Contributed to by Harison K. Kipkulei, as part of the Centre for Climate Resilience, this study highlights how generative artificial intelligence and large language models can effectively transform climate information services and empower vulnerable local communities through localized early warning systems and micro-level vulnerability mapping, specifically focusing on bridging the communication gap in data-scarce and developing regions.

Why climate services need a smart upgrade

The world is facing an escalating climate crisis that requires a fundamental shift in how Climate Information Services (CIS) are delivered to enhance resilience. While individual studies have shown the potential of Generative Artificial Intelligence (GenAI) and Large Language Models (LLMs) like ChatGPT, research remains fragmented. Current CIS often focus on scientific prediction systems rather than usability for end-users, leading to a lack of access for local communities. This paper aims to assess the potential and limitations of GenAI for CIS through a systematic literature review.

Mapping the Landscape of Generative Climate Tech

The authors conducted a systematic literature review following PRISMA guidelines, focusing on the period between 2022 and 2025 to coincide with the rapid growth of LLMs following the release of ChatGPT. Using the Scopus database, they initially identified 281 results. After applying rigorous inclusion and exclusion criteria, such as requiring peer-reviewed, open-access journal articles in English, 19 studies were selected for detailed qualitative content analysis.

Five ways GenAI protects communities

This chapter details five key areas where GenAI enhances climate services:

• Micro-level climate vulnerability mapping: GenAI can synthesize high-resolution, local-scale climate data (e.g., precipitation, wildfires) and bridge gaps where scientific data is scarce by incorporating qualitative, lived-experience data from communities.
• Enhancing community integration: GenAI reduces the "digital divide" between scientists and local populations through human-in-the-loop mechanisms and by translating complex concepts into local languages and plain text.
• Extreme weather events forecasting: Advanced models like LaDCast and GED improve the accuracy and diversity of short-to-medium-range forecasts, such as extreme precipitation, allowing for automated community advisories.
• Climate data generation: Technologies like Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) help fill gaps in observational networks by generating realistic synthetic data, which is crucial for planning in data-scarce regions.
• Advancing early warning systems: GenAI facilitates the "Early Warning for All" (EW4A) initiative by processing multi-source data and delivering timely, accessible alerts in regional languages.

Black boxes and digital barriers

Despite its potential, several challenges exist:

• Lack of transparency: Many models operate as "black boxes," making it difficult for experts to understand how outputs are generated, which can undermine trust.
• Misinformation and manipulation: GenAI can produce incorrect responses or fake media that threaten the reliability of climate assessments.
• Digital accessibility barriers: Benefits are often restricted by location disparities, as rural communities may lack the necessary internet or mobile connectivity to access these services.
• Ethical and technical concerns: Issues such as data privacy, high energy consumption of large models, and inherent biases in model design require further attention.

Why the future of climate AI must be human-centered

The authors highlight a significant disparity between technological advances and user-centered design. Much of the current literature focuses on technical performance rather than how communities actually trust and use these tools. Future research should move toward field-based evaluations, participatory pilots, and "living labs" to better understand adoption behavior and long-term impact.

A new chapter for climate resilience

The study concludes that while GenAI offers transformative opportunities to make CIS more user-friendly and inclusive, its deployment must be accompanied by bottom-up approaches involving local communities. Addressing challenges like transparency and misinformation is vital for building the trust necessary for widespread adoption. Ultimately, there is an urgent need for more empirical, user-centered studies to bridge the gap between technical potential and real-world resilience.

Source: Malekela, A. A., Lusiru, S., Kipkulei, H. K., Kimaro, P., Kabirigi, M., Sieber, S., & Ryo, M. (2026): Generative artificial intelligence for climate information services. Discover Applied Sciences 8, 684.