The research team, led by Lizzeth Vázquez-Caamal and Christian M. Appendini of the National Autonomous University of Mexico, together with colleagues from UNAM’s Morelia campus, the University of Iceland, and UNAM’s Institute of Engineering, took a fundamentally probabilistic approach. Rather than modeling a single worst-case earthquake, they generated a vast ensemble of plausible seismic events using stochastic rupture modeling. The work builds on the framework of probabilistic seismic hazard analysis, which integrates earthquake recurrence statistics, magnitude distributions, and variability in how ruptures unfold to estimate exceedance probabilities. This allows the assessment to capture the full spectrum of possible events, including ones larger than anything in the historical record.

At the heart of the methodology is the FakeQuakes module of the open-source MudPy code, which generates realistic synthetic earthquake slip distributions on a three-dimensional model of the subduction interface. The team used the Slab2.0 model to define the geometry of the plate boundary, discretizing it into a finite mesh extending to about 30 kilometers depth, the range where large thrust earthquakes typically occur in this region. Slip heterogeneity was sampled using a Karhunen-Loève expansion, producing physically plausible variability in how each rupture distributes its energy. In total, the researchers created 4,800 synthetic earthquakes spanning magnitudes from Mw 7.0 to Mw 8.5 in increments of 0.1, with 300 scenarios for each of 16 magnitude bins. Each scenario was weighted according to its annual occurrence rate derived from a regional Gutenberg-Richter relationship, with a correction applied to represent the smaller rupture domain evaluated in the study.

Before trusting the probabilistic results, the team validated the entire modeling chain against a known event. They generated 600 synthetic rupture models constrained to the characteristics of the 1995 Colima-Jalisco earthquake, a magnitude 8.0 event that produced strong tsunami currents in Manzanillo Bay. From this ensemble, 28 realizations with magnitudes within a narrow range of the target were retained for controlled comparison. The simulations were run using GeoClaw, a well-validated solver for the nonlinear shallow water equations that employs adaptive mesh refinement to efficiently handle both broad-scale wave propagation and fine-scale coastal inundation. The comparison with the observed tide gauge record from Manzanillo’s harbor showed that the model captured the first wave, a trough followed by a peak consistent with regional subduction movement, along with the trailing oscillations and their decay over time. The simulated arrival times were slightly delayed by about 3 to 10 minutes across all scenarios, likely due to small inaccuracies in the assumed rupture location or velocity model, but the wave period and overall duration were well reproduced.

The numerical setup was designed to resolve the complex coastal geography of Manzanillo in detail. Five levels of nested grids were used, with the coarsest covering a large portion of the Pacific at 5 arc-minute resolution and the finest centered on the bays and lagoons of interest at about 30 meters. This resolution was adequate to model water flow through narrow channels and around small islands or peninsulas. The team merged global bathymetry with high-resolution local surveys and topographic data to create a seamless digital elevation model capturing features like beach berms and channel connections between the ocean and lagoons. A Manning’s roughness coefficient of 0.025 was applied to account for energy losses from seafloor roughness and onshore terrain friction, a standard reference value for tsunami inundation modeling over mixed coastal environments.

The results reveal a spatially shifting hazard pattern that depends on earthquake size. For a 100-year return period, corresponding to a 40 percent exceedance probability in 50 years, the highest tsunami amplitudes among five virtual gauge stations were found at point P3, situated near the city center in Manzanillo Bay. This suggests that for relatively moderate events, the central bay experiences the largest waves. However, as return periods extend beyond 800 years, the hazard shifts to Santiago Bay, represented by point P2, which begins to show higher levels than the city center. Point P5, representing the open coast near Campos, also shows increased hazard by a 1,000-year return period, consistent with the idea that open-coast segments are prone to the impact of the largest tsunamis. This spatial variability underscores the importance of site-specific assessment rather than relying on regional averages.

The inundation maps generated for 50-, 100-, and 500-year return periods consistently identify low-lying areas near the coast as the most vulnerable. Santiago Bay, the downtown and port area around Manzanillo Bay, and the extensive wetlands around Cuyutlán Lagoon emerge as hotspots of inundation. For the 50-year scenario, roughly corresponding to a magnitude 7.3 earthquake, flood extents primarily affect zones immediately adjacent to the shoreline, though the Cuyutlán Lagoon area experiences the farthest inland flooding at about 5.6 kilometers from the coast, largely because the terrain there is very flat and low. Once water overtops the beach or enters the lagoon inlet, it can propagate far across the marshy land with little resistance. For the 100-year event, roughly a magnitude 7.74 scenario, flooding extends about 1.8 kilometers inland in Manzanillo Bay and up to 5.8 kilometers at Cuyutlán in the static maps, with dynamic simulations showing depths exceeding 6 meters within the port.

The 500-year scenario, approaching the upper magnitude considered at roughly Mw 8.3, delineates the outer envelope of tsunami flooding one might reasonably prepare for. The static map shows very extensive flooding at Manzanillo Bay, covering nearly all urban and port areas with inundation reaching up to 2.6 kilometers inland. Around Juluapan Lagoon, flooding may extend 2.5 kilometers inland, affecting surrounding roads and residential areas. At Cuyutlán, inundation reaches up to 6.2 kilometers, suggesting the entire lagoon would become part of the ocean temporarily. The dynamic simulation for a magnitude 8.49 event indicates extreme water depths, with about 6.2 meters in Cenicero Bay and 6.4 meters on the eastern side of Santiago Bay. The port area records about 7.8 meters of overtopping the breakwaters and 6.9 meters of inundation inside the port, which could cause severe disruption to operations and infrastructure.

A key methodological contribution of the study is the explicit comparison between simplified elevation-based mapping and physics-based hydrodynamic simulation. The elevation-based approach uses Green’s law to scale offshore wave amplitudes to nearshore reference depths and then floods all hydraulically connected terrain below a prescribed elevation. The hydrodynamic approach simulates representative events whose offshore peak coastal amplitudes match the target exceedance levels from the probabilistic curves. For the 500-year event, the static method yields an inundated area of approximately 54.16 square kilometers, compared with 10.82 square kilometers from the full hydrodynamic simulation, indicating the simpler estimate is roughly five times larger. This discrepancy highlights the importance of frictional dissipation, represented by the Manning roughness coefficient, in controlling inland flow. The static method floods all connected terrain below a prescribed elevation without accounting for finite wave duration, momentum losses, topographic steering, or energy dissipation during overland flow. The authors suggest the GIS approach is useful as a conservative screening tool but may substantially overpredict inland penetration in low-gradient lagoonal and urban environments.

The findings carry significant implications for coastal planning and preparedness in Manzanillo. The hazard exceedance curves indicate that moderate tsunami amplitudes up to 1 meter have a high probability of occurrence within a 50-year exposure window, while a 500-year return period event could inundate areas up to 5 or 6 kilometers inland through the city’s water bodies. Critical infrastructure, including the port, power facilities, and tourism centers, largely lie within the 100-year tsunami flood zone. The maps and exceedance curves can support preliminary screening of evacuation zones, land-use planning, and critical infrastructure exposure, though the authors caution they should not be interpreted as final operational evacuation products. The study focuses on earthquake-generated tsunamis from local and regional subduction sources and does not include far-field or volcanic-source events, which the authors note should be addressed in future assessments. They also recommend incorporating spatially variable roughness, sensitivity analyses, and more explicit treatment of epistemic uncertainty, along with interactions with other coastal hazards such as sea-level rise and storm surge, to support more integrated coastal hazard planning for Manzanillo.

Subject of Research: A site-specific probabilistic tsunami hazard assessment and flood modeling study for Manzanillo, Colima, Mexico, using stochastic earthquake sources and hydrodynamic simulations.

Article Title: Probabilistic tsunami hazard assessment and flood modeling: a case study of Manzanillo, Colima, Mexico

Article References: Vázquez-Caamal, L., Appendini, C. M., Solano-Hernández, E. A., Ruiz-Angulo, A., & Ordaz, M. (2026). Probabilistic tsunami hazard assessment and flood modeling: a case study of Manzanillo, Colima, Mexico. Bulletin of Earthquake Engineering. https://doi.org/10.1007/s10518-026-02701-8

Image Credits: AI Generated

DOI: 10.1007/s10518-026-02701-8

Keywords: tsunami, PTHA, Manzanillo, Mexico, flood modeling, GeoClaw, stochastic rupture, subduction zone, coastal hazards, inundation mapping, Cuyutlan Lagoon, hazard assessment

The stakes are enormous. In 2024, the United States experienced 27 separate billion-dollar weather and climate disasters, and tornadoes and thunderstorms accounted for more than half of them. Although violent tornadoes rated EF3 and above represent a small fraction of all occurrences, they dominate economic losses, infrastructure disruption and threats to life. Wind engineers, unlike meteorologists issuing real-time warnings, need long-term probabilistic estimates of peak wind speeds to inform design standards for buildings and infrastructure, and the broad EF categories obscure exactly the information they need most.

