, soil-specific background lines. The same fiill canopy green wheat point was used to define unit greenness vectors orthogonal to each individual soft line. Consequently, separate soft brightness-wheat canopy greenness planes were derived for each soil type. This was nec-

essary in order to (1) align individual soft lines parallel to the soft brightness axis, (2) maintain brightness as the axis of principal soft variation, and (3) bring out maximum variation between soft and vegetation spectra.

Results A positive linear relationship of soft spectral data was found among all four bands (Table 2). Band 2 (red) correlations with the two infrared bands were very high while band relationships involving band 1 were least correlated. A trace view of the distribution of soft data from the band 4 vs. band 2 plane is plotted in Fig. I, demonstrating a two-dimensional soil line.

Principal components analysis of all soft spectra verified the existence of a soft line in 4-space (Table 3). 98% of the soft spectral variance was explained by the first soil component while the second principal component, representing the greatest soil spectral variance orthogonal to the soil line, accounted for an additional 1.6% of the variance. In four-dimensional Euclidean space, the distribution of soil spectra forms an elongated ellipsoid with a length 8 times as long as it is wide, a width 3 times as wide as it is thick, and in a fourth dimension, a thickness 1.2 times as thick as thin. Thus, the width of the cloud of soil spectra varies from a maximum of 11 units to a minimum of 3 units (Table 3).

TABLE 2 Correlation Matrix of Soil Reflectance Data

Band 1 Band 2 Band 3 Band 4

BAND 1

Ba~ro 2

BAND3

BAND4

1.000

0.968 1.000

0.947 0.990 1.000

0.954 0.989 0.993 1.000

SOIL SPECTRAL EFFECTS ON 4-SPACE VEGETATION DISCRIMINATION

159

60-

.¢50-

°o

-o ¢ 40-

¢].:',:~ff

qr 3 0 ¢

..i:4:~.~'"

20-

slope 1.166 intercept 4.20 R-squared 0.98 data pairs 704

~-':"

Ib

2'o

sb

4'o

s'o

6b

Band 2 (red)

FIGURE 1. TABLE 3 Spectra

Trace view of mean soil line in the band 4 versus band 2 plane.

Principal Components Analysis for All Soil PERCENTAGE OF

COMPONENT

1 2 3 4

VARIANCE

98.0 1.6 0.2 0.1

AMPLITUDE

83.7 10.7 3.8 3.1

a

(length) (width) (thick) (thin)

aSquare root of eigenvalue.

Our results compare favorably with the analysis of Condit's data by Kauth and Thomas (1976). They described the 4space distribution of soil spectra as a flattened cigar-shaped ellipsoid with length seven times as long as wide and width two times as wide as thick. In both cases the first principal component is the soil line and the remaining components represent soil spectral deviations away from

the mean soil line. Table 4 shows a comparison between the mean soil line parameters presented here and those from Condit's data. The two soil lines are very similar and their slopes are identical in the band 4 versus band 2 plane. Spectral data from full canopy wheat were used to derive unit vector coefficients in the greenness direction, orthogonal to the mean soil line (Table 4). These coefficients, together with those defining the soil line, provide the appropriate transformation equations to plot 4-space spectral data onto a two-dimensional soil brightness-plant canopy greenness plane (Fig. 2). The width of the bare soil spectral data in this plane is critical to vegetation analysis because such deviations from the soil line could be falsely interpreted as green biomass.

TABLE 4 Brightness and Greenness Coefficients for the Mean Soil Line and Brightness Coefficients for Condit's Soil Line ~-SPACE COEFFICIENTS

Mean soil line Condit's soil line Greenness

A1

A2

0.428 0.398 - 0.397

2-SPACE a

A3

A4

SLOPE

0.476 0.486

0.532 0.532

0.555 0.567

1.166 1.165

- 0.606

0.165

0.669

--

a Band 4 versus band 2 plane. g4-space intersect that adjusts soft line to pass through origin.

INTERCEPT 0.042 -0.046 g

160

HUETE, POST AND JACKSON

strates the noise and uncertainty in discriminating bare soil from green wheat if an overall, mean soil line is used to measure greenness.

•wheat

-5o m 10

Skeletal soil cloud

-25 ...1~. o; ° * • -~: • "~.:Z~..''').. (

-10

•

2'o

"~.:'.....'~..~'.;..~

."

4b e'o Soil brightness

~..

s'o

1o o

FIGURE 2. Soil spectra in the mean soil brightnessgreenness plane.

