Evaluating the Efficacy of Targeted Pesticide Applications and Machine Learning Integration for Precision Turfgrass Management

Abstract

There is an increasing demand for sustainability in turfgrass pest management, emphasizing the need to reduce chemical inputs while maintaining effective pest control. Precision turfgrass management (PTM), a subsect of precision agriculture, addresses these sustainability goals by tailoring management practices to the site-specific variability of the landscape. Targeted pesticide applications are a fundamental practice within PTM, involving precisely delineating pest outbreaks to create pest maps. These pest maps are then used as a guide for Global Navigation Satellite System (GNSS)-guided sprayers to apply pesticides solely to pest-affected areas as opposed to traditional broadcast pesticide applications. However, significant barriers have hindered the widespread adoption of these practices. A prominent limiting factor to the adoption of GNSS-guided sprayers is turfgrass manager's skepticism of the technology's accuracy due to a lack of documented quantifiable data demonstrating its capabilities. Additionally, the considerable labor, technical skills, and resources needed to create spatially accurate pest maps create another substantial barrier for turfgrass practitioners. An investigation into the accuracy and precision of GNSS-guided sprayers examined the influence of operational parameters such as travel speed and target size on spray deposition patterns. Fluorescent dye imaging under ultraviolet illumination, coupled with digital image analysis, quantified spray accuracy, overspray, and overlap. Results showed that the sprayer applied solution less accurately when operated at 4.8 kilometers per hour (km/h) compared to 7.2 or 9.7 km/h. Target size had no significant impact on any metric tested, indicating that targeted pesticide applications made while traveling 7.2-9.7 km/h were highly accurate, no matter the target size. An automated solution for pest mapping using machine learning was developed to identify and quantify dollar spot (Clarireedia spp.), a common turfgrass disease. A semantic segmentation model (DeepLabV3+) was trained on diverse images featuring various turfgrass species, disease severities, and environmental conditions. The developed model successfully automated the identification and quantification of dollar spot, offering significant improvements in efficiency, consistency, and accuracy compared to manual visual assessments. Overall, these findings demonstrate the efficacy of GNSS-guided sprayers under specific operational conditions and validate machine learning as a viable method for automating pest mapping. Future research should focus on optimizing technological and operational aspects to improve the practicality and effectiveness of precision turfgrass management.