| 18 Aug 2026
A global dataset of bias-corrected waves and storm surge (1950–2023) with improved extremes for hazard mapping
Abstract. Consistent global datasets of storm surges and waves are essential for assessing coastal vulnerability under changing coastal hazard conditions. Building on the global coupled hindcast, we present an enhanced hazard-oriented dataset of storm surges and bulk and spectral peak wave parameters spanning 1950–2023. Modelled significant wave heights from the hindcast are bias-corrected using satellite altimetry observations, while storm-surge signals are derived from modelled sea surface heights through frequency-based filtering. The dataset includes significant wave height, wave period, wavelength, mean wave direction, directional spreading, and spectral characteristics of the three dominant energy peaks. We validate the corrected significant wave heights against observations from 485 buoy locations and the derived storm surges against 550 tide gauges, with particular emphasis on the upper tail of the distributions. For significant wave heights, the global median normalised bias beyond the 95th percentile is −0.88 %, indicating that systematic errors in extreme wave conditions are largely removed. At higher latitudes (30° N–65° N), where tide-gauge coverage is denser, storm surges show a median normalised bias beyond the 95th percentile of 2.62 % and a median correlation of 0.81, indicating that the modelling framework captures observed storm-surge variability effectively in well-observed extratropical regions, while larger uncertainties remain in tropical and equatorial regions. Additional event-based evaluation during tropical cyclones shows limited systematic bias in modelled peak wave heights and surge levels. The resulting dataset provides a consistent global baseline for coastal hazard mapping, non-stationary extreme value analysis, regional downscaling, compound-event assessment, and data-driven modelling applications. The dataset (Wadalkar et al., 2026) is available at https://doi.org/10.2905/JRC.W0YPDCR.
This paper applies an empirical quantile-based bias correction to the significant wave height output of the Mentaschi et al. (2023) SCHISM-WWM-V global hindcast, and presents an extensive validation of both Hs and storm surge residuals derived from modelled SSH against satellite and tide gauge observations. The validation effort is substantial, and the resulting dataset is likely to be widely used. Because the bias correction is the paper's central contribution, however, my comments focus on whether the corrected fields remain suitable for the applications the authors propose. I have the following main concerns, which I believe need to be addressed before publication. These comments require additional analysis; therefore, I recommend major revision.
Comment 1
Wave height and storm surge are both driven principally by the surface wind, so a low bias in Hs is evidence of a low wind bias and therefore of a low surge. Correcting Hs alone, post-hoc, leaves the SSH field unaltered and still carrying the uncorrected wave signature through drag and resolved setup. The Hs, Tm, SSH triplet is no longer a coupled solution, and the wave-surge magnitude relationship is distorted, in this case most severely in the tail. This directly affects the compound-event and boundary-forcing applications the dataset is presented for. The magnitude of this inconsistency is currently unquantified. An uncoupled sensitivity run over a subset of extreme events would establish how much of the modelled SSH derives from the wave field, and therefore how large the residual error is after Hs alone is corrected. The authors could correct period and surge consistently and report the effect on the joint distribution. At minimum, state the limitation explicitly for users.
Comment 2
The bias correction is derived from satellite altimetry, which is routinely flagged or discarded within roughly 10–30 km of the coast. How many usable observations support the BCF in coastal cells, and where coverage is insufficient, is the correction extrapolated from offshore?
Nearshore model bias arises from sheltering, refraction over unresolved bathymetry and depth-limited breaking, none of which the offshore correction factor represents. This has direct implications for the alongshore distribution of wave energy, and hence for coastal erosion applications.
On top of that, Figure 2 shows corrected Hs quantiles agreeing with altimetry. This is circular. Your correction is based on altimetry and then validated against altimetry. The Figure just validates the arithmetic, not the model.
Comment 3
The Hs BCF correction is applied pointwise to each 3-hourly value, with a factor that increases with percentile, so peaks are inflated more than the lows of the same storm. Event duration above threshold, integrated wave energy, and storm persistence are therefore altered by an unvalidated amount. These are the quantities that matter for erosion, structural fatigue, and operability analysis. These are all uses for which the dataset is well suited and likely to be adopted. The authors should quantify the change in duration above threshold and in integrated event energy before and after correction, and to note that the period is uncorrected, so wave steepness and derived energy flux are internally inconsistent.
Comment 4
This is more a comment on the paper scope than a technical aspect. Although waves are dynamically coupled, the surf zone is unresolved at 2 km, so shoreline setup is captured only in a small part. The product is a water level dataset with a partial setup contribution and, in my opinion, cannot be used for coastal hazard mapping. I recommend toning down such a claim and focusing more on non-stationary EVAs, regional downscaling, and ML applications.
Comment 5
The dataset is proposed for non-stationary extreme value analysis and climate trend analysis, but the correction is not homogeneous across the record. The BCF is derived from satellite altimetry, available from 1991, and applied to the full 1950–2022 period. Over roughly four decades, it therefore cannot be verified, and it is applied to a period in which the underlying wind error is structurally different: ERA5 assimilates no satellite data before 1991, and the observing system continues to change thereafter. Any trend fitted across the full record risks conflating a climate signal with changes in the reanalysis and in the validity of the correction itself. I ask the authors to test the homogeneity of the corrected fields across the record, to demonstrate that fitted trends survive restriction to the altimeter period, and to state clearly over which period the correction is verifiable.