Developing a Technique for Evaluating Weed-Specific Mapping Systems

Federal regulation and public awareness of agricultural chemical use have fueled precision agriculture research for the last decade. An extensive body of research on potential reduction of herbicide inputs by automated patch-spraying or site-specific management has developed. Two dominant methods have developed for site-specific application of herbicide. Map-based systems use predefined application maps to direct herbicide application and sensor-based systems use real-time weed sensors to identify and treat weeds as the sprayer moves through the field. Weed maps, generated for map-based application of herbicide are beneficial for out-of-field decision-making but are labor intensive to create and sensitive to many types of sampling errors. Real-time sensor-based systems are not as labor-intensive but have historically made no record of what parts of the field received herbicide and are subject to weed discrimination errors. The University of Tennessee Weed Mapping System (UTWMS) is made up of a digital event recorder and a WeedSeeker discrete herbicide application system. The overarching objective of this study was to evaluate the UTWMS under field conditions. Specific objectives included the use of georeferenced manually-sampled plots for evaluation of map accuracy; development of an automated documentation system for quantifying hits, misses, and false triggers of a real-time sensor-based spraying system; updating the logging software of the UTWMS to include a count of spray transitions; and investigate potentials for reducing number of sensors to reflect the existing spatial correlation of weeds. Manually sampled subplots at one-meter resolution did not correlate with weed maps and only weakly correlated when averaged by plot (8x30m). A video documentation system was successfully developed for evaluating discrimination accuracy of sensor-based sprayers. While investigating sensor resolution reduction to reflect spatial correlation of weeds, a sensor was replaced with a conditionally triggered solenoid valve during a simulation. More than 75% of the simulated weeds were accurately sprayed for all four conditional scenarios tested. A software modification to the UTWMS provided enumeration of spray transitions for weed scientist to investigate weed distribution during “percent time on” integration. The update rate of the GPS unit in the UTWMS should be increased if weed maps are to be representative of small research subplots.