The width of bare soil spectra along the greenness axis is approximately seven units, whereas the ftdl canopy green wheat point has a greenness of 28 units. A percent green canopy cover scale, assuming a linear relationship between greenness and percent green wheat cover, is included in Fig. 2. Such a relationship between greenness and green cover is illustrated in Fig. 3, utilizing simultaneous measurements of wheat spectra and percent green wheat cover. As can be seen, soil spectral variations along the green cover axis is equivalent to a percent green wheat cover range of 25%. This demon-

Figure 4 is the same plot as the previous figure with the cloud of soil spectra replaced by individual soil lines. The dry and wet reflectance data delineating each line are presented in Table 5. The skeletal view of soil spectra enables us to analyze the behavior of specific soils within the overall soil spectral cloud. The position along any individual soil line is dependent on the moisture status of that particular soil and the endpoints delineate the extreme dry and wet conditions. As can be seen in Fig. 4, most individual soil lines do not coincide with the mean soil line, and greenness measurements are therefore likely to vary with changes in soil moisture condition. When the horizontal mean soil line is used as a base to measure plant canopy greenness, bare soil spectra will only be interpreted as zero "green" at the particular moisture condition where an individual soil line

80-

60-

E o c 40-

20-

i 5

I 1

i 15

I 20

l 25

Greenness

FIGURE 3. Relationship between green canopy cover and greenness for wheat on Avondale loam soil.

30

SOIL SPECTRAL EFFECTS ON 4-SPACE VEGETATION DISCRIMINATION

12

1 C~oversprings 2. Comoro 3. Holtville

8. Superstition 9. Silica sand 10. Hayhook

-40

4. Whitehouse (A) 5 Whitehouse (8) 6 Avondale

11. Pimer 12. Mohave 13. Gila

-30

7. Vint

= et 0

-2o ;,

3

4 /

-

12 10 ,6

11 13

o t .... 3

10 mean s o i l l i n e /

g

0

. . . . . . . . . . . Ij

lO

2'o

4'o

6'o

do

,~o

Soil brightness

FIGURE 4. Individual soil lines in the mean soil brightness-greenness plane.

intersects the mean soil line. In general, individual soils would appear"greener" as their surface goes from a wet to a dry condition. Dry, red soils (4,5,10,12) would be interpreted as much "greener" than either dark or bright softs. Greenness measurements over the Whitehouse (B) sandy clay loam could vary by over 10% depending on soil moisture status. On the other hand, softs as diverse as the TABLE 5

161

Holtville silty clay and Superstition sand produced soil line slopes identical with that of the mean soil line. Consequently, ff the mean soil line is used, greenness measurements over these soils would appear independent of soil moisture condition; however, since their soil lines are located below the mean soil line, a 10% green wheat cover over these softs would be falsely interpreted as bare soft. In comparing Figs. 2 and 4, it is worth noting that the width of the overall cloud of soil spectra is determined by the dry and wet end points of individual soil lines. As a result, measurements taken over drying soils can be removed without affecting the width of soil spectra. Even if only dry soils were included, the width of the soil cloud would only be reduced from 7 to 6 units. Soil-specific background lines Increased sensitivity in vegetation assessment can be achieved if soil-specific

Dry and Wet Reflectance Data (%) Delineating Individual Soil Lines DRY

SOIL Cloversprings Comoro Holtville Whitehouse (B) Whitehouse (A) Avondale Grabe Pinaleno Vint Agua Pima Brazito Contine Laveen Pimer Hayhook Mohave Gila Superstition Silica sand

WET

B1

B2

B3

B4

B1

5.2 10.3 13.5 9.1 12.2 15.6 17.2 16.4 18.6 19.3 19.5 19.6 17.3 18.6 20.2 19.0 18.4 22.0 24.7 39.2

6.7 13.3 17.7 17.3 20.1 22.1 21.7 24.1 25.4 24.7 25.2 25.3 26.3 26.1 26.5 27.5 27.8 28.5 32.9 43.0

9.2 18.1 21.2 25.2 26.8 27.3 27.6 30.5 30.9 30.8 31.0 31.5 32.9 32.3 32.1 34.8 35.4 34.9 38.0 49.1

11.4 21.6 23.1 27.1 29.9 29.7 30.8 33.4 32.7 34.1 34.3 35.1 34.4 34.7 34.7 37.2 37.6 38.3 39.7 54.2

2.1 4.2 5.5 5.3 5.5 6.1 5.3 5.9 7.0 6.6 6.0 7.3 8.5 7.0 7.9 8.6 8.8 7.3 13.9 22.7

B2

B3

B4

2.6 6.1 8.1 10.6 10.5 9.9 7.8 9.9 10.7 9.4 8.9 10.4 14.5 11.1 11.4 13.1 14.8 10.7 20.1 25.6

