Skip to main content
ARS Home » Midwest Area » Columbia, Missouri » Cropping Systems and Water Quality Research » Research » Publications at this Location » Publication #140922
Title: ESTIMATING WITHIN-FIELD VARIATIONS IN SOIL PROPERTIES FROM AIRBORNE HYPERSPECTRAL IMAGES

Author
item HONG, S - KOREAN RURAL DEV ADM
item Sudduth, Kenneth - Ken
item Kitchen, Newell
item Drummond, Scott
item PALM, HARLAN - UNIV OF MO
item WIEBOLD, WILLIAM - UNIV OF MO

Submitted to: Pecora Conference Land Satellite Information in the Next Decade
Publication Type: Proceedings
Publication Acceptance Date: 11/10/2002
Publication Date: 11/15/2002
Citation: Hong, S.Y., Sudduth, K.A., Kitchen, N.R., Drummond, S.T., Palm, H.L., Wiebold, W.J. 2002. Estimating within-field variations in soil properties from airborne hyperspectral images. Pecora Conference Land Satellite Information in the Next Decade. American Society of Photogrammetry and Remote Sensing. Bethesda, Maryland

Interpretive Summary: Precision farming relies on the ability to efficiently and economically collect and interpret data describing the soil variability within cropped fields. Remote sensing from satellites or airplanes is one approach to obtaining the data needed for precision farming. Recently hyperspectral remote sensing data has become available to the agricultural research community. In contrast to conventional, or multispectral, remote sensing data where only a few data bands are obtained, hyperspectral data includes tens to hundreds of channels. While this large amount of data may allow us to develop better relationships between the remotely sensed image and actual field conditions, it also greatly increases the complexity of the data processing and interpretation procedure. In this study, we examined 120-band aircraft hyperspectral data that were collected over a Missouri corn/soybean field prior to planting in 2000 and 2001. A number of statistical procedures were employed to relate the hyperspectral data to soil properties. We found that these bare soil images were significantly correlated with soil properties, with blue wavelengths showing the strongest relationships. It was necessary to convert the 120-band data into a small number (less than 10) of summary variables for efficient data processing and interpretation. This research will benefit other researchers, producers, and consultants who may be interested in using hyperspectral data to understand soil differences for precision agriculture.

Technical Abstract: The ability of hyperspectral image (HSI) data to provide estimates of soil electrical conductivity (EC) and soil fertility levels without requiring extensive field data collection was investigated. Bare soil images were acquired using a prism grating pushbroom scanner in April 2000 and May 2001 for a central Missouri experimental field in a minimum-tillage corn-soybean rotation. Data were converted to reflectance using chemically-treated reference tarps with eight known reflectance levels. Geometric distortions of the pushbroom sensor images were corrected with a rubber sheeting transformation. A 5 m pixel size was selected by analysis of short-range variations in five sub-field areas. Statistical analyses including simple correlation, multiple regression (MR), and principal component (PC)analysis were used to relate HSI data and derived Landsat-like bands to field-measured soil properties. The blue wavelengths of the HSI and Landsat-like images showed the highest correlation with EC and soil chemical properties. With the exception of pH and P, the soil fertility data were negatively correlated to the HSI reflectance data. The highest correlations to the HSI bands were found for Mg and CEC. Stepwise multiple linear regression (SMLR) models using the full HSI dataset included too many variables, which increased the danger of overfitting. MR models using Landsat-like bands may be more practical than SMLR models for mapping soil properties. Analysis of principal components showed that PC 2 and PC 4 explained soil variability well for CEC, Mg, OM, K, and pH. Both approaches to data volume reduction, creating Landsat-like bands and principal component analysis, showed potential for developing relationships with soil properties. HSI analysis appears promising for quantifying soil property variability.