Estimation of the Leaf Area Index of Field Plants Based on Multi-Resolution Point Cloud Phenotype Reconstruction

Abstract

The Leaf Area Index (LAI) is a key parameter in crop yield estimation and growth monitoring. LAI is also an important canopy structural characteristic of crops that controls their biophysical processes and respiration. Researchers have mostly used remote sensing images and LiDAR to estimate the LAI; however, these two estimation methods do not consider estimation accuracy or efficiency. This study seeks to remedy the shortcomings of these methods using UAV photogrammetry with tobacco in the mature stage as an example. UAV photogrammetry is convenient and efficient in generating images that provide a good spatial description. In this study the collected UAV images were matched with feature points to generate a dense point cloud, which was then used to construct three-dimensional tobacco points. The cloud phenotype model uses the Lambert spherical coordinate system to convert the three-dimensional coordinates into spherical coordinates, and extracts the target plants to calculate the porosity, effective leaf area index, and clustering index to obtain the real leaf area index. The results calculated by the hemispheric photography method were used as reference values to examine the accuracy of the leaf area index calculation at different spatial resolutions at the individual plant and plot scales. The results show that: (1) The LAI estimated from the three-dimensional point cloud data was generally higher than the calculation results of the hemispheric image, but the overall calculation accuracy was higher. The calculation results of the four spatial resolution models were compared with those of the hemispheric images, yielding coefficients of determination R² of 0.959, 0.931, 0.967, and 0.985; the relative errors RE were 11.87%, 19.74%, 14.96%, and 11.79%; the root mean square errors (RMSE) were 0.150, 0.195, 0.136, and 0.094; and the rRMSE values were 20.81%, 26.97%, 18.87%, and 13.10%. (2) Of the four spatial resolutions, the three-dimensional point cloud model with the highest calculation accuracy (87.29%) had 2.15 cm spatial resolution. (3) The results of the leaf area calculation at the plot scale showed that the calculation results of the four three-dimensional point cloud models were all within 0.05 of the hemispheric image calculation results, and the 2.15 cm spatial resolution model showed the best calculation accuracy, 94.24%. (4) Therefore, it is feasible, accurate, and efficient to obtain orthophotos through UAV photogrammetry, match feature points, and build a three-dimensional point cloud phenotype model to calculate the leaf area index of field plants. This method can provide important technical support and methodological references for the management and accurate yield estimation of other field crops and can also provide a technical means and scientific basis for precision agricultural planting and high-quality agricultural development.

Keywords

• leaf area index
• uav point cloud
• porosity
• tobacco