Reading the tide off the beach: chart datum and tidal reduction implications

What the TUM–Oxford method actually does

Researchers at the Technical University of Munich (TUM) and the University of Oxford have published a tide-observation method that uses the beach itself as a measuring device. Rather than reading water depth, they combine the visible waterline in satellite imagery with the known slope of the beach to recover sea level, and from a sequence of images they reconstruct how the tide varies. Applied to a 90km bay in New Zealand, the approach resolved tidal differences of nearly one metre within that single stretch. The team reports data at a spatial resolution on the order of 100 metres, against roughly ten kilometres for earlier satellite tide products – a hundredfold gain in how finely the along-coast tide can be sampled.

Thomas Monahan of the University of Oxford frames the finding around flood risk: small-scale tidal differences can decide whether neighbouring stretches are spared or inundated in the same storm. Michael Hart-Davis of the German Geodetic Research Institute at TUM describes the method as combining the precision of local tide gauges with the coverage of satellite imagery, and so closing the data gap in coastal regions without gauges of their own – parts of Africa and Southeast Asia among them. The stated applications are coastal protection, flood forecasting, navigation and saltwater intrusion.

That is the grounded claim. What follows is our assessment of where it lands in the vertical referencing chain that hydrographers and metocean engineers actually work in, and what it does and does not change for chart datum, tidal reduction and offshore vertical control.

Spatial resolution is not vertical accuracy – read the number correctly

The headline figure, 100 metres, is a spatial sampling improvement: how densely the along-coast tide can be observed. It is not a vertical uncertainty. Those are different quantities, and conflating them will get a survey manager into trouble.

The physics of the technique sets the vertical error budget. On an idealised planar foreshore, water-level elevation relates to cross-shore waterline position through the foreshore slope. Differentiate that relationship and two error terms fall out. The first is horizontal waterline detection error multiplied by the slope. The second is waterline position multiplied by slope uncertainty. Both scale with how well the beach is characterised and how steep it is.

The consequence is that the method is beach-selective. On a gently sloping, dissipative beach – a foreshore of, say, 1:100 – the waterline sweeps a wide cross-shore band across the tidal cycle, so the tidal signal is large relative to a satellite pixel and easily detected, while a horizontal detection error of a pixel translates into a small vertical error. On a steep reflective beach of 1:10 or steeper, the waterline barely migrates between low and high water, the tidal signal is compressed into a few pixels, and the same pixel-level detection error becomes a large fraction of the tidal range. The technique will perform very differently on the dissipative flats of a macrotidal embayment than on a shingle bank, and any accuracy claim has to be quoted with the foreshore slope attached.

Two further physical effects sit on top of the geometry. Wave setup raises the mean water surface at the shore, and swash runup extends the instantaneous waterline further up the beach than the still-water line implies. Both bias the retrieved level upward, and on an energetic coast they can add tens of centimetres that must be modelled or filtered before the residual is treated as tide. This is the same nearshore hydrodynamics problem that constrains other remote-sensing approaches, and it is worth reading alongside how wave-inversion bathymetry is positioned next to multibeam – both are optical or radar techniques whose limits are set by wave processes, not by the sensor.

Where tidal observation sits in the vertical chain

To judge the value of denser tidal observation, put it where the errors accumulate. A charted or surveyed depth is a measured range reduced to a vertical datum, and every link in that reduction carries uncertainty.

Chart datum in most jurisdictions is Lowest Astronomical Tide, per the long-standing IHO resolution on the reference for tidal predictions and soundings. LAT is a tidal quantity: it is derived from harmonic constituents at a point, and it varies along the coast exactly as the tide does. A method that resolves along-coast tidal variation at 100m also, in principle, resolves how LAT itself shifts along the frontage. That matters wherever chart datum is transferred from a distant gauge into a survey area by simultaneous comparison or by co-tidal modelling.

The tidal reduction problem is where the finding bites hardest. Reduction of soundings has three broad approaches. A single gauge with a fixed offset assumes the tide at the vessel equals the tide at the gauge – acceptable only over short distances and small range gradients. Co-tidal zoning, whether by discrete zones or by continuous methods such as tidal constituent and residual interpolation, models the phase and amplitude change between control points. Ellipsoidally referenced survey removes tide from the measurement chain entirely by taking GNSS-derived heights and applying a separation model between the ellipsoid and chart datum.

Consider the New Zealand result as an error-budget statement. Nearly one metre of tidal difference over 90km is a gradient of roughly 11mm per kilometre. Reduce soundings 45km from your only gauge with a fixed offset, and an unmodelled linear gradient of that size implies about half a metre of reduction error at the far end. Set that against IHO S-44 Order 1a, where the allowed total vertical uncertainty at shallow depth is governed by a = 0.5m and b = 0.013. Half a metre of reduction bias would consume the entire vertical budget on its own before any sounder, motion or sound-velocity error is added. The point is not that surveyors ignore this – competent tidal zoning is designed precisely to catch such gradients – but that in gauge-poor regions the zoning is often built on thin observational support. Denser tidal observation is a direct input to constraining that model.

