SPATIAL AUTOREGRESSION MODELING OF SITE-SPECIFIC WHEAT YIELD

This paper illustrates spatial autoregressive response modeling applie d to estimating the relationship between site-specific wheat yield and selected terrain variables within a dryland production field in Monta na. Yield values for numerous irregularly spaced areal units are obtai ned with a mass flow yield sensor. Kriging is used to compute spatiall y weighted averages for new, larger areal units of a regular surface p artitioning as needed for implementing autoregression response (AR) mo deling. The new units were combined by systematic increments to produc e new regularly spaced units in scales ranging from 9.1 m to 329.2 m. Yield data are then related to elevation and a vegetation index comput ed through processed digital aerial imagery. Correlation and R-2 value s between yield and the terrain attributes are obtained at each scale from ordinary least squares (OLS) correlation and regression analyses. Statistical results are obtained from OLS regression and AR regressio n to assess the negative impact of spatial dependency, or autocorrelat ion, in these yield data. Though values of regression parameter estima tes were unaffected, the primary effect of increasing spatial scale by combining areal units of yield data was to systematically increase th e value of correlation coefficients and the R-2. A Slight amount of co mbining areal units did not seriously affect statistical results. The primary effect of autocorrelation was to underestimate standard errors resulting in too much significance being assigned to parameter estima tes of an OLS-based regression model. (C) 1998 Elsevier Science B.V. A ll rights reserved.