The starting point is COSMO-REA6, a regional reanalysis for the European CORDEX domain developed by the Hans-Ertel-Centre for Weather Research together with the universities of Bonn and Cologne and the German Meteorological Service. Reanalyses blend numerical weather models with historical observations to produce a physically consistent record of past atmospheric conditions, and COSMO-REA6 does so on a grid of roughly six kilometers with forty vertical levels. But even at this resolution, the model’s deterministic gust diagnostic struggles with the chaotic, localized nature of wind gusts, which are produced by turbulent eddies deflecting fast-moving air parcels down to the surface or by convective downdrafts spreading horizontally when they strike the ground. These mechanisms, first formalized in diagnostics by researchers such as Brasseur and Nakamura, are deterministic by design and carry no measure of uncertainty, which is precisely what forecasters and emergency managers need most.

Ertz and Friederichs attack this problem with the tools of extreme value statistics, the branch of mathematics built specifically for rare, high-impact events. Because a gust measurement represents the highest three-second average wind speed within an hour, the theoretically appropriate model for daily maxima is the generalized extreme value distribution. After fitting single-station models and finding that the shape parameter showed no clear signal or leaned slightly negative, the authors deliberately fixed it at zero, yielding a Gumbel distribution. This choice avoids imposing an artificial upper bound on gust speeds, a property of the Weibull-type distribution that would be physically implausible for storm-driven extremes, while keeping parameter estimates stable.

The real innovation lies in how the model handles space. At the top level of the hierarchy, gust observations follow a Gumbel distribution whose location and scale parameters vary in both space and time. These parameters are expressed as linear regressions on predictor variables drawn from COSMO-REA6, chiefly the maximum ten-meter wind speed and the mean horizontal wind, plus a covariate capturing the difference between the station altitude and the model’s representation of that altitude. Crucially, the regression coefficients themselves are not treated as fixed numbers. Instead, each coefficient is a realization of a two-dimensional Gaussian random field with a constant mean and an isotropic Matérn covariance function, meaning that stations close to one another tend to share similar gust behavior. This latent-process structure allows information to be pooled across the entire network of 109 synoptic stations, stabilizing estimates at data-sparse locations and enabling kriging-based interpolation to any unobserved point in the domain.

Perhaps the most elegant trick in the paper is a correction to the distance metric itself. Mountain stations such as the Zugspitze or the Brocken can sit within a few kilometers of valley towns like Braunlage, yet their gust climates are worlds apart. Under a purely geographic distance measure, the model would wrongly treat these stations as near neighbors, contaminating the estimated spatial dependency structure for the whole region. The authors solve this by adding a scaled elevation difference to the great-circle distance between stations, effectively pushing mountain and valley locations apart in the covariance calculation. The scaling factor is treated as an unknown parameter with a prior centered on the ratio of buoyancy to Coriolis forces, a value of roughly one hundred derived from quasigeostrophic scale analysis. The result is that mountaintop observations can remain in the training data without distorting the picture elsewhere, and the estimated correlation lengths become larger and more stable.

Training such a model is no small computational feat. The authors used the Stan probabilistic programming language via its Python interface, running the No-U-Turn Sampler, an adaptive form of Hamiltonian Monte Carlo, for chains of 1500 iterations with 500 discarded as burn-in. The full spatial model required about an hour of sampling time, and the subsequent interpolation of parameter fields across a 121-by-161 grid cell domain over Germany took roughly three hours on an ordinary desktop computer, using an iterative scheme that simulates horizontal slices of the country one after another, each conditioned on its southern neighbor. The authors emphasize that the entire pipeline is reproducible on standard hardware and that the code and preprocessed data are openly available.

To find out whether all this statistical machinery actually pays off, the team subjected the model to a demanding verification regime. Data from odd-numbered years between 2001 and 2018 trained the model, while even-numbered years served for evaluation, and a leave-one-out cross-validation ensured that the station being predicted was always excluded from training. Performance was judged with proper scoring rules: the Brier score for probabilities of exceeding warning thresholds of 14, 18, 25, 29 and 33 meters per second, and the quantile score for predictive quantiles ranging from the 75th to the 99.9th percentile. Both scores were decomposed into miscalibration, resolution and uncertainty components using modern isotonic-regression-based estimators, allowing the authors to see not just whether the model was skillful but why.

