ASPRS/ACSM (1994), copyright ASPRS/ACSM
This study relates Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) imagery to field measurements of corn and soybean residue and green crop covers in an Indiana study area. The primary research objective is to identify the types and amounts of field residue covers in order to provide vital data as input into models that focus on soil erosion such as the Revised Universal Soil Loss Equation (RUSLE). A 126-band subset of AVIRIS data was developed for research by excluding bands too severely affected by water absorption to be useful. Representative spectral curves of bare soils and soils containing varying amounts of corn or soybean residues and green vegetation were analyzed. Vegetation and soil indices were applied to selected red, near infrared, and mid-infrared bands to delineate pixels with mixed composition into their component parts. Pearson regression analyses (derived from indexed images) were used to identify the percentages of green and soil cover separately in study site fields. The R-squared values for these features were .99 and .88 respectively. Estimated percentages of green and soils were calculated using separate regression equations. The differentiation between corn and soybeans residues was determined from an index in which corn had distinctly higher values than soybeans. The predicted amounts of soil, residue amount (and types), and green vegetation were within 10 percent of known values from field measurements.
Understanding the /Spatial distribution and characteristics of soils throughout this century has been of increasing special interest to soil geographers, soil scientists, environmentalists and others, especially with the growing need for soil erosion detection and control. The accelerated development and use of remote sensing during the past three decades has provided a good tool for improving our understanding of the /Spatial distribution and characteristics of soils and land resources.
It is well known that crop residue and crop canopy cover are major factors
[End Page 731]
affecting soil erosion by their influence on raindrop impact and storm runoff (Khan et al., 1987; Bechman, 1991; Laflen et al., 1991; Moldenhauer et al., 1983; Stott et al., 1990). Both the Universal Soil Loss Equation (USLE) (Wischmeier and Smith, 1978) and its revised version (RUSLE) (Kenneth et al., 1991) consider crop residue cover conditions as an important subfactor. Many studies have been conducted on detecting the different /Spatial and spectral characteristics associated with field residue cover, living biomass, and soil using remotely sensed data (Gausman et al., 1975, 1977; Mustard 1993; Gamon et al., 1993; Zhuang et al., 1991; McNairn et al., 1990; Roberts et al., 1991; Brisco et al., 1991). However, most of these analyses have been qualitative.
In 1987 the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) was introduced for multipurpose scientific uses. The AVIRIS is a National Aeronautics and Space Administration (NASA) sponsored Earth-observing imaging spectrometer designed, built, and operated by the Jet Propulsion Laboratory (JPL). These imaging spectrometers provide spectral coverage over the visible and near-infrared wavelength regions from 0.40 to 2.45 micrometers in approximately 10 nm intervals (9.8 nm) for a total of 219 bands. Data were acquired with about 18 m by 18 m /Spatial resolution from an altitude of 20 km (Chrien et al., 1991). This hyperspectral data set with its high spectral resolution may be used to identify specific features on the earth surface such as soils with different field surface cover composition.
The use of selected AVIRIS-derived bands with an appropriate methodology may assist in accurate delineation of soil features for agricultural and nonagricultural applications. One of the more important and difficult agricultural applications which may require or benefit from using AVIRIS data is the delineation of differences in quantity and/or quality of surficial soil crop residues. This type of information is used as vital input into modified versions of the universal soil loss equation and other types of erosion modeling that focuses on the impacts of conservation tillage on soil loss.
Even though AVIRIS data may have the potential to make important advances in classification of spectrally complex features or identification of absorption/reflectance data which leads to better feature discrimination, as yet there is little documentation of methods by which these advances can be accomplished. Hence basic research in understanding and analyzing energy-matter spectral interactions in soil-vegetation (living and/or dead) complexes is needed. This research includes band reduction of hyperspectral data, selection of the best band(s) for differentiating features of interest, development of new or use of established data transformations to highlight or enhance particular features (i.e. green and dead crop covers), and development of effective methods to utilize 12 bit digitization of AVIRIS spectral responses. The focus of this research is to develop and implement methods suitable to delineate the types and amounts of field residue covers in a northern Indiana study site in order to provide data suitable as input into soil erosion models.
