The core idea behind downscaling is deceptively simple. Global climate models divide the planet into grid cells that are typically one hundred kilometres or more across, a resolution at which coastlines, mountain ranges, and individual thunderstorms are essentially invisible. Regional climate models take the large-scale output from these global models and re-simulate it over a smaller domain at much finer resolution, adding back the topographic and coastal complexity that shapes local weather. In Australasia, where the Great Dividing Range, the complex coastline of Bass Strait, and the steep rainfall gradients of the tropical north all exert powerful influences on climate, this added detail is not a luxury but a necessity for credible projections.

The CORDEX-Australasia programme has now completed successive ensembles based on the fifth and sixth phases of the Coupled Model Intercomparison Project, known respectively as CORDEX-CMIP5 and CORDEX-CMIP6. These coordinated collections of regional simulations, produced by modelling groups across Australia and beyond, allow scientists to separate the robust signals of climate change from the noise of model-to-model differences. The synthesis of this fifteen-year effort, published in PLOS Climate by Jason Evans of the University of New South Wales and a large team of collaborators from institutions across the region, documents both the scientific achievements and the formidable challenges that remain.

One of the clearest findings to emerge from the programme is that regional models demonstrate genuine added value over the global models that drive them. This is not merely a matter of sharper pictures. The high-resolution simulations represent temperature extremes, heavy precipitation events, coastal processes, and complex terrain with substantially greater fidelity, which matters because these are precisely the phenomena that cause the most damage to communities and infrastructure. When a regional model can resolve the sea breeze that moderates a coastal heatwave, or the orographic uplift that triggers a flash flood on a mountain slope, its projections of future extremes become far more useful to planners and emergency managers.

The CMIP6-based projections now available for Australasia show robust agreement on one central fact: substantial warming through the twenty-first century is essentially locked in across the region. The magnitude of that warming, however, depends overwhelmingly on which emissions pathway the world follows, making human choices the dominant source of uncertainty in temperature projections. Precipitation is a different story altogether. Rainfall responses vary considerably from one subregion to another, with some areas projected to become drier and others facing changes that are harder to pin down, reflecting the intricate interplay of atmospheric circulation, ocean temperatures, and local geography that governs water availability across the continent and its island neighbours.

Beyond the headline projections, the past decade and a half has seen significant methodological advances that have quietly revolutionised the field. Convection-permitting modelling, which runs regional models at kilometre-scale resolutions fine enough to simulate thunderstorms directly rather than approximating them with statistical formulas, has moved from research curiosity to operational tool. Benchmarking frameworks now allow systematic evaluation of model performance against observations, while improved bias-correction methods help translate raw model output into information that hydrologists, engineers, and farmers can actually use. Together these advances have sharpened both the physical realism of the simulations and the credibility of their evaluation.

The practical payoff of this scientific investment is now visible across nearly every sector of Australian and Pacific society. Regional climate projections underpin decision-making in water resource management, where reservoir operators need to know how inflows will shift in the decades ahead. Agricultural planners use the projections to assess how growing seasons and crop water requirements will change. Disaster risk reduction agencies, energy system planners, public health authorities, and national climate services all draw on the same coordinated datasets, meaning that a single scientific infrastructure quietly supports adaptation decisions worth billions of dollars across the region.

None of this would have been possible without sustained investment in shared computing infrastructure. The National Computational Infrastructure in Australia has provided the enormous processing power and data storage that regional climate modelling demands, and coordination has been strengthened further through the National Partnership for Climate Projections, which links state and federal programmes into a coherent whole. This model of collaboration, in which different levels of government and multiple research institutions contribute to a common international framework, has become a quiet success story in how complex scientific infrastructure can be built and maintained over decades.

Significant challenges nonetheless remain on the horizon. Regional climate modelling is computationally voracious, and the demand for ever-finer resolution continually pushes against available supercomputing capacity. Persistent model biases, particularly in simulating precipitation, continue to limit confidence in some projections. Designing ensembles that efficiently sample the range of possible futures without duplicating effort is an ongoing scientific puzzle, and perhaps the most difficult challenge of all is communicative rather than computational: translating sprawling multi-model datasets into decision-ready guidance that a council engineer or a hospital administrator can act upon without needing a doctorate in climate science.

