| The capability of applying spatially variable inputs to field crops has triggered new opportunities in crop management (Site-Specific Management, SSM). The success of SSM depends on the ability to relate the status of a crop system with its physical performance. One of the current challenges of SSM is the implementation of a tool for that purpose. The objective of this dissertation is to develop and validate procedures to predict the spatial pattern of yield. Two procedures deserve special attention: artificial neural networks and crop simulation models. In addition we aim to understand the mechanisms that lead to spatial yield variation under soil-water limiting conditions.; A field experiment was carried out in a 3.6 ha corn farm field in Michigan, USA, for two years. In 1997, a dry year, plant population and plant-available soil-water, as affected by effective soil depth, soil-water holding limits and seasonal amount of rainfall, accounted for as much as 82% of the total final grain yield spatial variability (range of 6.6--12.0 t/ha). In 1998, with average seasonal rainfall amounts and a more uniform stand, plant-available soil-water alone was able to explain 68% of final grain yield spatial variability (range of 9.8--12.6 t/ha).; Effective soil depth was more strongly related to topography than sand and clay content. The topographic attributes that were most related to these soil properties were downstream flow length and distance to flow network. Corn growth and yield were consistently related to topography in both a dry and a normal rainfall year. These relations were stronger for the dry year, especially for downstream flow length and distance to flow network.; The neural network model that included all variables in the study produced smaller predictive errors (average of 918 kg/ha). The agronomic and topographic neural networks had RMSEP over 1000 kg/ha. CERES-Maize accurately predicted the spatial variability of corn growth and yield for two years of independent data when accurate inputs were provided for soil properties and plant population (average RMSEP of 501 kg/ha). |
