Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Proceedings
Publication Acceptance Date: June 16, 2004
Publication Date: July 1, 2005
Citation: Drummond, S.T., Sudduth, K.A. 2005. Analysis of errors affecting yield map accuracy. In: Mulla, D.J., editor. Proceedings of the 7th International Conference on Precision Agriculture, July 25-28, 2004. Precision Agriculture Center, University of Minnesota, St. Paul, MN. [unpaginated CDROM] Interpretive Summary: Yield maps are a key component of precision agriculture, used both in developing and evaluating precision management strategies. Unfortunately, yield monitors produce complex datasets that often contain a large number of errors. These errors should be removed, or at least well understood, as they may cause serious problems when yield data are used for research and analysis tasks. Even when yield maps are merely used in an interpretive fashion, or to visualize patterns in the data, the conclusions drawn by the user may be negatively affected by these error sources. In this study, automated and manual filters to detect and remove yield errors were investigated through application of the 'Yield Editor' software tool to several datasets. The results verified the need to remove errors from raw yield datasets and provided information about the relative importance of the various filter types. This research is a step toward a standardized procedure to clean yield data, a result which would improve data quality for both researchers and producers involved in precision agriculture.
Technical Abstract: In this study, five raw yield monitor datasets were examined to identify a variety of operational and sensor-related errors. These errors were then removed from the datasets, using both automated and manual filtering techniques. These filtering procedures removed 13 to 27% of the observations from the datasets, depending on the complexity of the harvest pattern. As a result, they increased mean yield by 3 to 11% while reducing the standard deviation of yield by 22 to 56%. Individual filtering techniques were evaluated to provide insight into the relative importance of the filters. Most of the filters were able to detect a large number of errors in one or more of the five datasets, although redundancy of detections between filters was evident. Of the automated filtering techniques, the start delay and end delay filters were most critical, with each identifying an average of more than 3% of the raw yield observations left undetected by the other filters. The manual filtering procedure was also important, removing an average of 4% of the raw yield observations. These results emphasize the need for a thorough, standardized procedure for filtering raw yield monitor datasets.