Soil Science Society of America Journal 66:1922-1929 (2002)
© 2002 Soil Science Society of America
DIVISION S-5—PEDOLOGY
Estimation of Soil Organic Matter from Red and Near-Infrared Remotely Sensed Data Using a Soil Line Euclidean Distance Technique
Garey A. Fox*,a and
George J. Sabbaghb
a A205B Engineering, Dep. of Civil Engineering, Colorado State Univ., Fort Collins, CO 80523-1372
b Bayer Company, 17745 S. Metcalf, Stillwell, KS 66231 and Adjunct Faculty, Dep. of Agric. Engineering, Texas A&M Univ
* Corresponding author (gfox{at}engr.colostate.edu)
 |
ABSTRACT
|
|---|
The soil line is a well-known linear relationship between the near-infrared and red reflectance or image intensity of bare soil images. Remotely sensed estimations of soil surface properties can lead to improved representation of spatial heterogeneity. The objectives of this research are to develop an approach based on image soil lines that incorporates image intensity in the red and near-infrared bands for mapping surface organic matter (OM) and to provide guidance for soil sampling. The soil line concept is used to develop predictive relationships between the amount of OM within the surface horizon of the soil profile and intensity in the red and near-infrared bands. The soil line Euclidean distance (SLED) technique is based on relating a pixel's Euclidean distance of the red and near-infrared intensity value to the red and near-infrared intensity value for the bottom-most point on the soil line. The technique is evaluated for two fields in the U.S. Midwest. The technique performs as well as or better than a recently proposed technique, while at the same time relating to the parent material (i.e., soil series) within the field. A technique for significantly reducing the number of soil samples required in grid sampling is also introduced and evaluated. This technique utilizes pixels at various percentile locations along the soil line to characterize the predictive relationship between percentage surface OM and image intensity.
Abbreviations: GPS, global positioning system • MSL, measurement site location • OM, organic matter • SLED, soil line Euclidean distance • SOC, surface organic C
|
INTRODUCTION
|
|---|
THE ABILITY TO OBTAIN soil property estimations from remotely sensed images has been identified as a valuable technique in improving the understanding of the site-specific variation within the soil's surface horizon. Currently, grid-sampling techniques consist of soil measurements taken at randomly selected points throughout the field. Usually, that soil sample is assumed to represent the properties for a larger, homogeneous area. The philosophy behind obtaining soil properties from remotely sensed images is the recognition of improved representation of field heterogeneity.
The relationship between the humus content, or soil OM, and remotely sensed measurements has been the subject of considerable research (Chen et al., 2000; Shonk et al., 1991; Vinogradov, 1982; Al-Abbas et al., 1972; Baumgardner et al., 1970). Baumgardner et al. (1970) and Al-Abbas et al. (1972) performed the first airborne experiments to study the relationship between OM and reflectance in the visible and near-infrared ranges. Twelve different radiance regions (i.e., 0.46–0.48 µm, 0.52–0.55 µm) were investigated and a multi-factor linear equation that related OM to reflectance was developed. Correlation coefficients were in the range of 0.52 to 0.56. However, these early linear functions failed to incorporate spectral ranges between 0.64 and 0.72 µm, which has been suggested as the most influential in distinguishing variable OM content (Vinogradov, 1982). Both Baumgardner et al. (1970) and Al-Abbas et al. (1972) concluded that the OM largely controls the radiance of the soil when the surface OM content is >2% (wt./wt.). It was later suggested that when the OM content falls below that threshold, other soil properties (iron and manganese content) dominate the reflectance pattern. Vinogradov (1982) suggested that the largest optical differences in distinguishing between soils with varying OM contents are present in the red band between 0.64 to 0.72 µm, with optical differences in the near infrared disappearing at 1.2 µm. Vinogradov (1982) used spectral measurements in the 0.6- to 0.7-µm range and discovered that an exponential relationship was most appropriate. Smith et al. (1987) and Sudduth and Hummel (1988) demonstrated that OM content could be predicted using linear or curvilinear relationships in the visible and near-infrared range.
