Exploring the potential of machine learning for leaf angle distribution type identification from leveled digital photography: A case study for broadleaf tree and shrub species

Abstract

Leaf angle distribution (LAD) is an important plant structural trait that determines radiation interception, biomass production, rainfall interception, and evapotranspiration. LAD has remained one of the most poorly constrained parameters and a major source of uncertainty in ecological models due to difficulty in quantification. In this study, we look for a simple, affordable way to estimate leaf angle distribution type by anyone, anywhere. For the first time, we test the possibility of determining LAD type from photographs (816 in total) of single plants from various broadleaf tree and shrub species taken from field works. We used Google's TensorFlow and built models using a convolutional neural network to classify these images by LAD type. The best results, training accuracy of 95% and validation accuracy of 91%, were acquired using the two most distinct LAD types – planophile and erectophile. The accuracy decreased with the addition of categories, as expected. However, our results indicate that the involvement of machine learning may indeed hold the potential to remove the current bottleneck in retrieving the information on LAD.