loading page

gPCE Uncertainty Quantification Modeling for Bathymetric LiDAR and Earth Science
  • Alexandra Wise,
  • Kevin Sacca,
  • Jeffrey Thayer
Alexandra Wise
University of Colorado Boulder

Corresponding Author:[email protected]

Author Profile
Kevin Sacca
University of Colorado at Boulder
Author Profile
Jeffrey Thayer
University of Colorado at Boulder
Author Profile


NASA’s 2021 STV Incubation Study Report lists vertical (horizontal, geolocation) accuracy as an associated SATM product parameter for all (most) identified Science and Application Knowledge Gaps. The presented generalized Polynomial Chaos Expansion (gPCE) based method has wide ranging applicability to improve positioning, geolocation uncertainty estimates for all STV disciplines, but is presented for the bathymetric lidar use case, due to added complexity introduced by wave structure, roughness, and entry angle. Most LiDARs, though precise, are vulnerable to position, pointing errors as deviations from the expected principal axis lead to projection errors on target. While fidelity of location/pointing solutions can be high, determination of uncertainty remains relatively basic. Currently, the standard approach is the calculation of the Total Propagated Uncertainty (TPU), which is often plagued by simplifying approximations and ignored covariances. Additionally, uncertainty sources are often exclusively modeled as Gaussian, inaccurately capturing some variable distributions. Prominently, wave behavior is better described by Gamma distributions (which are supported under gPCE). This research addresses specific knowledge gaps in bathy-LiDAR measurement uncertainty through a more complete description of total aggregated uncertainties, from system level to geolocation, by applying a gPCE uncertainty quantification approach. gPCE intrinsically accounts for covariance between variables to determine the uncertainty in a measurement, without manually constructing a covariance matrix, through a surrogate model of system response. Determining point-wise positioning uncertainty using gPCE is less computationally expensive than Monte Carlo methods and more tractable for most dimensionalities of interest (roughly from 3 to 20+). The method also does not rely on simplifying assumptions used in typical TPU methods. Additionally, a key attribute of this approach is that global sensitivity analysis (GSA), after obtaining gPCE coefficients, is trivial and nearly costless to compute. Furthermore, GSA of system configurations/uncertainty is a powerful tool to design and develop LiDAR systems with the measurement requirements integrated into the design solution.