Automated algorithm to generate depth invariant bottom reflectance from multiple remote sensing plat

Session: Remote Sensing, Visualization, and Spatial Data Applications for the Great Lakes (2)

Reid Sawtell, Michigan Tech Research Inst., [email protected]
Mike Sayers, Michigan Tech. Research Inst., [email protected]

Abstract

Standard classification methods for bottom type in shallow water regions often rely on particular operating assumptions such as constant water optical parameters across a scene. When these assumptions fail, the algorithms report misleading results, such as confusing a dark bottom types with CDOM laden river plumes. Manual interpretation or manipulation of the algorithms may improve the results but requires a significant investment of person hours. This is especially problematic in light of the increasing temporal and spatial availability of high resolution satellite imagery such as Landsat-8 or Sentinel-2. To remedy this, an automated analysis algorithm was created to isolate small shoreline areas into overlapping windows to dynamically estimate water optical properties. Coastal Lidar is then used to generate depth-invariant bottom reflectance, which can be used to classify bottom types in a given image. Statistical methods are then applied to merge multiple satellite bottom type classifications into a single image, producing a product that is significantly more robust to variable input data that does not conform to assumptions. This new fully automated approach allows for the repeatable processing of multi-sensor and multi-date satellite imagery into robust depth invariant reflectance data with little to no human involvement.