[End Page 732]
A one square mile study site was selected in Tippecanoe County, Indiana. It is located in the northwest part of the county on a nearly level loess-mantled Wisconsin age glacial till plain. The native vegetation was prairie. Five major soil map units which contain both Alfisols and Mollisols occur in the sample fields (Table 1). The residue cover types in the site were mainly corn, soybean, and wheat. This study concentrated on corn and soybean residues only because they were the dominant cover types in the study site.
Six different types of data were developed or acquired for use in this research. These data include:
The methods developed in this study included band reduction, ground truth data collection, development and implementation of different AVIRIS data transformations, regression analyses, and the spectral differentiation of corn and soybean field residue covers. Data processing was conducted using MULTISPEC software (Landgrebe and Biehl, 1993) on a Mac Quadra 800 microcomputer.
Visual image interpretation of the 219 AVIRIS bands was conducted and a 126 band subset was used that was free from excessive noise or atmospheric water absorption. Seven spectral curves of representative soil, crop residue, and green vegetation features were selected and plotted using the 126-band data set (Figure 1). These features included: bare darker colored Mollisol soils; bare lighter colored Alfisol soils; approximately 50 percent soybean residue with a) darker and b) lighter soils; approximately 50 percent corn residue with a) darker and b) lighter soils; and 100 percent green vegetation cover. These curves provided insight into structure of the spectral distribution and degree of spectral separability among the selected ground features in the various AVIRIS bands.
[End Page 733]
Maximum spectral divergence of all seven features was found in band 27 (centered at 841 nm), thus it was used initially for study in this preliminary analysis.
A soil map of the study site was digitized based on the Tippecanoe County, Indiana Soil Survey. The digitized map units representing soil series and descriptive data for each series were used as base soil data. Green and dead crop measurements for each field in the study site were collected by a team of agronomists at Purdue University using a line-transect method. This method basically used a cord with 50 or 100 equally-spaces bead or knot markers, stretched diagonally across the rows in a field, and coincidences of the markers and pieces of crop residue on the soil surface were visually counted (Morrison et al., 1993).
A Normalized Difference Vegetation Index (NDVI) was applied using red (centered at 687 nm) and near infrared (centered at 783 nm) bands. A similar equation format was also applied on two middle infrared bands (centered at 2103 nm and 2192 nm). In both cases the bands selected were determined from analysis of the shape of spectral curves of soil and crop residues in which maximum separability was the primary consideration. The NDVI image (Figure 2) reduced the effects of ground residue cover features and enhanced only the distributions of green vegetation. The soil indexed image (Figure 3), on the other hand, reduced the effects of both green vegetation and crop residue, and concentrated mainly on the distributions of soils, so it was designated as a soil index. Averaged digital numbers (DN) were derived from both indexed images from selected fields with known cover conditions (types and percentages of green and dead crops). Pearson regression analyses were conducted between derived DN values and the corresponding percentages of green vegetation and soil cover. The R-squared values of these vegetation and soil indices were calculated. The estimated percentages of green vegetation and soils were also calculated based on the regression equations from the two sets of transformed data (Tables 2 and 3, Figures 4 and 5).
AVIRIS band 27 (centered at 841 nm) was masked and combined with a digitized soil map, and two images were produced and designated as: AVIRIS band 27 dark soil and AVIRIS band 27 light soil. The dark soil was defined as Munsell color 10YR 2/1 or lower and the light soil as 10YR 3/1 or higher (The Munsell color designation for each soil series was acquired from Tippecanoe County Soil Survey). Mean values for fields with different residue covers were listed using AVIRIS band 27 dark and AVIRIS band 27 light images separately (Table 4). The DN values per percent of soybean and corn residues were generated based on the following equation:
[End Page 734]
DN(mixed) = DN(green) * %(green) (1)
+ DN(soil) * %(soil)
+ DN(residue) * %(residue)
Where: DN(mixed) is the mean DN values of the composite reflectance of soil, green vegetation, and crop residue; DN(green) and DN(soil) are the highest values on both vegetation and soil indexed images; %(green) and %(soil) were derived from statistical estimation (Tables 2 and 3); and %(residue) was derived from subtracting percent of green and percent of soil from the 100 percent total. Differentiation of corn residue from soybean residue was determined by the DN(residue) values which are unique to each crop residue type.