The core innovation of the study lies in what it predicts. Rather than assigning tornadoes to EF categories, the researchers model the probability that a tornado’s maximum damage wind speed (DWS) exceeds specific thresholds: 136, 140, 145, 150, 155, 160 and 165 miles per hour, all within the EF3 range of 136 to 165 mph. This range matters because it is a transition zone. Most structures built to current codes are expected to withstand EF2-and-below winds of 135 mph or lower, while EF4 and EF5 events above 166 mph often produce near-total destruction. The EF3 band, in which tornadoes cause a broad spectrum of severe but not necessarily complete structural damage, is where prediction offers the greatest engineering value, and EF3 tornadoes occur far more frequently than their stronger counterparts, providing more data for model development.

Building a trustworthy dataset was the study’s crucial first step. The team quality-controlled and cross-validated records from three federal sources: the Storm Events Database maintained by the National Centers for Environmental Information, the Damage Assessment Toolkit (DAT), and the Storm Prediction Center (SPC) tornado database. The DAT, which provides the most comprehensive post-event damage information, served as the primary source. Records with missing, inconsistent or duplicated values were flagged and investigated rather than simply deleted, and corrections were made only when supported by at least two independent sources, including National Weather Service and NOAA field survey reports. The result is a curated historical dataset of more than 630 authenticated EF3 through EF5 tornadoes in the United States from 2007, when the EF scale was introduced, through 2025.

With the data curated, the researchers turned to a Random Forest (RF) classifier, an ensemble method that constructs many decision trees and aggregates their predictions to improve accuracy and reduce overfitting. The model uses just four geometric features of the tornado path: track length, maximum width, and the sine and cosine components of the path bearing angle. The authors deliberately minimized the feature set because earlier studies showed that most variables in tornado databases contribute little predictive power, and because correlated variables such as duration, which tracks closely with path length, introduce redundancy. The bearing angle, a circular variable where 0 and 360 degrees mean the same direction, was decomposed into sine and cosine components to avoid discontinuities that would confuse the algorithm.

Model development centered on Iowa, home to the National Testing Facility for Enhancing Wind Resiliency of Infrastructure in Tornado-Downburst-Gust Front Events (NEWRITE) project and part of the midwestern tornado alley. Because violent tornadoes are rare, only 10 EF5 tornadoes occurred nationwide between 2007 and 2025, the team had to confront severe class imbalance. They enriched the training data with above-threshold tornadoes from across the country and applied the Synthetic Minority Over-sampling Technique (SMOTE) where necessary. Because this rebalancing inflates the apparent probability of extreme events, the researchers calibrated the raw RF outputs using logistic regression, following a method proposed by Dankowski and Ziegler, which produced closer agreement with observed Iowa exceedance rates than traditional Bayes prior correction.

The results are striking. Across all seven DWS thresholds, the Iowa RF models and their independent validation datasets consistently achieved accuracy and weighted F1 scores above 85 percent, with recall scores for both above-threshold and below-threshold tornadoes generally exceeding 85 percent. Permutation feature importance analysis revealed that tornado path width was by far the most influential predictor, followed by path length, with the directional variables ranking third and fourth. This makes physical sense: stronger tornadoes carve wider damage paths, and the relationship proved more direct than the link between intensity and path length, since some EF3 tornadoes leave shorter tracks than weaker storms. Comparisons with logistic regression and XGBoost baseline models showed broadly comparable overall performance, but the RF models maintained consistently higher recall for the above-threshold class, the metric that matters most for hazard assessment, where missing a dangerous tornado is far worse than a false alarm.

Perhaps the most consequential finding concerns geography. When the Iowa 136 mph model was applied to other states, performance varied substantially. Neighboring states such as Illinois, Missouri, Nebraska, Wisconsin and Minnesota performed comparably to Iowa, with Illinois reaching 89 percent accuracy, but Kansas, the Dakotas and Florida showed lower recall for above-threshold tornadoes, while Mississippi and Tennessee caught dangerous tornadoes well but misclassified weaker ones. In Kansas, path length even showed slightly negative feature importance, apparently because several EF3 tornadoes there had unusually short tracks. State-specific models developed for Missouri, Mississippi, Tennessee and Texas generally achieved accuracy between 75 and 95 percent, suggesting that tornado geometry-damage relationships are not uniform across the country and that state-specific or regional models may capture local variability better than any universal model.

The framework’s engineering implications extend well beyond classification. Structural engineers use fragility functions, conditional probabilities that a given building type exceeds a specific degree of damage at a given wind speed, to assess tornado risk. By combining the RF model’s DWS exceedance probabilities with state-specific fragility functions, the approach could estimate the overall probability that a residential building suffers severe or life-threatening damage in a tornado, moving the field toward performance-based design. The authors note that previous engineering-based damage assessment tools require extensive calculations for each location, whereas the machine learning framework embodies multiple variables at once and can be updated as new event data arrive each year.

The study also candidly acknowledges its limits. Only 10 EF5 tornadoes occurred in the study period, so the rarest and most destructive events remain poorly represented, and historical damage records carry uncertainties from survey practices, varying state building codes and evolving rating methodologies. The authors caution that model predictions should be interpreted as data-driven estimates rather than direct representations of tornado wind fields, especially since surface wind speeds during tornadoes remain nearly impossible to measure directly. Still, by demonstrating that a tornado’s path geometry, above all its width, carries quantifiable information about its destructive potential, the research opens a practical path toward probabilistic tornado hazard maps, fragility analysis and, ultimately, more resilient building design in the communities that need it most.

Subject of Research: Machine learning prediction of tornado damage wind speed exceedance from tornado path geometry

Article Title: Tornado damage prediction using machine learning

Article References: Li, Z., Cristea, N., Sarkar, P., & Reed, D. (2026). Tornado damage prediction using machine learning. Frontiers in Built Environment, 12, Article 1971791. https://doi.org/10.3389/fbuil.2026.1971791

Image Credits: AI Generated

DOI: 10.3389/fbuil.2026.1971791

Keywords: tornado, machine learning, random forest, Enhanced Fujita scale, damage wind speed, wind engineering, path width, Iowa, hazard assessment, fragility functions, Storm Prediction Center, structural design

The study, led by Muhammad Naveed Zafar of Université Savoie Mont Blanc and colleagues across France, Italy, Switzerland, and the United Arab Emirates, reconstructs two catastrophic delta collapses in Lake Iseo with unprecedented detail. By combining seismic reflection profiles, sediment cores, historical manuscripts from the Vatican Library, archaeological evidence, and state-of-the-art numerical modeling, the team demonstrated that earthquakes in 1117 and 1222 CE triggered massive subaqueous landslides at the Oglio River Delta, generating tsunamis with maximum run-ups of roughly 50 meters and 13 meters respectively. These are among the largest landslide-induced tsunamis ever inferred in any lake setting worldwide.

The mechanism behind this hazard begins on the hillsides of Val Camonica, one of the largest valleys in the southern European Alps. Lake Iseo sits at the downstream end of this catchment, which drains 1,777 square kilometers of steep terrain through the Oglio River. Since the Roman period, intensified deforestation, agriculture, and pastoralism have dramatically accelerated hillslope erosion. Paleoenvironmental records show that sediment flux to the lake tripled compared with earlier Holocene background levels. All that extra sediment built up and steepened the delta at the lake’s northern margin, overloading the subaqueous slopes and making them increasingly susceptible to collapse during seismic shaking.

The geological evidence for what happened next is preserved in the deep basin of Lake Iseo as thick mass-wasting deposits, sometimes called megaturbidites. The 1117 CE Verona earthquake, magnitude 6.5, produced a deposit reaching a maximum thickness of about 7 meters and an estimated volume of 62 million cubic meters. This exceptional event represented roughly six centuries of accumulated human-driven sediment that had piled up at the delta front since the previous collapse in the sixth century. Just over a century later, the 1222 CE Brescia earthquake, magnitude 6.0, triggered a second collapse that left a thinner deposit about 2 meters thick with a volume of 10 million cubic meters.

