Abstract: Abstract: The fruit tree canopy structure determines the canopy illumination spatial distribution (CISD) in every growth stage, and especially in the stable stage, the CISD degree is one of the most important factors related to fruit yield and quality. In the fine management of orchards, in the quiescent stage, trees were trimmed to ideal structures in order to get the high quality CISD in the stable stage. As we all know, the trimming effect directly affects the distribution of the CISD. To analyze the CISD associated with the spring pruning in the apple tree canopy, spindle-shaped apple tree was taken as the research object, and a predicting methodology, the illumination spatial distribution prediction method considering apple tree canopy fractal features, was proposed based on the three-dimensional (3D) canopy structure. In this study, the canopy was divided into cell grids regularly. The Trimble TX8 was used to get the 3D apple tree canopy point clouds, and then the tree canopy point clouds were cut into cell grids as the same size as in the orchards, too. To describe the different structures within the canopy, each cell grid was projected on a horizontal surface, which was perpendicular to the tree trunk and parallel to the ground. The projection of each grid was different, but similar with each other. Because of these characteristics, a new box-counting dimension based on 3D point clouds projection approach of the cell grid was used to describe different cell grids' spatial structure. Onset light intensity acquisition system was used to obtain the illumination intensity in each cell grid in the stable growth stage from June to September. Further, the average illumination intensity of corresponding cell grid was calculated. In the prediction research work, a practical new hybrid model to predict trimming effect based on the relative CISD in apple tree canopy was proposed. The model was based on particle swarm optimization (PSO) in combination with support vector machines (SVMs). This optimization mechanism involved kernel parameter setting in the SVM training procedure, which significantly influences the regression accuracy. The kernel function was determined by analyzing the influence of 4 different kernel functions (linear, polynomial, RBF (radial basis function) and sigmoid) on prediction accuracy, mean squared error and squared correction coefficient. The comparative analysis result was that the RBF kernel function achieved the expected result. The prediction model was based on the statistical learning theory and goodness of fit to experimental data, and successfully used here to predict the relative CISD. This combination of 2 different descriptors, which represented 2 features of a cell-grid, was utilized for subsequent classification (invalid light area or high quality light area) by employing the model. In the field experiment, the cell grid size was 0.4 m × 0.4 m × 0.4 m. All the cell grids were divided into 2 groups, facing to the sun (FS) and backing against the sun (BS). The model in this paper was trained by the data in 2014, 2015 and 2016, and then predicted the relative CISD in 2017. At the same time, the model was trained by the data in 2015, 2016 and 2017, and then predicted the relative CISD in 2018, too. The experimental results showed that the classification prediction accuracies of the FS model and the BC model were 76.11% and 74.10%, respectively, which indicated the good performance of the proposed method. The specific method proposed in this paper can make a contribution to the fruit quality management of apple orchard.