More recent research has focused on the use of image intensity to derive soil properties. Chen et al. (2000) utilized image intensity in three bands (red, green, and blue) to develop a logarithmic linear relationship for organic C:
 | [1] |
where SOC is the surface organic C; R, G, and B are image intensity values in the red, blue, and green bands, respectively; and a, b, c, and d are curve-fit parameters. Chen et al. (2000) reported that the relationship between soil OM and reflectance was poor when measurements are taken across large geographical areas, suggesting that the difficulties may be the result of different types of parent materials.
The soil line concept has been utilized extensively in attempts to characterize vegetation growth (Richardson and Wiegand, 1977; Baret and Guyot, 1991; Campbell, 1996). The soil line concept, conceptualized in Fig. 1
, is a widely researched linear relationship between the near-infrared and visible reflectance of bare soil images (Richardson and Wiegand, 1977):
 | [2] |
where NIR and R are the image intensity or reflectance in the near-infrared and red bands, and 
and ß are the soil line slope and intercept, respectively. The soil line extends from a lower region consisting of the darker soils with low R and NIR reflectance (Point A in Fig. 1) to an upper region of bright soils with high reflectance values in both the R and NIR bands (Point B in Fig. 1). Campbell (1996) notes that Point C corresponds to a pure vegetation pixel and Point D corresponds to a partially vegetated pixel. According to Baret et al. (1993), the soil line for a particular soil type "results from the combined variations of its surface status characterized by its roughness and moisture". Jasinki and Eagleson (1989) showed that three unique soil lines exist: a soil mineral line (soil type), a soil moisture line, and a soil roughness line. Baret et al. (1993) suggests that soil type is the primary factor determining the slope and intercept of the soil line.
View larger version (14K):
[in this window]
[in a new window]
|
Fig. 1. Soil line concept demonstrating the observed linear relationship between red (R) and near-infrared (NIR) reflectance or image intensity of bare soil. The soil line extends from darker soils with low R and NIR image intensity (Point A) to an upper region of bright soils with high R and NIR image intensity (Point B). Point C represents a pure vegetation pixel and Point D represents a partially vegetated pixel. Modified from Campbell (1996).
|
|
|
OBJECTIVES
|
|---|
The objectives of this research are to develop an approach based on image soil lines that incorporates the image intensity in both the R and NIR bands for mapping surface OM and to provide guidance for soil sampling. The SLED technique is developed to derive the relationship between a pixel's Euclidean distance along the soil line and surface OM. The technique allows a relationship to be developed between image intensity and surface OM that relates to the soil line, a function of the soil series within a field. The use of the soil line to derive sampling locations is suggested as a mechanism for significantly reducing the required number of soil samples compared with grid sampling techniques. Note that the procedure is not limited to the use of intensity values. The reasoning for using intensity rather than reflectance in this paper was based on deriving a relationship that uses the limited information that might be made available to an agricultural producer. Apparent reflectance can be derived from intensity values by normalizing to exoatmospheric irradiance (Jensen, 1996).
|
MATERIALS AND METHODS
|
|---|
Soil Line Euclidean Distance Technique
The SLED technique utilizes intensity values in the R and NIR bands for pixels specifically located within the particular field of interest. Derivation of the SLED technique is based on soil measurements collected at specific collection locations. The spatial coordinates of these collection locations are known such that intensity value at these specific locations can be obtained. Using a least-squares regression technique, the SLED technique calculates the linear equation (slope and intercept) for the bare soil image using the R and NIR intensity values for pixels in the field. The linear equations are not determined manually but rather by a program similar to the routine proposed and evaluated by Fox (2000). The program identifies the intensity values in the R and NIR bands for each pixel within the field. An iterative procedure is used to locate the minimum value of the NIR band for each R band value present in the data. These minimum points are then used in a linear regression algorithm and and ß are determined. The program also includes an iterative procedure to make sure that the minimum points are occurring in the expected increasing, linear trend and ignores points causing the calculated slope and intercept to go astray. Note that standing water, waterways, and heavily vegetated areas within the field should be masked from the images in order to ensure accuracy in development of image soil lines.