What makes the second event particularly striking is how quickly the delta rebuilt itself. Using a mean deep-basin sedimentation rate of about 3 millimeters per year across an 18-square-kilometer depositional area, the researchers estimate that roughly 5.7 million cubic meters of sediment accumulated between the two earthquakes. That figure is conservative, because it only counts fine-grained material preserved in the deep basin and excludes coarser sediment stored at the delta front where the failure actually originated. The accommodation space carved out by the 1117 collapse likely enhanced sediment trapping, accelerating the rebuild. In other words, the delta had barely recovered before it was loaded and primed to fail again.

To translate these geological deposits into wave heights and inundation distances, the team employed a two-stage numerical approach. First, they simulated the landslide dynamics using BingClaw, a two-layer depth-averaged viscoplastic debris-flow model based on Herschel-Bulkley rheology, specifically designed for cohesive subaqueous landslides. The model represents the failed mass as an upper plug layer moving en bloc and a lower shear layer where deformation occurs once applied stress exceeds the yield strength. Yield strength values were constrained by observational runout distances from seismic profiles, with sediment density and water content taken directly from core data. The 1117 slide used a yield strength of 25 pascals and a bulk density of 1,600 kilograms per cubic meter, while the 1222 slide, being wetter and coarser, used a lower yield strength of 7 pascals.

Second, the time-dependent bathymetric changes from the landslide simulation were fed into GeoClaw, a nonlinear shallow-water equation model running on a 10-meter grid. The results are sobering. The 1117 event produced wave heights exceeding 50 meters in the northern part of the lake, inundating nearby shores within just one minute and propagating up to 2.7 kilometers inland. The 1222 event generated waves peaking at about 13 meters with inundation extending roughly 1 kilometer inland. Monte Isola, the large central island and the largest inhabited lake island in Europe, acted as a natural barrier that split the incoming waves and partially shielded cities behind it. Even so, the town of Iseo still experienced maximum wave heights of about 5.5 meters in 1117 and 2 meters in 1222, arriving roughly 7 minutes after the landslides began.

The numerical reconstructions find powerful corroboration in written history. A twelfth-century manuscript preserved in the Vatican Library describes the 1117 event in vivid terms, stating on folio 184r that the lake overflowed so much that very few inhabitants of nearby villages escaped the danger, and that many castles and fortresses collapsed miserably with the destruction of their inhabitants. Archaeological evidence adds further constraints. The Pieve church of Sant’Andrea d’Iseo underwent major reconstruction shortly after 1117, and the simulations indicate the tsunami reached the shoreline in front of the church with a maximum wave height of about 3.5 meters and an inundation distance of 165 meters, sufficient to flood the church area. Extensive structural damage documented around the lake shore following the 1222 earthquake supports the second tsunami as well.

The implications extend far beyond one Italian lake. A global synthesis of 632 lake-watershed records shows that human-driven soil erosion has now reached its highest level since the Industrial Era, progressively overloading deltas worldwide. Thick megaturbidites in the deep basins of nearby Lakes Garda and Como suggest those steep, sediment-overloaded slopes also collapsed during medieval seismic activity. Further afield, Lake Geneva, Lake Sils, Lake Brienz, and Lake Lucerne in the Alps have all experienced recurrent tsunamigenic delta collapses during periods of intensified erosion. Outside Europe, lakes Eklutna, Skilak, and Kenai in south-central Alaska suffered large tsunamigenic delta collapses during the 1964 megathrust earthquake, and delta collapses have been identified as the primary source of repeated lake tsunamis in New Zealand.

Today, the shores of Lake Iseo are among the most densely developed in the Italian Alps, and the large perialpine lakes of northern Italy together host more than 370,000 permanent residents and attract tens of millions of visitors annually. The researchers emphasize that warning times for near-field locations would be negligible, with waves reaching the northern shore in less than a minute, leaving no realistic window for evacuation even with an early warning system. The team argues that tsunami hazard assessments in densely inhabited lacustrine environments must now incorporate human-driven sediment dynamics, particularly in tectonically active regions. The hidden legacy of land use, they conclude, means that today’s land management decisions will shape the disasters of tomorrow, and communities around Lake Iseo and similar settings worldwide face a low-probability but high-consequence threat that has been quietly building for two millennia.

Subject of Research: Human-accelerated erosion of lake deltas increasing earthquake-triggered tsunami hazard, reconstructed from Lake Iseo sediment records and numerical modeling

Article Title: Human-enhanced tsunami hazard in large lakes

Article References: Zafar, M. N., Sabatier, P., Dutykh, D., Rapuc, W., Paris, R., Piccin, A., Vuillemin, P., & Anselmetti, F. S. (2026). Human-enhanced tsunami hazard in large lakes. Nature Communications, 17(1), Article 10531. https://doi.org/10.1038/s41467-026-78272-8

Image Credits: AI Generated

DOI: 10.1038/s41467-026-78272-8

Keywords: lake tsunamis, delta collapse, subaqueous landslides, Lake Iseo, soil erosion, land use, paleoseismology, numerical modeling, Alps, natural hazards, sediment flux, hazard assessment

Glaciologists have long known that proglacial lakes act as powerful accelerants on glacier retreat. Lake water promotes thermal undercutting and calving at the ice front, while the buoyant support of the water reduces the effective pressure at the glacier bed, lowering basal friction and allowing the ice to slide faster. Lake-terminating glaciers in the Himalaya and elsewhere consistently show faster flow, stronger velocity variability, and more pronounced thinning near their termini than their land-terminating neighbors. But while the hazards of glacial lake outburst floods, or GLOFs, have been studied extensively from the perspective of downstream destruction, almost nothing has been quantified about what a sudden lake drainage does to the glacier that feeds the lake. The new study of Jinwu Glacier, published in The Cryosphere, provides the first quantitative answer.

Led by Yunyi Luo and Qiao Liu of the Institute of Mountain Hazards and Environment at the Chinese Academy of Sciences, the research team assembled a remarkable archive of satellite observations spanning 1990 to 2025. Landsat imagery at 30-meter resolution documented the early decades, while high-resolution PlanetScope images at 3 meters captured the critical years around the flood. Using the COSI-Corr image correlation technique, the team extracted annual surface velocity fields from 2017 to 2025, tracking the displacement of surface features between carefully selected image pairs acquired in September and early October to avoid terrain shadows. Digital elevation models built from ASTER stereo data for 2002, 2014, and 2025 revealed patterns of surface elevation change, and manual mapping of the terminus position and lake area completed the picture.

The headline result is striking. Within 100 to 200 meters of the glacier terminus, mean ice-flow velocity fell from 42.97 meters per year during 2017 to 2020 to 24.09 meters per year during 2021 to 2025, a slowdown of roughly 44 percent. The annual record shows the decline in detail: velocities near the front dropped from 48.43 meters per year in 2017 to 38.53 meters per year in 2020, then plunged to 27.03 meters per year after the flood, accompanied by a fundamental change in the along-flow velocity gradient. Before the GLOF, ice moved rapidly and almost uniformly along the lower tongue; afterward, speeds clearly decreased toward the terminus, a signature of stronger resistive forces acting on the ice front.

Equally revealing is what did not change. In the section of the glacier tongue 650 to 1600 meters upstream, mean velocity shifted only from 37.55 to 35.24 meters per year, a reduction of about 6 percent that is barely distinguishable from natural variability. The reason lies in the glacier’s topography. Jinwu Glacier’s lower tongue is interrupted by a pronounced icefall, a steep cascade of ice with an elevation drop of about 300 meters and slopes exceeding 35 degrees. Previous work at Alaska’s Yahtse Glacier and elsewhere has shown that such icefalls can decouple glacier sections dynamically, attenuating stress perturbations before they propagate upstream. At Jinwu, the icefall appears to have acted as a dynamic firewall, confining the response to the flood almost entirely to the terminus sector below it.

The frontal morphology also transformed. PlanetScope imagery shows that the terminus ice cliff degraded and collapsed after the flood, with ice-cliff collapse documented by UAV surveys and field investigations in October 2021. Intriguingly, the mapped terminus position advanced by about 23 meters between 2020 and 2021, before resuming retreat at a modest 8.47 meters per year through 2025. The authors interpret this apparent advance with caution: rather than a genuine surge of the glacier, it likely reflects short-term morphological adjustment of a destabilized ice front, as the cliff collapsed and debris redistributed while the glacier adjusted to its new, lake-free boundary conditions. The physical mechanism is well understood in glaciological theory. Removing the lake eliminated hydrostatic pressure and buoyant support at the front, increasing basal and frontal resistive stresses and slowing the ice, while simultaneously undermining the stability of the unsupported cliff.

