Koji SHIMADA

Enhancement of a Vertical Distribution Estimation Model for Rice Canopy Point Clouds Acquired by Drone LiDAR Measurements

Kazuyoshi TAKAHASHI

　In recent years, with the advancement of smart agriculture, three-dimensional measurement of rice paddy fields using drone LiDAR has gained increasing attention. Drone LiDAR is an effective measurement technique that is less affected by weather and lighting conditions and enables non-destructive, comprehensive capture of entire field structures. However, previous studies have reported that the vertical distribution of rice canopy point clouds obtained through drone LiDAR strongly depends on variations in laser spot size caused by differences in flight altitude and LiDAR systems. Therefore, interpretation of such data must account for differences in observation conditions. Although earlier studies proposed vertical distribution estimation models considering laser spot size and demonstrated their effectiveness under specific conditions, limitations remained regarding applicability during early growth stages and under varying observation environments.
　To address these issues, this study developed a vertical distribution estimation model capable of representing laser spot size effects in greater detail and evaluated its reproducibility under different LiDAR systems and flight altitude conditions. Compared with the existing model, the laser reflection process was refined, and a new metric was introduced that treats laser spot size as a continuous variable rather than a discrete one. This modification enables the model to represent continuous variations in spot size resulting from differences in flight altitude and sensor characteristics.
　The model was evaluated using drone LiDAR data collected in fiscal years 2024 and 2025. Similarity between measured and estimated vertical point cloud distributions was assessed using RMSE. Results showed that for data acquired in June 2024, corresponding to the early growth stage, the proposed model reduced RMSE by approximately 60% on average compared with existing models, significantly improving reproducibility. Furthermore, stable performance was confirmed under different flight altitudes and LiDAR systems, demonstrating the robustness and practical potential of the proposed approach.