The verdict is a qualified but meaningful success for the spatial approach. Against a spatially constant baseline model, which itself already achieved skill scores of 10 to 45 percent relative to local climatology, the spatial Bayesian model delivered median improvements of 2 to 5 percent for prediction quantiles, with the largest gains at the extreme upper quantiles that matter most for warnings. When the evaluation was restricted to the top quartile of forecast gust events, the spatial model’s advantage grew even more pronounced, and for the 99.9th percentile it nearly matched the skill of a model trained locally at each station, the theoretical ceiling of the approach. The score decomposition revealed that the gains came primarily from better resolution rather than better calibration, meaning the spatial model discriminates more effectively between windy and calm situations. For threshold exceedance probabilities the improvement was marginal, though the spatial model showed a slight edge at the highest 33 meters per second threshold.

The study is candid about its limits. At sheltered inland locations with weak mean winds, exemplified by Garmisch-Partenkirchen, even the locally trained model has no skill against climatology, because the reanalysis predictors simply lack explanatory power in complex, low-wind terrain. The spatial interpolation can also introduce bias where a station’s gust character differs sharply from its surroundings. Yet the overall message is compelling: by letting the data itself determine how gust behavior varies across the landscape, and by treating mountains as the statistical islands they truly are, the Bonn team has produced a proof of concept that spatially aware extreme value modeling can extract more reliable gust information from reanalysis archives. With extensions already envisioned for ensemble forecasts, diurnal and annual cycles, and land-sea contrasts, the framework could soon help weather services issue wind warnings that are not only more accurate but accompanied by the probabilities that decision-makers increasingly demand.

Subject of Research: Spatial Bayesian hierarchical extreme value post-processing of wind gusts from the COSMO-REA6 reanalysis over Germany

Article Title: Post-processing of wind gusts from COSMO-REA6 with a spatial Bayesian hierarchical extreme value model

Article References: Ertz, P., & Friederichs, P. (2025). Post-processing of wind gusts from COSMO-REA6 with a spatial Bayesian hierarchical extreme value model. Advances in Statistical Climatology, Meteorology and Oceanography, 11(2), 229-256. https://doi.org/10.5194/ascmo-11-229-2025

Image Credits: AI Generated

DOI: 10.5194/ascmo-11-229-2025

Keywords: wind gusts, extreme value statistics, Bayesian hierarchical model, COSMO-REA6, reanalysis, post-processing, spatial statistics, Gaussian random fields, Gumbel distribution, forecast verification, Germany, meteorology

The numbers behind the finding are remarkable. Microplastic concentrations in Plum Run ranged from 1.2 to 23.0 particles per liter, averaging 7.6 particles per liter with a standard deviation of 4.9, meaning the stream’s plastic burden swung wildly from one sampling trip to the next. When the researchers split the year into warm and cool seasons, the pattern was unmistakable: mean concentrations were approximately 4.7 particles per liter between May and September, but 10.6 particles per liter from November through March, roughly a twofold difference. A statistical comparison using Welch’s unequal-variances t test confirmed that winter concentrations were significantly higher than summer values, with a difference of 6.8 particles per liter and a 95 percent confidence interval of 3.1 to 10.5.

What makes this study unusual is not just the seasonal signal but the statistical machinery used to detect it. Rather than relying on a single regression approach, the team built three separate multiple linear regression models: a stepwise selection model guided by the Akaike information criterion, a LASSO regression that shrinks weak predictors to zero, and an elastic net model that balances variable selection with the retention of correlated predictors. All three approaches converged on the same answer. Higher wind gusts were positively associated with microplastic counts, with a coefficient of 0.14 and a p-value of 0.037, while higher temperatures were negatively associated, with a coefficient of -0.34 and a p-value below 0.0002. The stepwise model alone explained roughly 60 percent of the variation in concentrations, with an adjusted R-squared of 0.60.

