Abstract: Digital modeling of fruit trees has been one of the most fundamental technologies in smart orchards of intelligent agriculture. The procedure has evolved from geometric reconstruction toward an integrated framework with multi-source sensing, structural reconstruction, physiological simulation, and data-driven decision support. This study aims to systematically review the status, representative applications, and future trends of data-driven fruit tree modeling in the orchard. Three dimensions included multi-source data acquisition and fusion, digital model construction, and intelligent applications. (1) Sensing technologies were examined to support orchard digitalization. An integrated "satellite-aerial-ground-station" sensing network gradually emerged to realize orchard observation at multiple scales. Typical data included point clouds for explicit 3D structural representation, RGB imagery for texture and multi-view geometric constraints, spectral and thermal data for physiological and biochemical state inversion, and environmental parameters for dynamic driving and boundary conditions. Major fusion pathways were summarized as the cross-platform structure, 2D-3D semantic and attribute, structure-physiology, and multi-temporal process-driven fusion. Digital representations were obtained from spatially explicit, semantically rich, and temporally evolving fruit trees. (2) Digital models were classified into physical structural, process mechanistic, and data-driven models. Physical reconstruction was selected as LiDAR, RGB-D imaging, SfM/MVS photogrammetry, NeRF, and 3D Gaussian Splatting. Branches, fruits, canopies, and roots were reconstructed to quantify canopy architecture, organ distribution, and operational space. Process models were used to simulate the growth, photosynthesis, respiration, water transport, nutrient uptake, and carbon allocation under different environments, particularly for yield prediction and irrigation scheduling in the orchard. Functional-structural plant models also coupled physiological processes with explicit 3D structures to simulate organ-level interactions between plant form and function. Meanwhile, statistical learning, machine learning, deep learning, and multimodal collaborative modeling performed best to capture nonlinear relationships among orchard structure, physiological status, and environmental variables. Together, these model types gradually formed a coordinated modeling framework in smart orchards. (3) Digital fruit tree models were applied in growth monitoring and diagnosis operations. Structural, physiological, and environmental indicators were extracted from individual trees to whole orchards, including water stress detection, nutrient assessment, disease and pest identification, and interpretation of yield and fruit distribution, particularly in robotic harvesting, precision spraying, pruning assistance, localization, navigation, and semantic mapping. Agricultural IoT and end-edge-cloud architecture were integrated to promote the transition from static digital representation to continuously updated digital twins. Edge computing supported real-time local inference and control, while cloud platforms provided long-term data governance, model training, and global optimization. The challenges remained, including sensing robustness and semantic consistency under complex orchard conditions, limited integration of mechanistic and data-driven models, weak adaptation to dynamic growth, and immature large-scale deployment under edge-resource constraints. Future research should focus on embodied intelligent sensing and active perception, physics-informed large-model decision-making, real-time interactive digital twins, and efficient end-edge-cloud collaboration. These findings can be expected to improve the accuracy, adaptability, and practical value of orchard digital modeling. Technical support can also be provided for intelligent equipment in a sustainable smart orchard.