The longer-term context matters too. Jinwu Glacier has been retreating for decades, shrinking from 5699 meters in length in 1990 to 4938 meters in 2025, a cumulative loss of more than 760 meters. Its lake grew by roughly 56 percent between 1990 and 2019, part of the broader proliferation of glacial lakes across High Mountain Asia as the climate warms. Surface elevation data show sustained thinning, with the terminal 550 meters of the glacier lowering at rates that accelerated from 2.90 meters per year in 2002 to 2014 to 3.71 meters per year in 2014 to 2025, while thinning in an upstream section actually slowed. The team is careful to note that the elevation-change record cannot isolate an event-scale response to the flood, since the 2014 to 2025 differencing period straddles the event, but the pattern of pronounced terminal thinning is consistent across independent datasets.

The study is also notable for its rigorous treatment of uncertainty. The team validated their satellite-derived velocities against manually traceable surface features, achieving a correlation of 0.96 and a root-mean-square error of just 2.14 meters per year in the near-terminus zone where the slowdown was strongest. They acknowledge that image matching is noisier in the crevassed, shadow-prone icefall zone, and that elevation-change estimates disagree more near the steep terminus than upstream, but targeted checks confirm that the central finding, the post-flood deceleration gradient, is robust. This methodological transparency matters, because the conclusion carries weight for hazard science: GLOFs are not merely downstream disasters but event-driven perturbations that can reorganize the dynamics of the source glacier itself.

As climate projections point toward warmer, wetter monsoon-influenced conditions in High Mountain Asia and a rising frequency of moraine-dammed lake outbursts, the implications are clear. Models of lake-terminating glacier evolution, and the hazard assessments built upon them, should treat sudden lake drainage as a discrete forcing event capable of rapidly altering ice flow, frontal stability, and calving behavior. Jinwu Glacier offers a rare, well-documented glimpse of what happens when the lake that once drove a glacier’s acceleration vanishes overnight: the ice slows, the cliff crumbles, and a new dynamic equilibrium begins to take shape, one icefall away from the rest of the glacier.

Subject of Research: Glacier dynamic response to abrupt lake disconnection following a glacial lake outburst flood at Jinwu Glacier, southeastern Tibet

Article Title: Glacier response to the transition from lake- to land-terminating following a glacial lake outburst flood at Jinwu Glacier, southeastern Tibet

Article References: Luo, Y., Liu, Q., Fu, Y., Lu, X., Yin, Y., Yang, J., Zheng, G., & Lu, X. (2026). Glacier response to the transition from lake- to land-terminating following a glacial lake outburst flood at Jinwu Glacier, southeastern Tibet. The Cryosphere, 20(10), 5883-5894. https://doi.org/10.5194/tc-20-5883-2026

Image Credits: AI Generated

DOI: 10.5194/tc-20-5883-2026

Keywords: glacial lake outburst flood, Jinwu Glacier, Tibetan Plateau, glacier dynamics, ice velocity, proglacial lake, ice cliff collapse, remote sensing, The Cryosphere, High Mountain Asia, glacier retreat, hazard assessment

The research, led by Chenxi Hu of the University of Texas at Austin together with Xinxin Sui of the National Science Foundation National Center for Atmospheric Research, Dev Niyogi, Clint N. Dawson, and Zong-Liang Yang, analyzed 885 observed tropical cyclone landfall events spanning the years 1979 through 2024. That forty-five-year record allowed the team to compute how the rate of post-landfall weakening has changed over time for storms of different intensities. The results were striking. Strong storms, defined in the study as those with maximum sustained winds of at least 29 meters per second, now weaken 18.6 percent faster after landfall than they did in the earlier part of the record. Weak storms, by contrast, persist 14.5 percent longer once they move inland, holding onto their circulation and their rainfall-producing capacity well beyond where past storms of similar strength would have fizzled.

This split behavior matters because the two trends create opposite hazard profiles. A strong hurricane that decays rapidly after landfall concentrates its destructive winds and coastal surge near the shore, intensifying the already devastating impacts on coastal communities. A weak storm that lingers, however, becomes a slow-moving rain engine, pushing flooding rainfall and inland wind damage into regions that may have little experience preparing for tropical systems. The study’s authors describe the result as a dual-threat landscape: coastal impacts from faster-dying strong storms on one side, and expanding inland risks from increasingly persistent weak systems on the other.

What makes the divergence particularly interesting from a physical standpoint is that the two storm classes respond to fundamentally different aspects of the changing environment. The researchers found that weak storms are governed primarily by thermodynamic factors. Warmer sea surfaces pump more energy into storms before landfall, an enhanced atmospheric moisture supply keeps their circulations fueled after they leave the coast, and increased oceanic influence along storm tracks means that even over land, weak systems can draw on wetter conditions that slow their decline. All three of these thermodynamic trends favor prolonged inland persistence, and weak storms, it turns out, are exquisitely sensitive to them.

Strong storms tell a different story. Their post-landfall behavior is more tightly coupled to dynamic and land-surface conditions rather than to heat and moisture alone. Over the study period, storms have been translating faster, and a faster-moving cyclone spends less time over any given stretch of terrain, but it also experiences more rapid changes in the forces acting on it. At the same time, atmospheric stability has been increasing, which suppresses the deep convective activity that a strong cyclone needs to maintain its inner core once its oceanic energy supply is cut off. Drier soils compound the problem, reducing the evaporation and latent heat fluxes from the land surface that can partially sustain a storm’s circulation. Together, these dynamic disruptions promote faster decay in strong systems, which the study shows are more vulnerable to such disruption than their weaker counterparts.

To move beyond correlation and probe the underlying mechanisms, the team conducted numerical experiments that isolated the different environmental influences. Those simulations confirmed the observational picture: weaker storms respond more strongly to thermodynamic enhancement, gaining longevity when moisture and surface heating conditions improve, whereas stronger storms are disproportionately damaged by dynamic disruption. In other words, the same changing climate is pushing the two ends of the tropical cyclone intensity spectrum in opposite directions after landfall, not because the environment is inconsistent, but because the storms themselves have different sensitivities depending on how powerful they are.

The implications for risk assessment are immediate. Most existing hazard models and building codes implicitly assume that environmental change affects storms of all intensities uniformly, so a trend identified for the average landfalling cyclone is applied across the board. The new findings show that this uniform-response assumption can misallocate protective resources in both directions. Coastal defenses and evacuation planning may need to account for strong storms that deliver their worst impacts in a compressed window near the shore, while inland flood management, dam operations, and emergency response networks in regions far from the coast face a growing burden from weak but persistent tropical systems that historically would have decayed before reaching them.

The geographic dimension of the problem is equally consequential. Regions that have historically been spared the worst of tropical cyclone impacts, precisely because weak storms tended to die quickly after landfall, may now find themselves in the path of systems that survive longer over land. These vulnerable and often unprepared areas may lack the drainage infrastructure, floodplain zoning, and public awareness campaigns that coastal populations have developed over generations of exposure. The study’s authors urge refined hazard assessment and adaptation strategies that explicitly account for intensity-dependent behavior rather than relying on aggregate storm statistics that blend strong and weak systems into a single, misleading average.

The methodological foundation of the work rests on a careful observational record. By assembling nearly nine hundred landfall events across four and a half decades and stratifying them by intensity, the researchers could separate long-term trends in decay behavior from the considerable natural variability that characterizes individual storm seasons. The combination of global observations with targeted numerical experiments gives the findings a robustness that purely statistical studies often lack, and the open-access publication ensures that the underlying analysis can be scrutinized and extended by other research groups. The work was supported by the U.S. Department of Energy’s Advanced Scientific Computing Research Program, with additional support from the National Science Foundation’s National Center for Atmospheric Research.

As the climate continues to warm, the divergence documented in this study is likely to become more pronounced rather than less. Continued sea-surface warming and a moister atmosphere will keep strengthening the thermodynamic lifeline that sustains weak storms over land, while trends toward faster storm motion, greater stability, and drier soils in many landfall regions will continue to erode the resilience of strong systems once they leave the ocean. For forecasters, the message is that intensity must become a first-order variable in post-landfall prediction, not an afterthought. For the public, the message is subtler but no less important: the storms that linger quietly over the interior, stripped of their category ratings but not of their rainfall, may pose the fastest-growing tropical cyclone threat to the largest number of people.