To guard against overfitting in a dataset of only 49 samples, the researchers subjected their models to repeated tenfold cross-validation, repeating the process 100 times so that every subset of the data served as both training and validation material. The results were strikingly consistent. The stepwise model achieved a cross-validated R-squared of 0.75 with a root mean squared error of 3.1 particles per liter, while the LASSO and elastic net models both reached cross-validated R-squared values of 0.79 with root mean squared errors of 2.9 particles per liter. Taken together, the models explained approximately 80 percent of the variability in microplastic concentrations, an unusually high figure for environmental field data and a strong indication that temperature and wind gusts are the dominant environmental controls on plastic abundance in this stream.

The physical story the researchers propose is one of wind-driven atmospheric delivery. Fibrous microplastics, which made up 96 percent of the particles identified in Plum Run, have high aspect ratios and low effective settling velocities, allowing them to remain suspended in air far longer than fragments or films of equivalent mass. Laboratory and field studies have shown that small fibrous particles are more readily entrained into the atmosphere as wind speeds increase, and short-lived gusts may exceed the threshold velocities needed to lift fibers that stay put under average conditions. Once aloft, these particles can travel considerable distances before turbulence subsides and they settle back to Earth, potentially landing directly on the stream surface or on the riparian corridor beside it. The mean aspect ratio of fibers measured in this study was 14, squarely within the range that model simulations identify as having enhanced atmospheric transport potential.

Two mechanisms may explain why cold weather amplifies the effect. First, vegetation acts as a seasonal filter: deciduous trees trap airborne particulates on their leaves, and the epidermal waxes that enable this trapping peak during spring and summer before declining sharply in autumn and winter. Forested zones in metropolitan areas are estimated to capture roughly 2.2 billion airborne microplastic particles per year, so when leaves fall, the pathway from air to stream becomes far less obstructed. Second, and perhaps more provocatively, cold weather changes human behavior. People wear heavier, more densely woven synthetic garments in winter, wash them more often, and run their tumble dryers more frequently, and mechanical agitation of synthetic textiles in dryers is a known source of airborne microfibers. A single dryer cycle can release more than 500,000 synthetic microfibers into the environment.

The polymer chemistry points in the same direction. Using scanning electron microscopy with energy-dispersive X-ray spectroscopy, the team screened 75 individual particles and found that 93.3 percent were polyester, with the remainder identified as polyvinyl chloride and polyolefins. Because EDS cannot definitively distinguish among carbon-rich polymers, the researchers confirmed a subset of 30 particles with Fourier transform infrared spectroscopy coupled to a diamond attenuated total reflection crystal. Every fiber that had shown oxygen peaks in its EDS spectrum, 28 particles in all, was confirmed as polyester by FTIR, while the two remaining carbon-dominated particles proved to be polypropylene. Polyester fibers are the signature morphotype of synthetic textiles, and with residential land use covering roughly half of the Plum Run watershed, the authors argue that laundering and drying of synthetic clothing is the most plausible dominant source.

To test whether wind-driven delivery is physically plausible rather than merely statistically suggestive, the team regressed wind gust data against daily plastic flux in the stream, obtaining a relationship with an R-squared of 0.42. The wind-sensitive component predicted by this regression averaged 1.28 times ten to the seventh particles per day, closely matching the mean observed stream flux of 1.38 times ten to the seventh particles per day. Scaling this wind-associated export over a 50-meter riparian buffer along the stream’s 19.3-kilometer length, and assuming a conservative one percent same-day delivery ratio, yields an implied atmospheric deposition rate of 663 particles per square meter per day. A sensitivity analysis spanning riparian widths of 10 to 150 meters and delivery ratios of 1 to 10 percent produced deposition rates from 22 to 3,161 particles per square meter per day, all of which fall within the range of atmospheric microplastic deposition previously measured in urban, suburban, and rural settings worldwide.

