St. Petersburg Coastal and Marine Science Center
ALPS also includes modules for the creation of bare-Earth, canopy-top, and submerged topography Digital Elevation Models (DEMs). The EAARL system uses an Earth-centered coordinate and reference system that removes the necessity to reference submerged topography data relative to water level or tide gages (Nayegandhi and others, 2006).
For waveform-resolving instruments such as EAARL, the range is determined in post-processing. Processing algorithms have been developed to extract the range to the first and last significant return. The shape of the waveform is determined by a number of sensor parameters and backscattering properties of targets.
Some important sensor parameters include the shape of the laser pulse, the receiver impulse function, and parameters describing pulse spreading (Wagner and others, 2007). Algorithms are adjusted to tasks to account for waveform complexity (Wagner and others, 2004).
ALPS uses the following algorithms to differentiate between returns: the zero crossing of the second derivative is used to detect the first return and the trailing edge algorithm is used to detect the range to the last return; i.e., the algorithm searches for the location prior to the last return where direction changes along the trailing edge (fig. 2). In submerged environments, effects of refraction and change in speed of light as it enters the water column are accounted for in the “submerged topography” algorithm. The exponential decay of the return signal through the water column is also determined based on the clarity of the water column. These corrections are performed by examining sample waveforms from spatially distributed locations in the survey area to define constants for exponential decay of the laser at the water surface and within the water column. A selection of constants is defined in ALPS for different water column and depth conditions: ranging from deep and clear water column to shallow and turbid water column. Data processed for submerged topography are referenced to an ellipsoid datum, which is independent of the elevation of the water surface (Nayegandhi and others, 2004).
ALPS applies semi-automated statistical filtering methods to remove false bottom returns and other outliers from the EAARL lidar data. Erroneous (i.e., outlier) points might include reflections from objects such as birds, multiple atmospheric effects (e.g., dust, moisture), or multiple reflections from bright targets.
Two filtering methods within ALPS are used to extract ground (bare-Earth) elevations from a point cloud of processed last returns: Random Consensus Filter (RCF) and Iterative Random Consensus Filter (IRCF). The RCF is based on the Random Sample Consensus (RANSAC) paradigm, which was originally published by Fischler and Bolles (1981). The filter uses a grid of non-overlapping square cells of user-defined size overlaid onto the original point cloud.
The user defines the grid cell size and vertical tolerance based on the topographic complexity and point sampling density of the data. The maximum allowable elevation range within a cell is established by the vertical tolerance. An iterative process searches for the maximum concentration of points within the vertical tolerance, and removes those points outside of the tolerance (fig.3).
The IRCF algorithm uses the RCF algorithm and a triangulated irregular network (TIN) model iteratively to progressively densify the output point cloud. The RCF paradigm is used to label the initial point cloud that represents the ground. All labeled ground points are triangulated using Delauney's Triangulation to create a TIN model. The points rejected from the first RANSAC iteration are treated as potential ground points. Each triangulated facet within the TIN model is defined as a three-dimensional plane, the equation of which is determined from the vertices of the triangulated facet. All potential ground points above or below each facet are classified as ground if they fall within the user-defined vertical range (also called the TIN vertical width) from the 3-D plane. The TIN model, created in the subsequent iteration with the new set of classified ground points, is further densified by adding all potential ground points within the vertical width for each triangulated facet. This process continues for a pre-defined number of iterations, or until less than 2% of potential ground points are added to the final population of ground points.