Randomized trial evaluates biomass production and nutritional value in tropical pastures, indicating effective forage management tools.
Key Points
This research aims to assess the effectiveness of machine learning models in predicting biomass and nutritional value of various pasture species using multispectral data.
Evaluated machine learning models including linear and polynomial models, tree-based algorithms, and support vector machines.
Used a multispectral sensor on a DJI Matrice 350 RTK platform, collecting agronomic and nutritional data.
Applied Group Shuffle Split for model validation, considering pasture species as the grouping variable.
Extra Trees model achieved the highest coefficients of determination (R²) for yield-related variables.
VARI significantly contributed to yield predictions while NDRE was consistent for nutritional attributes.
Overall predictive performance for yield was better than for nutritional quality.