3.6 8.9 10.5 15.4 15.0 13.5 11.1 13.9 13.8 13.5 12.5 14.4 18.8 15.1 14.4 17.2 19.6 15.5 23.6 30.2

5.0 11.5 12.0 16.4 17.4 15.4 13.8 16.1 14.8 15.9 15.4 17.2 19.8 16.9 16.6 19.1 21.5 18.0 25.0 34.3

162 TABLE 6

HUETE, POST AND JACKSON Soil Line Parameters for 20 Soils in 2- and 4-space, in Order of Increasing Brightness (SBI)a 4-SPACECOEFFICIENTS

SOIL Cloversprings Comoro Holtville Whitehouse (B) Whitehouse (A) Avondale Grabe Pinaleno Vint Agua Pima Brazito Contine Laveen Pimer Hayhook Mohave Gila Superstition Silica sand

2-SPACE(b4-b2)

SBI (DRY)

A1

A2

A3

A4

SLOPE

17.0 32.8 38.4 41.9 46.5 48.5 49.7 53.7 55.0 55.5 56.0 56.9 57.1 57.2 57.8 61.0 61.4 62.9 68.6 90.5

0.311 0.366 0.399 0.234 0.321 0.376 0.394 0.351 0.375 0.396 0.399 0.393 0.350 0.372 0.386 0.339 0.346 0.406 0.407 0.455

0.408 0.429 0.487 0.408 0.459 0.485 0.465 0.475 0.474 0.478 0.481 0.475 0.472 0.481 0.474 0.465 0.468 0.490 0.483 0.478

0.560 0.556 0.542 0.596 0.569 0.549 0.552 0.558 0.551 0.542 0.547 0.543 0.564 0.550 0.553 0.571 0.568 0.533 0.542 0.516

0.650 0.610 0.557 0.651 0.603 0.567 0.569 0.583 0.575 0.566 0.557 0.570 0.580 0.572 0.566 0.586 0.581 0.558 0.554 0.546

1.596 ' 1.426 1.146 1.597 1.316 1.169 1.222 1.226 1.213 1.185 1.157 1.198 1.226 1.190 1.195 1.261 1.242 1.140 1.148 1.143

INTERCEPT 0.7 2.7 2.8 - 0.6 3.5 3.9 4.2 3.9 1.8 4.8 5.1 4.7 2.1 3.7 3.0 2.5 3.1 5.8 1.9 5.0

aSoil brightness index = 5~4=1A iX % reflectance, band i.

background lines are used as a base to measure greenness. The unit vector coefficients defining the individual soil lines are summarized in Table 6. Soil spectral variances in the first two principal components for individual soils are shown in Table 7. In 14 of the 20 soils, over 99% of the spectral variance fell onto the first principal component, and, in comparison to the mean soil line, maximum spectral deviations away from individual soil lines (second principal component) were smaller. Consequently, the spectral deviations from the soil lines in their respective greenness planes were also smaller and ranged from a low of 0.7 units to a high of 3.4 units. The average width of individual bare soil spectra in the direction of greenness corresponded to a green wheat cover range of 7%, an improvement by a factor of 4 over that of the mean soil line. The width of a soil line in the direction of greenness determines a vegetation den-

sity limit below which vegetation cannot be discriminated from bare soil. Another source of variation which influences vegetation analysis at all density levels involves the slope and location of soil lines in 4-space, since it is the perpendicular distance from a green point to a soil line that is used to measure greenness. To demonstrate this influence the orthogonal distance from each soil line to a htll canopy wheat point was calculated (Table 7). The perpendicular distance or GVI to the full canopy green wheat point varied from 24 units to 30 units, depending on the selection of the soil line. Thus, a GVI of only 24 units corresponds to 100% green wheat cover over the Whitehouse (B) sandy clay loam, while a GVI of nearly 30 units would be needed to represent total green cover over the Holtville silty clay or Superstition sand. If these differences in GVIs from individual soil lines are normalized by the

SOIL SPECTRAL EFFECTS ON 4-SPACE VEGETATION DISCRIMINATION TABLE 7

163

Individual Soil Line Properties in Their Greenness Planes

SOIL

PRINCIPALa COMPONENT 1 2

Cloversprings Comoro Holtville Whitehouse (B) Whitehouse (A) Avondale Grabe Pinaleno Vint Agua Pima Brazito Contine Laveen Pimer Hayhook Mohave Gila Superstition Silica sand

99.4 98.5 99.5 97.7 99.0 98.9 99.6 99.7 99.8 99.5 99.2 99.7 98.4 99.3 99.8 98.6 99.2 99.3 98.7 99.7

