GCA Framework: A GCC Countries-Grounded Dataset and Agentic Pipeline for Climate Decision Support

Muhammad Umer Sheikh; Khawar Shehzad; Salman Khan; Fahad Shahbaz Khan; Muhammad Haris Khan

cs.LGApr 14, 2026

GCA Framework: A GCC Countries-Grounded Dataset and Agentic Pipeline for Climate Decision Support

Authors: Muhammad Umer Sheikh, Khawar Shehzad, Salman Khan, Fahad Shahbaz Khan, Muhammad Haris Khan

Organizations: Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI), UAE · University of Missouri, USA · Australian National University, Australia · Linköping University, Sweden

Abstract

Climate decision-making in the GCC states increasingly demands systems that can translate heterogeneous scientific and policy evidence into actionable guidance, yet general-purpose large language models (LLMs) remain weak both in region-specific climate knowledge and grounded interaction with geospatial and forecasting tools. We present the GCA framework, which unifies (i) GCA-DS, a curated multimodal dataset grounded in the GCC states, and (ii) Gulf Climate Agent (GCA), a tool-augmented agent for climate analysis. GCA-DS comprises 200k question--answer pairs spanning governmental policies and adaptation plans, NGO and international frameworks, academic literature, and event-driven reporting on heatwaves, dust storms, and floods, complemented with remote-sensing inputs that couple imagery with textual evidence. Building on this foundation, the GCA agent orchestrates a modular tool pipeline grounded in real-time and historical signals and geospatial processing that produces derived indices and interpretable visualizations. Finally, we benchmark open and proprietary LLMs on climate tasks in the GCC states and show that domain fine-tuning and tool integration substantially improve reliability over general-purpose baselines.

Figures & tables

Figure 1: Overview of Gulf Climate Agent (GCA) Framework. We curate a Gulf-focused multimodal QA dataset, GCA-DS and fine-tune a tool-augmented LLM that routes user queries to specialized climate tools to produce grounded, interpretable outputs

Dataset	Size	Automated	Validated	Multimodal	Region	Remote Sensing	Topic Covered
Climate Crisis QA (2024)	19,241	✓	✗	✗	General	✗	5
Pirá 2.0 (2024)	2,250	✗	✓	✗	General	✗	4
Climate-FEVER (2020)	1,535	✗	✓	✗	General	✗	3
CPIQA (2025)	54,612	✓	Partial	✓	General	✗	5
ELLE (2025)	1,130	✗	✓	✗	General	✗	6
ClimaQA-Gold (2025)	566	✓	✓	✗	General	✗	5

Table 1: Comparison of climate- and environment-focused QA datasets. “Partial” denotes limited human validation (e.g., expert-guided prompting or subset review). Topic Covered reports the number of represented topic types.

Figure 2: Example samples from the gca-ds dataset spanning text-grounded QA and visual-temporal QA over Gulf cities.

Figure 3: Distribution of GCA-DS across question types and categories.

Figure 4: Gulf Climate Agent (GCA) framework. The figure summarizes multimodal dataset curation for text and visual-temporal sources, producing the gca-ds dataset, and the tool-augmented inference stack in which a fine-tuned LLM routes user requests through a tool controller to specialized climate services to generate grounded responses.

Model	Format Err. (%)	Arg. Err. (%)	N/A (%)
GPT-5	19.8	11.9	6.4
Claude 4.5 Sonnet	18.6	14.7	7.1
Gemini 2.5 Pro	21.3	16.5	6.8
Qwen2.5-VL 7B	46.7	38.2	21.5
Pixtral-12B	39.4	29.8	15.2
GCA (ours)	12.7	8.3	3.9

Table 2: Percentages of tool-use error types on GCA. Format errors: invalid tool-call structure; argument errors: incorrect/missing schema fields; N/A: no actionable tool call when required

Model	Step-by-Step Mode $\uparrow$				End-to-End Mode $\uparrow$
	InstAcc	ToolAcc	ArgAcc	SummAcc	AnsAcc	AnsAcc+I
GPT-5	88.6	92.3	90.2	87.4	86.3	87.0
Claude 4.5 Sonnet	86.8	90.2	88.3	85.6	84.3	84.8
Gemini 2.5 Pro	85.9	89.4	87.2	84.7	83.2	84.1
Qwen2.5-VL 7B	60.5	62.2	58.3	55.6	52.3	54.0
Pixtral-12B	66.9	68.3	64.2	60.8	56.2	58.5

Table 3: Tool-use benchmark results on GCA. Step-by-step mode reports instruction-following accuracy (InstAcc), tool selection accuracy (ToolAcc), argument name accuracy (ArgAcc), and step-conditioned summary accuracy (SummAcc). End-to-end mode reports final execution answer accuracy (AnsAcc) and answer accuracy with image generation enabled (AnsAcc+I).

