GEOSTATISTICAL ANALYSIS OF ON-FARM TRIALS IN PRECISION AGRICULTURE ALEXANDER BRENNING (1), HAGEN PIOTRASCHKE (2) and PEER LEITHOLD (2) (1) Department of Geography, University of Waterloo, Ontario, Canada. (2) Agri Con GmbH, 04749 Jahna, Germany. . ABSTRACT Geostatistical methods are important tools for the assessment of site‐specific management (SSM) approaches in on‐farm research (OFR) on variable rate technology (VRT) based on high-resolution yield and environmental data. As a case study we analyze an on-farm trial assessing an SSM procedure for winter wheat. A simulation study is used to evaluate different spatial linear models (generalized-least-squares - GLS regression and spatial autoregressive - SAR error models) and ordinary-least-squares regression in terms of estimation bias, efficiency and computational challenges. All spatial linear models produce comparable results in the estimation of linear model coefficients with small differences in efficiency, but in some cases considerable bias in the estimation of autocorrelation parameters; they are clearly superior to the non-spatial model. Regression by GLS with a variogram fitted to OLS residuals is computationally the least demanding approach and is comparable to the other spatial models. INTRODUCTION Precision Agriculture uses geophysical techniques, remote sensing, terrain attributes and yield monitor data to implement and improve site-specific management (SSM) strategies. On-farm research (OFR) is conducted to assess the economic potential of different SSM approaches. First we describe a case study to illustrate the application of spatial linear models in this context and also some of the challenges associated with OFR data. We then compare different statistical analysis approaches in a simulation study. This expands upon previous related studies that dealt mainly with small data sets or a smaller set of methods (e.g. Gotway-Crawford et al., 1997; Long, 1998; Ver Hoef and Cressie, 2001; Lambert et al., 2003; Anselin et al., 2004; Liu et al., 2006).