The significance of the work lies in what it makes possible. High-resolution air temperature data are the raw material for effective, localised heat adaptation strategies, informing everything from where to plant trees to which neighbourhoods need cooling shelters first. Yet obtaining such data at city scale has remained stubbornly difficult, because conventional weather stations are expensive to install and maintain, and their placement rarely reflects the diversity of urban surfaces, building densities, and vegetation cover that drive temperature differences block by block. The new framework, developed by Marzie Naserikia, Melissa Anne Hart, and colleagues, offers a way to leapfrog that infrastructure bottleneck using data that many cities already possess or can access cheaply.

At the heart of the approach is a convolutional neural network, a class of machine learning model originally developed for image recognition. The researchers treated the city itself as an image, feeding the model a stack of spatially aligned layers that describe Sydney in fine detail. These included satellite-derived land surface temperature, which captures how hot roofs, roads, and parks appear from orbit; datasets describing urban form and fabric, such as building heights, footprints, and materials; and standard meteorological variables that anchor the estimates in the day’s actual weather. The model learned the complex, nonlinear relationships between these predictors and the air temperature recorded at ground level, and then applied those relationships across the entire city grid.

The training data came from a combination of official weather stations and crowdsourced observations, a growing resource in urban climatology thanks to the proliferation of low-cost personal weather stations connected to networks such as those run by citizen science platforms. By blending these sources, the team assembled enough ground truth to teach the network how the city’s surface characteristics translate into near-ground air temperature at around 10:00 a.m. local time, the snapshot the study targeted. The resulting maps resolve temperature variation at a scale fine enough to distinguish one side of a street from the other, a level of detail that coarse regional models and sparse station networks cannot approach.

The accuracy figures are striking. On an independent held-out test set, data the model had never seen during training, the framework achieved a coefficient of determination of 0.97 and a root mean square error of just 0.91 degrees Celsius. In practical terms, that means the model’s estimates deviated from observed temperatures by less than a degree on average, a level of precision that rivals or exceeds what has been reported for any comparable machine learning approach to urban air temperature mapping. For city planners and public health officials, such accuracy transforms temperature maps from illustrative graphics into decision-grade evidence.

Accuracy on familiar territory, however, is only half the story. A model that performs beautifully where it was trained but fails elsewhere has limited value, particularly for cities that lack dense observation networks of their own. The researchers therefore tested how well the Sydney-trained model generalised to unseen locations across the metropolitan area, and the results were encouraging: the network maintained strong predictive accuracy, with coefficients of determination between 0.91 and 0.93 and root mean square errors between 1.27 and 1.44 degrees Celsius. The model, in other words, had learned genuinely transferable relationships between urban form, surface temperature, and air temperature, rather than memorising quirks of its training sites.

The team also stress-tested the framework’s dependence on ground stations, a critical question for any city hoping to adopt the method without first building an expensive monitoring network. When the number of air temperature stations used for training was slashed by roughly 80 percent, model performance remained stable, suggesting that the approach does not require vast quantities of local observations to deliver reliable maps. This resilience matters enormously for cities in data-sparse regions, where even a modest station network may be out of reach. The message is that a small, well-distributed set of ground measurements, combined with globally available satellite and geospatial data, may be sufficient to generate city-scale temperature maps of genuine quality.

Not every test was a clean sweep. When the researchers evaluated the model on unseen days, that is, weather conditions not represented in the training data, performance became more variable, with coefficients of determination ranging from 0.66 to 0.93 and root mean square errors between 1.52 and 2.60 degrees Celsius. This dip carries an important lesson for anyone deploying machine learning in environmental science: models are only as good as the diversity of conditions they have been shown. A network trained on mild spring days cannot be trusted to map a brutal heatwave, and the study makes clear that training datasets must deliberately span the full range of weather a city experiences, from cool, cloudy mornings to the extreme heat events where accurate temperature data matter most for protecting lives.

Perhaps the most consequential finding concerns the ingredients themselves. High-resolution, city-descriptive datasets, the kind that only well-mapped cities in wealthy countries tend to possess, proved beneficial but not essential. The researchers found that comparable accuracy could be achieved using only globally available predictors, such as satellite-derived land surface temperature and worldwide geospatial products. This single result dramatically widens the framework’s potential reach. Cities in developing regions, which often face the most severe urban heat risks and the least monitoring infrastructure, could in principle adopt the same approach without waiting for detailed local surveys of building fabric and land cover. The barrier to entry drops from years of data collection to a modest set of ground observations and freely available global datasets.

