Photogrammetry and 3D Gaussian Splatting for 4D plant growth capture and greenhouse digital twins

Plants are inherently three-dimensional organisms, yet many biologically important traits remain difficult to document using two-dimensional photography because of self-occlusion and complex plant architecture. Image-based three-dimensional (3D) digitisation offers an accessible alternative using standard RGB cameras, while recent advances in 3D Gaussian Splatting (3DGS) enable photorealistic real-time visualisation of reconstructed scenes. We present a practical workflow that combines Structure-from-Motion/Multi-View Stereo (SfM/MVS) photogrammetry with 3D Gaussian Splatting for botanical digitisation across multiple spatial scales. The workflow was evaluated using a 12-timepoint ("4D") fern growth series, a greenhouse digital twin assembled from fourteen independently reconstructed sections and additional datasets including a smartphone-photographed flower, two macro-imaging workflows, a potted plant, a drone-based tree reconstruction and a controlled wind experiment. Global multi-day image alignment was used to establish a common coordinate system for the fern time series, allowing individual daily reconstructions to share the same scale and orientation without repeated fixed camera positions. Across the fern time series, photogrammetric reconstructions produced 95,698-325,415 tie points with RMS re-projection errors of 1.36-1.72 pixels. Corresponding 3DGS models achieved PSNR values of 21.99-26.35 dB, SSIM values of 0.726-0.842 and LPIPS values of 0.223-0.339. Controlled macro-imaging produced the highest perceptual quality, whereas a dedicated wind experiment demonstrated a marked reduction in reconstruction fidelity caused by leaf movement. The greenhouse sections were manually assembled into an interactive digital twin suitable for visual exploration, although not intended as a metrically validated spatial survey. The study demonstrates that consumer-grade cameras, photogrammetry and 3D Gaussian Splatting support reproducible botanical digitisation from macro-scale flowers to greenhouse environments. The presented workflow provides practical guidance for generating scalable 3D datasets for plant documentation, living collections, digital exhibitions and future quantitative analyses.