Prediction of biomass and nutritional quality of tropical pastures using multispectral analysis and machine learning models

Summary

This study demonstrates the application of machine learning models integrated with multispectral remote sensing to predict tropical pasture biomass and nutritional quality without destructive sampling. Extra Trees algorithms achieved the strongest predictive performance for yield, whilst SHAP analysis revealed differential importance of vegetation indices—VARI for yield prediction and NDRE for nutritional variables. The findings support adoption of non-destructive, data-driven approaches for optimising forage management in tropical livestock systems.

Regional applicability

This research is directly applicable to UK temperate pasture systems where similar remote-sensing and machine learning approaches could improve grassland monitoring and forage quality assessment, though the specific vegetation indices, species composition, and tropical agroclimatic context mean direct transfer would require validation with UK pasture species and phenology.

Key measures

Coefficients of determination (R²); vegetation indices (VARI, NDRE); biomass yield; nutritional attributes; model interpretability via SHAP analysis

Outcomes reported

The study evaluated machine learning models' ability to predict biomass production and nutritional value of tropical pasture species using multispectral vegetation indices. Model performance was assessed through coefficients of determination (R²), with interpretability analysed via SHAP framework.

Topic tags