Los puntos clave no están disponibles para este artículo en este momento.
Precision agriculture predominantly relies on aboveground data, limiting the direct detection of growth constraints caused by subsurface soil conditions. To improve understanding of spatiotemporal root systems dynamics, we aimed (1) to evaluate the relationship and relative performance of a minirhizotron method using deep learning-based image analysis in comparison with the trench profile method, with soil core sampling used as a common reference, based on statistical comparisons of association strength, regression structure, and error-based accuracy metrics; and (2) to determine whether two-dimensional root distribution dynamics can be captured using images from five minirhizotron tubes installed perpendicular to the planting row. Deep learning-based image analysis accurately extracted roots from minirhizotron images, eliminating interference from water droplets and soil color variation. The minirhizotron method showed a stronger association with core sampling than the trench profile method, as confirmed by statistical comparison of dependent correlations. Error-based accuracy metrics derived from Deming regression consistently demonstrated lower prediction errors for the minirhizotron method across soil depths and growth stages, indicating superior predictive performance relative to the trench method. Furthermore, we successfully predicted two-dimensional root distribution dynamics (approximately 90 cm in width and 60 cm in depth). Although the minirhizotron method tended to overestimate roots below 30 cm after the heading stage, previous studies suggest that root observations within the top 30 cm during stem elongation are sufficient to detect cultivar and environmental differences. These findings demonstrate that deep learning-assisted minirhizotron imaging provides a reliable, non-destructive approach for monitoring root distribution dynamics in agricultural fields.
Hashimoto et al. (Fri,) studied this question.