LINKING MULTIPLE LAYERS OF INFORMATION FOR DIAGNOSING CAUSES OF SPATIAL YIELD VARIABILITY IN SOYBEAN

Soybean yields are highly variable across fields because of the complex interaction of many factors, including weather, management, soil properties, fertility, pests, and weeds. However, only limited progress has been made on techniques for diagnosing reasons for yield variability and for identifying and managing different areas of the field to maximize profit or minimize environmental risk. The objective of this study was to develop a crop model–based technique to attribute yield losses due to water stress, soybean cyst nematodes (SCN), soil pH, and weeds that may cause soybean yield variability. The procedure computes yield as a function of: (a) site–specific soil water parameter inputs to a crop model, and (b) residual site growth and yield–reducing factors not accounted for by the crop model and its spatially variable soil water parameters. We used a two–year data set from Iowa (1995 and 1997) to estimate parameters and evaluate yield predictions using a combined crop model–statistical regression approach. The data were modified by imposing pH, SCN, and weed stress to 8 of 60 grids and a combination of these stresses to 8 grids. Thus, the dataset was constructed based on the both natural (field measured) and artificial yield variation. This was done to evaluate the accuracy with which the proposed technique could quantify losses. The model indicated that water stress reduced soybean yields by an average 1092 and 710 kg ha –1 for 1995 and 1997, respectively. After taking into account water stress, pH, SCN, and weeds, the combined approach was able to explain 96% of yield variability over two years of data. The RMSE was 143 kg ha –1 . Our technique was able to reproduce the quantities of yield loss for each grid and attribute them to the correct causes. Standard errors of attribution accuracy were 185, 220, and 163 kg ha –1 for soybean yield losses due to soil pH, SCN, and weeds, respectively. The combined crop model–regression technique was able to reproduce an observed spatial grid of yield according to both natural and imposed yield variation, and to quantify losses due to several factors.