During the Fall 2018 soybean harvest, soybean samples were collected from 32 country elevator locations belonging to one Iowa-based cooperative which has its own elevator locations and processing plant. This was done to update historical data about geographic variations in protein and oil content of Iowa soybeans, and to assist the cooperative in making more informed decisions about their soybeans to maximize value potential. These samples were analyzed using near-infrared spectroscopy (NIR) to determine protein and oil contents. The data were accumulated and sorted to look for geographic variations in protein and oil content of soybeans throughout Iowa. The data were run through an Estimated Processing Value (EPV) model to determine value differences of soybeans between elevator locations. The cooperative source soybeans for processing from the elevator locations closest to the plant to mitigate trucking costs. They wanted to know whether this strategy was maximizing their net processing value. Results showed that significant variability between locations did exist, which represented a $0.23/bushel EPV spread. Additionally, it was found that 15 samples were needed to accurately represent an elevator location, and that two weeks was a sufficient period to characterize the data to be able to make a marketing decision. Lastly, the soybean quality was not found to vary significantly over the course of harvest, so marketing decisions can made at the beginning of the season.

An error analysis was also performed to find the effects of potential error on location separation, because errors would reduce the certainty of any marketing decisions based on measured value differences. Both random and systematic errors were possible with the use of NIR analyzers. Random errors were simulated using an Excel-based model that created random values with a specified standard deviation and mean, which were then added to the original data points. This simulation was performed for three test cases – one with typical standard deviations for protein and oil contents, one with higher-than-average standard deviations, and one with typical standard deviations but with a bias element added to a subset of the locations. The introduction of random error made any value gaps between locations smaller, which made discrimination of high-value locations from average or low-value locations difficult. These results showed the importance of having standards for measuring instruments if the soybean supply chain is ever to move to a protein and oil pricing basis, because one of the largest sources of error in a commodity-based market system is inconsistency of measuring units with each other.

Overall, geographic variability across the cooperative’s locations was evident, and testing inbound loads with an NIR analyzer, even during busy harvest days, was feasible to characterize soybean protein and oil content. However, the validity of marketing decisions made using the resulting data depends highly on the amount of error involved in sample analysis. Future studies should identify specific sources of error and attempt to eliminate them, because maximizing potential value capture will not be possible unless the value differences between locations are characterized as precisely as possible.