Post-tile terrain outputs are compared directly with their source surfaces at defined validation areas. The programme covers global GEBCO bathymetry at Level 10 and higher-detail regional topo-bathymetric sources at Level 13, reporting completion, sample counts and residual statistics in metres.
Three Level 13 outputs were checked against established regional elevation and bathymetry datasets at defined survey coordinates. Each run completed all 256 scheduled tiles and 1,081,600 elevation comparisons. These statistics describe fidelity of the produced Quantized Mesh terrain to the selected source surface; they are not independent estimates of absolute seabed or ground-truth accuracy.
| Data source | Survey coordinate | Source block | Tiles | Valid samples | Failed | Invalid | RMSE | MAE | Bias | P95 abs. | Max abs. |
|---|---|---|---|---|---|---|---|---|---|---|---|
| AusBathyTopo 250 m (Australia) 2023 | 37°S, 124°E | DATASET_22_z13_x13824_y2400 | 256 / 256 | 1,081,600 | 0 | 0 | 1.1729 | 0.3322 | +0.0017 | 1.6900 | 58.2439 |
| EMODnet Bathymetry DTM 2024 | 52°N, 30°W | DATASET_13_z13_x6816_y6448 | 256 / 256 | 1,081,600 | 0 | 0 | 0.3576 | 0.1450 | −0.0008 | 0.7000 | 12.1366 |
| NOAA NCEI U.S. Coastal Relief Model | 28°N, 90°W | DATASET_18_z13_x4096_y5360 | 256 / 256 | 1,081,600 | 0 | 0 | 0.0024 | 0.0021 | 0.0000 | 0.0000 | 0.0067 |
All three Level 13 runs completed without tile or sample loss. Residual magnitudes differ substantially because the sources have different native resolutions, vertical references and terrain regimes; the results are therefore within-dataset fidelity checks rather than a ranking of the underlying datasets.
AusBathyTopo uses WGS 84 / EPSG:4326 with MSL heights and depths. The EMODnet run uses the MSL-converted DTM variant. The tested NOAA Gulf Coastal Relief Model uses WGS 84 horizontally and EGM2008 vertically.
Three completed Level 10 validation blocks compare smoothed panDEM bathymetric terrain with the native GEBCO reference. Together they cover 3,072 tiles and 12,979,200 valid comparisons, with zero failed tiles and zero invalid samples. The test measures changes introduced by smoothing, resampling, tiling and encoding; it does not treat GEBCO as independent ground truth.
| Data source | Coordinate | Source block | Tiles | Valid samples | Failed | Invalid | RMSE | MAE | Bias | P95 abs. | Max abs. |
|---|---|---|---|---|---|---|---|---|---|---|---|
| GEBCO · Run 1 | Not specified | GEBCO_z10_x896_y448 | 1,024 / 1,024 | 4,326,400 | 0 | 0 | 4.9582 | 2.9287 | −2.8856 | 10.2000 | 128.1911 |
| GEBCO · Run 2 | Not specified | GEBCO_z10_x896_y352 | 1,024 / 1,024 | 4,326,400 | 0 | 0 | 5.6905 | 3.6983 | −3.6621 | 11.7900 | 93.2711 |
| GEBCO · Run 3 | Not specified | GEBCO_z10_x1280_y288 | 1,024 / 1,024 | 4,326,400 | 0 | 0 | 6.3210 | 3.8729 | −3.8291 | 12.8300 | 275.6236 |
Pooled RMSE is 5.6839 m, mean MAE is 3.5000 m, pooled bias is −3.4589 m and mean regional P95 is 11.6067 m. The cumulative maximum absolute residual is 275.6236 m. Under the project-specific QA rubric, each submitted run scores 4.85/5.00 (97.0%).
The recurring signal is a small negative vertical bias: all three GEBCO runs are lower/deeper than the reference on average. The supplied analysis identifies this as a trend to monitor as geographic validation coverage expands, while typical and P95 errors remain tightly controlled.
The validation programme is designed to quantify how faithfully the production pipeline carries source elevation through smoothing, resampling, triangulation, tile construction and encoding. Source resolution, datum and terrain character differ by dataset, so cross-source residuals should be read in context rather than as direct quality rankings between providers.