Cotton Growth Modelling Using UAS-Derived DSM and RGB Imagery

Abstract

Modeling cotton plant growth is an important aspect of improving cotton yields and fiber quality and optimizing land management strategies. High-throughput phenotyping (HTP) systems, including those using high-resolution imagery from unmanned aerial systems (UAS) combined with sensor technologies, can accurately measure and characterize phenotypic traits such as plant height, canopy cover, and vegetation indices. However, manual assessment of plant characteristics is still widely used in practice. It is time-consuming, labor-intensive, and prone to human error. In this study, we investigated the use of a data-processing pipeline to estimate cotton plant height using UAS-derived visible-spectrum vegetation indices and photogrammetric products. Experiments were conducted at an experimental cotton field in Aliartos, Greece, using a DJI Phantom 4 UAS in five different stages of the 2022 summer cultivation season. Ground Control Points (GCPs) were marked in the field and used for georeferencing and model optimization. The imagery was used to generate dense point clouds, which were then used to create Digital Surface Models (DSMs), while specific Digital Elevation Models (DEMs) were interpolated from RTK GPS measurements. Three (3) vegetation indices were calculated using visible spectrum reflectance data from the generated orthomosaic maps, and ground coverage from the cotton canopy was also calculated by using binary masks. Finally, the correlations between the indices and crop height were examined. The results showed that vegetation indices, especially Green Chromatic Coordinate (GCC) and Normalized Excessive Green (NExG) indices, had high correlations with cotton height in the earlier growth stages and exceeded 0.70, while vegetation cover showed a more consistent trend throughout the season and exceeded 0.90 at the beginning of the season.