Artificial intelligence is helping cities identify neighborhoods threatened by extreme heat and find the people most vulnerable to high temperatures. However, a study published in the journal Climate Risk Management revealed a major imbalance: scientists are far more likely to develop AI for predicting heat risks than for organizing aid to those affected. The next step is learning how to turn accurate forecasts into concrete decisions that truly protect people.

The problem is becoming more acute as extreme heat intensifies amid climate change. According to a Yale University study published in 2025, the average annual number of heat-related deaths in the United States rose from 2,670 in 2000–2009 to more than 4,000 in 2010–2020—about a 53% increase, Axios reports.

Research focuses on forecasts, not assistance

The authors analyzed existing studies on the use of artificial intelligence and machine learning for managing extreme heat risks. The results showed that scientists’ main focus is on identifying danger, not on how to act after receiving a forecast.

The distribution of studies was as follows:

· 98.4% are devoted to forecasting heat and related risks;

· 59.2% address prevention;

· 10.4% examine emergency response;

· 3.2% are devoted to recovery after impacts;

· 1.6% focus on preparedness for hazardous events.

The categories in the study may overlap, so these figures should not be added into a single distribution structure. But the overall conclusion is clear: much less attention is paid to the practical use of AI for responding to heat than to forecasting it.

The researchers describe this gap as a problem in moving from scientific development to real-world implementation. An algorithm can show with high accuracy where temperatures will be especially dangerous and which population groups face the greatest risk. But a forecast alone does not open a cooling center, organize water deliveries, or help emergency services allocate resources.

According to Karishma Patel, a RAND researcher and co-author of the study, AI is especially valuable in situations where a decision depends on large volumes of data or requires very detailed territorial analysis.

Why an accurate forecast is not enough

To use forecasts in the work of city services, not only good models are needed but also reliable data infrastructure. Information must come from different sources, be updated in a timely manner, and be presented in a form understandable to the people making decisions.

Cornell University researcher Jenna Gosciak, who studies the use of AI in heat response, considers the creation of such systems one of the key conditions for success. Even the most sophisticated model will be of little help if its results cannot be quickly integrated into the daily work of municipal services.

There is another nuance: a more complex algorithm is not always more useful than a simple vulnerability indicator.

Gosciak compared traditional indices that assess population vulnerability to heat with more complex predictive algorithms that can be used in government decision-making. According to her, the choice depends on the task and the time frame in which action must be taken.

When complex AI loses to simple indices

For rapid response tied to a specific event, algorithms that process constantly updated data may be especially useful. For example, they can help determine where risk is rising quickly and where assistance should be directed first.

But for long-term planning, simple vulnerability indices may sometimes be better suited. They allow authorities to explicitly indicate which population groups need to be protected first, instead of relying entirely on the conclusions of a complex model.

This is an important distinction: forecasting shows where danger is likely, but the decision about allocating resources also depends on public priorities. The more complex the algorithm, the more important it is to understand what data it uses and why it highlights certain neighborhoods or groups of people rather than others.

Thus, the goal is not to replace all existing methods with artificial intelligence, but to choose the right tool for a specific decision.

Google uses AI to find overheating urban areas

One example of the practical use of such technologies is the Google Heat Resilience project. In June, the company announced the expansion of its heat resilience data system from an initial pilot project in 14 cities to more than 50 cities worldwide.

The service uses artificial intelligence and high-resolution imagery to assess the reflectivity of the roofs of individual buildings. This information helps city planners identify areas where measures to reduce overheating could provide the greatest benefit.

According to Google, the results of the pilot projects have already been taken into account in decision-making in several cities, including the development of cool roof requirements and climate adaptation plans.

This approach makes it possible to move from general assessments of urban temperatures to a more detailed analysis of specific buildings. However, even here, having a detailed map does not guarantee that the necessary changes will be made: this requires funding, administrative decisions, and coordinated action by city services.

The next stage is linking forecasts to action

In Patel’s view, a promising direction for AI development is systems that combine changing data and help authorities make decisions almost in real time. Highly detailed local forecasts that make it possible to assess risks at the level of individual neighborhoods could be especially useful.

In addition, such systems can take into account the combination of several threats. Heat is often accompanied by air pollution, wildfire smoke, or drought. Joint analysis of these factors can give services a more complete picture of where the population needs protection.

But the main criterion for success is not the complexity of the algorithm or the accuracy of the forecast in themselves. What matters is whether the technology helps remove a specific obstacle: directing aid faster, allocating resources more effectively, or preparing the most vulnerable areas in advance.

As Patel emphasizes, it is not enough simply to create AI-based technology—it must solve a real problem. In the case of extreme heat, this determines whether scientific advances will be transformed into practical protection for people.