Subject of Research: Intensity-dependent trends in the post-landfall decay of tropical cyclones

Article Title: Divergent trends in post-landfall decay of strong and weak tropical cyclones

Article References: Hu, C., Sui, X., Niyogi, D., Dawson, C. N., & Yang, Z.-L. (2026). Divergent trends in post-landfall decay of strong and weak tropical cyclones. Nature Communications. https://doi.org/10.1038/s41467-026-77958-3

Image Credits: AI Generated

DOI: 10.1038/s41467-026-77958-3

Keywords: tropical cyclones, landfall, post-landfall decay, hurricanes, climate change, inland flooding, storm intensity, thermodynamics, atmospheric moisture, land surface, hazard assessment, Nature Communications

The research, led by Erin L. Harvey of Durham University together with colleagues from institutions in the United Kingdom, the United States, Nepal, and New Zealand, set out to answer a deceptively simple question: when you watch a mountain landscape for more than two decades, what actually determines where landslides appear? The team compiled and analysed multi-temporal landslide inventories at national scale, allowing them to track not just where landslides occurred, but which sites failed repeatedly, which scars healed and vanished from view, and how individual triggering events stamped their signature onto the landscape. The work was supported by the UK Global Challenges Research Fund through the Sajag-Nepal project and the UKRI-DFID SHEAR programme, as well as by NASA’s Land-Cover and Land-Use Change grant.

The headline finding is striking: recurrent or persistent landslides dominate the national-scale pattern of landsliding in Nepal. At any given point in time, roughly 80 percent of all visible landslide areas are not fresh failures on previously stable slopes, but sites with a documented history of movement. In other words, the landslide hazard that mountain communities face on an everyday basis is overwhelmingly concentrated in places that have already failed. Slopes, it turns out, have long memories, and the scars of past events remain active, evolving features rather than one-off blemishes that quickly disappear.

This persistence has profound implications for how hazard maps are drawn and how mitigation budgets are spent. Traditional hazard assessment often treats each landslide as an isolated event, cataloguing fresh failures after a major earthquake or an exceptional rainfall episode and using them to model where future failures might occur. But if four out of every five visible landslides at any moment are reactivations or continuations of pre-existing features, then a hazard assessment that ignores the existing inventory of unstable slopes is systematically missing the majority of the present-day danger. The study’s authors argue that pre-existing landslides must be given far greater weight when assessing hazard, particularly in regions where the resources available for mapping and mitigation are limited.

The spatial statistics reinforce that message. Of the new landslide areas recorded in the dataset, only half occurred in locations more than 30 metres away from other landslides. That means a substantial fraction of fresh failures sprouted in the immediate vicinity of existing ones, in zones where slopes were already weakened, scarred, and stripped of stabilising vegetation. Landslides, in this sense, beget landslides. The edges of old scars, the debris deposits left by previous failures, and the disrupted drainage patterns around them all create conditions that favour subsequent movement nearby. For planners deciding where to route roads, site buildings, or prioritise slope stabilisation, the neighbourhood of a known landslide is demonstrably not safe ground.

Perhaps the most technically significant contribution of the study is its demonstration that individual triggering events leave distinctive, measurable footprints at the national scale. By computing metrics such as landslide density, persistence, and recurrence across the 22-year record, the researchers showed that the fingerprint of a major earthquake differs recognisably from that of rainfall-driven failure episodes. Earthquakes tend to trigger widespread, synchronous failure across broad regions in an instant, while rainfall-driven landsliding accumulates more gradually and follows different spatial patterns tied to hydrological conditions. These signatures, visible only because the record spans decades and the entire country, would be invisible in the short, local datasets that have dominated the field until now.

The methodological lesson is that multi-temporal landslide inventories are far more valuable than the sum of their parts. A single post-event inventory tells you where slopes failed once. A time series of inventories tells you which slopes keep failing, how long scars remain visible before vegetation and erosion erase them, how quickly hazard re-cumulates after a major event, and how the overall hazard landscape shifts between quiet years and catastrophic ones. Metrics such as persistence and recurrence, the authors argue, add significant value to such inventories and provide a practical basis for targeting mitigation efforts. In a country like Nepal, where thousands of kilometres of roads cross landslide-prone terrain and where maintenance budgets are chronically stretched, knowing which slopes are chronic offenders is exactly the information that makes limited resources go furthest.

The timing of this work matters. As mountain populations grow and infrastructure expands into increasingly marginal terrain, the number of people and assets exposed to landslide hazard is rising steeply across the Himalaya and other high mountain ranges worldwide. Climate change is intensifying rainfall extremes in many mountain regions, and seismic hazard in Nepal remains high more than a decade after the devastating 2015 Gorkha earthquake, which itself triggered tens of thousands of landslides. Understanding whether the hazard landscape after such an event decays back to normal, or whether it leaves a lasting legacy of weakened, repeatedly failing slopes, is central to realistic long-term risk management. The 22-year Nepali record suggests that legacy is both real and dominant.

There is also a broader scientific payoff. Geomorphologists have long debated how landslide activity couples to landscape evolution: whether landslides are stochastic, essentially random responses to individual storms and shakes, or whether they cluster in persistent hotspots controlled by geology, slope geometry, and hillslope material. The Nepal record comes down firmly on the side of persistence and clustering. The dominance of recurrent failures and the tight spatial association of new failures with old ones indicate that the landscape itself, not merely the timing of triggers, governs where hazard accumulates. That insight should improve the next generation of landslide susceptibility models, which can now be trained not only on where landslides have happened but on how long they last and how likely they are to return.

For the millions of people who live on and beneath Nepal’s unstable slopes, the practical takeaway is refreshingly concrete: the slopes that have already failed are the ones most likely to fail again, and the ground near them is hardly safer. Mapping, monitoring, and treating known landslide sites, rather than chasing only the aftermath of each new disaster, offers the most efficient path to reducing losses. The study’s authors hope that the metrics of density, persistence, and recurrence they developed will be adopted in other mountain countries, wherever long satellite archives and national inventories can be combined. Two decades of watching an entire mountain nation fail, heal, and fail again have shown that landslide hazard is not a series of isolated accidents but a dynamic, persistent, and mappable property of the landscape itself.

Subject of Research: Long-term national-scale landslide persistence and dynamic hazard in Nepal

Article Title: Long-term, national-scale records of landslide persistence and dynamic hazard in a mountain landscape

Article References: Harvey, E. L., Rosser, N. J., Kincey, M. E., Densmore, A. L., Chen, T.-H. K., Dhital, M. R., Seto, K. C., Basyal, G. K., Robinson, T. R., Shrestha, R., Oven, K. J., & Arrell, K. (2026). Long-term, national-scale records of landslide persistence and dynamic hazard in a mountain landscape. Communications Earth & Environment. https://doi.org/10.1038/s43247-026-04109-3

Image Credits: AI Generated

DOI: 10.1038/s43247-026-04109-3

Keywords: landslides, Nepal, hazard assessment, landslide persistence, recurrence, multi-temporal inventories, geomorphology, natural hazards, earthquake triggering, rainfall triggering, mountain landscapes, risk mitigation

The research was triggered by real events in the Swiss Alps. In June 2022, in the Gruobenwald forest near Klosters, a storm with heavy rain and wind felled a barkless, branchless dead tree standing on a slope of roughly 35 degrees. The snag, about 30 metres long with a diameter at breast height of 75 centimetres, slid approximately 120 metres downhill and came to rest on a road of national importance. Two days later, another log of similar dimensions slid down and lodged in a wooden palisade next to a house. During a subsequent helicopter operation to remove dangerous snags, two more logs were mobilised and slid onto the road and a hiking trail. In February 2022, near Rothenbrunnen, a tree slid some 200 metres in altitude, blocked both a road and a railway line, and apparently launched off a small terrace to jump clean over a 3.5-metre-high rockfall barrier.

Until now, no modelling framework existed to describe how mobilised deadwood moves on steep slopes. The team recognised that sliding logs share key mechanical characteristics with rockfall, including gravity-driven acceleration, impact, sliding, and deflection by obstacles, and therefore adapted the RAMMS::Rockfall simulation software. Their approach rests on nonsmooth rigid-body dynamics, combining hard contact laws with Coulomb friction. Each log is modelled as a rigid, convex polytope shaped like a truncated cone, defined by its diameter at breast height, stem length, and wood density, with the top diameter approximated at half the base diameter. The log’s position and orientation are tracked with generalised coordinates and unit quaternions, while its motion obeys the balance of linear momentum and Euler’s equations for rotation.

