A Pipeline for Modeling Point-Sampled Surfaces: From Preprocessing to Quality Assessment

The monitoring of the Earth’s surface is increasingly driven by dense, point-sampled observations from techniques such as LiDAR, photogrammetry, multibeam echosounders (MBES), and satellite radar products like persistent scatterer interferometry (PSI). Despite their different sensing physics, these techniques typically generate large point clouds whose quality varies spatially and temporally. These point clouds are difficult to interpret because discrete points do not provide information about connectivity. These measurements are often non-uniformly distributed and contain data gaps. They are also contaminated by noise and outliers. Therefore, reliable use requires cleaning, surface modeling, and quality assessment. However, these steps are often handled separately with manual intervention and limited uncertainty reporting, which reduces reproducibility and weakens trust in the resulting products. This cumulative dissertation presents an end-to-end pipeline for modeling point-sampled surfaces, from preprocessing to quality assessment. It synthesizes four publications that collectively develop and validate the proposed pipeline, including three peer-reviewed journal articles and one published conference paper. The pipeline has been designed to be scalable, reproducible, and uncertainty-aware. The work focuses on 2.5D surfaces, z = f(x, y), balancing geometric suitability for large-scale geospatial products with computational tractability. The pipeline is built around a single multiresolution surface representation based on the Multilevel B-Spline Approximation (MBA) approach. This approach provides a consistent backbone for outlier detection, surface approximation, and uncertainty-aware evaluation. The methodologies are tested using PSI for deformation monitoring applications and MBES data for mapping, navigation, dredging, and infrastructure monitoring. The preprocessing stage provides two complementary, surface-based outlier detection methods. One method uses a Data Adaptive, coarse-to-fine strategy to target sparse, isolated anomalies in large, heterogeneous deformation point sets, such as those in PSI. For MBES bathymetry, the Iterative Estimation of Surface Parameters (IESP) framework identifies clustered contamination by combining robust estimation and iterative, distribution-guided trimming until the residual behavior stabilizes. In the Hannover PSI case study, the Data Adaptive Outlier Detection method shows high agreement with a published reference filter, achieving a balanced accuracy of 0.95. For Kiel Canal MBES data, IESP achieves a balanced accuracy of 0.99 and reduces manual cleaning time from approximately two days to about 30 minutes. The output is a cleaned observation set for subsequent modeling and assessment. The modeling stage uses MBA as the common surface model and the geometric backbone of the pipeline. Its hierarchical refinement adds detail stepwise and remains robust under irregular sampling and data gaps. The quality assessment stage provides spatially varying uncertainty under two regimes. When observation variances are unknown, nonparametric bootstrapping yields prediction uncertainty and confidence intervals for MBA-based surfaces. When system-level error models are available, a forward uncertainty framework is used to estimate pointwise measurement uncertainty in MBES and propagate it into surface estimation. Simulations with known ground truth demonstrate that uncertainty-aware weighting reduces model error. For example, it decreases model error (1σ) from 0.044 m to 0.023 m. The output consists of uncertainty maps and confidence measures that make the local uncertainty transparent and facilitate interpretation of the results. In sum, the dissertation presents a coherent pipeline that connects automated cleaning, MBA-based surface modeling, and uncertainty-aware quality assessment for point-sampled surfaces. The dissertation reduces subjectivity in preprocessing, improves the transparency of modeling decisions, and generates surface products with defensible uncertainty information to inform decisions in deformation monitoring and hydrographic applications.