0.4 1.0 0.3 1.5 0.7 0.8 0.3 0.1 0.1 0.4 0.7 0.2 1.3 0.5 0.2 1.0 0.6 0.5 0.8 0.2

SOIL LINE LENGTH

GREEN b LINE WIDTH

10.0 16.7 19.9 16.5 20.9 25.2 30.0 29.8 21.1 32.1 33.9 31.4 25.2 31.3 32.0 31.0 27.8 36.5 26.6 35.0

0.7 1.8 1.3 1.1 1.9 3.2 1.7 1.5 1.1 2.0 1.9 1.8 2.4 1.8 1.2 3.4 2.5 2.5 3.1 1.9

% GREEN LINE (WIDTH)

2.8 7.1 4.4 4.4 7.4 11.2 6.2 5.5 3.9 7.0 6.7 6.6 8.4 6.6 4.4 12.2 9.2 9.0 10.6 6.6

GVI c (WHEAT)

NORMALIZED PERCENT GREENd

25.4 26.4 29.9 24.2 25.7 28.1 27.8 27.0 29.1 27.8 28.0 27.7 28.0 27.9 28.8 27.6 27.1 27.5 29.9 29.2

90.0 93.8 106.1 85.9 91.2 99.9 98.8 96.1 103.3 98.6 99.3 98.3 99.3 99.1 102.4 98.0 96.2 97.6 106.1 103.7

a Percentage of variance.

bSoil spectral variations in the direction of greenness. ~Green vegetation index of htll-canopy wheat point. dNonnalized to GVI from mean soil line.

corresponding GVI from the mean soil line (28 units), one finds that the greenness of a single fltll-canopy wheat point varies from 86% to 106%. Consequently, separate soil-dependent greenness scales would have to be developed since the fiflly canopy wheat point, with no soil showing, is independent of soil background.

Discussion

At first glance the results of this research appear to confirm the existence of a single, universal soil line. 98% of the soil spectral variance was explained by a 4space soil line, even though the data consisted of soils with widely ranging spectral properties. With all soil spectra,

dry or wet, plotting near this line, it is easy to imagine a series of parallel, individual soft lines extending from end to end on the mean soil line. If this were the case, vegetation assessment would only be limited by bare soil spectral deviations in the direction of greenness and only low vegetation densities would be affected. The complex behavior of soil background on vegetation analysis became apparent by studying the skeletal, individual soil lines that made up the overall soft cloud. Most soft lines were not parallel but intersected the mean soil line at various inclinations and soil lines that were parallel varied in position in the mean soil-greenness plane. As a result, when the mean soil line was used, greenness measurements were sensitive to both soil type and soil moisture condition.

164

The influence of soil type would be critical in many wildland areas where numerous soils are often present in complex spatial patterns and vegetation densities are low. Development of local or soil-specific background lines would greatly enhance vegetation discrimination from bare soil. In agricultural areas, varying soil moisture conditions may limit greenness measurements, and soil-dependent background lines would be crucial in the isolation of vegetation development. The use of individual soil lines would not only reduce the influence of soil moisture on greenness measurements, but may also minimize those effects due to variations in soil brightness caused by crop canopy shadowing and soil tillage operations. In either case, the development of appropriate soil background lines is userdependent. In certain situations a universal, scene-independent background line is all that is needed for vegetation classification. In other situations, scene- or temporal-dependent background lines may be needed for improved vegetation discrimination. Finally, for more intense levels of vegetation analyses, multiple soil background lines may be developed based upon scene stratification schemes, landuse, or soil types. The level to which vegetation can be measured, however, may become limited by such factors as sensor noise, atmospheric conditions, and spatial and radiometric resolution. The orthogonal distance from a vegetation point to individual soil lines varied considerably and required the development of soft-dependent vegetation indices. Relative to a full canopy wheat point, some soft lines were located at greater orthogonal distances than others. An interpretation of this result is that certain soil materials have better contrast

