Image-derived constraints for inversion of physics-based methods in satellite-derived bathymetry

点击翻译按钮获取中文摘要

Two groups of methods exist to estimate water depth from satellite imagery: empirical methods are simple toimplement but require in-situ calibration data, while physics-based methods are more complicated but can beimplemented with the need for in-situ data. The latter are thus necessary for mapping areas where no in-situdata exist and none can feasibly be acquired, such as for the remote Arctic or foreign territory. Physics-basedmethods simultaneously derive information on water depth, seafloor reflectance, and water optical quality, on apixel-by-pixel basis, by inverting a forward radiative transfer model parameterized with these quantities.However, when used with multispectral satellite data suffering from imperfect atmospheric correction,inversion algorithms must be robust to spectrally correlated and spatially uncorrelated noise. Robust algorithmsmay rely on the reasonable assumption of locally constant water quality, and on image-based indications of biasfrom atmospheric correction. In this research, we propose to develop and test robust inversion algorithms thatimprove bathymetry retrievals from multispectral satellite data. We will do this by using low-altitude airbornehyperspectral data coincident with satellite overpasses to identify and isolate atmospheric correction biases, andby developing stronger constraints on inversion, for example deriving from assumptions of regionallyhomogeneous water quality ranging from an optically deep area (where the seafloor does not influence thecolour of the area as seen from space) to a nearby optically shallow area targeted for bathymetry mapping. Theresearch will improve bathymetry mapping from multispectral satellite images, increasing the value ofsatellite-derived bathymetry in Canada and beyond.