Tuesday, November 6, 2007
231-45
A Geostatistical Approach for Prediction of Underlying Characteristics of Coal Mined Lands in Reclamation.
K. L. Armstrong, D. G. Bullock, G. A. Bollero, and R. E. Dunker. Crop Sciences Department, University of Illinois at Urbana-Champaign, 1102 South Goodwin Avenue, Urbana, IL 61801
Characterizing the spatial variability of a crop field requires prior knowledge of yield affecting factors. Often, many factors simultaneously affect crop variability (e.g., weather, soil and crop properties, and pests) and may contribute with differing magnitudes in describing yield patterns. Unlike undisturbed soils, disturbed mine soils can exhibit differences in soil properties not only due to natural variation, but also due to reclamation techniques. Predicting crop yield on such lands is made even more difficult when soil characteristics are not well described or even known. In our experiment, georeferenced corn and soybean (only Lewis Mine) yield data, cone penetrometer test (CPT) at different soil depths, elevation (plus terrain derivatives), and apparent electrical conductivity (ECa) data were collected for two fields at the Cedar Creek Mine site in western, IL and one field at the Lewis Mine site in southwestern, IN. Previously, cluster and canonical discriminant analyses were performed to assess how our measured variables discriminated among yield categories (high, medium, low). Results showed that ECa, elevation, terrain derivatives, and CPT variables linked to soil compaction discriminated the most among yield categories. These variables were subsequently used in a geostatistical approach to map variable distribution throughout fields, beyond sampled points. The spatial structure of these variables was investigated through semivariography. Ordinary kriging, and ordinary cokriging were used as interpolation methods for prediction of variables of interest. Map error was assessed through cross validation and root mean square error (RMSE). Results showed that spatial structure exists for all variables of interest except elevation which is best fitted by a trend surface. Ordinary cokriging results in smaller map error compared to ordinary kriging. Our experiment shows benefits to using a geostatistical approach for describing mined land soil variation and subsequently providing detail about problematic field areas not meeting bond release standards.