When the Ground Beneath Your Feet Lies: Why Land Surface Temperature Isn’t Telling You the Whole Story About Urban Heat
Land surface temperature from satellites often fails to capture the actual heat stress experienced by people on urban streets.
26 Apr 2026, Yanjiang
We think of urban heat as something we can see. The shimmering asphalt on a July afternoon. The dark rooftop baking under an unrelenting sun. The satellite image that paints the city in shades of red and orange, each pixel a measurement of how hot the surface has become. It feels intuitive: the hotter the ground, the hotter we feel.
A single square kilometer of the city shows fine-grained patterns of land use, building heights, and tree canopy that are invisible at the broader island scale. This reveals how local urban structure directly shapes human heat stress, guiding precise interventions to cool neighborhoods.
But what if that intuition is wrong? What if the ground beneath your feet is cool, yet you are still sweltering? And what if, in some places, the ground is scorching while the air around you remains surprisingly bearable?
A team led by Rudi Stouffs at the National University of Singapore — working with Shengao Yi, Xiaojiang Li, Pengyuan Liu, Zhiwei Yang, and Ronita Bardhan — has confronted this question directly. Their preprint (arXiv:2604.22433) builds a comprehensive framework to compare two very different ways of measuring urban heat: the familiar land surface temperature (LST) captured by satellites, and a far more human-relevant metric called the Universal Thermal Climate Index (UTCI) , which attempts to measure what a person actually feels when standing on the street.
The results are not subtle. They are, in fact, a warning.
Two Thermometers, Two Worlds
Heat builds up intensely on dark rooftops and asphalt, while shaded parks and narrow streets stay far cooler. This local-scale map reveals exactly where urban design can be changed to protect people from dangerous heat stress.
Land surface temperature is the standard tool for urban heat studies. Satellites like Landsat measure the infrared radiation emitted by roofs, roads, and soil — essentially, how hot the physical surfaces of the city have become. It is a useful metric, but it has a fundamental blind spot: it tells you nothing about how that heat transfers to a human body standing in the shade, or walking through a narrow canyon of skyscrapers, or breathing air that has been trapped by poor ventilation.
The Universal Thermal Climate Index, by contrast, is a human-centric model. It takes into account air temperature, humidity, wind speed, and — crucially — the mean radiant temperature, which captures the net effect of all the radiation (solar and thermal) that a person absorbs from their surroundings. In dense urban environments, this is dominated by shading from buildings and trees, and by the trapping of longwave radiation in street canyons. UTCI is what your skin and lungs actually experience, not what a satellite sees from orbit.
The Singapore team’s first major finding is a stark spatial mismatch between these two metrics. Using 30-meter resolution LST from Landsat and 1-meter resolution UTCI computed via GPU-accelerated models, they produced maps that tell two very different stories about the same city.
In some areas, high LST coincided with high UTCI — the intuitive case, where hot surfaces mean hot people. But in other zones, the two metrics diverged dramatically. The team identified four distinct thermal profiles: cool surfaces with comfortable conditions (the ideal); hot surfaces with high human stress (the worst case); hot surfaces where human stress was mitigated — for example, by strong ventilation or shade; and, most surprisingly, cool surfaces where human stress remained severe, due to trapped radiation or lack of shade.
This last category is the most troubling. It suggests that a neighborhood could look safe on a satellite image — cool roofs, green spaces, moderate surface temperatures — while its residents are actually suffering from significant heat stress. The ground is lying to us.
What Drives the Divergence? The Surprising Role of Sky View Factor
To understand why LST and UTCI diverge, the team employed a sophisticated machine learning workflow. They used a geographically weighted XGBoost (GW-XGBoost) model — essentially, a gradient-boosted decision tree that allows the relationships between urban features and thermal metrics to vary across space. Unlike a global model that assumes a single relationship holds everywhere, GW-XGBoost lets the model learn that, for example, building height might matter a lot in one neighborhood and hardly at all in another.