Results in this study showed that the two indexed images were critical in determining the percentage of residue cover in the field. Between them, NDVI is a well-known equation, and it has been applied in many studies. The key part of this particular application on AVIRIS was to select the most representative red and near infrared bands, and this was easily done by plotting and examining spectral curves of selected ground features. Soil indices, on the other hand, were generated using two middle infrared bands. Due to the timing of the AVIRIS data (June, 1992), both corn and bean residues had been well-blended with the soils since the last harvest season. The condition of these residues apparently was not conducive to highlight the known mid-IR cellulose and lignin absorption bands identified in some other studies (Roberts et al., 1993). This wavelength region (2103 nm - 2192 nm) was the only region where the spectral difference between soil and residue was revealed no matter how subtle was the difference (Figure 6). Therefore, NDVI and an equation with elements of a NDVI format were generated and the derived R-squared values for green vegetation and soil indices were 0.99 and 0.88 respectively, which confirmed that the indexed images were reasonable and helpful in determining the amount of green vegetation and crop residue covers in this study site. The predicted amounts of soil, residue amount, and green vegetation were within 10 percent of known values from field measurements.
Results from equation 1 (DN(mixed)] indicated that the corn residue DN values per percent of their occurrence were greater than 20, while the soybean residue DN values per percent of their occurrence were lower than 20. These results were confirmed by Figure 1 where corn residue always has higher spectral reflectance than soybean residue in the same soil color background environment. The equation was applied to both dark colored and light colored soils, and the output results were mostly consistent with only about 2 DN values per percent different from one soil to the other.
The above discussions may be explained more clearly by the following examples (assume dark colored soil DN values were used):
[End Page 735]
1. The data needed for the calculation are:
DN(mixed) = 1843 (from Table 4 in field 6) DN(green) = 3154 (from Figure 2) DN(soil) = 1851 (from Figure 3) %(green) = 2.36 (from Table 2) %(soil) = 89.54 (from Table 3) %(residue) = 8.10 (from 100% - %(soil) - %(green))
Thus the equation used is:
DN(residue) = DN(mixed)-DN(green)*%(green)-DN(soil)*%(soil)/100
_____________________________________________
%residue
= 1843-(3154*2.36%)-(1851*89.54)/100
______________________________
8.10%
= 13.76 DN per percent
The ground truth information confirmed that the residue cover type in this field was soybean.
2. The data needed for the calculation are
DN(mixed) = 2233 (from Table 4 in field 2) DN(green) = 3154 (from Figure 2) DN(soil) = 1851 (from Figure 3) %(green) = 7.72 (from Table 2) %(soil) = 37.02 (from Table 3) %(residue) = 44.74 (from 100% - %(soil) - %(green))
Thus the equation used is:
DN(residue) = DN(mixed)-DN(green)*%(green)-DN(soil)*%(soil)/100
_____________________________________________
%residue
= 2233-(3154*7.72%)-(1851*37.02%)/100
_______________________________
44.74%
= 29.14 DN per percent
The ground truth information confirmed that the residue cover type in this field was corn.
June 1992 AVIRIS data were applied in a study site in Tippecanoe County, Indiana, to delineate different amounts and types of crop residue cover
[End Page 736]
conditions in Mollisol and Alfisol soils. The original 219 bands were reduced to the best 126 bands based on visual interpretation of information context in individual bands. Vegetation and soil indices were used and Pearson regression analyses were performed to ensure the regression was significant. The results showed that the R-squared values for vegetation and soil indices were 0.99 and 0.88 respectively, and the predicted percentages of soil, green vegetation, and residue in test fields were within 10 percent of known values from field measurements. The different types of crop residue covers (corn and soybean in this study) were successfully differentiated spectrally using methods developed in this research. The results derived from implementation of these methods suggest a possible approach to estimating the percentage of crop residue cover and also the different spectral characteristics of corn and soybean residues without the aid of laboratory measurements. However, these results are site-specific and maybe valid only for a narrow range of soils. Thus the methods used must be tested in other study sites with a different range of soil backgrounds and percentages of corn and soybean residue covers in order to validate them for widespread applications. This will become a main goal in the further research.
The authors wish to thank the following institutes and people for their support and help to this research: The Jet Propulsion Laboratory at the California Institute of Technology for providing the AVIRIS data and Purdue University agronomists for collecting ground truth information; and Mr. Larry Biehl in the Department of Electrical Engineering, Purdue University for technical support.
Bechman, T. 1991, More residue cover needed to meet compliance rules. Prairie Farmer, January 15, 1991, pp: 10-12.
