The problem the researchers set out to solve is a stubborn one. Peatlands are among the largest terrestrial carbon reservoirs on the planet, and their expansion after the last ice age helped shape the concentration of methane and carbon dioxide in the Holocene atmosphere. Yet peat is perishable. Radiocarbon dating, the workhorse of peatland science, reaches back less than 50,000 years, and in formerly glaciated northern landscapes, ice sheets destroyed nearly all peat that formed before the Last Glacial Maximum. In old, undisturbed bogs, decomposition eventually balances the input of fresh plant litter, so the modern peat column no longer records the full history of peat on the landscape. Everything scientists believe about peatlands before the last glacial cycle therefore rests on a handful of exceptionally well-preserved deposits and on computer models.

The Belgian team’s solution exploits a property of quartz that has become one of geomorphology’s most powerful tools. Cosmic rays striking the uppermost meters of the Earth’s surface produce rare isotopes, including beryllium-10 and aluminum-26, inside mineral grains. The deeper a grain sits beneath overlying material, the fewer cosmic rays reach it, so the concentration of these cosmogenic nuclides in a sample carries a memory of how much mass, in this case peat, shielded it over its exposure history. Crucially, the two isotopes decay at different rates: aluminum-26 has a half-life of about 717,000 years while beryllium-10 persists for about 1.39 million years. When a landscape erodes slowly enough, the mismatch between these clocks becomes measurable, and the ratio of the two isotopes can be unpicked to yield both the rate of landscape lowering and the average thickness of the blanket above it simultaneously.

The Hautes Fagnes turned out to be an almost ideal natural laboratory. Perched on the highest ground in the Belgian Ardennes, the plateau receives about 1,440 millimeters of rain per year while mean annual temperatures hover near 6.7 degrees Celsius, and persistent orographic cloud keeps evapotranspiration low, conditions that favor peat accumulation. The study hillslope sits upstream of a mid-Pleistocene knickzone on the Hoëgne river, in a landscape so sluggish that it has barely responded to the uplift that rejuvenated rivers elsewhere in the Ardennes. The team cored through the modern peat blanket, which ranges from 0.2 to 2.1 meters deep along the transect, to collect saprolite, the rotted bedrock beneath, at six positions from the interfluve down to the channel. They also gathered stream sediment at the base of the slope and spent a full year gauging discharge and sampling stream water for dissolved silicon.

The measurements delivered a surprise. In quartz formed at the surface, aluminum-26 and beryllium-10 accumulate in a characteristic production ratio, about 7.2 at this site. Every sample the team analyzed fell below that ratio, from 3.29 at the summit to 6.06 at the toe-slope, and all plotted below the steady-state denudation line on a two-nuclide diagram by more than the modern peat cover could explain. Because the site sits near a local topographic high in a region never overridden by ice, deep burial by sediment or glaciers could not account for the deficit. The only plausible interpretation was that the samples had spent their exposure histories under substantially more shielding than the present-day peat provides, in other words, that the peat used to be much thicker.

To turn that qualitative signal into numbers, the researchers built a sophisticated inverse model. They accounted for the fact that saturated peat, which is roughly 90 percent water, attenuates cosmic radiation differently from mineral rock, adopting a shorter absorption mean free path of about 110 grams per square centimeter in peat compared with 160 in saprolite. They also modeled muon-induced production, which matters greatly under thick cover because muons penetrate far deeper than ordinary cosmic-ray nucleons. A Markov Chain Monte Carlo inversion then explored millions of parameter combinations to find those consistent with the measured isotope concentrations, yielding probability distributions rather than single answers. The results were cross-checked against an entirely independent estimate of the watershed-scale denudation rate, about 7.2 tons per square kilometer per year, derived from dissolved silicon fluxes and the weathering intensity of the regolith, which agreed in order of magnitude with the cosmogenic values.

The modeled denudation rates were astonishingly low, between 0.3 and 4.9 tons per square kilometer per year, equivalent to a landscape lowering of roughly 0.1 to 1.9 meters per million years, placing the Ardennes plateau among the slowest-eroding landscapes on Earth, comparable to ancient surfaces in Brazil, Australia, and South Africa. Because erosion is so slow, the isotope signal integrates over 0.7 to 1.8 million years, spanning every glacial-interglacial cycle since the mid-Pleistocene. Against that long baseline, the median long-term overburden thicknesses exceeded modern values by 190 to 350 grams per square centimeter along the transect, equivalent to roughly 1.8 to 3.4 meters of saturated peat. The inferred thicknesses, between about 2.5 and 4.2 meters, are physically sensible: they fall below the roughly 8.5-meter maximum observed in plateau raised bogs such as Misten, and their relationship to hillslope gradient obeys the classic groundwater-mound physics that governs steady-state bog morphology.

