Modeling local terrain attributes in landscape-scale site-specific data using spatially lagged independent variable via cross regression

Abstract

Analysis methods for landscape-scale site-specific agricultural datasets have been adapted from a wide range of quantitative disciplines. Due to spatial effects expected at landscape scales with respect to yield affecting factors, inference from aspatial analyses may lead to inefficient statistical inference. When spatial correlation exists within a random variable e.g. explanatory variables such as elevation or soil characteristics, spatial statistical methods can provide unbiased and efficient estimates on which to base economic analyses and farm management decisions. Simple continuous terrain variables derived from spatially lagged independent variable transformation of relative terrain position allowed models to be estimated using familiar linear aspatial models without introducing the problems associated with interpolated data in inferential spatial statistics. Using site-specific data from three example fields, cross regressive elevation variables complemented topographic attributes, rather than replacing them in a range of statistical models. Results indicated that cross regressive elevation variables, especially relative elevation, reduced estimation problems due to correlation among independent variables and bias arising from spatially interpolated data in statistical analysis.