Brisco, B., R. Brown, S. Snider, G. Sofko, J. Koehler and A. Wacher. 1991, Tillage effects on the radar backscattering coefficient of grain stubble fields. Int. J. of Remote Sensing, 12(11): 2.2283-2298.
Chrien T., M.L. Eastwood, C.M. Sarture, R.O. Green, and W.M. Porter. 1991, Current instrument status of the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). The Third AVIRIS Workshop, Pasadena, California. pp. 302-313.
Gamon, J.A., C.B. Field, D.A. Roberts, S.L. Ustin, and R. Valentini. 1993, Functional patterns in an annual grassland during an AVIRIS overflight. Remote Sensing of Environ., 44:239-253.
Gausman, H., R. Learner, J. Noriega, R. Rodrignez, and C. Wiegand. 1977, Field-measured spectroradiometric reflectance of disked and nondisked soil with and without wheat straw. Soil Sci. Soc. Am. J., 41: 793-796.
[End Page 737]
Gausman, H., A. Gerbermann, C. Wiegand, R. Leamer, R. Rodriguez, and J. Noriega. 1975, Reflectance differences between crop residues and bare soil. Soil Sci. Soc. Am. Proceedings, 39: 752-755.
Kenneth G.R, G.R. Foster, G.A. Weesies, and J.P. Porter. 1991, RUSLE -Revised universal soil loss equation. Journal of Soil and Water Conservation, 46(1): 30-33.
Khan, M.J., E.J. Monke, G.R. Foster. 1988, Mulch cover and canopy effect on soil loss. Transactions of the ASAE, 31(3): 706-711.
Laflen, J.M., L.J. Lane and G.R. Foster. 1991, WEPP: A new generation of erosion prediction technology. Journal of Soil and Water Conservation, 46(1): 34-38.
Landgrebe, D., and L. Biehl. 1993, An Introduction to MULTISPEC, Purdue University, West Lafayette, Indiana.
McNairn, H. and R. Protz. 1990, Evaluation of the use of Satellite imagery to measure crop residue cover on fields in the Lake Erie watershed. Proceedings of Nonpoint Source Planning, Chicago, Illinois.
Moldenhauer, W.C., G.W. Langdale, W. Frye, D.K. McCool, R.I. Papendick, D.E. Smika, and D.W. Fryrear. 1983, Conservation tillage for erosion control. Journal of Soil and Water Conservation, 38(3): 144-151.
Morrison, J.E., C. Huang, D.T. Lightle, and C.S.T. Daughtry. 1993, Residue measurement techniques. Journal of Soil and Water Conservation. 48(11): 479-483.
Mustard, J.F. 1993, Relationship of soils, grass, and bedrock over the Kaweah serpentinite melange through spectral mixture analysis of AVIRIS data. Remote Sensing of Environ., 44:293-308.
Roberts, D.A., M.O. Smith, J.B. Adams. 1993, Green vegetation, nonphotosynthetic vegetation, and soils in AVIRIS data. Remote Sensing of Environment, 44: 255-269.
Roberts, D.A., M.O. Smith, J.B., Adams, and A.R. Gillespie. 1991, Leaf spectral types, residuals, and canopy shade in an AVIRIS image. The Third AVIRIS Workshop, Pasadena, California. pp. 43-50.
Stott D.E., H.F. Stroo, L.F. Elliott, R.I. Papendick, and P.W. Unger. 1990, Wheat residue loss from fields under no-till management. Soil Sci. Sci. Am. J. 54(1): 92-98.
[End Page 738]
Ulrich, H.P., T.E. Barnes, and B.A. Krantz. 1959, Soil Survey of Tippecanoe County, Indiana. Soil Conservation Service, U.S. Department of Agriculture. Series 1940, No. 22. Washington, D.C.
Wischmeier, W. H. and D.D. Smith. 1978, Predicting rainfall erosion losses. Agr. Handbook. No. 537. U.S. Dept. Agr., Washington, D.C.
Zhuang X., B. Engel, M. Baumgardner, and P. Swain. 1991, Improving classification of crop residues using digital land ownership data and Landsat TM imagery. Photogrammetric Engineering and Remote Sensing, 57 (11): 1487-1492.
[End Page 739]
[End Page 740]
[End Page 741]
[End Page 742]
[End Page 743]