The strategic priorities identified for the coming years point toward an ambitious future. Kilometre-scale convection-permitting modelling is expected to become the new standard, capturing storm-scale processes that current models still approximate. Enhanced coupling between regional models and Earth system components such as oceans, vegetation, and carbon cycles promises more complete simulations. Coordinated super-ensembles combining many models and scenarios, the responsible integration of artificial intelligence techniques to accelerate both simulation and analysis, stronger observational constraints to narrow uncertainty, and expanded support for vulnerable remote Australian islands and Pacific Island nations all feature prominently in the roadmap. For the low-lying atoll nations of the Pacific, whose very existence is threatened by rising seas and intensifying storms, the stakes of this work could hardly be higher.

The fifteen-year story of CORDEX-Australasia is ultimately a story about turning planetary-scale physics into locally meaningful knowledge. As the synthesis makes clear, delivering robust, high-resolution climate information for adaptation and resilience across Australasia will require sustained investment, continued collaboration, and relentless innovation. But the foundation has been laid, the datasets exist, the methods have matured, and the region now possesses a scientific capability that its founders could scarcely have imagined when the first coordinated downscaling experiments began. The task ahead is to ensure that this capability keeps pace with a changing climate and the growing urgency of the decisions it must inform.

Subject of Research: Regional climate downscaling and high-resolution climate projections for Australasia under the CORDEX programme

Article Title: Fifteen years of regional climate downscaling in CORDEX-Australasia

Article References: Fifteen years of regional climate downscaling in CORDEX-Australasia. (n.d.). https://doi.org/10.1371/journal.pclm.0001066

Image Credits: AI Generated

DOI: 10.1371/journal.pclm.0001066

Keywords: CORDEX-Australasia, regional climate modelling, climate downscaling, CMIP6, convection-permitting models, climate extremes, precipitation projections, Australasia, Pacific Islands, climate adaptation, bias correction, climate services

The core problem the team set out to solve is known as downscaling. Physics-based dynamical downscaling, in which regional climate models are nested inside global models, captures the rich spatiotemporal correlations of real weather, but running enough simulations to sample rare extremes is computationally brutal. Statistical downscaling methods are far cheaper, but they often fail to capture the multivariate dependencies that drive compound events, such as the deadly combination of heat and humidity, or wildfires fed simultaneously by dry vegetation and strong winds. Machine learning approaches developed in recent years promised a middle path, but most of them share a crippling assumption: they require temporally aligned pairs of low-resolution and high-resolution data for training. That alignment exists when you compare a weather forecast against observations at the same instant, but it is fundamentally impossible for free-running climate projections, where the chaotic climate system means no simulated day will ever correspond to a specific observed day.

GenFocal sidesteps this obstacle with a two-stage generative architecture that learns statistical correspondences between distributions rather than between individual samples. In the first stage, a bias-correction operator built on a technique called rectified flow maps the distribution of coarse, biased climate model output onto the distribution of coarse-grained real weather. Rectified flows work by learning a neural-network vector field that transports one probability distribution into another along an ordinary differential equation, which the researchers solve with a fourth-order Runge-Kutta solver. Because the method matches distributions rather than individual events, it never needs to know which simulated day corresponds to which real day. The second stage is a conditional diffusion model, the same family of generative techniques behind modern image and video synthesis, which performs super-resolution: it learns to statistically invert the coarse-graining operation and add plausible fine-grained spatiotemporal detail, raising the resolution from daily data at 1.5 degrees to 2-hourly data at 0.25 degrees across a localized region.

The training setup reflects the intended use case. The team trained on twenty years of data from 1980 to 1999, using the ERA5 reanalysis as the high-resolution target and the Community Earth System Model Version 2 Large Ensemble as the coarse source. Hyperparameters were tuned on the decade from 2000 to 2009, and the final evaluation covered 2010 to 2019, with the full large ensemble downscaled. This chronological split mirrors how the framework would be applied in practice, grounding near-term risk assessments in the statistical character of past observations. The model ingests ten daily averaged variables and outputs four variables sampled every two hours, and a domain decomposition procedure at inference time stitches together shorter generated segments into long, temporally coherent sequences spanning months.