So where did the missing meters of peat go? The team systematically ruled out alternatives. Snow cover, even at persistent winter maxima, would add only a few grams per square centimeter of shielding on average, two orders of magnitude too little. A continuous loess blanket thick enough to explain the signal would have to exceed a meter and a half, yet no loess survives at the site and the geochemistry of the saprolite shows it formed in place from the underlying bedrock. Ground ice during periglacial episodes was too local and too short-lived. That leaves human activity. Peat cutting for fuel is documented in the Hautes Fagnes since the sixteenth century and intensified dramatically in the nineteenth, when records show nearly 200,000 cubic meters extracted annually. Medieval mowing, grazing, slash-and-burn cultivation, and wildfires shifted the bogs into negative mass balance, and drainage ditches plus spruce plantations in the early twentieth century accelerated subsidence at rates of one to two centimeters per year reported for comparable bogs. Together, the authors calculate, these pressures could account for more than two meters of peat loss, explaining most of the gap between the long-term average and today’s thin blanket.

The summit sample held one final revelation. Its model clearly resolved a step-change in overburden thickness, a thickening event with peak probability between 0.8 and 1.4 million years ago, which the authors interpret as the onset of the peat blanket itself. Remarkably, that timing coincides with a well-documented pulse of tectonic uplift in the Ardennes, recorded by the abandonment of the Younger main terrace of the Meuse and Rhine rivers around 0.7 million years ago. Before that uplift, the summit stood near 530 meters elevation, just below the threshold where orographic cooling and cloudiness begin to create the cool, wet conditions peat requires. As the massif rose roughly 100 meters, the site crossed into peat-forming territory, tying the birth of this carbon-rich ecosystem directly to the slow breathing of the Earth’s crust.

Beyond its local story, the study carries a warning for anyone estimating erosion rates in peatland terrain. If the beryllium-10 concentration in the stream sediment were interpreted the conventional way, assuming the modern peat cover represents long-term steady state, the resulting erosion rate would overestimate landscape lowering by an order of magnitude, because depressed aluminum-26 to beryllium-10 ratios masquerade as deep burial rather than thick organic cover. The paired-nuclide approach demonstrated here should work in any slowly eroding, unglaciated landscape with peat, from the moorlands of southwestern Britain to tropical cratonic peatlands in Amazonia and the Congo basin. And pairing beryllium-10 with cosmogenic carbon-14, whose short half-life suits faster-eroding and formerly glaciated terrain, could soon extend the method to peatlands worldwide. At a moment when drained and degraded peatlands are recognized as ticking carbon bombs, the Hautes Fagnes offer a sobering baseline: what we see today is not what these landscapes once were, and the atmosphere remembers the difference.

Subject of Research: Using paired cosmogenic 26Al and 10Be nuclides to reconstruct long-term peat thickness and denudation rates in the Hautes Fagnes upland peatland, Belgian Ardennes

Article Title: Long-term peat thickness from cosmogenic 26Al and 10Be, Hautes Fagnes, Belgian Ardennes

Article References: Moore, A., Henrion, M., Li, Y., du Bois d'Aische, E., Gautschi, P., Christl, M., Jonard, F., Lambot, S., Van Oost, K., Opfergelt, S., & Vanacker, V. (2026). Long-term peat thickness from cosmogenic 26 Al and 10 Be, Hautes Fagnes, Belgian Ardennes. Earth Surface Dynamics, 14(5), 763-780. https://doi.org/10.5194/esurf-14-763-2026

Image Credits: AI Generated

DOI: 10.5194/esurf-14-763-2026

Keywords: cosmogenic nuclides, peatlands, beryllium-10, aluminum-26, Hautes Fagnes, Belgian Ardennes, denudation rates, carbon cycle, peat degradation, land use history, mid-Pleistocene uplift, Earth Surface Dynamics

The challenge the researchers set out to solve is a familiar one in aeolian geomorphology. Traditional approaches, such as drift potential calculations and mobility indices, do a good job of describing wind-driven sediment transport at a given moment, but they offer limited ability to forecast whether dunes will migrate or become fixed over time. That gap matters increasingly in a changing climate, where shifts in wind regimes, precipitation patterns, and storm frequency may alter dune behavior with real consequences for ecosystems, roads, and homes. Complex numerical simulations could in principle fill the gap, but they demand data and expertise that many coastal regions simply do not have.

Toffani and Casadío turned instead to binomial logistic regression, a statistical technique that predicts the probability of a binary outcome, coded as zero or one, from a set of continuous predictor variables. First developed in the early 1980s, the method has been applied successfully to landslide susceptibility, soil mapping, and vegetation dynamics, yet its use for coastal dune mobility remained notably limited. The elegance of the approach lies in its ability to fold multiple climatic variables, including wind speed, precipitation, and large-scale atmospheric indices, into a single probabilistic framework that outputs a forecast probability of dune activation rather than a static description.

The natural laboratory for the study was the northern coast of the San Matías Gulf, stretching roughly 180 kilometers from the mouth of the Negro River to San Antonio Oeste. The region is a cold semi-arid environment where strong westerly winds blow hardest during the austral summer, exacerbating the aridity created by the Andean rain shadow. Its dunefields are remarkably varied: the Bahía Creek–Caleta de los Loros system extends over 36 kilometers with active dunes up to 16 meters high migrating east-northeast at 6 to 10 meters per year, while dunes near El Cóndor advance at around 7 meters per year. The area is also home to thousands of residents and a booming tourist trade, with Las Grutas alone averaging 118,000 visitors annually between 2006 and 2023, making dune encroachment a genuine practical concern.

