Combination of Lidar Elevations, Bathymetric Data, and Urban Infrastructure in a Sub-Grid Model for Predicting Inundation in New York City during Hurricane Sandy

We present the geospatial methods in conjunction with results of a newly developed storm surge and sub-grid inundation model which was applied in New York City during Hurricane Sandy in 2012. Sub-grid modeling takes a novel approach for partial wetting and drying within grid cells, eschewing the conventional hydrodynamic modeling method by nesting a sub-grid containing high-resolution lidar topography and fine scale bathymetry within each computational grid cell. In doing so, the sub-grid modeling method is heavily dependent on building and street configuration provided by the DEM. The results of spatial comparisons between the sub-grid model and FEMA’s maximum inundation extents in New York City yielded an unparalleled absolute mean distance difference of 38m and an average of 75% areal spatial match. An in-depth error analysis reveals that the modeled extent contour is well correlated with the FEMA extent contour in most areas, except in several distinct areas where differences in special features cause significant de-correlations between the two contours. Examples of these errors were found to be primarily attributed to lack of building representation in the New Jersey region of the model grid, occluded highway underpasses artificially blocking fluid flow, and DEM source differences between the model and FEMA. Accurate representation of these urban infrastructural features is critical in terms of sub-grid modeling, because it uniquely affects the fluid flux through each grid cell side, which ultimately determines the water depth and extent of flooding via distribution of water volume within each grid cell. Incorporation of buildings and highway underpasses allow for the model to improve overall absolute mean distance error metrics from 38m to 32m and area comparisons from 75% spatial match to 80% with minimal additional effort.

💡 Research Summary

The paper introduces a novel sub‑grid hydrodynamic modeling framework that integrates high‑resolution LiDAR‑derived topography, fine‑scale bathymetry, and detailed urban infrastructure (buildings, streets, highway underpasses) to simulate storm‑surge‑induced flooding in New York City during Hurricane Sandy (2012). Traditional 2‑D flood models rely on coarse computational grids, which smooth out critical urban features and cannot accurately represent partial wetting and drying within a cell. The sub‑grid approach overcomes this limitation by nesting a 1‑m resolution “sub‑grid” inside each larger computational cell (typically 30 m). This sub‑grid contains the full LiDAR DEM, bathymetric data, and a vectorized representation of buildings and underpasses, allowing the model to calculate fluxes across cell faces based on the actual wetted area rather than the full cell width.

Model construction proceeded in four steps: (1) acquisition and merging of city‑wide LiDAR DEM with coastal bathymetry to produce a seamless 1‑m resolution elevation surface; (2) extraction of building footprints, heights, street widths, and underpass locations from open‑data portals, which were rasterized into the DEM as “building blocks” and “transparent corridors”; (3) definition of the sub‑grid hierarchy and assignment of high‑resolution topographic attributes to each parent cell; (4) solution of the shallow‑water equations on the parent grid while the sub‑grid supplies dynamically updated wetted‑area fractions for flux calculations.

The model was calibrated and validated against FEMA’s Maximum Inundation Extent for Hurricane Sandy. Two quantitative metrics were used: Mean Absolute Distance Error (MADE) between modeled and FEMA flood boundaries, and Percentage of Area Match (the proportion of overlapping flooded area). The initial configuration, which omitted building and underpass details, yielded a MADE of 38 m and an area match of 75 %. After incorporating building heights and rendering highway underpasses as permeable gaps, the MADE improved to 32 m and the area match rose to 80 %. This 6‑meter reduction in distance error and 5‑percentage‑point gain in spatial agreement demonstrate the substantial benefit of representing urban infrastructure in a sub‑grid context.

Error analysis identified three dominant sources of discrepancy: (i) missing building data in the New Jersey portion of the domain, leading to artificial flow barriers; (ii) underpasses that were represented as solid terrain in the DEM, erroneously blocking water that in reality passes beneath roadways; and (iii) differences in vertical datum between the DEM used for the model and the DEM underlying FEMA’s flood maps, causing systematic elevation offsets in low‑lying coastal zones. The study shows that even modest enhancements—adding a few thousand building polygons or correcting underpass elevations—can markedly improve flood extent predictions without a proportional increase in computational cost.

The authors argue that the sub‑grid method offers a cost‑effective pathway to high‑fidelity urban flood modeling. It retains the efficiency of coarse‑grid hydrodynamics while capturing sub‑cell heterogeneity critical for accurate water depth and extent estimates. Potential applications include real‑time flood warning systems, urban planning for climate‑resilient infrastructure, and scenario analysis for future sea‑level rise. Future work is suggested in three areas: (1) automated pipelines for ingesting up‑to‑date LiDAR and bathymetric surveys; (2) extension of the sub‑grid to include subsurface networks such as storm‑drainage and subway tunnels; and (3) integration of machine‑learning techniques to quantify and correct model uncertainties. Overall, the paper demonstrates that precise representation of urban form—particularly buildings and underpasses—substantially enhances the predictive skill of storm‑surge models in densely built environments.