Perhaps the most striking demonstration involves tropical cyclones, the storms responsible for thousands of deaths and tens of billions of dollars in damage every year. Coarse climate models poorly resolve the fine-scale processes that drive cyclone genesis and intensification, and the storms in the raw CESM2 Large Ensemble output show both lower frequency and excessively long durations compared with reality. Remarkably, GenFocal was never specifically trained to produce hurricanes; the storms emerged on their own from the learned statistics. When the researchers downscaled projections for the North Atlantic basin over the test decade, the generated cyclones reproduced the statistics of the ERA5 reanalysis in track density, cyclogenesis locations, landfall counts, storm frequency, intensity distributions on the Saffir-Simpson scale, and even track morphology. The framework accomplished this even when the input projections contained little recognizable storm structure, generating plausible cyclones purely from the large-scale environmental conditions.

The framework also proved its worth on compound heat extremes, the kind of interacting hazards that conventional downscaling systematically underestimates. Humid heatwaves, combining high temperature with high humidity, strain human health and power grids alike, and their severity depends on correlations between variables, across space, and over time. Evaluating summer heat index extremes across the contiguous United States, GenFocal reduced the average bias in the 99th percentile of the heat index by more than 35 percent relative to two established statistical baselines: BCSD, a method routinely used for downscaling ensembles from the Coupled Model Intercomparison Project, and STAR-ESDM, the state-of-the-art method recommended for the US Fifth National Climate Assessment. For the tail dependence between temperature and humidity extremes, the joint behavior that matters most for compound risk, GenFocal cut average errors by 44 percent compared with those baselines, with notable gains across the Midwest and the American West.

The model’s grasp of spatial structure proved equally important. In California, where diverse topography produces sharply contrasting microclimates, GenFocal captured the summertime decorrelation of the heat index between San Francisco and inland areas, a pattern driven by the coastal cooling effect of sea breezes that intensifies as inland temperatures climb. It also reproduced the complex spatial correlations of wind speed shaped by terrain. BCSD and STAR-ESDM, which do not model spatial correlations explicitly, failed to recover this rich structure from the coarse simulations. Duration matters as much as intensity, since heat-related mortality rises with heatwave length. When the team assessed five-day streaks of daily maximum heat indices exceeding 305 kelvin, the threshold for the National Oceanic and Atmospheric Administration’s extreme caution advisory, GenFocal produced largely unbiased estimates along the East Coast and in the Midwest, where the statistical baselines tended to overestimate risk. Across these duration metrics, GenFocal reduced average bias by 44 percent against BCSD and 57 percent against STAR-ESDM.

Crucially, the framework appears to preserve climate change signals rather than distorting them, a known weakness of statistical methods trained to correct biases over a historical reference period. Comparing projected changes in the top decile of daily maximum temperature across western US cities between 2020 and 2080, GenFocal produced regional warming trends consistent with physics-based dynamical downscaling from the Weather Research and Forecasting model: relatively weak warming in coastal San Diego but much stronger warming in inland cities such as Albuquerque, Phoenix, and Portland. The statistical baselines, by contrast, predicted quasi-uniform warming that ignored this modulation by regional processes. For tropical cyclones, the team generated the equivalent of 8,000 August-to-October seasons for each of two periods, 2010 to 2019 and 2050 to 2059, by downscaling trajectories from the large ensemble with eight samples per trajectory. The projections showed an increase in tropical storm and hurricane landfalls along the US East Coast, aligning with forecasts from other downscaled projections, alongside subtropical intensification and tropical weakening consistent with the observed poleward migration of storms at maximum intensity, with the strongest projected intensification over the Carolinas and the Mid-Atlantic.

