Employing K-means clustering to reconstruct missing water surface elevations in LiDAR digital elevation models for...

Accurate Digital Elevation Models (DEM) data are essential for reliable hydrodynamic modeling of runoff processes, particularly for estimating runoff depth. However, DEM data inherently contain uncertainties over river and lake surfaces and along riverbanks. To address this issue, this study proposes a reconstruction approach in which river and riverbank sites are identified using a k-means clustering algorithm, riverbank elevations are reconstructed with inverse distance weighting, and river and lake elevations within a specified range are replaced with the minimum riverbank elevation to estimate water surface elevations. Validation indicates that estimated water surface elevations closely match observed values (R² = 0.98) and correspond to historical normal levels, although slight overestimations occur at a few locations relative to historical peak water levels. Comparing hydrodynamic simulation results before and after DEM reconstruction, river storage was overestimated due to negative uncorrected river surface elevations, leading to underestimated runoff depths near rivers. Overall, this study provides a method for estimating water surface elevations based on DEM data, offering a reference for more accurate runoff simulations and urban flood risk prediction in area with data scarcity or uncertainty in the future.