Effect of water column on water depth derived from unmanned aerial vehicle multispectral image

Abstract

The term water depth refers to the depth of the water body relative to the level of the water surface. In the remote sensing approach, water depth is determined using indirect methods that retrieve the bottom level of the water body without physically touching it. In the context of satellite imaging, water depths were mapped by using two radiative transfer models, namely Depth Invariance Index (DII) and Bottom Reflectance Index (BRI). However, the estimation of water depth by using airborne and satellite-borne pose error of water column. The water depth mapping by using UAV was recently conducted through the Structure from Motion (SFM). Therefore, this study presented the effects of water column correction on the multispectral Unmanned Aerial Vehicle (UAV) image to derive water depth. The following objectives were realised, firstly to retrieve the water-leaving radiance from all target points of different depths; secondly to model the water depth by applying the radiative transfer model to water-leaving radiance and; thirdly to determine and assess the effects of the water column on depths derived from UAV image. A total of six different sets of targets which consisted of forty four different depths had been deployed at Universiti Teknologi Malaysia’s swimming pool. DII and BRI radiative transfer models were used to minimise the column error on the imagery. The results showed that both radiative transfer models produced lower accuracy than direct modelling with water column correction. The best depth modelled was obtained by using log regression with band 1, which reported an accuracy of 0.042m, compared to the images corrected with BRI and DII with 0.162m and 0.128m, respectively. In conclusion, the outcomes of this study should serve as a basis for enhancing water column effect on depth estimation by using UAV multispectral image, hence, proof beneficial to assist the application of data in coastal monitoring.