The SLED technique requires the identification of the minimum point along the calculated soil line (i.e., Point A in Fig. 1). The minimum point refers to the pixel with the lowest R and NIR intensity values, corresponding to the left-most extreme point on the soil line and representing the darkest soils within the field. The routine then calculates the distance (D) of each pixel's intensity values away from the soil line's minimum point:
 | [3] |
where NIRCL = NIR intensity value at the collection location, NIRmin = NIR intensity value for minimum point on soil line, RCL = R intensity value at the collection location, and Rmin = R intensity value for minimum point on soil line.
For example, if the collection location was located at Point B in Fig. 1, the technique would calculate the distance along the soil line from Point A (Rmin, NIRmin) to Point B (RCL, NIRCL). This distance measurement has units of image intensity value (I) and provides an index of the position of pixels along the soil line. The final steps in the SLED technique involve correlating the Euclidean distance with the corresponding measurements of OM. Regressions are then performed to develop predictive equations that can be used to estimate OM throughout a field.
Site and Data Description
Development and evaluation of the SLED technique focused on two fields (designated Field 1 and Field 2) in the Midwest USA. Field 1 is a 32.4-ha, tiled field located in Buchanan County, Iowa. The field is generally under a corn and soybean rotation. Map units representing eight soil series, the field boundary, and soil sampling locations are shown in Fig. 2
. Eight soil series are located within this field: Burkhardt silty loam (241B; sandy, mixed, mesic Typic Hapludolls), 2 to 5% slopes; Flagler sandy loam (284; coarse-loamy, mixed, superactive, mesic Typic Hapludolls), 0 to 2% slopes; Clyde clay loam (391B; fine-loamy, mixed, superactive, mesic Typic Endoaquolls), 1 to 4% slopes; Readlyn loam (399; fine-loamy, mixed, superactive, mesic Aquic Hapludolls), 1 to 3% slopes; Olin fine sandy loam (408B; coarse-loamy, mixed, superactive, mesic Typic Hapludolls), 2 to 5% slopes; Sparta loam fine sand (41B; sandy, mixed, mesic Entic Hapludolls), 2 to 5% slopes; Schley variant sandy loam (807B; fine-loamy, mixed, superactive, mesic Udollic Endoaqualfs), 1 to 4% slopes; and Kenyon loam (83B; fine-loamy, mixed, superactive, mesic Typic Hapludolls), 2 to 5% slopes. Soil property measurements, including surface OM, were measured at 123 measurement site locations (MSLs). Surface measurements were made from the top inch of the soil surface following procedures outlined by Page et al. (1982) for OM and cation exchange capacity and by Brown and Wancke (1980) for pH. Ranges of soil surface horizon properties were 1.4 to 8.9% (wt./wt.) for surface OM, 5.4 to 29.3 cmol kg-1 for cation exchange capacity, and 5.6 to 7.3 for soil pH.
View larger version (160K):
[in this window]
[in a new window]
|
Fig. 2. Field boundary, soil series, and soil sampling locations of a 32.4-ha field in Buchanan County, Iowa (Field 1).
|
|
Field 2 is a 42.9-ha field located in Fremont County, Iowa. Field 2 is also managed under a corn and soybean rotation. Map units representing the two prevalent soil series within this field [McPaul silty loam (70; coarse-silty, mixed, superactive, calcareous, mesic Mollic Udifluvents), 0 to 2% slopes; and Moville silty loam (275; coarse-silty over clayey, mixed, superactive, calcareous, mesic Aquic Udifluvents), 0 to 2% slopes] and soil sampling locations are shown in Fig. 3
. Surface OM and other characteristics soil properties were measured at 113 MSLs using the same procedures as documented for Field 1. Ranges for surface soil horizon properties were 1.2 to 2.9% (wt./wt.) for surface OM, 11.0 to 22.0 cmol kg-1 for cation exchange capacity, 6.9 to 7.8 for soil pH, and 8 to 20% for percentage clay content.