HUETE, POST AND JACKSON

with greenness. The sharper the spectral contrast between green plants and soil, the easier it is to discriminate the two materials, and greenness sensitivity is improved. In this study, greatest spectral contrasts occurred between green wheat and Superstition sand and HoltviUe silty clay. The red Whitehouse (B) sandy clay loam provided least contrast. The use of an overall, mean soft line totally ignores contrast differences between soils and green vegetation. The variations in soft spectra presented here are also likely to affect other vegetation indices. The two-dimension PVI, although smaller in magnitude, behaves in a similar manner as the 4-space GVI. Whereas the full canopy wheat GVI varied from 24 to 30 units for separate soft lines, the corresponding PVI in the band 4 versus band 2 plane ranged from 22 to 27 units. Since nearly all soil lines intersected the band 4 axis above the origin (Table 6), the I R / R e d ratio changes as a soft goes from a wet to dry surface condition. Furthermore, bare soils with steeper band 4 versus 2 slopes possess higher IR/Red ratios. Thus the I R / R e d ratio may vary not only with green canopy cover, but also with soil type and soft moisture condition. In conclusion, the use of a mean soil line was found to hamper both the accuracy and sensitivity of vegetation analysis. The width of the mean soft line precluded greenness assessment below 25% cover. The slope and location of individual soil line components within the overall soil cloud rendered vegetation analysis highly sensitive to soil background influences. The use of individual soft lines reduced soil background sensitivity and increased the accuracy of greenness measurements, especially at low vegetation densities. This

SOIL SPECTRALEFFECTSON 4-SPACEVEGETATIONDISCRIMINATION

165

sphere, Remote Sens. Environ. 13:187-208. Kanth, R. J., and Thomas, G. S. (1976), The tasseled cap--A graphic description of the spectral-temporal development of agricultural crops as seen by Landsat in ProceedReferences ings of the Symposium on Machine Processing of Remotely Sensed Data, LARS, Angstrom, A. (1925), The albedo of various Purdue University, West lafayette, Indisurfaces of ground, Geografiska Ann. 7: ana, pp. 41-51. 323-342. Bowers, S. A., and Hanks, R. J. (1965), Re- Kauth, R. J., Lambeck, P. F., Richardson, W., Thomas, G. S., and Pentland, A. P. (1979), flection of radiant energy from soils, So/1 Feature extraction applied to agricultural Sci. 100: 130-138. crops as seen by Landsat, in Proceedings of Colwell, J. E. (1974), Vegetation canopy rethe LACIE Symposium, Johnson Space flectance, Remote Sens. Environ. 3: Center, Houston, Texas, pp. 705-721. 175-183. Kollenkark, J. C., Daughtry, C. S. T., Bauer, Condit, H. R. (1970), The spectral reflectance M. E., and Housley, T. L. (1982), Effects of of American soils, Photogramm. Eng. 36: cultural practices on agronomic and reflec955-966. tance characteristics of soybean canopies, Fukuhara, M., Hayashi, S., Yasuda, Y., et al. Agron. 1. 74:751-758. (1979), Extraction of soil information from Rao, V. R., Brach, E. J., and Mack, A. R. vegetated areas, in Proceedings of the Sym(1979), Bidirectional reflectance of crops posium on Machine Processing of Remotely and the soil contribution, Remote Sens. EnSensed Data, LARS, Purdue University, viron. 8: 115-125. West Lafayette, Indiana, pp. 242-252. Richardson, A. J., and Wiegand, C. L. (1977), Hutchinson, C. F. (1982), Remote sensing of Distinguishing vegetation from soil backarid and semi-arid rangeland, paper ground information, Photogramm. Eng. Represented at International Geoscience and mote Sens. 43: 1541-1552. Remote Sensing Symposium, Munich, Stoner, E. R., and Banmgardner, M. F. (1981), Federal Republic of Germany, June. Characteristic variations in reflectance of Jackson, R. D. (1983), Spectral indices in sudace soils, Soil Sci. Soc. Am. 1. n-space, Remote Sens. Environ. 45:1161-1165. 13:401-429. Thompson, D. R., and Wehmanen, O. A. Jackson, R. D., Reginato, R. J., Pinter, P. J., (1980), Using Landsat digital data to detect Jr., and Idso, S. B. (1979), Plant canopy moisture stress in corn-soybean growing information extraction from composite regions, Photogramm. Eng. Remote Sens. scene reflectance of row crops, Appl. Opt., 46: 1087-1093. 18: 3775-3782. Wiegand, C. L., and Richardson, A. J. (1982), Jackson, R. D., et al. (1980), Hand-held radiComparisons among a new soil index and ometry, ARM-W-19, U. S. Department of other two- and four-dimensional vegetation Agriculture, Science and Education Adminindices, in Proceedings of the 48th Annual istration, Oakland, California, 66 pp. Meeting of the American Society of PhotoJackson, R. D., Slater, P. N., and Pinter, P. J., grammetrg, pp. 211-227. Jr. (1983), Discrimination of growth and water stress in wheat by various vegetation Received 4 April 1983;revised26 October1983. indices through a clear and turbid atmo-

improvement was gained at the expense of having to calculate separate greenness indices for different soil lines.