A cost-effective approach of building plant digital twin combining plant phenotyping and growth models: A case study on rapeseed plants

High-throughput phenotyping platforms enable dense temporal monitoring of plant growth, yet field-based image collection remains labor-intensive with sparse sampling intervals, leading to missing information on intermediate developmental stages. Functional–structural plant models (FSPMs) offer dynamic growth simulation but often lack realistic 3D architecture derived from actual plants. This study proposes a lightweight framework that bridges sparse phenotyping time points by integrating 3D reconstruction with the GreenLab plant growth model, enabling the construction of plant digital twins for rapeseed ( Brassica napus L.). Multi-angle smartphone images of four rapeseed cultivars at two key growth stages were used to reconstruct 3D plant architecture via 3D Gaussian Splatting (3DGS). Leaves were segmented from the point cloud to calculate area. GreenLab source and sink parameters were estimated from organ-level fresh weights (leaves, internodes, and roots) measured at two growth stages. The goodness–of–fit was high, with weighted sum of squares (WSS) between 5.63 and 9.33 and Pearson correlation coefficients ( r ) ranging from 0.93 to 0.95 across the four cultivars. The calibrated model reproduced leaf areas consistent with 3D reconstructed static structures. Reconstructed organ geometries were used for visualization and functional analysis of light interception. The integrated model thus recovered the complete dynamic development and growth trajectory between sparse sampling points. Temporal hold-out validation showed reasonable agreement between reconstructed leaf area and length–width–derived reference estimates, with a single-leaf mean absolute percentage error (MAPE) of 18.42%. Light interception analysis revealed cultivar-specific differences in canopy light use efficiency attributable to contrasting morphological architectures. Simulated and reconstructed canopy light interception showed strong agreement ( r = 0.98 ). This framework provides a cost-effective approach to reconstructing plant growth dynamics and evaluating light capture characteristics from minimal field data. It offers practical utility for breeders in cultivar evaluation and for optimizing environmental control strategies in precision agriculture.