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Tool	Purpose	Inputs	Outputs
Remote sensing and land surface
get_satellite_ image	Retrieve a multispectral satellite image for a coordinate and date, used for downstream index and change analyses.	lat, lon, date	image + metadata.
calculate_ ndvi	Compute NDVI from an image to quantify vegetation condition (scalar map + summary).	image	ndvi_map + stats.
calculate_ ndwi	Compute NDWI from an image to highlight water/moisture signals (scalar map + summary).	image	ndwi_map + stats.
desertification_ analysis	Compare two images and return land-degradation indicators (e.g., index deltas and affected area).	image1, image2	change_map + metrics.
Biodiversity and species

Appendix

Table 4: Tool Suite Details. GCA tool suite. Each tool is exposed to the LLM with typed inputs/outputs; outputs are normalized (units, timestamps) to support multi-step composition.

LLM Grounding

Large language model (LLM) grounding studies how models anchor answers and actions in evidence, structured knowledge, and real-world context.

See all 319 papers

ClimAgent: LLM as Agents for Autonomous Open-ended Climate Science Analysis
Apr 18, 2026Hao Wang, Jindong Han, Wei Fan +1AI Agents for Scientific Discovery Climate Science
1The Hong Kong University of Science and Technology (Guangzhou) · 3Shandong University · University of Auckland +1
GeoNatureAgent (GNA): A Framework and Benchmark for Pre-Production Evaluation of Tool-Using Agents on Geospatial and Environmental Tasks
Oct 6, 2026Gabriel Diaz-Ireland, Diego Prieto-Herráez, Mario García Peces +3Tool-Use Evaluation LLM Agent Evaluation
Universidad Católica de Ávila (UCAV) Ávila, Spain · Independent Researcher Madrid, Spain · Department of Earth and Planetary Sciences, Johns Hopkins University Baltimore, MD, USA +1
CLIMATEAGENT: Multi-Agent Orchestration for Complex Climate Data Science Workflows
Nov 25, 2025Chenyue Li, Hyeonjae Kim, Wen Deng +4AI for Science Multi-Agent Orchestration
1The Hong Kong University of Science and Technology

Apr 18, 2026J/K move · Enter open · S save

ClimAgent: LLM as Agents for Autonomous Open-ended Climate Science Analysis

Hao Wang, Jindong Han, Wei Fan, Hao Liu

1The Hong Kong University of Science and Technology (Guangzhou) · 3Shandong University · University of Auckland+1

AI Agents for Scientific Discovery Climate Science LLM Agents Large Language Model-Guided Scientific Discovery

Open paper AskPDF arXiv

Climate research is pivotal for mitigating global environmental crises, yet the accelerating volume of multi-scale datasets and the complexity of analytical tools have created significant bottlenecks, constraining scientific discovery to fragmented and labor-intensive workflows. While the emergence Large Language Models (LLMs) offers a transformative paradigm to scale scientific expertise, existing explorations remain largely confined to simple Question-Answering (Q&A) tasks. These approaches often oversimplify real-world challenges, neglecting the intricate physical constraints and the data-driven nature required in professional climate science.To bridge this gap, we introduce ClimAgent, a general-purpose autonomous framework designed to execute a wide spectrum of research tasks across diverse climate sub-fields. By integrating a unified tool-use environment with rigorous reasoning protocols, ClimAgent transcends simple retrieval to perform end-to-end modeling and analysis. To foster systematic evaluation, we propose ClimaBench, the first comprehensive benchmark for real-world climate discovery. It encompasses challenging problems spanning 5 distinct task categories derived from professional scenarios between 2000 and 2025. Experiments on ClimaBench demonstrate that ClimAgent significantly outperforms state-of-the-art baselines, achieving a 40.21% improvement over original LLM solutions in solution rigorousness and practicality. Our code are available at https://github.com/usail-hkust/ClimAgent.

Abstract

LLM Grounding

Explore similar work

ClimAgent: LLM as Agents for Autonomous Open-ended Climate Science Analysis

GeoNatureAgent (GNA): A Framework and Benchmark for Pre-Production Evaluation of Tool-Using Agents on Geospatial and Environmental Tasks

CLIMATEAGENT: Multi-Agent Orchestration for Complex Climate Data Science Workflows