View larger version (105K):
[in this window]
[in a new window]
|
Fig. 3. Field boundary, soil series, and soil sampling locations of a 42.9-ha field in Fremont County, Iowa (Field 2).
|
|
Bare soil images of the two fields were acquired using a digital camera system during the 1998 growing season. The digital camera system utilizes area array technology (digital charged coupled devices) with a spatial resolution of 0.5 m. The imaging system combines the digital camera with a targeting camera and global positioning system (GPS), allowing a full frame capture (1500 by 1000 pixels) of eight bit data using high quality imaging optics. The images were georeferenced into the Universal Transverse Mercator projection using GPS measurements of ground targets. The location error was estimated to be between 1 and 5 m for both fields, using GPS measurements of several locations within and around the fields. The aerial images provide four bands of data: blue (400–500 nm), green (500–600 nm), red (600–700 nm), and near infrared (700–1000 nm).
At the time of image acquisition, the surface roughness was assumed to be uniform, and even though soil moisture variations were present in the field, these variations were generally small at image acquisition. Soil moisture was measured at twelve locations in each field. Soil moisture variations were deemed insignificant by comparing the coefficient of variation (i.e., mean divided by standard deviation) in soil moisture vs. the coefficient of variation in percentage OM. The coefficient of variation in soil moisture was 0.10 and 0.07 in Fields 1 and 2, respectively. The coefficient of variation in surface OM was 0.40 and 0.20 in Fields 1 and 2, respectively.
Methodology
Relationship between Euclidean Distance and Organic Matter Content
The relationship between a pixel's Euclidean distance along the soil line (D) to surface OM was initially investigated. Soil lines were calculated for each field and the minimum point (Rmin, NIRmin) along the soil line was identified. Soil lines were derived using all pixels within the field boundaries of each image. Pixels outside of the field boundaries were not used in the derivation of soil line parameters. The coefficient of variation in soil moisture was less than the coefficient in variation in surface OM content in both fields. Therefore, it is assumed that the soil moisture variations had no significant impact on the results. The deviation of a pixel away from the soil line was not expected to be significant in terms of affecting the Euclidean distance measurements. Lines formed between the bottom-most point on the soil line and a pixel's NIR and R image intensity resulted in angles between this line and the soil line that were generally <5° for most pixels. Furthermore, the soil lines developed within the fields were strongly correlated and showed no significant deviations along its length.
Relationships were derived between the Euclidean distance (D) of the MSL's intensity values and (Rmin, NIRmin), as given by Eq. [3], and the surface OM measurements using the SLED technique. Governing equations were developed for the D vs. OM relationship. Several functional forms (i.e., linear, exponential, quadratic) were investigated. The derived relationships were then used to develop predicted surface OM maps based on a pixel's Euclidean distance along the soil line.
The SLED technique was then compared with a technique recently proposed by Chen et al. (2000), assuming the following relationship between surface OM and SOC as given by Page et al. (1982):
 | [4] |
The model proposed by Chen et al. (2000) relates SOC content to R, B, and B image intensity. Intensity values in the R, G, and B bands for each MSL were extracted from the image and related to organic C using the logarithmic linear equation given by Eq. [1]. A nonlinear regression was performed to derive the coefficients (a, b, c, and d) in Eq. [1].
Predictive Capability of Soil Line Euclidean Distance
The predictive capability of the SLED technique was then investigated through the use of two separate data sets within the MSLs of each field. The first group was used for development of the D vs. surface OM predictive equations. The second group was used for evaluating the predictive capability of the derived equations. The MSLs were assigned into a particular group using a random number generator. Field 1 possessed 123 MSLs, with 62 MSLs assigned to the first group and 61 MSLs assigned to the second group. Field 2 possessed 113 MSLs, with 57 MSLs in the first group and 56 MSLs in the second group. Relationships developed from all MSLs and the first group of MSLs were compared. The predictive equations for the first group of MSLs were then utilized to predict OM at each MSL in the second group based on image intensity. The predictive ability was assessed based on a linear regression of predicted vs. observed OM. The same groups of MSLs were also used to derive governing relationships and predict OM based on the technique proposed by Chen et al. (2000). The predictive ability of the SLED technique was compared with the Chen et al. (2000) model.
