Quality Control of GNSS/Leveling-Derived Geoid Heights and Implications for LiDAR-Based Coastal Low-Elevation Classification

Abstract

Global Navigation Satellite System (GNSS)/leveling-derived observed geoid heights are widely used to validate gravimetric geoid models and to construct hybrid geoid models for practical height transformation. However, in tectonically active and subsidence-prone regions, these observations may contain errors caused by vertical deformation, benchmark instability, reference-frame differences, and temporal mismatch between GNSS and leveling measurements. This study evaluates the quality of recent GNSS/leveling-derived observed geoid heights in Taiwan and investigates the effects of these inconsistencies on hybrid geoid construction and Light Detection and Ranging (LiDAR)-based coastal low-elevation classification. A vertical velocity field was first used to correct the height changes occurring between the observation epochs of the GNSS-derived ellipsoidal heights and the leveling-derived orthometric heights. The resulting observed geoid heights were then assessed using Leave-One-Out Cross Validation (LOOCV) and an iterative 3σ outlier screening procedure to identify spatially inconsistent observations. The refined dataset was used to construct a hybrid geoid by modeling a correction surface between the gravimetric geoid and the corrected observed geoid heights. A 30% random holdout evaluation shows that the hybrid correction substantially reduces residuals within the screened GNSS/leveling network, decreasing the mean residual from −0.222 m to 0.003 m and the RMSE from 0.236 m to 0.041 m. The two geoid models were subsequently applied to a 20 m LiDAR-derived elevation dataset to evaluate their effects on orthometric-height conversion and coastal low-elevation classification. Although the broad pattern of coastal low-lands remains similar, local centimeter- to decimeter-level height differences can alter the classification of pixels near critical elevation thresholds, particularly in Taiwan’s flat western coastal plain. These results demonstrate that deformation-aware quality control of GNSS/leveling-derived geoid heights and hybrid geoid refinement are important not only for height modernization but also for reliable remote-sensing-based coastal elevation assessment.