Spatial statistical analysis of on-site crop yield and soil observations for site-specific management

摘要

From an agricultural engineering point of view, a lot of progress has recently been made with respect to satellite-based positioning systems, automatic yield mapping, and remote sensing of the land surface. However, concepts for the analysis of on-site field processes of soil and crop state are mainly limited to ordinary regression approaches that are based on deterministic relations. In this study alternative statistical techniques are evaluated to better understand spatial processes as a basis for expectation spatial crop yield patterns. A technique that seems predestinated in this context is state-space analysis, bacause it is not limited to a singular response function applicable for the entire field. Data sets with different spatial resolution and variable combinations were examined. Although a relatively intensive soil sampling campaign was undertaken in one field site, the properties selected contributed only little to an explanation of yield variation, because crop response changed spatially, and sampling resolution was too coarse. In another site, yield data for different years were available as well as remote sensing images of crop state and soil mapping information. With the higher resolution these relatively simply obtainable variables in combination with state-space analysis provided a good basis for predicting the crop yield pattern. In a third experiment classical ANOVA was affected by soil heterogeneity leading to erroneous conclusions in a comparison of fertilizing strategy treatments. Only after removing local correlation structure by nearest neighbor analysis treatment effects could be identified.