The study arrives at a moment when urban heat has become one of the most urgent climate challenges, driving excess mortality, straining energy grids, and eroding the liveability of cities worldwide. By demonstrating that a convolutional neural network can convert sparse observations and satellite data into accurate, scalable, 30-metre temperature maps, the Sydney team has provided a template that other cities can follow. The framework’s combination of high accuracy, robustness to reduced station coverage, and independence from locally specialised datasets suggests that city-scale air temperature mapping could soon become a standard tool in urban climate assessment, climate adaptation planning, and public health preparedness, extending the benefits of fine-grained heat intelligence to the cities that need it most.

Subject of Research: Machine learning-based high-resolution urban air temperature mapping in Sydney

Article Title: Machine learning-based temperature mapping in Sydney: Accuracy, transferability, and scalability

Article References: Machine learning-based temperature mapping in Sydney: Accuracy, transferability, and scalability. (n.d.). https://doi.org/10.1371/journal.pclm.0001050

Image Credits: AI Generated

DOI: 10.1371/journal.pclm.0001050

Keywords: urban heat, machine learning, air temperature, convolutional neural network, land surface temperature, Sydney, climate adaptation, crowdsourced data, transferability, public health, urban climate, Machine

Land surface models are the bridge between the atmosphere and the ground in climate simulations. They calculate how sunlight, heat, and water move between the soil, the vegetation, and the air, and they feed those exchanges upward into the atmospheric component of a climate model. Historically, the models built for global circulation ran on grids tens to hundreds of kilometers wide, where a city was too small to matter. But computing power has changed the game. Global models are now pushing toward kilometer-scale grids, and regional models are running climate-length simulations over huge domains. At those resolutions, cities are no longer invisible specks; they are resolved features whose energy and water budgets can distort temperature, rainfall, and runoff if they are represented badly.

ORCHIDEE is a heavyweight in this world. It simulates coupled water, energy, and carbon cycles and serves as the land component of the IPSL climate model. Its architecture divides each grid cell into a mosaic of plant functional types, with separate soil columns for bare ground, high vegetation, and low vegetation. Until now, however, anything urban was simply assigned to the bare-soil category. That meant no reflective rooftops, no tall buildings stirring the wind, no concrete storing the afternoon sun, and no sealed asphalt blocking rain from soaking into the earth. The model could simulate a forest or a wheat field with sophistication, but a metropolis looked like a patch of exposed dirt.

The new scheme is deliberately parsimonious. Rather than adding an elaborate three-dimensional canyon geometry with separate energy budgets for roofs, walls, and roads, the team introduced a single urban tile, or PFT, into the existing mosaic. This tile carries urban-specific values for albedo, building height, thermal conductivity, and heat capacity, all of which can be prescribed from global or local datasets. The approach follows a lesson from the Urban-PLUMBER intercomparison project, which evaluated thirty urban land surface models and found that simpler slab schemes often outperform more complex ones, provided that vegetation, water availability, and imperviousness are represented realistically. Geometry, in other words, is not everything; what the surface is made of matters just as much.

The genuine novelty lies in how the scheme handles imperviousness. Most urban models treat sealed ground as a surface phenomenon: rain hits the pavement and runs off. Lalonde and colleagues went deeper, literally. Instead of splitting the urban tile into pervious and impervious sub-tiles, they modify the saturated hydraulic conductivity of the entire soil column beneath the city. Two options were tested. The first, called Urban1, slashes conductivity to ten percent of its reference value throughout the column, representing a heavily sealed city. The second, Urban2, scales the reduction linearly with the impervious fraction, so a city that is fifty-five percent sealed sees its conductivity cut proportionally. This single parameter change ripples through the whole hydrology: it suppresses infiltration, boosts surface runoff, reduces soil water storage, and limits the water available for evaporation, all through the model’s existing Richards-equation soil physics rather than any bolted-on machinery.

