AN INTEGRATED INFORMATION SYSTEM FOR ON-FARM PRECISION AGRICULTURE EXPERIMENTATION DATA USING MACHINE LEARNING APPROACHES

ABSTRACT Precision agriculture has grown along with advances in farming, engineering, and computing. Modern farm equipment is now capable of spatially sensing yields and spatially varying seeding rates and input application rates. High resolution sensors capture these data, creating a detail log of on-farm agronomic operations. Combining this with remote sensing and environmental data presents many opportunities for gaining insight into and improving modern agronomic practices. While these large volumes of data hold great potential, there are several associated challenges that must be overcome before they are truly useful. One of the most difficult is transforming these mountains of data into information. This must occur before it is possible for farmers to gain knowledge from their data. This project addresses this challenge through the development of a prototype agronomic information system. A focus is on process automation, freeing researchers from spending large amounts of time performing repetitive tasks, such as finding environmental data that is spatially and temporarily relevant to agronomic sensor data. Another is presenting the meaningfully combined data in a manner that is useful to farmers and other researchers. An automated spatial data framework was developed to collect, filter, and combine agronomic and relevant environmental datasets into a spatial data matrix. Various predictive yield modeling techniques were evaluated using the custom matrix data. The most efficient modeling technique was determined. A web portal was developed to collect and display agronomic data and to serve as a front end to the spatial data framework and predictive yield modeling processes, allowing farmers to view spatial maps of their data and to request on-demand predictive yield models throughout the crop cycle, transforming disparate, raw datasets into agronomic intelligence