Enhancing snap bean yield prediction through synergistic integration of UAS-Based LiDAR and multispectral imagery

• A UAS-based LiDAR and MSI system effectively predicted crop yield as early as the bloom stage with reduced labor costs. • The MSI-LiDAR fusion model outperformed LiDAR-only, MSI-only, and traditional NDVI or lidarLAI benchmark models. • Novel LiDAR features, such as the M2VR metric, improved yield prediction over traditional methods. Unmanned aerial system (UAS)-based remote sensing technologies have seen increasing application in precision agriculture and crop management. By collecting and analyzing remote imagery of crops, we can extract valuable information about the nutrient, structure, and growth variation of crops at scales < 0.05 m, and then perform yield prediction at the sub-plot level. Here we utilized data from 14 unmanned aerial system (UAS) flights, equipped with both light detection and ranging (LiDAR) and multispectral cameras, to monitor snap bean field trials over two years. Our approach involved constructing predictive models using LiDAR-derived structural features, multispectral-derived spectral features, and a fusion of these methodologies. We benchmarked our results against the normalized difference vegetation index (NDVI) and LiDAR-derived leaf area index (LAI). The models based on LiDAR and multispectral imagery (MSI) individually outperformed these benchmarks, each showing unique strengths at different crop growth stages. The fusion method, integrating LiDAR and MSI, significantly enhanced model performance, achieving an adjusted R-squared (Adj.R2) of 0.827 and a root relative squared error (rRMSE) of 0.094 at harvest in the second year’s data. Notably, this fusion-based approach attained good accuracy (Adj.R2 > 0.74, rRMSE < 0.13) by the full bloom stage, at least a week before harvest. This early prediction capability is crucial for effective agricultural planning and management. Our findings underscore the potential of combining multiple UAS-based remote sensing technologies for efficient and accurate yield assessment and prediction in agriculture.