The thermal side of the scheme is equally consequential. Urban materials conduct and store heat very differently from soil. In grid cells where urban cover exceeds fifty percent, the model now prescribes a thermal conductivity of 3.24 watts per meter per kelvin and a heat capacity of 1890 kilojoules per cubic meter per kelvin, values borrowed from the Noah-MP land surface model. Compare that with the median thermal conductivity of natural soils in the study region, around 1.20 watts per meter per kelvin, and the physical implication is clear: cities soak up far more energy during the day and release it slowly through the night. That stored heat is the engine of the urban heat island, and the model now captures it.

To test the scheme, the team ran standalone simulations at twenty urban flux tower sites drawn from the harmonized Urban-PLUMBER dataset, spanning thirteen countries and climates from tropical Singapore to dry Phoenix and continental Europe. Each simulation was spun up for forty years until soil moisture and temperature reached equilibrium, then run over twelve to sixteen years of observations, with fluxes computed every thirty minutes. The benchmark was the old bare-soil treatment of cities. The verdict, measured by mean absolute error against tower observations, was unambiguous: the new Urban1 configuration reduced the average error in latent heat flux from 30.3 to 27.3 watts per square meter and in sensible heat flux from 41.9 to 36.9 watts per square meter across all sites. At sites with more than fifty percent impervious cover, the gains were larger still, with sensible heat error falling from 46.3 to 38.2 watts per square meter.

The detailed physics behind those numbers is telling. At the reference site of AU-Preston in Melbourne, the urban configurations reproduced the observed summer sensible heat flux far better than the old scheme, both during the day and at night. The improved daytime heat storage, driven by the higher thermal conductivity, left less energy to be dumped into the atmosphere as sensible heat, cutting a summer overestimation that reached a median bias of 77 watts per square meter in the bare-soil runs down to roughly 14 to 18 watts per square meter in the urban runs. Nocturnal sensible heat, which cities keep elevated by releasing stored warmth, also improved in both seasons. Not everything was fixed: daytime sensible heat remained underestimated in winter, and latent heat stayed generally underestimated in summer, likely because the scheme lacks active urban vegetation, irrigation, and anthropogenic heat from traffic and buildings, which can reach nearly 93 watts per square meter at highly urbanized sites like Seoul’s Jungnang district.

The hydrological results reveal both the promise and the limits of the new imperviousness parameterization. Under the strong Urban1 hypothesis, annual surface runoff roughly doubled at most stations, and the balance flipped so that surface runoff exceeded subsurface runoff at fifteen of the twenty sites, exactly the signature urbanization leaves on real catchments. But the moderate Urban2 option produced smaller and sometimes non-monotonic responses, and the simulated reduction in subsurface runoff sits awkwardly with observational studies showing that urban groundwater recharge can actually increase, thanks to leaking pipes and irrigation. The authors are candid that without runoff or soil moisture observations at the tower sites, the water-budget analysis remains a sensitivity assessment rather than a validation.

What comes next is where the story gets ambitious. The team plans convection-permitting simulations over Paris at resolutions below three kilometers, where cities are explicitly resolved and an inaccurate urban energy balance would feed wrong signals straight into the atmosphere. They also intend to add anthropogenic heat fluxes, refine the thermal parameters against measured data, embed active vegetation with seasonal phenology inside the urban tile, and exploit datasets like WUDAPT and the Copernicus Urban Atlas to make albedo, building height, and imperviousness vary across real cityscapes. Notably, most Earth system models contributing to CMIP still ignore cities altogether, a gap that becomes untenable as climate risk assessments increasingly focus on the built environment. A computationally cheap, physically grounded urban tile that works from the neighborhood scale to the globe is exactly the kind of tool that can close it, and ORCHIDEE now has one.