The authors are careful to frame these results as an indirect inference rather than a proven mechanism. The wind connection rests on statistical association, not on direct measurements of atmospheric deposition, and polymer identification by FTIR was limited to particles larger than 250 micrometers, leaving the smallest fibers characterized only by EDS screening. Still, the study carries a pointed message for anyone who assumes that rain and runoff are the whole story of riverine plastic pollution. In temperate, non-monsoonal catchments like Plum Run, the atmosphere itself may be a major conveyor of microplastics, and the rhythm of pollution may follow the calendar of human comfort: when the wind picks up and the temperature drops, the fibers of our winter wardrobes may be riding the gusts straight into the water.

Subject of Research: Environmental and meteorological controls on microplastic concentrations in a freshwater stream

Article Title: Seasonality and maximum wind gusts correlate with microplastic concentrations in a freshwater stream system

Article References: Mattie, C., Ashman, I., Fork, M., & Arnold, T. E. (2026). Seasonality and maximum wind gusts correlate with microplastic concentrations in a freshwater stream system. Environmental Monitoring and Assessment, 198(11), Article 1173. https://doi.org/10.1007/s10661-026-15985-9

Image Credits: AI Generated

DOI: 10.1007/s10661-026-15985-9

Keywords: microplastics, freshwater streams, atmospheric deposition, wind gusts, polyester fibers, synthetic textiles, seasonality, regression modeling, FTIR spectroscopy, SEM-EDS, urban watershed, Pennsylvania

The team used the ERA5 reanalysis, the European Centre for Medium-Range Weather Forecasts’ fifth-generation global dataset, which provides hourly estimates of wind gusts and snowfall on a roughly 30-kilometre grid. Because reanalysis offers complete and consistent coverage, including over the Baltic Sea where direct observations are sparse or absent, it was the natural choice for a study spanning nearly seven decades. Rather than applying a single fixed threshold, the researchers defined extremes relative to the local climate, using the 95th and 98th percentiles of hourly snowfall and wind gusts at each grid point. An hour was classified as a compound event, labelled SWG, only when both variables exceeded their respective thresholds simultaneously. The stricter SWG98 definition isolates the most severe tail of these events, while SWG95 captures a broader sample relevant to operational preparedness.

Geographically, the results show a clear pattern. With the 95th percentile threshold, compound events clustered in a zone stretching from the shores of the Bothnian Bay in the northwest towards Lake Ladoga in the southeast. When the stricter 98th percentile was applied, the hot spots shifted towards the south-facing shores of the Baltic Sea and Lake Ladoga, underlining how open water and onshore flow conspire to produce the most intense simultaneous snowfall and winds. Northern Finland and the northwest-facing shore of the Bothnian Bay saw the fewest events. Winter storms typically reach Finland from the southwest, crossing the Baltic Sea, and when that sea remains ice-free the storms pick up heat and moisture that can supercharge snowfall along the coast, a mechanism that helps explain the coastal concentration.

Year-to-year variability was striking. At the four study locations, the annual number of compound event days ranged from zero to as many as 12 days per year under the SWG98 definition, and up to 19 days per year under SWG95. Crucially, the researchers found no statistically significant long-term increase or decrease in compound event frequency over the 65-year record. This stability held even though heavy snowfall events in isolation, those exceeding 10 millimetres of accumulated snow, have actually declined across most coastal areas at a rate of roughly 0.1 to 0.4 events per decade. The message is subtle but important: while substantial snowfall alone may be becoming less common in southwestern Finland, the snowstorms that combine heavy snow with strong winds have not shown a corresponding decline.

Duration proved to be one of the most operationally significant findings. Most compound events are remarkably short: roughly half lasted less than five hours, and single-hour events were the most common. Mean durations ranged from 3.4 to 4.5 hours depending on the threshold, with the stricter SWG98 definition naturally capturing shorter, more intense cores of the same storms. Yet the record contains genuine outliers. The longest SWG95 event persisted for 35 hours at Tornio Torppi, while Loviisa Hästholmen endured two events exceeding 30 hours. Events lasting at least 12 consecutive hours, which the authors flagged as long-lasting, were rare, occurring on average only 0.2 to 0.4 times per year at the 95th percentile and far less often at the 98th. For infrastructure operators, however, these rare marathons matter most, because sustained loading of structures, blocked access routes and prolonged strain on ventilation systems scale with duration, not just peak intensity.

