Dense bathymetry in turbid coastal zones using Airborne Hyperspectral Images

Abstract

We used airborne hyperspectral images to generate a dense survey of bathymetric data in the Bahía Blanca estuary (Buenos Aires Province, Argentina). This estuarine area is characterized by intense sediment transport turning the water muddy, and thus optical bathymetric estimations are dif- ficult. We used 24 spectral bands in a range of 500?900 nm acquired with a hyperspectral camera aboard an unmanned aerial vehicle, together with 100 bathymetry data points surveyed with a sonar sensor aboard an unmanned surface vehicle, covering an area of about 800 m2. Random-forest and support-vector-machine regressors were trained with this data set. The resulting model yielded a determination coefficient of 0.815 with unseen data, a root-mean-square error of 0.166 m, and an absolute average error less than 2%. These results allow dense and accurate reconstructions of the underwater profile in wide, muddy, shallow regions of the Bahía Blanca estuary, showing the feasibility of hyperspectral imagery combined with sonar data in turbid shallow waters.