Classification of Grass and Forb Species on Riverdike Using UAV LiDAR-Based Structural Indices

Abstract

Herbaceous vegetation on riverdikes plays an important role in preventing soil erosion, which, otherwise, may lead to the collapse of riverdikes and consequently, severe flooding. It is crucial for managers to keep suitable vegetation conditions, which include native grass species such as Imperata cylindrica, and to secure visibility of riverdikes for inspection. If managers can efficiently find where suitable grass and unsuitable forb species grow on vast riverdikes, it would help in vegetation management on riverdikes. Classification and quantification of herbaceous vegetation is a challenging task. It requires spatial resolution and accuracy high enough to recognize small, complex-shaped vegetation on riverdikes. Recent developments in unmanned aerial vehicle (UAV) technology combined with light detection and ranging (LiDAR) may offer the solution, since it can provide highly accurate, high-spatial resolution, and denser data than conventional systems. This paper aims to develop a model to classify grass and forb species using UAV LiDAR data alone. A combination of UAV LiDAR-based structural indices, V-bottom (presence of vegetation up to 50 cm from the ground) and V-middle (presence of vegetation 50–100 cm from the ground), was tested and validated in 94 plots owing to its ability to classify grass and forb species on riverdikes. The proposed method successfully classified the “upright” grass species and “falling” grass species / forb species with an accuracy of approximately 83%. Managers can efficiently prioritize the inspection areas on the riverdikes by using this method. The method is versatile and adjustable in other grassland environments.