Intensity analysis revealed a fascinating asymmetry around Finland’s coastline. The highest average wind gusts during compound events occurred over the Bothnian Sea, reaching 20 to 24 metres per second under SWG95 conditions and 21 to 26 metres per second under SWG98, while the largest average snowfall rates concentrated along the southern coast, at 0.8 to 1.0 millimetres per hour for SWG95 and 1.1 to 1.3 for SWG98. In other words, the western coast tends to take the brunt of the wind while the southern coast takes the snow. The joint distribution of hourly values showed a triangular shape: moderate wind gusts accompanied a wide range of snowfall intensities and vice versa, and the peak values of the two variables rarely coincided within the same hour. Long-duration events, notably, were not the most intense ones hour by hour; their danger lay in persistence.

Using absolute thresholds aligned with Finnish warning practice, the study quantified the truly extreme tail. Compound events with heavy snowfall rates exceeding 2.5 millimetres per hour, a level corresponding to visibilities of roughly 0.2 to 0.8 kilometres, occurred only one to three times per decade under SWG95 and less than once per decade under SWG98. Events piling up more than 20 millimetres of snowfall were rarer still. Wind gusts above the first Finnish warning level of 20 metres per second accompanied compound events 8 to 20 times per decade, but gusts beyond 25 metres per second appeared only 0.3 to 2 times per decade. Blizzard conditions, defined as gusts above 17 metres per second with at least 10 millimetres of accumulated snow, occurred three to six times per decade, most frequently at Tornio Torppi. Under the harshest blizzard criteria, gusts above 20 metres per second with more than 20 millimetres of snow, only four events were detected across all four sites in 65 years, and all lasted more than 11 hours.

Among the standout cases, one December 1966 snowstorm simultaneously struck Tornio Torppi, Eurajoki Olkiluoto and Turku Artukainen, with SWG98 conditions persisting for up to 20 hours, snowfall rates peaking at 3 millimetres per hour, gusts reaching 26 metres per second and nearly 26 millimetres of snow accumulating. The longest SWG98 event at Eurajoki Olkiluoto produced the second-largest daily snow accumulation ever recorded in Finland, 50 centimetres on 21 November 1971, attributed to sea-effect snowfall fed by moisture from the ice-free sea. Across the whole record, the highest hourly snowfall rate, 3.7 millimetres per hour, and the largest accumulation, 28.5 millimetres, occurred at Turku Artukainen, while the strongest gust, 29 metres per second, struck Loviisa Hästholmen during a 31-hour event whose moderate snowfall still accumulated nearly 23 millimetres.

The framework proved its worth in November 2024, when two powerful cyclones, Lyly and Jari, hit Finland within three weeks. Storm Lyly brought the first officially observed hurricane-force winds in Finnish marine areas, a ten-minute mean of 33.5 metres per second at the Kylmäpihlaja lighthouse, with coastal gusts to 39.6 metres per second, felling trees, collapsing power line poles and cutting electricity to roughly 67,000 customers. Yet because Lyly’s peak winds did not coincide with heavy snowfall, it failed the study’s compound-event definition. Storm Jari, with lower peak gusts of 34.2 metres per second, instead produced persistent snowbands over the open Bothnian Bay that swept into the coast, dropping up to 28 centimetres of snow, knocking out power to more than 70,000 households and even driving a 1.6-metre sea-level withdrawal near Kemi. Jari qualified as a genuine compound event, including an 18-hour SWG95 episode at Eurajoki Olkiluoto.

The contrast between the two storms crystallises the study’s central lesson: the severity of a winter storm is not determined by the magnitude of its individual hazards alone, but by where and when those hazards overlap. Simultaneous strong winds and heavy snowfall drive snow drifting far beyond what snowfall totals alone suggest, bury roads and ventilation intakes, fell snow-laden trees onto power lines and reduce visibility to hazardous levels. Even short compound bursts can trigger traffic accidents and overwhelm snow removal, while the rare prolonged events pose serious challenges for energy production and nuclear facility operations. With no detectable trend in compound event behaviour over 65 years, the authors argue that embedding compound-event analysis into hazard assessments and preparedness planning is essential for a warming North where, paradoxically, heavy snowfall days may persist even as ordinary snowfall declines.