The practical implications reach well beyond atmospheric science. Accurate spatial correlation modeling can improve energy grid planning by quantifying the risk of concurrent heat extremes that simultaneously raise demand and threaten transmission lines. Capturing temperature-humidity interdependencies feeds directly into heat index predictions relevant to public health, food production, and disaster preparedness. And because GenFocal delivers a full probabilistic characterization rather than a single deterministic answer, it enables risk assessment for compound hazards involving any number of interacting extremes. By making it feasible to downscale large ensembles of climate projections, a task computationally out of reach for physics-based approaches, the framework arrives at a moment when AI-accelerated climate simulation and ever-larger model ensembles are expanding the raw material available for regional risk analysis. Pretrained model weights, training data, and source code have been released openly, inviting the broader community to stress-test and extend what may become a new standard tool for turning global climate science into local, actionable intelligence.

Subject of Research: Probabilistic machine learning for regional climate risk assessment by downscaling global climate model projections

Article Title: Regional climate risk assessment from climate models using probabilistic machine learning

Article References: Wan, Z. Y., Lopez-Gomez, I., Carver, R., Schneider, T., Anderson, J., Sha, F., & Zepeda-Núñez, L. (2026). Regional climate risk assessment from climate models using probabilistic machine learning. Nature Machine Intelligence. https://doi.org/10.1038/s42256-026-01308-7

Image Credits: AI Generated

DOI: 10.1038/s42256-026-01308-7

Keywords: climate downscaling, generative AI, diffusion models, tropical cyclones, heatwaves, compound events, climate risk, machine learning, climate models, extreme weather, probabilistic forecasting, climate adaptation

The research, led by Fatemeh Salehipour Bavarsad of the Czech Technical University in Prague together with colleagues at Shahrood University of Technology and Osaka Metropolitan University, tackles this blind spot head-on. The team developed high-resolution, site-specific urban weather datasets that explicitly account for the urban heat island effect, the well-documented phenomenon in which cities run several degrees warmer than their rural surroundings because asphalt, concrete and building mass absorb and re-emit heat while vegetation and moisture are scarce. Rather than treating the city as a passive backdrop, the researchers coupled statistically downscaled future climate projections with a physics-based urban canopy model known as the Urban Weather Generator, or UWG, to produce multiple urbanized future weather files tailored to a specific urban site.

The Urban Weather Generator works by taking an upstream rural weather signal and transforming it through a physically grounded representation of the urban fabric. It accounts for how much of the site is covered by buildings, how tall those buildings are, how closely they are packed, the thermal properties of their walls and roofs, and the heat released by human activity and traffic. As solar radiation heats horizontal surfaces during the day and those surfaces release stored energy at night, the model computes the resulting air temperature differences between the rural reference and the urban site, hour by hour. This means the output is not a generic city correction but a weather file that reflects the actual three-dimensional geometry and material composition of the neighborhood surrounding a given building.

To disentangle the many forces at play, the team designed a structured three-scenario framework. The first scenario isolates regional climate change, asking how a future atmosphere alone alters building behavior. The second adds the urban heat island effect on top of that changed climate, capturing the extra thermal burden that urbanization imposes. The third goes further and incorporates the surrounding urban morphology, meaning the actual arrangement of neighboring buildings and their radiative and shading interactions with the building under study. By comparing results across these scenarios, the researchers could attribute changes in building performance to each driver separately, and, crucially, observe how the drivers interact with one another in ways that simple additive assumptions miss.

The test case is a representative mid-rise office building in Osaka, Japan, a metropolis that combines a hot, humid summer climate with intense urban density, making it an ideal laboratory for studying the collision between global warming and urban heat. The team simulated the building’s energy use and indoor thermal conditions under conventional weather files and under their new urbanized future weather files, tracking metrics that include cooling energy demand, indoor overheating degree, indoor overcooling degree, indoor discomfort degree and thermal autonomy, a measure of how much of the year a building maintains comfort without active systems. The contrast between the datasets was stark and, in places, counterintuitive.

The headline finding is that conventional weather datasets systematically underestimate indoor thermal discomfort in dense urban settings. Because the reference data comes from cooler rural or airport locations, simulations driven by such files paint an optimistic picture of future comfort, understating both the frequency and the intensity of overheating episodes. For a building designer, that optimism is dangerous. It can lead to undersized cooling systems, insufficient attention to shading and facade design, and a false sense of thermal resilience that evaporates during the first serious heat wave. As climate change pushes regional temperatures upward, the penalty for ignoring the urban heat island grows larger, because the extra urban warming compounds the baseline shift rather than merely adding to it.