Identification of Soil Sampling Locations
The SLED technique was then extended to automatically determine soil sampling locations within a field. Current methods of soil analysis utilize randomly selected soil samples assumed to represent the heterogeneity within the field. The SLED technique allows samples to be obtained at critical locations along the length of the soil line. Measurements at these critical locations yield a descriptive curve relating D to surface OM. A seven-point percentile method was selected to govern grid sampling of surface OM. The seven-point percentile method was based on the 1, 10, 25, 50, 75, 90, and 99% percentiles between the minimum and maximum Euclidean distance along the soil line for all pixels within the field. Such percentiles selected the median Euclidean distance and several locations on the extremes of the soil line.
These percentile distances were used to develop the governing relationships between D and surface OM for the MSLs within Fields 1 and 2. The seven MSLs with D closest to the percentile Euclidean distances were selected to develop the governing relationships. These relationships were then used to predict surface OM at all other MSLs throughout Fields 1 and 2, with a linear regression performed between predicted and observed OM. Predictive equations derived from all MSLs and those using the seven-point percentile method were also compared.
An automated routine was developed to determine the Euclidean distance corresponding to these percentiles and then locate pixels within the field that most closely matched the percentile Euclidean distances, requiring no preliminary data to develop regression coefficients. Input into the routine includes grid files of pixel R and NIR image intensity. The routine includes a buffer to ensure that soil sampling locations are not selected on the edges of the field. The default buffer is 10 m from each edge of the field, but this buffer is user selectable within the automated routine. No manual intervention is required in determining soil line parameters, the Euclidean distance percentiles, or soil sampling locations. Output of the automated routine includes a soil sampling location file that can be directly imported into GIS software and overlain on image files to locate exact soil sampling locations. The automated routine was used to determine soil sampling locations for Field 1.
|
RESULTS AND DISCUSSION
|
|---|
Relationship between Euclidean Distance and Organic Matter Content
The strongest correlation between OM and image intensity was observed in the R and NIR bands in both fields, with lower correlation for the B and G bands. Table 1 quantifies the correlation between OM and image intensity in each band. In most cases, the strongest correlation was observed when using a two-parameter exponential decay function:
 | [5] |
where a and b are regression parameters and 
is the image intensity in a particular band. NIR demonstrated the strongest correlation in Field 1; R demonstrated the strongest correlation in Field 2. These conclusions suggested that a technique using both the R and NIR bands (i.e., the soil line) could be most valuable in developing relationships between image intensity and surface OM. Scatter plots showing the distribution of R and NIR and the resulting soil lines derived from the bare soil images of Fields 1 and 2 are shown in Fig. 4 .
View this table:
[in this window]
[in a new window]
|
Table 1. Relationship between image intensity in the blue (B), green (G), red (R), and near-infrared (NIR) bands and surface organic matter content (OM) by weight for all measurement site locations (MSLs) within Fields 1 and 2.
|
|
View larger version (17K):
[in this window]
[in a new window]
|
Fig. 4. Scatter plots of red (R) and near-infrared (NIR) image intensity and the resulting soil lines for (a) Field 1 and (b) Field 2.
|
|
After determination of image soil lines, the relationship between surface OM and Euclidean distance (D) was investigated through the use of the MSLs within each field. A governing equation in the form of a two-parameter exponential decay function was determined to most appropriately characterize the relationship between D and OM for both Fields 1 and 2:
 | [6] |
where a and b are regression parameters with units of percentage OM and inverse image intensity value, respectively. Table 2 outlines the soil line parameters, (Rmin NIRmin) points, OM vs. D curve-fit parameters, and regression coefficient for the relationship between OM and D. Nonlinear regressions used to derive the coefficients for the Chen et al. (2000) model resulted in governing relationships with similar to slightly less favorable correlation coefficients to the SLED technique (i.e., R2 = 0.71 in Field 1 and R2 = 0.69 for Field 2).