Contact between the log and the terrain is detected whenever a vertex of the log’s point-cloud geometry touches or penetrates the digital elevation model. At each contact point, the Signorini condition governs the normal force, and a spatial Coulomb friction law determines whether the log sticks or slips. Impacts are handled with a Newtonian impact law using a normal restitution coefficient of zero, consistent with the underlying rockfall model. Standing trees, whose locations and dimensions were extracted from airborne laser scanning data using single-tree detection, are represented as rigid truncated cones that block or deflect sliding logs. Rockfall barriers are included as thin, rigid, platy obstacles. The equations of motion are solved iteratively with Moreau’s time-stepping scheme and Gauss-Seidel iteration on a high-resolution 0.5-metre digital elevation model.

Calibrating the model against the documented events yielded a remarkably narrow range for the effective sliding friction coefficient, between 0.2 and 0.3, which the authors interpret as the typical contact friction between wet terrain and a wet, barkless log. With this value, the simulations reproduced the observed deposition patterns at both sites. In the Klosters case, 20,000 simulated logs were released and 1,829, or 9.1 percent, reached the national road, with average velocities of 14 metres per second and maxima of 30 metres per second. At Rothenbrunnen, the model confirmed that jumping the barrier was physically plausible: fitting a flight parabola between the take-off and landing traces gave a take-off velocity of 24 metres per second, and 178 of 20,000 simulated logs, or 0.9 percent, cleared the barrier, a small but realistic frequency given that such an event had never before been recorded there.

The simulations distilled three prerequisites for sliding deadwood hazards. First, friction must be low, which occurs when terrain or the log itself is wet, whether from rain, wet snow, or foliage cover. Second, slopes must be very steep, with all documented events occurring on gradients of 35 degrees or more; although the calibrated friction coefficient suggests mobilisation is theoretically possible above about 18 degrees, the macroscopic roughness captured by the terrain model raises the practical threshold. Third, and perhaps most intriguingly, the wood must be in a narrow decay-stage window: late stage II or early stage III in the standard classification, when bark and branches have been lost, slashing sliding friction, but the stem still retains enough structural strength to survive impacts without shattering. For spruce killed by bark beetles, this critical window opens roughly two to five years after death, while beech decays significantly faster.

Forest structure emerged as a decisive factor in the hazard equation. When the researchers removed all standing trees from the Klosters simulation, the fraction of logs reaching the road exploded from 9.1 percent to 74.5 percent; at Rothenbrunnen it rose from 23.3 percent to a full 100 percent. The forest, in other words, is a highly effective shield against hazards created by its own deadwood. Yet trees cut both ways: while they arrest logs, they also deflect them, increasing lateral spread. In the Klosters simulations, lateral spread shrank from values of 40 to 60 degrees with the real forest to 18 to 31 degrees without it, meaning that deadwood can traverse slopes far more widely than rocks and strike infrastructure that rockfall models would deem safe. Notably, logs slid further when the simulation used real laser-scanned tree positions rather than a randomly generated forest of identical density, because natural gullies, which are often sparsely vegetated, form preferential sliding corridors that random tree placement would artificially block.

The findings carry pointed implications for forest management, especially as deadwood volumes in Swiss mountain forests have doubled over three decades, from 17 to 35 cubic metres per hectare, and climate-driven disturbances promise still more standing and lying dead timber. The study reveals a genuine trade-off over time: young deadwood is strong and protective against rockfall and avalanches, mid-stage deadwood in the two-to-five-year window is both still protective and maximally prone to sliding, and old, rotten wood loses its protective value while its ecological worth as habitat and seedling substrate keeps climbing. Spatially, the same gullies that channel sliding logs are also prime locations for intercepting rockfall and stabilising snow. The authors argue that targeted interventions, such as felling and securing snags across the slope on wire-cable-supported stumps, or prioritising removal in the lower, steepest sections of forested slopes where logs have little forest left to traverse, can minimise the sliding hazard while preserving most of deadwood’s protective and ecological benefits.

The model is not without limits, and the authors are candid about them. Logs are treated as rigid bodies that never fracture, standing trees never break or dissipate energy by swaying, and the single effective friction coefficient is tied to the 0.5-metre terrain resolution, rising to an 18 percent reach frequency when a coarser 5-metre model was tested. Initial conditions are idealised, releasing logs upright within a 10-metre release polygon rather than simulating the actual toppling process. Yet because the framework is explicitly conditioned on a prior assessment that mobilisation is possible, these simplifications yield deliberately conservative, worst-case estimates of runout and impact frequency. The researchers also note that slender logs concentrate impact loads on small frontal areas, a concern for ring-net barriers with large mesh openings, and that logs left uncleared in nets can shorten the braking distance available for future rockfall. With well-documented events still rare, the team calls for systematic documentation of future incidents, but their message is already clear: sliding deadwood is a real, quantifiable, and increasingly relevant mountain hazard, and for the first time it can be simulated, mapped, and managed.

Subject of Research: Physics-based simulation of sliding deadwood logs as a natural hazard in steep mountain forests

Article Title: Simulation of sliding deadwood logs in mountain forests: towards a quantitative hazard assessment

Article References: Borner, J., Bebi, P., Bührle, L. J., Ringenbach, A., & Leine, R. I. (2026). Simulation of sliding deadwood logs in mountain forests: towards a quantitative hazard assessment. Natural Hazards and Earth System Sciences, 26(10), 4843-4859. https://doi.org/10.5194/nhess-26-4843-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4843-2026

Keywords: deadwood, mountain forests, natural hazards, rockfall simulation, Coulomb friction, nonsmooth dynamics, forest management, wood decay, Swiss Alps, airborne laser scanning, hazard assessment, bark beetle

The research, published in the open-access journal Microplastics and Nanoplastics, focuses on a technical but crucial distinction in laboratory toxicology: the difference between the administered dose and the delivered dose. The administered dose is simply the amount of plastic a scientist pours into a cell culture dish. The delivered dose is the fraction of that plastic that actually settles onto, or otherwise contacts, the cells over the course of an experiment. When particles are suspended in liquid culture medium, they do not all reach the cell layer at the same speed, and some may never fully deposit at all. Ignoring this distinction, the researchers show, can distort toxicity rankings and lead to conclusions that are wrong by a factor of two or more.

To investigate the problem systematically, the team, led by M. Kloukinioti and colleagues including co-senior authors H. Braakhuis and L. A. Parker, used fluorescent aminated polystyrene microspheres of two very different sizes: one micrometer and ten micrometers in diameter. Polystyrene is one of the most widely used plastics in consumer products and is also the standard material for laboratory test particles, making it a sensible model for the broader microplastics problem. The fluorescent labels allowed the researchers to track exactly where the particles went over time, simply by measuring how much fluorescence appeared at the bottom of the wells in a cell-free system.

The experimental results were striking. Over a standard 24-hour exposure period, neither particle size achieved complete deposition. In other words, even after a full day of settling, a significant portion of the plastic added to the dish never made contact with the surface where cells would normally grow. This finding alone challenges a widespread assumption in in vitro toxicology: that the amount of material pipetted into a well is a reasonable proxy for the amount the cells experience. For small particles suspended in viscous biological media, reality is considerably more complicated.

The researchers then turned to two computational models that attempt to predict how particles settle in liquid. The first is Stokes’ Law, the classic nineteenth-century physics equation that describes the sedimentation of spherical particles in a fluid, taking into account the particle’s radius and density, the density of the fluid, and its viscosity. The second is the RiskGONE model, a more recent dosimetry tool developed within a European nanosafety project, which incorporates additional parameters such as particle size distributions and can account for more complex settling behavior. Both models require detailed input about the culture system, including the density and viscosity of the medium and the physical characteristics of the particles themselves, which the team measured carefully using techniques such as static light scattering.

What emerged was a surprising split verdict. The sedimentation behavior of the smaller, one-micrometer particles was accurately predicted by the RiskGONE model, while the settling of the larger, ten-micrometer particles aligned closely with the classical Stokes’ Law. This size-dependent divergence matters because it means there is no single, one-size-fits-all formula for calculating delivered dose in microplastics experiments. A researcher who applies the wrong model to the wrong particle size could introduce systematic errors into their dose calculations, and those errors would propagate directly into their toxicity conclusions.

The most consequential part of the study came when the team connected their dosimetry findings to actual biological measurements. In parallel with the cell-free sedimentation experiments, the researchers exposed THP-1 cells, a widely used human immune cell line derived from monocytes, to the same polystyrene particles under identical conditions of medium composition, plate setup, and exposure volume. They then assessed cell viability using standard assays and compared the hazard estimates obtained when calculations were based on the administered dose versus the delivered dose.