Subject of Research: Development and benchmarking of a new urban land surface scheme with imperviousness-dependent soil hydraulics in the ORCHIDEE 2.2 land surface model

Article Title: Benchmarking a new urban scheme in the ORCHIDEE 2.2 land surface model

Article References: Lalonde, M., Bastin, S., Oudin, L., Arboleda-Obando, P. F., & Ducharne, A. (2026). Benchmarking a new urban scheme in the ORCHIDEE 2.2 land surface model. Geoscientific Model Development, 19(19), 9463-9488. https://doi.org/10.5194/gmd-19-9463-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-9463-2026

Keywords: ORCHIDEE, land surface model, urban climate, imperviousness, saturated hydraulic conductivity, surface energy balance, Urban-PLUMBER, sensible heat flux, latent heat flux, urban hydrology, convection-permitting simulations, Geoscientific Model Development

The stakes are far from academic. Global mean air temperature has already risen by roughly 1.1 degrees Celsius, and urban areas suffer additional warming from the urban heat island effect, in which asphalt, concrete, and dense building fabric trap heat long after sunset. Epidemiological investigations in Paris, Chicago, and Shanghai have documented substantial spikes in heat-related mortality during extreme heat events, and global assessments estimate that around three billion people now experience heat stress conditions exceeding safe physiological thresholds. Against this backdrop, dozens of indices have been developed to translate raw meteorological variables such as air temperature, humidity, wind speed, and radiation into a single integrated value of human thermal load. They fall into four broad families: empirical indices like the Heat Index and Humidex, bioclimatic indices such as the Wet Bulb Globe Temperature, rational heat-balance indices like the Physiological Equivalent Temperature and the Universal Thermal Climate Index, and physiological strain indices such as Predicted Heat Strain.

The trouble, as comparative studies have repeatedly shown, is that these indices disagree. Work in humid Hong Kong, Singapore, and Guangzhou found that humidity-sensitive indices such as UTCI and the Heat Index best captured perceived thermal stress, while studies in arid Tehran, Phoenix, and Riyadh concluded that solar radiation and wind speed dominate thermal strain, favoring PET and WBGT. The same weather conditions can therefore land in different heat stress categories depending on which index is applied. Yet most urban climate and occupational health studies never justify their choice of index at all, defaulting to whatever is conventional in their region or discipline. Middle Eastern researchers tend to lean on WBGT because it is embedded in occupational guidelines, while European investigators often champion PET and UTCI. The new study set out to replace this habit with a transparent, reproducible method.

The framework unfolded in four stages across three carefully chosen cities: Abadan, a hot-humid city near the Persian Gulf; Isfahan, a hot-dry city at 1,570 meters altitude; and Gorgan, a temperate-humid city near the Caspian Sea. First, the team compiled a pool of twenty candidate indices through a systematic search of Scopus, Web of Science, PubMed, and Google Scholar. Second, they applied the Fuzzy Delphi Method, a structured consensus technique in which twenty-six experts with at least five years of experience in heat stress, thermal comfort, occupational health, or climatology rated each index on a five-point linguistic scale. The verbal judgments were converted into triangular fuzzy numbers, aggregated, and defuzzified using the center-of-gravity method. Only indices achieving a defuzzified score of at least 0.70 and sufficient expert consensus survived; exactly ten did, led by WBGT with 96.2 percent consensus, followed by PHS, UTCI, and PET, all above 88 percent. Simpler empirical indices clustered near the bottom of the accepted list.

Third, the researchers used the Fuzzy Analytic Hierarchy Process to rank the ten surviving indices against five evaluation criteria: physiological relevance, climatic sensitivity, outdoor applicability, data requirements, and scientific validation support. The expert panel judged physiological relevance the most important criterion, with a normalized weight of 0.370, followed by scientific validation support at 0.229 and climatic sensitivity at 0.210. Pairwise comparisons were performed separately for each city, yielding climate-specific global weights and rankings. The results were unambiguous. In steamy Abadan, WBGT took first place with a final weight of 0.178, followed by PHS and UTCI. In arid Isfahan, UTCI led at 0.171, ahead of PET and mPET. In temperate Gorgan, PET ranked first at 0.166, with UTCI and SET close behind. Across all three climates, the simplest empirical indices, the Heat Index and Humidex, languished near the bottom of the table.

Crucially, the team did not stop at expert opinion. They subjected the rankings to a one-at-a-time sensitivity analysis, perturbing each criterion weight by plus or minus 10, 20, and 30 percent while rescaling the remaining weights. The top-ranked index in every city survived every perturbation, and Spearman rank correlations between baseline and perturbed rankings ranged from 0.88 to 0.97, indicating that the climate-specific hierarchies are structurally embedded in the decision framework rather than artifacts of particular expert judgments. All pairwise comparison matrices also met the standard consistency threshold, with consistency ratios below 0.10, reinforcing confidence in the aggregated expert judgments.