View this table:
[in this window]
[in a new window]
|
Table 2. Parameters for soil lines and relationships between Euclidean distance (D) and percentage surface organic matter (OM) content based on all measurement site locations (MSLs).
|
|
Predictive Capability of Soil Line Euclidean Distance
The governing equations that were developed based on the first group of MSLs are shown in Fig. 5
for Fields 1 and 2, respectively. The parameters (a and b) of the predictive equations utilizing only the first group of MSLs were similar to the equations for all MSLs (Table 2). This first group of MSLs was also utilized to develop predictive equations for the relationship between SOC and image intensity proposed by Chen et al. (2000). The following regression coefficients were derived for Field 1: a = 2.84, b = -0.03, c = -0.014, and d = 0.031. The relationship between image intensity in the B, G, and R bands and observed SOC had a regression coefficient of 0.68, equivalent to the SLED technique. For Field 2, the regression coefficients were a = 3.37, b = -0.012, c = -0.0024, and d = -0.006, yielding a regression coefficient of 0.76, slightly less than the SLED technique.
View larger version (17K):
[in this window]
[in a new window]
|
Fig. 5. Relationship between Euclidean distance (D) along the soil line and percentage surface organic matter (OM) content based on half of the measurement site locations for (a) Field 1 and (b) Field 2 using the soil line Euclidean distance technique.
|
|
The linear regression of predicted vs. observed OM for the SLED technique is shown in Fig. 6
for both fields. Note that the slope and intercept of the regression equation for Field 2 are more accurate than in Field 1. Such results may be the result of the more uniform soil properties within Field 2 and/or the small number of soil series in Field 2 compared with Field 1. The technique that was proposed by Chen et al. (2000) resulted in similar results to the SLED technique for predicted vs. observed SOC. For Field 1, the regression resulted in a slope of 0.57, intercept of 0.82 in units of SOC, and regression coefficient of 0.68, slightly less than the results using the SLED technique. For Field 2, the regression resulted in a slope of 0.72, intercept of 0.25 in units of SOC, and regression coefficient of 0.77, again slightly less than the SLED technique.
View larger version (16K):
[in this window]
[in a new window]
|
Fig. 6. Predicted (pred) vs. observed (obs) percentage surface organic matter (OM) content utilizing predictive equations based on half of the measurement site locations for (a) Field 1 and (b) Field 2 using the soil line Euclidean distance technique.
|
|
The SLED technique has been shown to predict OM as well as or slightly better than the model proposed by Chen et al. (2000) for SOC. The Chen et al. (2000) technique is field specific in that the regression parameters change values depending on parent material. To develop the predictive relationship between organic C content and R, B, and G image intensity, soil samples representative of the variability in soil characteristics would be required a priori. In other words, there is no suggested methodology for determining where to sample within a field to most appropriately describe the relationship between image intensity and organic C or OM. The SLED technique, on the other hand, is more robust and efficient since it relates back to the image soil line. The technique is universal, not in the values of the regression parameters, but in the method to calculate those values. The SLED technique uses the soil line, which can be defined from only the bare soil image, to identify the most appropriate soil sampling locations in the field. Thus, it provides an efficient approach for determining the optimum regression parameters that defines the relationship between OM and Euclidean distance.