The result was unambiguous: relying solely on the administered dose led to an underestimation of the microplastics’ hazard potential by approximately a factor of two. Because fewer particles reached the cells than had been pipetted into the wells, the true toxicity per delivered particle was roughly twice as high as the administered-dose calculation suggested. In practical terms, a plastic particle that appears relatively harmless when judged by the amount poured into the dish may in fact be considerably more damaging to the cells it actually contacts. For regulators and risk assessors trying to rank different plastics by hazard, such a systematic bias could shuffle the entire leaderboard.

This finding has implications that extend well beyond polystyrene spheres. Efforts to compare the toxicity of different polymer types, including polypropylene, polyvinyl chloride, polyethylene terephthalate, and polyamide, are accelerating worldwide, and many of these comparisons rest on in vitro data. If dosimetry is not properly accounted for, a plastic that settles quickly might appear more toxic than one that stays suspended, not because it is intrinsically more hazardous but simply because more of it reached the cells. Conversely, a genuinely dangerous polymer could be wrongly exonerated if most of its particles never arrived at the cell layer during the experiment. The study provides a technical roadmap for avoiding these pitfalls, emphasizing that researchers should characterize their particles’ size distribution and effective density, measure their medium’s properties, and apply an appropriate sedimentation model before interpreting any biological endpoint.

The work also highlights a broader lesson about the limits of computational models in new scientific domains. Both Stokes’ Law and the RiskGONE model were validated here against direct experimental measurements, and neither was universally correct. The authors’ approach, testing model predictions against fluorescence-based deposition data before trusting either one, offers a template that other laboratories can follow. As the field of microplastics toxicology matures and moves toward benchmark dose modeling and quantitative risk assessment, the distinction between what is poured in and what lands on the cells may prove to be one of the most important numbers in the entire experiment. The study was conducted as part of the Dutch MOMENTUM research program on microplastics and health, with funding from ZonMw, Health-Holland, and Plastics Europe, though the funding organizations had no influence on the work. The authors note that the article was shared early as a citable, peer-reviewed accepted manuscript, with a final version of record to follow, and the research stands as a reminder that in the fast-moving world of microplastics science, the most important measurements are sometimes the ones nobody thought to make.

Subject of Research: In vitro dosimetry of microplastics and the distinction between administered and delivered dose in cell toxicity testing

Article Title: In vitro dosimetry of microplastics: the importance of understanding the delivered dose in toxicity testing

Article References: Kloukinioti, M., Neijenhuis, F., Nijman, J. E., Kooter, I. M., van de Steeg, E., Braakhuis, H., & Parker, L. A. (2026). In vitro dosimetry of microplastics: the importance of understanding the delivered dose in toxicity testing. Microplastics and Nanoplastics. https://doi.org/10.1186/s43591-026-00234-0

Image Credits: AI Generated

DOI: 10.1186/s43591-026-00234-0

Keywords: microplastics, dosimetry, in vitro toxicology, polystyrene, delivered dose, administered dose, Stokes' Law, RiskGONE model, THP-1 cells, sedimentation, hazard assessment, nanotoxicology

The study, conducted by Mehmet Sari of Aksaray University, focused on a slope above Belisirma where a rock mass composed of hard ignimbrite, soft ignimbrite, and soft tuff serves as a persistent source of falling blocks. The upper section of the source rock consists of hard ignimbrite cut by persistent joints, while the lower portion is softer ignimbrite with nonpersistent joints, and the base is joint-free tuff. Stereographic analysis of the rock mass revealed three major joint sets: two sub-vertical orthogonal sets created by thermal cracks during cooling of the ignimbrite, and a near-horizontal bedding plane formed by different outflow facies of the ignimbrite deposition. Because joint spacing largely controls the size of blocks that can detach, this geological architecture determines whether the slope releases small debris or massive boulders.

To quantify what the slope could actually produce, Sari used the RocSlope3 program to reconstruct the fractured rock mass in three dimensions. Synthetic traverses defined on the upper slope generated a probabilistic joint network in which joint planes were represented as circular discs with known spacing and radius. The program then discretized the rock mass into individual blocks, identifying 354 discrete blocks with volumes ranging from a mere 0.01 cubic meters to a substantial 18.16 cubic meters. This in-situ block size distribution, a cumulative frequency curve capturing the natural variability of joint spacing and orientation, became the statistical backbone for the subsequent trajectory simulations, ensuring that the modeled rocks reflected what the cliff face could genuinely deliver.

With the block inventory established, the study turned to the central question: how do different impact models change the predicted behavior of falling rocks? The research compared two fundamentally different approaches available in the RocFall3 software. The lumped-mass model treats every rock as a dimensionless point mass, an infinitesimal particle with no physical volume, shape, or true rotation. Energy loss during impacts is governed entirely by a preset coefficient of restitution, and rotation is reduced to a kind of pseudo-rotation derived from an equivalent radius. The rigid-body model, by contrast, uses non-smooth multi-contact dynamics that explicitly account for the rock’s exact three-dimensional geometry, volume, and mass distribution. Impacts are computed at the actual instantaneous contact points on the rock’s surface rather than at its center of mass, allowing the software to capture the eccentric, chaotic behavior of real boulders.

The contrast between the two models was stark. In the lumped-mass simulations, most particles stopped close to the slope toe, and not a single one reached the edge of the model where the endangered houses stand. The rigid-body model told a very different story: most blocks crossed the road, and some reached the downslope houses, successfully reproducing the rockfall paths observed in previous events. The blocks accumulated in three distinct zones along the slope, with rounded, more angular blocks traveling furthest and sharp-edged, less angular blocks coming to rest closer to the toe. Larger blocks consistently traveled farther than smaller ones, carrying greater kinetic energy throughout their descent while also possessing the mass to propel themselves onward after each impact.

Shape emerged as a surprisingly powerful variable. Spherical blocks traveled the longest distances, as expected, but were excluded from the main simulations because freshly detached blocks rarely remain perfectly round. Among the fifteen custom shapes tested across seven size groups spanning 5 to 46,000 kilograms, blocks with extruded octagonal geometry exhibited unexpected lateral spread compared with all other shapes. This finding echoes large-scale experiments by the WSL Institute for Snow and Avalanche Research and ETH Zurich, in which wheel-shaped concrete blocks diverged laterally and occupied far larger impact areas than cube-shaped blocks, which followed the steepest descent path in narrow corridors. Lateral dispersion matters enormously because it controls the width of hazard zones and determines how long a protective barrier must be.

The physics behind these differences is rooted in rotational dynamics. When an irregular rock strikes the slope, the contact point is usually offset from its center of mass, producing an impulse torque that converts translational kinetic energy into rotational energy and back again. Flat edges can trigger sudden sliding or dampening, while sharp vertices cause chaotic, high-velocity angular deviations. Size also determines the moment of inertia, the resistance of a body to changes in its rotation. Larger, more elongated blocks have higher moments of inertia, meaning that once they begin to roll or spin, they store substantial rotational energy that stabilizes their trajectory and drives longer runout distances and higher bounce paths. None of these effects can be represented in a model where the rock has no shape at all.

Armed with the rigid-body results, the study proceeded to design an actual protection system. The optimal fence location was chosen at the end of the first accumulation zone, where block energies and bounce heights are relatively low, minimizing the size of the structure needed. Barrier length was calculated from the geometry of lateral spread using a similarity ratio between the distances from the fence to the source and to the houses, yielding a design length of 20 meters that was doubled for safety. Simulations of a 1400-throw scenario then tested barriers of increasing energy capacity. Without a barrier, 595 of the 1400 blocks, or 42.5 percent, passed the proposed fence location. A 100-kilojoule barrier captured only small blocks, letting 19.6 percent through, while 500-kilojoule and 1000-kilojoule barriers still allowed a handful of large blocks to reach the model edge where the houses stand.

The decisive insight came when specific block geometries were excluded from the analysis. Large, rounded, wheel-like blocks proved to be the highest-risk shapes for bypassing barriers, driven either by excessive energy or by eccentric motion arising from their shape. When these round and large blocks were systematically removed from the simulations, both 1000-kilojoule and 2000-kilojoule barriers achieved complete capture. A sensitivity analysis across 175 to 1400 throws confirmed that barrier performance was independent of sample size. The final recommendation was clear: a 2000-kilojoule fence barrier achieves a 99 percent catching ratio under the full range of realistic block shapes and sizes, while a 1000-kilojoule barrier suffices only if large, rounded blocks are excluded from the hazard scenario.

