Both ends of the gap
Nature published an addendum on 25 August to the March paper reporting that sea level in coastal hazard assessments sits higher than assumed. The addendum announces no new result; it gives the configuration of the geoid models the meta-analysis rests on. It states that it expands the description in the original publication to ensure full reproducibility of the results, and sets out that the data came from the ICGEM calculation service at the GFZ Helmholtz Centre for Geosciences, that the zero-degree term was included, that a tide-free system was used and that no filters were applied.[1]
The weight of these details comes from the shape of the original finding. The paper reported measured coastal sea level standing on average 0.27 and 0.24 metres above two widely used geoids, with standard deviations of 0.76 and 0.52 metres. The reported quantity is a difference: measurement at one end, a computed surface at the other. Which model the surface comes from, how far it is expanded and how finely it is sampled all shape the difference itself.[1]
The distinction this column drew a week ago earns its keep again here: openness and runnability answer different questions. The addendum opens the geoid side. The intermediate step is an interpolation with multiquadric radial basis functions in ArcGIS Pro version 3.1.3, with an optimized kernel parameter and a neighbourhood set automatically to between 10 and 15 neighbours in a single sector. A setting chosen automatically may not produce the same number in another pair of hands, to the extent that it depends on the version of the software.[1], [3]
The main analysis and the models added later
The addendum reports two different sampling densities. For EGM96, EGM2008 and GOCO06s the global data were downloaded on a 0.085 degree step, 8,971,848 points for each model. For GOCO2025s and EGM-DIR R4, described as added in the later process of the article, the step is 1.000 degrees and the point count 65,341. The first three were downloaded on 22 August 2023 and 10 January 2024, the last two on 3 October 2025.[1]
Sampling the two later models on a much coarser grid than the main analysis means the robustness check was not built at the same spatial detail as the headline result. The addendum also gives the maximum degree used for each model, and the comparison sharpens there: the maximum degree of GOCO2025s is 300, and that of GOCO06s is also 300. Of two models expanded to the same degree, one was sampled on a 0.085 degree grid and the other on a 1.000 degree grid.[1]
The same constraint in another coastal study
Another coastal study published the same day shows where the constraint comes from. A method reading four decades of Landsat shoreline images finds the tidal range varying by close to 1 metre along the South Taranaki Bight in New Zealand, a difference that stayed invisible because conventional satellite radar has a resolution near the shore measured in tens of kilometres. The constraint the two studies share is that a reported coastal number depends on the reference surface and the spatial step on which it is produced.[1], [2]
A testable expectation follows. Since the addendum gives the configuration of the geoid side, an independent team can now recompute the reported mean offsets of 0.27 and 0.24 metres with the same ICGEM settings. If such a recomputation finds the mean offset for the same geoids inside the reported standard deviations, the result can be treated as independent of the configuration choices. If the mean moves once the two later models are resampled on the main analysis grid, the robustness check is sensitive to the grid.[1]