Identification of Soil Sampling Locations
The MSLs within the 1998-bare soil images of Fields 1 and 2 were utilized to evaluate the point-percentile procedure. The MSL with a distance closest to each percentile distance was used to derive the relationship between D and OM. The seven-point percentile method resulted in exponential decay functions with parameters a = 9.270 and b = 0.014 for Field 1 and a = 3.318 and b = 0.014 for Field 2. These relationships were then used to predict surface OM at all other MSLs throughout Fields 1 and 2, with a linear regression performed between predicted and observed surface OM. Results of this investigation are shown for Fields 1 and 2 in Fig. 7
. In terms of a mathematical comparison, the predictive equation derived from all MSLs and the equation derived using the seven-point percentile method are compared for Fields 1 and 2 in Fig. 8
. For both fields, the seven-point percentile method resulted in exponential equations not significantly different statistically from the equations derived from all MSLs. The percentage difference in predicted OM based on the two predictive equations shown in Fig. 8 for Fields 1 and 2 is <5% throughout the range of Euclidean distances. To statistically verify that generated surface OM maps would be equivalent to maps developed based on grid sampling, the relationship between OM and D derived from the seven-point percentile method was used to generate predicted OM. A paired t test was performed between the predicted and measured OM using a 95% confidence interval. This statistical test resulted in the predicted and measured OM not being significantly different, with a P-value of 0.2. As such, the seven-point percentile method provides a mechanism for significantly reducing the number of soil samples required to represent the spatial heterogeneity in surface OM, and possibly other surface soil properties. The benefit of the technique is that it provides simpler and less expensive methods to develop maps of surface OM content that are not significantly different from maps generated by grid sampling.
View larger version (17K):
[in this window]
[in a new window]
|
Fig. 7. Predicted (pred) vs. observed (obs) percentage surface organic matter (OM) content using predictive equations developed from the seven-point percentile method for (a) Field 1 and (b) Field 2.
|
|
View larger version (24K):
[in this window]
[in a new window]
|
Fig. 8. Comparison of predictive relationships between Euclidean distance along soil line (D) to percentage surface organic matter (OM) content for all measurement site locations and using seven-point percentile method for (a) Field 1 and (b) Field 2.
|
|
Pixel R and NIR reflectance for Field 1 was then input into the automated, seven-point percentile routine. The automated routine determined optimal soil sampling locations for Field 1 as shown in Fig. 9
. The result file is overlain on top of the bare soil image. As expected, the routine selected pixels at the median and on the extremes of the soil line, corresponding to the brightest and darkest locations within the field. Using measurements obtained at these soil sampling locations, a surface OM map can be developed for the entire field.
View larger version (114K):
[in this window]
[in a new window]
|
Fig. 9. Identified soil sampling locations using the seven-point percentile method on the remotely sensed bare soil image of Field 1.
|
|
|
SUMMARY AND CONCLUSIONS
|
|---|
A technique to derive predictive relationships between surface OM content and Euclidean distance along an image soil line is derived. The technique is developed for the purpose of improving the spatial representation of within-field heterogeneity of surface OM content and to direct soil sampling. The SLED technique is derived using the Euclidean distance measurement of a pixel's near-infrared and red intensity values from the minimum near-infrared and red intensity values for the soil line. The percentage surface OM is significantly correlated (i.e., R2 > 0.68) to this Euclidean distance measurement for two fields in the Midwest USA. The SLED technique is compared with a recently proposed predictive relationship between SOC content and image intensity (Chen et al., 2000). The two techniques are similar (i.e., within 90%) in predicting OM or organic C content. The advantage of the SLED technique is that image intensity can be related to the parent material or soil type, providing a mechanism for determining soil sampling locations within a field. A seven-point percentile method is suggested to identify the most critical Euclidean distances along the soil line. The predictive relationships derived from the seven-point percentile method are not significantly different (i.e., percentage differences <5%) from those derived for all measurement locations, with a P-value of 0.2 from a paired t test. Also, the percentile method has been automated to output soil sampling location files that can be directly input into GIS software. The automated grid-sampling identification routine can be obtained from the authors upon request.
|
ACKNOWLEDGMENTS
|
|---|
The authors acknowledge the support of the EPA Science to Achieve Results (STAR) Graduate Fellowship Program, Grant No. U915329-01-1. The authors also acknowledge the support and guidance of Dr. Stephen Searcy, Department of Biological and Agricultural Engineering, Texas A&M University.
|
REFERENCES
|
|---|
- Al-Abbas, A.H., P.H. Swain, and M.F. Baumgardner. 1972. Relating organic matter and clay content to the multispectral radiance. Soil Sci. 114:477–485.
- Baret, F., and G. Guyot. 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sens. Environ. 35:161–173.