The implications reach well beyond one Turkish village. The study acknowledges limitations, including input parameters derived from back analysis of only two observed rockfall events, and the simplification of barriers as flat, rigid, undeformable bodies when real fences are flexible structures that deform under repeated impacts. Even so, the message for engineers and heritage managers is unambiguous: nominal barrier energy ratings alone may be insufficient, and design protocols must account for block geometry and rotation beyond standard vertical drop tests. For historic sites like Belisirma, where tourism, cultural heritage, and human lives converge beneath unstable cliffs, the difference between a point mass and a real boulder is not a technical footnote. It is the difference between a barrier that looks adequate on paper and one that actually stops the rock.

Subject of Research: Three-dimensional rockfall trajectory modeling and protection barrier design for a slope above the historic Belisirma village in Cappadocia, Türkiye

Article Title: Application of 3D rockfall impact models for a slope in the historical Belisirma village (Cappadocia) considering block shape and size effects on the design of protection barriers

Article References: Sari, M. (2026). Application of 3D rockfall impact models for a slope in the historical Belisirma village (Cappadocia) considering block shape and size effects on the design of protection barriers. Discover Geoscience, 4(1), Article 324. https://doi.org/10.1007/s44288-026-00699-9

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00699-9

Keywords: rockfall, 3D modeling, rigid-body dynamics, lumped-mass model, Cappadocia, Belisirma village, Ihlara Valley, protection barriers, block shape, block size, hazard assessment, geotechnical engineering

The research team, led by Suhail Ahmad of the University of Jammu, tackled a problem that has long frustrated glaciologists. Glacial lakes sit in remote, often inaccessible terrain, and mapping them by hand across decades of satellite imagery is slow, expensive, and inconsistent. Instead, the team harnessed Google Earth Engine, a cloud-computing platform capable of processing vast archives of satellite data in parallel. They drew on Landsat 5 Thematic Mapper and Landsat 8 Operational Land Imager imagery, both at 30-meter resolution, restricting their analysis to post-monsoon images from September through November, when seasonal snow cover is minimal and lake boundaries are most clearly visible against the surrounding terrain.

At the heart of the automated workflow lies a spectral technique called the Modified Normalized Difference Water Index, or MNDWI, which contrasts the reflectance of water in green and shortwave-infrared bands. Water absorbs strongly in the shortwave infrared, so lakes stand out sharply, while the modification helps suppress confusion with terrain shadows that plague ordinary water indices in steep mountain topography. The researchers refined the approach with an additional threshold on red-band reflectance, exploiting the fact that snow reflects more than half of incoming red light while clear water reflects very little. A slope filter derived from the ASTER Global Digital Elevation Model excluded pixels steeper than 20 degrees, since lake surfaces are inherently flat, and morphological opening operations stripped out spurious noise pixels before manual refinement removed river channels and other non-glacial water features.

The validation was rigorous. For each benchmark year, the team generated 500 stratified random validation points, half lakes and half non-lakes, and interpreted each one against high-resolution imagery in Google Earth Pro. The 1990 map achieved an overall accuracy of 88.6 percent with a Kappa coefficient of 0.772, while the 2025 map improved to 92.4 percent accuracy and a Kappa of 0.848. Field observations at the Durung Drung Glacier in 2024 provided qualitative confirmation, revealing a well-developed proglacial lake in direct contact with the glacier terminus, complete with debris-covered ice, exposed ice cliffs, and supraglacial meltwater channels, all signs of active ablation in terrain ringed by unstable moraine slopes.

The headline numbers tell a story of proliferation rather than dramatic swelling. Total lake area grew only modestly, from 34.81 to 37.72 square kilometers, an average gain of about 0.083 square kilometers per year. Yet the count of lakes surged by 205.5 percent, and 520 of the lakes mapped in 2025 did not exist at all in 1990. Remarkably, every one of the 255 lakes identified in 1990 still persisted through 2025, meaning the entire increase came from new formation rather than loss. The size distribution is starkly skewed: 99.5 percent of lakes in 2025 were smaller than one square kilometer, while just four lakes exceeded that threshold yet contributed disproportionately to the total water stored on the landscape.

This pattern, the authors argue, reflects how deglaciating mountains actually evolve. As glacier snouts retreat and ice thins, overdeepenings and depressions in freshly exposed bedrock and moraine fill with meltwater, spawning swarms of small proglacial and supraglacial ponds. Because each new pond is tiny, the cumulative area gain stays modest even as the lake count explodes. Elevation analysis reinforced the topographic control on this process: the overwhelming majority of lakes cluster between 4,000 and 5,000 meters above sea level, a band where glacier retreat, surface melting, and suitable basin geometry converge. Below that zone, glaciers are scarce; above it, cold limits meltwater supply.

The proliferation matters because glacial lakes are the reservoirs behind potential disasters. When a natural dam of ice or loose moraine fails, the impounded water can surge downstream as a glacial lake outburst flood, or GLOF, destroying roads, bridges, hydropower infrastructure, and villages in its path. The Hindu Kush-Karakoram region has a long history of such events, with compiled inventories documenting hundreds of occurrences across High Mountain Asia, concentrated heavily in the Karakoram, and Pakistan recording more GLOF events than any other nation. Many lakes are known to expand rapidly in the period preceding an outburst, making expansion rate itself a diagnostic signal.

To screen the region’s lakes systematically, the team built a hazard framework combining the Analytic Hierarchy Process with a Potential Risk Score Index. Four indicators were weighted by expert judgment grounded in documented GLOF mechanisms: lake area, expansion rate, distance to the nearest active glacier terminus, and topographic potential, represented by the extent of slopes steeper than 30 degrees within each lake’s contributing watershed, terrain capable of generating ice and rock avalanches. Avalanche susceptibility received the highest weight because mass movements plunging into lakes are the principal trigger of past outbursts. Applying the framework to all 775 lakes, the results were broadly reassuring: about 95 percent fell into the very low, low, or moderate hazard classes. But 36 lakes scored high and four scored very high, together just over 5 percent of the inventory, sharing the dangerous combination of rapid expansion, close glacier contact, large upstream catchments, and steep surrounding walls.

The study’s authors are careful about attribution. The Western Himalaya has warmed by roughly 0.9 degrees Celsius since the early twentieth century, with most of the increase after the 1970s, and warming rates within the Upper Indus Basin range from 0.07 to 0.55 degrees per decade, with projections suggesting median increases of about 3.3 degrees Celsius by century’s end under a moderate emissions scenario. Yet the team explicitly notes that it did not statistically test the relationship between lake evolution and temperature, precipitation, or glacier mass balance, so warming should be viewed as a plausible contributing driver interacting with local geomorphology rather than a proven sole cause. Gridded climate data for the Zanskar region confirm the strongly continental, cold-arid regime, with melt concentrated in the July-August ablation season.

The researchers also acknowledge the limits of their screening tool. Thirty-meter Landsat pixels blur the boundaries of very small lakes, lake volume and depth remain unknown without bathymetry, and moraine-dam composition and stability cannot be assessed from orbit. The Potential Risk Score Index is therefore a first-order prioritization, not a prediction of flood probability or magnitude. Future work, the authors suggest, should integrate higher-resolution imagery from Sentinel-2 and PlanetScope, surface elevation measurements from ICESat-2 and SWOT, field-based bathymetry, and hydraulic flood modeling to convert hazard rankings into concrete estimates of inundation extent and downstream risk. For now, the inventory establishes a critical baseline: forty lakes in the Upper Indus Sub-Basin now demand priority monitoring, and the number of such lakes will likely only grow as the region’s glaciers continue their retreat.

Subject of Research: Multi-decadal automated mapping of glacial lakes and outburst flood hazard assessment in the Upper Indus Sub-Basin, Western Himalaya

Article Title: Automated mapping and multi-decadal changes in glacial lakes in the Upper Indus Sub-Basin, Western Himalaya, using Google Earth Engine

Article References: Ahmad, S., Jasrotia, A. S., Singh, K. K., Sharma, A., & Ridwan, Q. (2026). Automated mapping and multi-decadal changes in glacial lakes in the Upper Indus Sub-Basin, Western Himalaya, using Google Earth Engine. Discover Geoscience, 4(1), Article 392. https://doi.org/10.1007/s44288-026-00761-6

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00761-6

Keywords: glacial lakes, Upper Indus Basin, Western Himalaya, Google Earth Engine, GLOF, remote sensing, Landsat, MNDWI, glacier retreat, climate change, hazard assessment, cryosphere