Advancing pasture biomass prediction with integrated proximal, multispectral, topographic and SAR data fusion

Accurate pasture biomass prediction is central to precision grazing and sustainable land management. This study presents a multi-source biomass prediction model for Mid-Missouri test-site pasture by integrating field-based proximal height sensing, multispectral satellite derived vegetation indices and weather variables from 2024 - 2025. A ridge regression framework with L2 regularization addressed predictor multicollinearity, with cross-validated tuning yielding an R² of 0.92 and a mean absolute error of 388 kg/ha, representing an approximately 50 percent improvement over height-only models. These results confirm the effectiveness of fusing proximal, spectral, and meteorological data for paddock-scale biomass estimation. Further gains in prediction accuracy can be achieved through systematic expansion of the predictor space within the existing multi-source framework. Incorporation of synthetic aperture radar (SAR) metrics from Sentinel-1, including backscatter coefficients and spatial texture measures derived from gray-level co-occurrence matrices, is expected to improve sensitivity to canopy structure, surface roughness, and moisture dynamics while maintaining robustness under cloud cover. In addition, terrain-based variables, including elevation and slope, will further explain spatial variability in pasture growth. This integrated framework is expected to reduce residual uncertainty, improve model stability across seasons, and enhance species specific calibration, providing a scalable foundation for highly accurate pasture biomass prediction and advance sustainable pasture management practices.