Time Series Analysis for Statistical Inferences in Tillage Experiments
- R. J. Roseberg and
- E. L. McCoy
Spatial dependence of soil physical properties often invalidates parametric inferencing of tillage treatment effects. Spectral analysis was examined as a possible means to generate inferences on tillage treatments from an observation set that is spatially dependent. The method was tested using surface elevation data from a long-term tillage study. Three tillage treatments were imposed on a continuous corn cropping system in a randomized complete block design beginning in 1962. Surface elevations were measured in 1971, 1976, 1980, and 1987 along four transects oriented perpendicular to the imposed treatments with each transect running the length of the field. The power spectrum from the 1971 observations yielded a significant spectral peak at a wavelength equal to one-plot width. The power spectra from the 1976, 1980, and 1987 observations yielded a significant spectral valley at a wavelength equal to one-plot width. The significant spectral valleys in the 1976, 1980, and 1987 power spectra represent low variance within the plot spacing as compared to variance between plots. In a randomized complete block design, this significant spectral valley represents an inference on the effect of tillage on surface elevation using spectral analysis. Calculation of number of observations needed to show significant spectral differences at wavelength equal to one-plot width compared to within and beyond plot wavelengths indicate that 75, 161, 223, and 149 observations were necessary for 1971, 1976, 1980, and 1987 sampling dates, respectively.
Copyright © .