Subject of Research: Compound extreme snowfall and wind gust events in Finland from 1960 to 2024

Article Title: Characteristics of extreme snowfall–wind-gust events in Finland (1960–2024): frequency, duration, and intensity

Article References: Olsson, T., Laurila, T. K., Aaltonen, A., & Jylhä, K. (2026). Characteristics of extreme snowfall–wind-gust events in Finland (1960–2024): frequency, duration, and intensity. Natural Hazards and Earth System Sciences, 26(9), 4621-4640. https://doi.org/10.5194/nhess-26-4621-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4621-2026

Keywords: compound weather events, snowfall, wind gusts, Finland, ERA5 reanalysis, winter storms, blizzards, nuclear power plants, critical infrastructure, coastal hazards, risk assessment, Baltic Sea

The study, published in Theoretical and Applied Climatology, pitted the Weather Research and Forecasting model, known universally as WRF, against its younger rival, the Model for Prediction Across Scales–Atmosphere, or MPAS-A. Both were developed at the National Center for Atmospheric Research in collaboration with the U.S. Department of Energy, but they represent fundamentally different philosophies of how to chop the atmosphere into computable pieces. WRF, the long-reigning workhorse of operational forecasting in Brazil and worldwide, divides space into a regular grid of quadrilateral cells using finite-difference mathematics. MPAS-A instead wraps the planet in an unstructured Voronoi mesh — a honeycomb of hexagonal cells solved with finite-volume methods — that can transition smoothly from coarse global coverage to razor-sharp regional detail without the awkward seams of nested grids.

That architectural difference matters because the São Paulo storm was a textbook case of scale interaction. Brazilian Navy synoptic charts from the day show a high-pressure system of 1026 hPa parked over the South Atlantic, east of the state, while a deepening low-pressure area associated with a trough near southern Brazil and Paraguay intensified rapidly. The clash between warm pre-frontal air and cold post-frontal air, supercharged by daytime heating and oceanic moisture, ignited a cluster of cumulonimbus clouds. GOES-16 satellite imagery shows isolated convective cores forming between 5:30 and 6:30 in the evening, organizing into a full-blown convective system by 8:30, and then sweeping out to sea after 9:00. A pre-frontal low-level jet, channeled by valleys and hills toward the coast, added fuel to the fire.

To see which model could capture this chaos, the team, led by Rosales Aylas Georgynio Yossimar of SENAI CIMATEC’s Supercomputing Center in Salvador, ran both models at 3-kilometer resolution over a domain covering southeastern Brazil and a generous slice of the South Atlantic. WRF used its classic two-way nested setup at 9 and 3 kilometers, while MPAS-A employed a quasi-uniform mesh at roughly 3 kilometers across the entire domain. Both simulations shared 41 vertical levels and equivalent physics suites — the tropical package in WRF and the mesoscale reference package in MPAS-A — and both were initialized and driven by the ERA5 reanalysis from the European Centre for Medium-Range Weather Forecasts. Three separate experiments, initialized at different times before the event, tested how forecast skill decayed with lead time.

The verdict on the big picture was encouraging: both models successfully reproduced the synoptic skeleton of the event, including the trough, the frontal organization, and the intensification of low-level winds. At the 850 hPa level, roughly one to two kilometers above the surface, both simulations captured wind speeds exceeding 25 meters per second — Beaufort force 10, a strong gale — sweeping over the central and northern parts of São Paulo state near Bauru before the wind corridor shifted southeastward and weakened after 8:00 in the evening. WRF painted broader wind fields, while MPAS-A drew more compact, concentrated cores of extreme wind, a signature the authors link to its finite-volume formulation on the Voronoi mesh.

