Title: Issues in analysis of soil-landscape effects in a large regional yield map collection Authors
Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Proceedings
Publication Acceptance Date: May 18, 2012
Publication Date: July 16, 2012
Citation: Myers, D.B., Kitchen, N.R., Sudduth, K.A. 2012. Issues in analysis of soil-landscape effects in a large regional yield map collection. Proceedings of the Annual Precision Ag Conference. 11th International Conference on Precision Agriculture, July 15-18, 2012, Indianapolis, Indiana, 2012 CDROM. Interpretive Summary: Producers now commonly use global positioning equipment and yield measuring systems called yield monitors during harvest to make maps of grain yield. Some producers have been yield mapping for over a decade and large collections of yield maps are accumulating. Precision agriculture service providers, seed, and chemical companies also accumulate these maps in data-warehouse sized collections with hundreds of thousands of acres of yield maps over many years. A key goal in using these yield maps is to understand how soils and landscapes affect grain yield and year-to-year risk in grain production. Past research has followed this track using statistical procedures to relate crop yield to topography and soil maps, but only for one or a few fields at a time. Our research investigated the use of data mining procedures to understand soil landscape impacts on grain yield and year-to-year risk on a large collection of yield maps over many counties. However we learned there are three major issues that cause difficulty in making this assessment. First, the soil-landscape effect on grain yield is different each year due to different climate. Second, there are data integrity issues with yield maps that cause problems for computer analysis. Third, yield maps generally do not carry important contextual details about the crop such as planting date, tillage system, irrigation, rainfall, or crop variety. These contextual details could be used along with topography and soil information to improve data mining results. We demonstrated the difficulty of using data mining and discuss how the three types of problems with yield maps could be causing this. We also recommend the use of a more complete data collection strategy for yield maps called a yield map data model. Better yield map data could improve farmers’ capability to optimize grain production practices and better manage year-to-year productivity risk, potentially increasing profitability and environmental sustainability.
Technical Abstract: Yield maps are commonly collected by producers and precision agriculture service providers and are accumulating in warehouse scale data-stores. A key goal in analysis of yield maps is to understand how climate interacts with soil landscapes to cause spatial and temporal variability in grain yield. However, there are many issues that limit utilization of yield map data for this purpose related to i) yield-landscape inversion between climate years, ii) sensor system malfunction and inaccuracy, iii) poor data management practices and operator error, iv) field configuration and logistical limitations, v) spatial, temporal, and producer variability in agronomic management, and vi) incomplete target and predictor dataspace. Each of these issues requires a significant effort to understand and then address by the commercial and research precision agriculture community. A key goal of this investigation was to use a regional extent yield map data warehouse to model the effects of soil landscape properties on site specific mean yield and yield risk. Data mining technologies were used to examine relationships between yield map data and soil landscape attributes. Our initial results indicate challenges in training data mining algorithms that produce stable estimates when applied to independent testing data both within and across years. We found the above factors reduce the effectiveness of data mining approaches. To improve this situation, we propose a more agronomically complete yield map data model to better populate important predictive information in yield map databases.