Deploying UAV multispectral imaging sensors to assess vegetation and identify a historic river channel in the Eddleston Water catchment, Scottish Borders

Floodplain depressions, such as historic river channels, are frequently characteristic of high water levels, causing vegetation aeration stress. Therefore, vegetation health changes are visible across historic channels. Unmanned aerial vehicle (UAV) multispectral imaging has successfully been used to classify vegetation species, however, success has often been limited to the use of coarse resolution data, as using fine-scale data results in overlapping spectral signatures between species. Analysing histogram separability of spectral responses is a technique applied in archaeological literature to identify fine-scale spectral variations. This technique has not been applied to applications such as identifying vegetation changes over varying floodplain topography. This study assesses the effectiveness of UAV multispectral data to detect a historic river channel in the Eddleston Water catchment, Scottish Borders. Supervised classification of vegetation species and spectral separability of different multispectral band combinations across regions of the historic channel compared to the surrounding field was assessed. Aims of the study were met with spectral separability identified between regions, particularly using red-edge and near-infrared bands, highlighting the potential of multispectral data to identify the channel. Best performance was found using the red-edge chlorophyll vegetation index map. Digital elevation models (DEMs) produced from UAV red, green, blue (RGB) structure-from-motion and LiDAR point clouds were required to confirm historic channel location as results were not statistically confirmable. Therefore, further work is recommended to investigate the best performing combination of multispectral bands to detect fine-scale changes in vegetation and confirm detection of historic river channels without requirement of DEMs.