Get reliable rangeland science

Controlling the spatial distribution of cattle dung pats in a pasture can enhance nutrient utilization and mitigate greenhouse gas emissions resulting from dung in grazing management. Dung pats are distributed in accordance with cattle behavior and location in a pasture, thus frequent monitoring of cattle location and dung distribution is essential for effective dung control. Recently, the detection of d ung pat distribution has been achieved using unmanned aerial vehicle (UAV) images. Additionally, a global positioning system (GPS) can provide continuous monitoring of cattle locations. Therefore, in this study, we monitored the effect of cattle location on dung distribution under strip stocking using UA V images and a deep learning approach. Five dairy cows, equipped with GPS collar s, grazed for five days under a strip stocking condition. A 3.5 ha pasture was divided into four paddocks, one of which was expanded each day. UAV images were captured before grazing each day. The training data generated were used t o estimate dung pat distribution using YOLO (YOLOv8x), an object detection algorithm. The accuracy of the dung distribution was assessed using the confusion matrix. The paddock was further divided into 10 m grids, and a generalized linear model was employed to evaluate the relationship between cow location and dung pat count within each grid. The detection accuracy of dung distribution was 0.793 (precision), 0.222 (recall), 0.210 (accuracy), and 0.347 (F-value), indicating the need to improve the accuracy of detecting undetected dung pats. As the pasture area increased, the cows spread out their location, resulting in an expanded dung distribution. However, the cow location did not correlate with the dung location (R2 = 0.053). This is presumably due to the insufficient recall of the dung distribution estimates. Additionally, not only the location, but also the cow's behavior, such as resting and lying, should be assessed.