267. UAV-based Multispectral Monitoring to Predict Pasture Nutritional Value, Grazing Behavior, and Methane Production in Grazing Heifers.

Abstract

Abstract Pasture nutritive value is an important indicator of production efficiency in grazing systems. Remote sensing using unmanned aerial vehicles (UAVs) equipped with multispectral sensors could be an alternative to estimate pasture nutritional value through vegetation indices (VIs) and monitor animal feeding behavior. Such variables could also be used to predict methane production. Therefore, this study aimed to (1) predict pasture nutritional value and diurnal grazing time using UAV multispectral images and (2) predict methane production in grazing heifers using vegetation indices, body weight, image-based body features, and grazing time obtained through UAV multispectral images. Multispectral images were captured using a DJI Mavic 3M UAV equipped with a 5-megapixel camera featuring four multispectral bands: Green, Red, Red Edge, and Near Infrared. Data collection occurred over nine consecutive days, with the UAV flying at an altitude of 20 m between 8:00 a.m. and 7:00 p.m., with flights conducted at 10-minute intervals. To predict pasture nutritional value, VIs computed from multispectral images were used as inputs, and a leave-one-pasture-out cross-validation approach was employed for model validation. To predict the diurnal grazing time, the following inputs were included in the model: body weight, body features extracted from UAV images (e.g., body area, width, and length), pregnancy status as a categorical variable, and VIs measured before the grazing period. To predict methane production, the model incorporated additional inputs including diurnal grazing, lying and standing times, manually annotated from UAV observations, VIs measured at the individual animal-level for each observed grazing event, and the average and standard deviation of CP, NDF, ADF, and IVOMD predicted using the model developed in objective 1. Models for grazing behavior and methane prediction were validated using a leave-one-animal-out cross-validation approach. To predict CP, NDF, ADF, and IVOMD, the model achieved R2 values of 0.71, 0.67, 0.62, and 0.65, with RMSE values of 1.93%, 2.54%, 2.46%, and 3.03%, respectively. For predicting diurnal grazing behavior, the model achieved an R2 of 0.41, CCC of 0.57, and RMSEP of 9.5%, representing 16.7% of the average observed value. The methane prediction model achieved an R2 of 0.42, CCC of 0.59, and RMSEP of 41.8 g/d, corresponding to 22.9% of the average observed methane production. These results demonstrate the potential of UAV-based monitoring to improve the estimation of pasture nutritional value, grazing behavior, and methane emissions, supporting more efficient and sustainable grazing systems.