The results were striking. The GW-XGBoost model achieved a global out-of-bag R² of 0.855 for LST and 0.905 for UTCI — meaning it could explain about 86% and 91% of the spatial variation in each metric, respectively. But the drivers of that variation were completely different.
To unpack what the model had learned, the team used SHAP (SHapley Additive exPlanations) , a game-theoretic method that attributes each feature’s contribution to the model’s output. SHAP values tell you, for each individual location, exactly how much each urban factor pushed the predicted temperature up or down.
And here is where the story gets counterintuitive.
The single most important factor for UTCI was the sky view factor (SVF) — a measure of how much of the sky is visible from a given point on the ground. A low SVF means the view is blocked by buildings or trees; a high SVF means open sky. In dense urban canyons, a low SVF provides shade during the day but traps longwave radiation at night. The SHAP analysis showed that SVF dominated UTCI variability across most of Singapore, with a clear nonlinear relationship: low SVF (enclosed spaces) was associated with lower daytime UTCI (shading), but higher nighttime UTCI (trapped radiation).
Yet when the same model was applied to LST, SVF was nearly irrelevant. The satellite sees roofs and road surfaces, not the human-scale experience of shade and trapped radiation. The ground simply does not “know” whether you are standing in a sunlit plaza or a shaded alley.
This is the central insight of the paper: land surface temperature is a poor proxy for human heat stress precisely because it cannot capture the radiative processes that govern how people actually experience heat in cities.
The Albedo Paradox: When Painting Things White Makes You Hotter
Perhaps the most surprising finding involves albedo — the reflectivity of surfaces. Conventional wisdom in urban heat management says: paint roofs and pavements white to reflect sunlight and keep cities cool. This works for surface temperature, and indeed, the team found that higher albedo was associated with lower LST.
But for UTCI, the relationship reversed. Higher albedo was associated with increased human heat stress.
Why? Because reflective surfaces do not simply bounce sunlight back to space. In a dense urban canyon, they bounce it onto pedestrians. The energy that would have been absorbed by the road is instead redirected to your skin. The SHAP dependence plots with Generalized Additive Model (GAM) smoothers — which capture nonlinear relationships — showed this transition clearly: below a certain albedo threshold, increasing reflectivity actually increased UTCI, because the reflected radiation was trapped within the urban geometry.
This is a profound challenge for urban design. The same intervention that cools the city’s surface can make its residents hotter. The solution is not to abandon reflective materials, but to deploy them selectively — in open areas where reflected radiation escapes, not in enclosed canyons where it bounces onto people.
What This Means for the Cities We Build
The implications extend far beyond Singapore. Every city that relies on satellite-derived surface temperature to guide heat adaptation is potentially making decisions based on incomplete information. A neighborhood that looks cool from space may be a hidden heat trap. An intervention that lowers surface temperature may paradoxically worsen human comfort.
The team’s framework — combining explainable machine learning with human-centric thermal metrics — offers a path forward. It suggests that urban heat management must become spatially explicit and human-aware. Shade, ventilation, and geometry matter as much as surface reflectivity. The sky view factor is not a niche parameter; it may be the single most important urban design variable for human thermal comfort.
And the albedo paradox reminds us that there are no silver bullets in complex systems. Every intervention must be evaluated in context, with the full radiative and aerodynamic environment taken into account.
As cities around the world grapple with rising temperatures, the tools we use to measure the problem will shape the solutions we pursue. This work suggests that we have been looking at the wrong thermometer. The ground, it turns out, is not always a reliable witness.
The question now is: will we listen to what human bodies are telling us, instead?
Yanjiang is an online editor of Loom Science
References
- Shengao Yi et al., Beyond Land Surface Temperature: Explainable Spatial Machine Learning Reveals Urban Morphology Effects on Human-Centric Heat Stress, arXiv:2604.22433