Climate Co-Pilot

A Multimodal AI Agent for Climate Resilience in the MENA Region

Project Focus
Multimodal Earth system learning Intelligent agent systems Water and food security applications

Project Overview

Climate change poses significant challenges across MENA, where extreme heat, water scarcity, and agricultural stress affect ecosystems, resources, and livelihoods. Climate analysis increasingly requires combining heterogeneous information, including gridded meteorological data, satellite observations, ground-based measurements, scientific models, and domain knowledge. However, these sources are often analyzed through separate tools and workflows, making integrated scientific analysis and decision support difficult.

Climate Co-Pilot aims to develop a conversational, scientifically grounded AI system built around a multimodal foundation model and an agentic pipeline. The project investigates how AI can jointly reason over physical Earth system data, imagery, and textual knowledge, while coordinating data retrieval, model inference, scientific tools, simulation, and result synthesis. The longer-term objective is to support interactive analyses and scenario-based decision support for water resource management and agricultural adaptation in MENA.

Research Directions

View research page →
01

Multimodal Foundation Models

How can AI jointly represent heterogeneous Earth system data and natural language?

  • Gridded meteorological data
  • Earth observation imagery
  • Scientific and contextual knowledge

Current work: EarthAlign & MeteoLIP

Explore →
02

Intelligent Agent Systems

How can AI agents decompose complex Earth system questions and reliably use scientific tools to answer them?

  • Tool-augmented reasoning
  • Data and model orchestration
  • Traceable scientific workflows

Current work: TerraAgent & TerraBench

Explore →

Work to Date

Current research has established initial capabilities in agentic Earth system reasoning and meteorology-language representation learning.

Agentic AI

TerraAgent & TerraBench

An executable agentic framework and benchmark for evaluating grounded reasoning across heterogeneous Earth system data, scientific tools, and simulations.

Learn More
Multimodal AI

EarthAlign & MeteoLIP

A large-scale Earth system-language dataset and representation-learning framework for aligning ERA5 meteorological data with natural language.

Learn More

Project Framework

Climate Co-Pilot connects heterogeneous Earth system data, multimodal representation learning, and agentic reasoning within a unified research workflow.

Climate Co-Pilot project framework figure
01Data Integration

Satellite observations, in-situ measurements, reanalysis, climate-model data, and textual knowledge.

02Multimodal Learning

Models learn representations connecting physical, visual, and language information.

03Agentic Analysis

Agents coordinate data retrieval, scientific tools, simulations, and model inference.

04Applications

Capabilities extend toward water resource management and agricultural adaptation.

Data Sources

Climate Co-Pilot considers complementary Earth system and domain-knowledge sources spanning satellite observations, in-situ measurements, gridded climate data, and scientific literature.

Planned Climate Co-Pilot data sources figure
Planned data sources supporting the Climate Co-Pilot research framework.

Partners

Climate Co-Pilot brings together expertise in multimodal learning, foundation models, agentic AI, Earth system science, and climate applications.

Updates

Research outputs, project milestones, collaboration, capacity building, and knowledge-transfer activities.

October 2026
Knowledge Transfer

Climate Co-Pilot at the ADIA-Lab Symposium

The Climate Co-Pilot team will participate in the ADIA-Lab Symposium, sharing project progress through talks and poster presentations.

October 2026
Research Release

EarthAlign dataset and MeteoLIP weights released on Hugging Face

The EarthAlign dataset and MeteoLIP model weights were released publicly to support reproducible research on language-aligned meteorological representation learning.

August 2026
Invited Contribution

Invited contribution to the Annual ADIA Lab Transactions in Data Science

The Climate Co-Pilot team was invited to contribute the chapter Tool-Augmented Language Agents for Earth Observation.

June 2026
Research Release

TerraAgent and TerraBench released on arXiv

TerraAgent and TerraBench were released publicly, presenting an executable framework and benchmark for tool-augmented Earth system reasoning.

June 2026
Knowledge Transfer

Climate Co-Pilot session at ADIA Lab

The project team presented the ongoing agentic and multimodal research directions to ADIA Lab summer interns and prepared introductory Google Earth Engine tutorial materials.

December 2025
Funding / Project

Climate Co-Pilot project launched

Climate Co-Pilot was launched with support from ADIA Lab to develop a multimodal AI agent for climate resilience in MENA, combining Earth system data, multimodal foundation models, and agentic reasoning to support scientifically grounded analysis and decision support.