For ellipsoidally referenced survey, the relevant product is the separation surface. The retrieved along-coast tidal field feeds the realisation of chart datum relative to the ellipsoid in the nearshore, which is the hardest zone to model because it is where hydrodynamic detail is greatest and where offshore tidal models lose fidelity. A validated separation model is what lets a contractor drop tide from the reduction chain, and better nearshore tidal observation is one way to test and refine those surfaces where no gauge exists to check them.

Finally, the S-100 framework turns this into product terms. S-104 carries water-level information for surface navigation and S-111 carries surface currents, alongside S-102 for the bathymetric surface and S-101 for the ENC. A gridded, along-coast water-level field is exactly the kind of input those water-level products are meant to distribute. As the industry works through the practicalities of the new standards, the change in how water-level data can be sourced and served belongs in the same discussion as what S-100 asks of offshore contractors.

The failure modes that decide whether you can trust it

A method is only useful once its failure modes are understood well enough to bound them. Several apply here, and each maps to a check before the data is admitted into a survey or a design study.

Beach slope is an input, not a constant. The technique depends on knowing the foreshore slope, and foreshores are mobile. Storm cut and post-storm recovery change beach face gradient seasonally, and a slope measured once will not describe the beach a year later. Any operational use needs a current slope from recent topo-bathymetric LiDAR or equivalent, with the morphological variability carried through as an uncertainty, not assumed away.

Wave setup and runup contaminate the level. As above, the instantaneous waterline is not the still-water line on an energetic coast. Without a setup and runup correction driven by the concurrent wave field, retrieved levels carry a sea-state-dependent high bias that aliases into apparent tidal and surge signals.

Sun-synchronous sampling aliases the tide. Optical missions such as Sentinel-2 and Landsat cross at near-fixed local solar time. The semidiurnal lunar constituent M2 is sampled at slowly drifting tidal phases across many passes, which is what makes harmonic recovery possible at all, but the solar constituents S2, K1 and P1 sit close to the sampling rhythm and are poorly separated. Estimating LAT, which depends on the full constituent set, from a constituent estimate with weak solar terms is a genuine limitation. Multi-year image runs and explicit checks on constituent separability are needed before the derived datum is treated as reliable.

Cloud and image availability set the sampling floor. Optical waterline extraction fails under cloud. On persistently cloudy coasts the usable image count collapses, lengthening the record needed to build a stable harmonic solution and leaving seasons under-sampled. This is not a fatal flaw, but it changes the confidence you can place in a short record.

It observes the shore, not the offshore. The waterline lives at the coast. Extrapolating an along-coast tidal field into an offshore survey block still requires a hydrodynamic model; the method improves the boundary constraint, it does not replace the model.

What to do with this now

We see this as a tool for constraining tidal models in data-poor coasts and for validating separation surfaces, not as a drop-in replacement for gauges or for proper zoning. Handled that way, it is a real addition to the toolkit. Concrete steps:

Quantify the along-coast tidal gradient before you fix zone boundaries. Where an embayment shows the kind of variation reported here, roughly 11mm/km, size discrete tidal zones so residual reduction error stays inside the fraction of the S-44 TVU you have allocated to reduction – typically hold the tidal reduction contribution well below the a-term for your Order at survey depths. Use the satellite-derived field to test whether a single-gauge assumption survives across the frontage.
Prefer ellipsoidally referenced survey where GNSS coverage allows, and use waterline observations to test the separation model. In gauge-absent regions, cross-check the modelled LAT-to-ellipsoid separation against the satellite-derived nearshore tidal field before you rely on it to drop tide from the reduction chain.
Attach the foreshore slope and its date to every derived level. Require a slope from topo-bathymetric LiDAR within a defined window of the imagery, and carry the slope uncertainty and expected morphological change into the vertical error budget explicitly.
Demand a wave correction and a record-length justification. Do not accept retrieved levels without a setup/runup treatment tied to the concurrent sea state, and require enough multi-year passes to separate the solar constituents. Where cloud limits imagery, state the sampling gaps rather than smoothing over them.
For flood and extreme-level work, stop assuming one gauge represents the whole frontage. A sub-kilometre tidal field changes the tide component of joint probability with surge and rainfall along a coastline. This belongs in the validation set metocean leads run when they check extreme sea level for offshore and coastal design, particularly where the still-water datum feeds crest levels and inundation lines.
Route the output into S-104 workflows, not into ad hoc files. If the field is to inform navigation, structure it for the water-level product from the outset so it can be validated and versioned rather than reconciled later.

The engineering value here is specific and defensible: better observation of how the tide changes along a coast, at a scale that matters for reduction and for datum realisation, in exactly the places where the observational network is thinnest. It does not lift the burden of understanding beach slope, wave setup, tidal aliasing and the difference between spatial resolution and vertical accuracy. Treat it as a constraint on your models and a check on your separation surfaces, and it earns its place. Treat the 100-metre figure as a vertical accuracy, and it will mislead you.

Based on: New method fills the gaps in coastal tide monitoring