The fourth and most novel stage brought real human bodies into the equation. Between April and September 2025, the researchers surveyed 300 adults, one hundred per city, recruited through stratified convenience sampling to represent walking, standing, and light work in streets, parks, and commercial areas. Calibrated instruments measured air temperature, humidity, wind speed, black globe temperature, and WBGT at roughly 1.1 meters above ground, the center of gravity of a standing adult, synchronized with each participant’s Thermal Sensation Vote on the standardized ASHRAE seven-point scale from minus three, cold, to plus three, hot. Perceived heat differed dramatically across the cities: the mean TSV was plus 2.10 in Abadan, where 68 percent of respondents reported feeling hot or very hot, compared with plus 1.52 in Isfahan and just plus 0.84 in Gorgan, where only 27 percent reached the hot end of the scale.

When the continuous index values were correlated against these field votes, the climate dependence of index performance became vivid. In Abadan, WBGT showed the strongest agreement with thermal sensation, a Spearman coefficient of 0.79, with PHS close behind at 0.74. In Isfahan, UTCI peaked at 0.76 and PET at 0.73. In Gorgan, PET led at 0.72, followed by UTCI at 0.69 and SET at 0.67. Grouped by category, physiological indices achieved a mean correlation of 0.70 and thermal comfort indices 0.68, while empirical indices trailed at 0.49. Most tellingly, when the expert-derived FAHP rankings were compared with rankings built purely from the perception data, the agreement was strong in all three cities, with Spearman coefficients of 0.82 in Abadan, 0.79 in Isfahan, and 0.81 in Gorgan, and the top three indices were identical in both rankings everywhere. The mean values of the top-ranked indices also rose monotonically across successive thermal sensation categories, showing they can discriminate progressive levels of perceived strain, an essential property for heat-warning systems.

The physiological logic behind these patterns is coherent. In humid climates, high moisture content of the air suppresses evaporative heat loss, so indices that explicitly track humidity-driven heat storage and sweat evaporation, like WBGT and PHS, track real strain best. In arid environments, intense solar radiation and wind dominate the energy balance, which is precisely what the radiation-sensitive, heat-balance indices UTCI and PET capture. In milder temperate-humid settings, severe physiological strain is less dominant and cumulative discomfort and behavioral adaptation matter more, favoring comfort-oriented indices that integrate energy balance with perceptual dimensions. The authors acknowledge limitations, including a predominantly Iranian expert panel, convenience sampling, a warm-season-only data window, and the omission of micro-scale urban morphology variables such as sky-view factor and vegetation shading. They call for extending the framework across seasons, adding urban form parameters, and testing dynamic indices under projected climate-change scenarios.

The practical message for city planners, public health authorities, and occupational safety bodies is direct: stop assuming one standard indicator fits everywhere. A heat-health warning system in a humid coastal city that relies on a dry-climate comfort index may understate danger precisely when evaporative cooling fails, while an arid inland city using a humidity-centric index may miss the dominant role of radiant heat. By fusing fuzzy expert consensus, multi-criteria weighting, sensitivity testing, and ground-truth human perception into a single transferable methodology, the study offers a template any city can adapt. As global warming pushes more urban populations past safe physiological thresholds, knowing which thermometer of human strain to trust may prove as important as the measurements themselves.

Subject of Research: Climate-sensitive prioritization of human heat stress indices in urban environments

Article Title: A climate-sensitive decision-support framework for prioritizing human heat stress indices: Integrating fuzzy Delphi, FAHP, and thermal perception in Abadan, Isfahan, and Gorgan, Iran

Article References: Mousavi, S. M., Dehghan, H., & Yazdanirad, S. (2026). A climate-sensitive decision-support framework for prioritizing human heat stress indices: Integrating fuzzy Delphi, FAHP, and thermal perception in Abadan, Isfahan, and Gorgan, Iran. Environmental and Sustainability Indicators, 32, Article 101529. https://doi.org/10.1016/j.indic.2026.101529

Image Credits: AI Generated

DOI: 10.1016/j.indic.2026.101529

Keywords: heat stress, thermal comfort, WBGT, UTCI, PET, urban climate, Fuzzy Delphi, FAHP, thermal sensation, climate change, Iran, decision support