Artificial intelligence (AI) records rapid progress and its applications range from education and healthcare to areas requiring complex analytics. AI is a broad term describing a variety of useful technologies that mimic human intelligence, including judgment and subsequent decision-making. Recent developments in AI, such as OpenAI’s ChatGPT models, machine vision and deep learning technologies, can be deployed in a variety of contexts, particularly in relation to climate change.

Particular interest is focused on Foundation Models (FMs), which can help broaden understanding of climate change and reduce the social risks of adaptation and mitigation initiatives.

Characteristics and strength of FMs

FMs are models trained on large, unlabeled datasets and can be adapted to perform a wide range of contemporary tasks, taking into account a variety of variables. They are enabled primarily through transfer learning, when knowledge gained from one task is applied to another, and through scaleThe scope is driven by three main features: improved hardware, transformer model architecture, and the availability of extensive training data.

These models, such as the Generative Pretrained Transformer (GPT), have revolutionized natural language processing (NLP) tasks. They are also successfully used in key sectors such as healthcare (e.g., when working with various combinations of medical data) and business (process optimization).

Application of FMs in the context of climate change

The distinctive features of FMs make them ideal for climate change mitigation and adaptation initiatives. They can flexibly help to better understand key causal relationships in the performance of climate systems. Potential contributions of FMs to adaptation and mitigation include:

1. Data analysis: FMs are capable of processing vast volumes of climate data, including temperature records, atmospheric measurements, and satellite imagery.
2. Climate modeling: AI models trained on vast amounts of data can simulate and predict Earth's climate systems, improving resolution and speeding up predictions. These models, such as ClimaX, can support decision-making for policymakers and disaster response teams.
3. Risk assessment: FMs can assess climate-related vulnerabilities and risks across different sectors, regions and societies.
4. Decision support: They provide policymakers and business managers with evidence-based knowledge that guides policy formulation and strategic management.

In addition, FMs assist in visual understanding phenomena such as sea level rise, and can increase user engagement by replicating human intelligence, which is useful in raising awareness about the impacts of climate change.

Industrial initiatives

Leading technology companies are actively developing FMs for climate change management:

• Google deploys the Environmental Insights Explorer (EIE) tool to reduce carbon emissions and Project Green Light to optimize traffic light timing.
• IBM is developing a geospatial FM in collaboration with NASA that converts satellite observations into maps of natural disasters and estimates risks to infrastructure.
• Microsoft in collaboration with UCLA presented ClimaX, a generalizable FM for weather and climate modeling.
• NVIDIA launched an initiative Earth-2 with the aim of building a digital twin of the Earth to improve extreme weather predictions and support adaptation strategies.

Ethical and technical constraints

Despite their enormous potential, FMs face significant challenges. Among ethical concerns belongs to reinforcing harmful stereotypes (bias amplification) present in the training data, the risk misuse for disinformation or deepfakes, and non-transparency decision-making processes (their "black-box" nature), which makes accountability difficult.

Between technical limitations their tendency is to generating incorrect outputs due to a lack of real understanding (a phenomenon known as “hallucination”). In addition, FMs are associated with rising environmental costs as a result high energy intensity and subsequent carbon emissions.

To maximize the positive impact of FMs on climate adaptation and mitigation, it is essential interdisciplinary effort to address these ethical and technical limitations, ensure transparency, and deploy AI responsibly. JRi

Walter Leal Filho et al. (2025) v Environmental Sciences Europe