Date of Award

Spring 8-1-2004

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Geography, Geology, and Anthropology

First Advisor

Paul W. Mausel

Second Advisor

Ryan R. Jensen

Third Advisor

S. L. Brian Ceh

Abstract

Soil classification is a complex task that involves several factors. Soil classifications that have been published usually need to be updated as soil series definitions and boundaries change over time. Most existing soil classification surveys in the United States, although often very good, lack the required accuracy for selected applications such as precision agriculture. 111 Hyperspectral remote sensing data, with its 100+ bands and very high spectral resolution can help facilitate soil classification detail not achieved through traditional remote sensing methods. The study area selected for research was Wolf Field, near Homer, Champaign County, Illinois, which is a very agriculturally productive area. One of the objectives of this research was to differentiate soils into discrete spectral regions, more finely delineated than soil series, based on analysis of hyperspectral remote sensing data. A second objective focused on establishing quantitative relationships between selected soil parameters, soil nutrients and hyperspectral data. Also of interest was developing suggestions for formulating soil sampling schemes, based on hyperspectral data analysis. These objectives were addressed employing a variety of techniques. The techniques were divided into two parts--remote sensing and quantitative analyses. Remote sensing techniques involved performing supervised classification, ratio and principal component analysis, as well as a fuzzy-c-means classification. Quantitative analyses involved establishing relationships among hyperspectral data, selected nutrients, soil texture classes, and cation exchange capacity, as well as between soil sample points IV and spectral data. These analyses were accomplished by implementing several statistical techniques, such as multiple regression among soil texture classes parameters, selected soil nutrients, cation exchange capacity and spectral reflectances of bands, factor analysis on the spectral dataset, and cluster analysis of the spectral dataset and soil sample points. Based on the results derived from using these techniques, the hypothesis was supported that soils can be differentiated into discrete spectral regions using hyperspectral remote sensing data, as well as the hypothesis that supports the existence of significant statistical spectral relationships between soil texture classes, selected soil nutrients, and cation exchange capacity and hyperspectral data at 95% confidence interval for multiple regression ( enter method) with R2 values greater then 0.68. The resulting classifications developed in this research were more detailed than ones usually found in county soil surveys. Analysis of these results helped in formulating guidelines for devising a soil sampling scheme potentially important for use in precision agriculture, as well as improving the precision of soil databases potentially useful for rural taxation purposes.

Download

Share

COinS