Yet the study also delivers a more nuanced and arguably more surprising message: urban form is not simply an enemy of thermal comfort. When the researchers included the surrounding three-dimensional urban morphology in their weather generation, they found compensating effects. Neighboring buildings shade one another from direct solar radiation, and surfaces exchange longwave radiation in ways that can moderate conditions at the facade. A building in a tight urban canyon may suffer from the heat island’s warmer air, but it also receives less sun on its walls and windows than an isolated building in an open field. These interactions are nonlinear, meaning their combined effect cannot be predicted by summing individual contributions, and they can materially alter both energy use and indoor thermal conditions depending on the specific geometry of the site.

This nonlinearity is precisely why the researchers argue that urban microclimate must be treated as an integral part of building performance simulation rather than an optional refinement. The interplay between regional warming, urban heat island intensity and urban geometry produces outcomes that no single-factor analysis can anticipate. A design decision that looks prudent under a rural weather file, such as a generous window-to-wall ratio to capture daylight, may perform very differently when the same glazing faces a heat-soaked urban canyon. Conversely, dense urban form that raises local air temperatures may simultaneously deliver shading benefits that reduce cooling loads. Capturing these trade-offs requires weather data that carries the fingerprint of the actual urban environment, which is exactly what the new framework produces.

The practical implications extend well beyond Osaka. Building performance simulation underpins energy codes, green building certification, HVAC system sizing and climate adaptation planning worldwide, and all of these applications currently inherit the rural bias of standard weather files. The framework proposed by the Bavarsad team offers a workflow that practitioners can follow: take future climate projections, downscale them statistically, pass them through an urban canopy model parameterized with local morphology, and feed the resulting urbanized weather files into standard building simulation tools. The approach is deliberately compatible with existing simulation practice, using formats and tools already familiar to the industry, which lowers the barrier to adoption. It also supports climate-responsive urban design, because by varying morphological parameters, planners can test how different configurations of street width, building height and density shape both neighborhood climate and building performance before anything is built.

The broader stakes are considerable. Heat is already the deadliest weather-related hazard in many regions, and the buildings people occupy are the first line of defense between urban populations and dangerous temperatures. As the Intergovernmental Panel on Climate Change has emphasized, even modest levels of global warming translate into substantial increases in extreme heat exposure, and cities concentrate that exposure. Studies of temperature-related mortality under Paris Agreement scenarios have shown that the difference between warming targets is measured in lives, and indoor conditions during heat waves are a critical part of that equation. A simulation methodology that understates urban heat therefore does not just mispredict energy bills; it risks misinforming the very decisions that determine whether buildings shelter people or endanger them during the extreme events of the coming decades.

What makes this study resonate beyond its technical contribution is the clarity of its central lesson: the city itself is a climate machine, and pretending otherwise corrupts every downstream calculation. The researchers demonstrate that accurate representation of urban microclimate, particularly the combined effects of the urban heat island and urban geometry, is essential for reliable assessment of building performance and thermal resilience under future climate conditions. Their three-scenario framework gives the field a reusable template for separating and recombining the effects of global climate change, urban warming and urban form, and their Osaka case study shows concretely how much conventional methods miss. As cities race to retrofit existing stock and design new districts capable of withstanding a hotter century, tools like the Urban Weather Generator, coupled with downscaled climate futures, offer a way to stop designing for a countryside that the building will never experience, and start designing for the city as it truly is, and as it is becoming.

Subject of Research: Integrating urban heat island effects and urban morphology into future weather files for building performance simulation

Article Title: Integrating urban heat island effects into future weather data for building performance simulation

Article References: Bavarsad, F. S., Mohajerani, M., Tywoniak, J., & Yuan, J. (2026). Integrating urban heat island effects into future weather data for building performance simulation. Theoretical and Applied Climatology, 157(10), Article 630. https://doi.org/10.1007/s00704-026-06571-7

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06571-7

Keywords: urban heat island, building performance simulation, Typical Meteorological Year, Urban Weather Generator, climate downscaling, urban morphology, overheating risk, thermal comfort, Osaka, future climate scenarios, energy demand, urban microclimate