Block-by-block AI maps reveal urban air temperatures in 380 U.S. cities

New research from the University of Illinois Urbana-Champaign reveals that urban heat islands in the U.S. may not be as severe as previously thought. Published on July 8, 2026, the study developed a high-resolution dataset that accurately measures urban air temperatures across 380 cities in the contiguous United States.

Understanding Urban Heat Islands

Urban areas are commonly referred to as "heat islands" due to the significantly higher temperatures in certain neighborhoods compared to others. This phenomenon is often reported in media, suggesting temperature differences can reach as high as 20° F (7° C). However, these figures are frequently based on satellite data, which does not reflect the actual conditions experienced by residents.

According to the research team, consisting of professor Lei Zhao, graduate student Yiwen Zhang, and Pierre Gentine from Columbia University, the lack of near-surface observations in urban areas contributes to this discrepancy. Standard weather stations are often located in open areas away from urban heat, leading to an underrepresentation of temperatures in densely populated neighborhoods.

High-Resolution Temperature Mapping

To address the data gap, the researchers created the Urban High-Resolution Air Temperature dataset (U-HAT), utilizing a physics-informed transfer learning model. This innovative approach combines physical atmospheric understanding with data-driven methods to estimate near-surface air temperatures at a very high spatial resolution.

The U-HAT dataset enables detailed mapping of air temperatures across cities, allowing for a block-by-block analysis. This data is crucial for various applications, including:

• Public health studies
• Urban climate research
• Energy planning
• Machine-learning applications

As Zhao states, "This data allows for pixel-by-pixel comparison between satellite land surface temperature and true urban air temperature." The findings suggest that satellite-based data often exaggerates heat stress and disparities, potentially misleading the public.

Global Implications of the Research

The implications of this research extend beyond U.S. cities. The methodology can be applied globally, particularly in regions lacking sufficient weather data, such as those in the Global South. Many of these areas are highly vulnerable to climate change and extreme heat, yet they often lack the infrastructure for reliable climate observations.

By employing physics-informed AI, researchers can generate more accurate climate information for communities that currently rely on limited data. This approach emphasizes global equity, ensuring that even the most vulnerable regions have access to essential climate and weather information.

The study, titled "Transfer learning reveals large discrepancies between air and land surface temperatures in cities," is published in Nature Communications and can be accessed using DOI: 10.1038/s41467-026-73716-7.