Beneath that broad agreement, however, the models diverged in revealing ways. Against ERA5, MPAS-A showed stronger spatial correlation over most of the analyzed period, particularly over the ocean, suggesting its unstructured grid handles mesoscale gradients and frontal features with more numerical grace. Yet higher spatial correlation did not translate into smaller errors: during the storm’s intensification phase, MPAS-A’s mean absolute error and root mean square error actually grew, hinting that the model may have over-intensified wind maxima or displaced the storm’s core. WRF, meanwhile, showed a more heterogeneous bias pattern, overestimating winds over land while underestimating them over the southern ocean sector. The team is careful to note that these differences cannot be pinned on mesh geometry alone — dynamics, physics coupling, and land-surface representation all conspire.

One unexpected confound emerged from the ground up. Because of limitations in MPAS-A’s preprocessing tools, the two models were fed different land-use datasets: WRF received the current 2019 MapBiomas classification of Brazil, while MPAS-A relied on the older MODIS-IGBP dataset from the early 2000s. Grasslands covered about 35 percent of WRF’s domain but only 7 percent of MPAS-A’s, with corresponding shifts in savanna and cropland fractions. Since vegetation categories control roughness length, albedo, and evapotranspiration — and therefore the surface energy balance that drives boundary-layer turbulence — this discrepancy adds a layer of ambiguity to any model-versus-model comparison, and the authors flag it as a caveat that future studies must untangle.

Comparisons against Brazil’s INMET weather stations exposed the models’ Achilles’ heel: complex topography. At Campos do Jordão, perched at 1,628 meters in the Serra da Mantiqueira, WRF produced its worst errors, with root mean square errors and mean biases exceeding 6 meters per second. The reason is almost cartographic: WRF’s nearest grid point sat closer to the mountain slope, where terrain-induced flow acceleration inflated its wind estimates, while MPAS-A’s representative point fell in the calmer urban core. At Cachoeira Paulista, tucked into the Paraíba Valley at the foot of the same range, similar problems appeared. Near the storm center around Bauru, by contrast, both models performed respectably, with errors under 3 meters per second and biases below 2.

Temperature told a story of mirror-image biases. WRF tended to overestimate 2-meter air temperature, sometimes by more than 1.5 degrees Celsius, while MPAS-A generally underestimated it, especially during daytime heating. Yet both models faithfully reproduced the diurnal cycle — the pre-storm warming on October 10, the abrupt frontal cooling on the 11th, and the recovery on the 12th — though they sharpened the storm-driven temperature drop more than observations showed. The most troubling statistic came from hourly wind correlations: no model exceeded 0.78 at any station, roughly 63 percent of stations fell between 0.2 and 0.6, and some correlations were near zero or negative, exposing a fundamental struggle to capture hour-by-hour wind variability during the event.

The authors attribute these timing failures to a familiar culprit: the sparse observational network of the Southern Hemisphere, which degrades the initial conditions fed into every model, compounded by uncertainties in turbulence parameterization and the use of hourly mean winds rather than true gust measurements. Their prescription is clear — better physics, denser observations, and complementary techniques such as data assimilation and statistical post-processing. As forecasting agencies worldwide weigh a transition from WRF’s rigid squares to MPAS-A’s flexible hexagons, this storm-soaked Brazilian case study offers a sobering reminder: the grid may be getting smarter, but the atmosphere still keeps some of its wildest moments just beyond the model’s grasp.

Subject of Research: Comparative simulation of an extreme wind gust event in São Paulo using the WRF and MPAS-A atmospheric models

Article Title: Simulation of an anomalously high wind gust event in São Paulo: A comparison of the MPAS-A and WRF models

Article References: Yossimar, R. A. G., da Silveira, P. W. M., dos, S. T. S., Duarte, J. W., de, L. F. J. L., Cotta, W. A. L., Cavalcante, A. A., Rodrigues, S. A., da Silva, R. D. N., & Martins, M. D. (2026). Simulation of an anomalously high wind gust event in São Paulo: A comparison of the MPAS-A and WRF models. Theoretical and Applied Climatology, 157(10), Article 644. https://doi.org/10.1007/s00704-026-06537-9

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06537-9

Keywords: WRF, MPAS-A, wind gusts, extratropical cyclone, São Paulo, numerical weather prediction, mesoscale modeling, Voronoi mesh, ERA5 reanalysis, boundary layer, severe weather, Brazil