To build their model, the researchers drew on hourly meteorological records from two stations, Viedma Aero and San Antonio Oeste Aero, covering the standard 1991 to 2020 reference period. Wind speeds and directions were normalized to a standard height of 10 meters, and winds exceeding 6.17 meters per second were flagged as capable of transporting sand, a threshold calculated from local grain sizes using Bagnold’s classic equations for the onset of saltation. From these data the team computed drift potential values and combined two established mobility measures, the Tsoar and Lancaster indices, into a new integrated index they dubbed TsoLa. Dunes were coded as active only when both indices simultaneously indicated mobility and the resultant drift direction aligned with the coast’s general eastward migration pattern.

The statistical screening was rigorous. Temperature, potential evapotranspiration, and soil moisture were excluded because of multicollinearity, while wave height, cattle populations, and the Southern Oscillation Index were dropped because they added little predictive power. What survived was a lean trio of predictors: monthly average wind speed, total monthly precipitation, and the Southern Annular Mode, or SAM, the dominant mode of extratropical climate variability in the Southern Hemisphere. When SAM is positive, the westerly wind belt contracts poleward and mid-latitude conditions turn warmer and drier; when it is negative, storm tracks shift north and the study region around 40 degrees south experiences stronger, stormier weather.

The results were strikingly consistent across both study sites. Wind speed emerged as a positive predictor of dune migration, with coefficients of 0.64 at San Antonio Oeste and 0.32 at Viedma, while both precipitation and SAM carried negative coefficients, meaning higher rainfall or a positive SAM phase reduces the probability of migration. In practical terms, the model suggests dune movement becomes unlikely when monthly precipitation exceeds 150 millimeters, when wind speeds fall below 20 kilometers per hour, or when SAM values are high. The models achieved area under the ROC curve values of 0.77 and 0.78, comfortably above the 0.5 threshold of random guessing, and correctly classified between 73 and 77 percent of observations in training and test datasets. A bootstrap analysis of 1,000 resamples confirmed that all three predictors had stable, statistically significant effects.

Validation against the real world came from nearly four decades of satellite and aerial imagery. The team digitized dune fronts in images spanning 1985 to 2023, placing reference points every 50 meters along each dune crest and measuring perpendicular distances between successive front positions. In the Viedma area, average migration rates ranged from 2.92 to 10.12 meters per year and correlated strongly with wind intensity values, with a correlation coefficient of 0.84. In San Antonio, migration was slower, between 2.51 and 6.01 meters per year, and showed little correlation with wind, likely because limited sand supply and encroaching vegetation, rather than wind energy alone, control movement there. Vegetation cover in the San Antonio area has in fact surged from 16 percent in 1961 to 60 percent in 2021, a greening trend the authors link to weakening winds and a trend toward more positive SAM values.

That contrast between the two dunefields carries an important lesson: geology matters. Viedma’s proximity to the Negro River mouth provides abundant sand that feeds large transverse dunes whose migration tracks wind capability closely, while San Antonio’s narrower beaches deliver a more limited sediment budget. The authors argue that integrating such geomorphological context into predictive models is essential, and they suggest that future refinements could incorporate higher-resolution vegetation indices, buoy-based wave data, better livestock records, and estimates of monthly sediment supply. They also note that lagged effects of El Niño events, which influence regional precipitation by up to 50 millimeters, deserve further study with properly calibrated time lags.

What makes the study resonate beyond Patagonia is its accessibility. Every input, from weather station records to the SAM index published by the British Antarctic Survey, is freely available, and the R scripts and datasets are posted in a public Zenodo repository. Because the framework relies on standard meteorological observations, it can be adapted to other coastal and inland aeolian systems with minimal modification, offering a scalable pathway for integrating climate forecasting with geomorphic hazard assessment. For coastal communities watching dunes advance toward their roads and homes, the message is empowering: the data needed to anticipate the next phase of dune activity may already be sitting in the nearest weather station’s archive.

Subject of Research: Statistical forecasting of coastal dune mobility using meteorological data and climate indices in northeastern Patagonia

Article Title: Forecasting coastal dune mobility: a logistic regression model driven by meteorological data and climate indices

Article References: Toffani, M., & Casadío, S. (2026). Forecasting coastal dune mobility: a logistic regression model driven by meteorological data and climate indices. Earth Surface Dynamics, 14(5), 801-820. https://doi.org/10.5194/esurf-14-801-2026

Image Credits: AI Generated

DOI: 10.5194/esurf-14-801-2026

Keywords: coastal dunes, dune mobility, logistic regression, aeolian geomorphology, Patagonia, Southern Annular Mode, wind erosion, climate indices, remote sensing, coastal management, sediment transport, Earth Surface Dynamics