- Baret, F., S. Jacquemond, and J.F. Hanocq. 1993. The soil line concept in remote sensing. Remote Sens. Rev. 7:65–82.
- Baumgardner, M.F., S.J. Kristoff, C.J. Johannsen, and A.L. Zachary. 1970. The effect of organic matter on multispectral properties of soils. Proc. Indiana Acad. Sci. 79:413–422.
- Campbell, J.B. 1996. Introduction to remote sensing, 2nd ed. Guilford Press, New York.
- Chen, F., D.E. Kissell, L.T. West, and W. Adkins. 2000. Field-scale mapping of surface soil organic carbon using remotely sensed imagery. Soil Sci. Soc. Am. J. 64:746–753.[Abstract/Free Full Text]
- Fox, G.A. 2000. Image use in the characterization of field parameters: Incorporation of remote sensing with hydrologic simulation modeling. M.S. Thesis. Texas A&M Univ., College Station, TX.
- Jasinki, M.F., and P.S. Eagleson. 1989. The structure of red-infrared scattergrams of semivegetated landscapes. IEEE Trans. Geosci. Remote Sens. 27:441–451.
- Jensen, J.R. 1996. Introductory digital image processing: A remote sensing perspective. Prentice Hall, Upper Saddle River, NJ.
- Brown, J.R., and D. Wancke. 1980. Recommended soil test procedures for the north central region. p. 5–6. In North Central Region Publ. No. 221 (revised). North Dakota Exp. Stn. Bull. 499 (revised). North Dakota State Univ., Fargo, ND.
- Page, A.L., R.H. Miller, and D.R. Keeney (ed.) 1982. Methods of soil analysis. Part 2—Chemical and microbiological properties. 2nd ed. Agron. Monogr. 9. ASA and SSSA, Madison, WI.
- Richardson, A.J., and C.L. Wiegand. 1977. Distinguishing vegetation from soil background information. Photogramm. Eng. Remote Sens. 43:1541–1552.
- Shonk, J.L., L.D. Gaultney, D.G. Schulze, and G.E. Van Scoyoc. 1991. Spectroscopic sensing of soil organic matter content. Trans. ASAE 34:1978–1984.
- Smith, D.L., C.R. Worner, and J.W. Hummel. 1987. Soil spectral reflectance relationship to organic matter content. ASAE Paper 87-1608. American Society of Agricultural Engineers, St. Joseph, MI.
- Sudduth, K.A., and J.W. Hummel. 1988. Optimal signal processing for soil organic matter determination. ASAE Paper 88-7004. American Society of Agricultural Engineers, St. Joseph, MI.
- Vinogradov, B.V. 1982. Remote sensing of the humus content of soils. Soviet Soil Sci. 13:103–113.
This article has been cited by other articles:
|
|
|
|
 
F. Chen, D. E. Kissel, L. T. West, W. Adkins, D. Rickman, and J. C. Luvall
Mapping Soil Organic Carbon Concentration for Multiple Fields with Image Similarity Analysis
Soil Sci. Soc. Am. J.,
January 11, 2008;
72(1):
186 - 193.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
S. A. Wills, C. L. Burras, and J. A. Sandor
Prediction of Soil Organic Carbon Content Using Field and Laboratory Measurements of Soil Color
Soil Sci. Soc. Am. J.,
March 12, 2007;
71(2):
380 - 388.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
M. R. Nanni and J. A. M. Dematte
Spectral Reflectance Methodology in Comparison to Traditional Soil Analysis
Soil Sci. Soc. Am. J.,
February 2, 2006;
70(2):
393 - 407.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
G. A. Fox and R. Metla
Soil Property Analysis using Principal Components Analysis, Soil Line, and Regression Models
Soil Sci. Soc. Am. J.,
September 29, 2005;
69(6):
1782 - 1788.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
G. A. Fox, G. J. Sabbagh, S. W. Searcy, and C. Yang
An Automated Soil Line Identification Routine for Remotely Sensed Images
Soil Sci. Soc. Am. J.,
July 1, 2004;
68(4):
1326 - 1331.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
