The rapid growth of multimodal data—such as text, images, sensor streams, graphs, and structured records—has made cross-modal integration critical for modern large-scale data analysis. However, the heterogeneous nature of multimodal sources and the limitations of conventional fusion techniques hinder effective semantic alignment, representation learning, and scalable analytics.

Although large language models (LLMs) offer strong capabilities in reasoning, abstraction, and cross-domain understanding, current data pipelines still lack efficient mechanisms to incorporate LLM-driven semantics into multimodal fusion workflows. This thematic series aims to bridge this gap by exploring innovative approaches that leverage LLMs to enhance multimodal data fusion and enable more powerful, comprehensive data-driven insights.

This collection focuses on advancing LLM-Augmented Multimodal Data Fusion for Large-Scale Data Analysis, encouraging research on:

LLM-enhanced representation learning Semantic alignment across heterogeneous modalities Generative or retrieval-assisted fusion strategies Scalable system designs for real-world applications The goal is to promote new analytical paradigms where LLM-driven intelligence reshapes multimodal integration and utilization in complex scientific and industrial ecosystems.

The topics include, but are not limited to:

LLM-augmented cross-modal semantic alignment for large-scale analytics

Generative and retrieval-assisted fusion for multimodal data integration

Representation learning for heterogeneous and multi-source data fusion

Knowledge grounding and reasoning across diverse data modalities

Scalable fusion architectures for large-volume multimodal datasets

Foundation-model-assisted modality completion and data annotation

Graph–text–sensor fusion for scientific and engineering data analysis

Temporal–spatial multimodal fusion for real-world big data applications

Self-supervised learning for multimodal representation and alignment

Benchmarking, datasets, and evaluation protocols for multimodal fusion

Efficient fusion mechanisms for high-dimensional industrial and IoT data

Domain-specific multimodal fusion applications for data-driven intelligence

SDG 9: Industry, Innovation & Infrastructure

We invite researchers, practitioners, and industry experts to submit original contributions to this special issue/track focused on AI‑driven data engineering for Earth and Space Sciences. This collection addresses the rising need for intelligent techniques that support scalable, interoperable, and FAIR (Findable, Accessible, Interoperable, Reusable) data ecosystems across environmental, geospatial, planetary, and space science domains.

Scope and Themes

This track welcomes submissions that integrate Artificial Intelligence, data engineering, and scientific workflows to tackle challenges in managing, harmonizing, analyzing, and reusing complex Earth and Space Science data.

Topics of interest include, but are not limited to:

AI‑enabled data pipelines for Earth observation, satellite systems, and space mission data. Semantic, ontology‑driven, and knowledge‑based frameworks for metadata enrichment and interoperability. Intelligent workflow orchestration, hybrid AI systems, and agent‑based automation. Representation learning, self‑supervised learning, and foundation models for geospatial and environmental data. FAIR principles, reproducible workflows, Open Science frameworks, and transparent data governance. Scientific applications in climate analytics, disaster prediction, planetary exploration, environmental sustainability, and more.

Key Focus Areas

The special issue/track also highlights emerging AI‑driven paradigms for next‑generation scientific data engineering, including:

Foundation models and self‑supervised learning for Earth observation.

Knowledge graphs, semantic mapping, and retrieval‑augmented generation (RAG).

Agentic workflow orchestration and enhanced digital twin accessibility.

AI‑based multimodal data fusion, explainability, and domain adaptation.

Synthetic data generation and quality assessment.

Multivariate time‑series modeling and forecasting for environmental and planetary monitoring.

Why Submit?

This track aims to showcase innovations, architectures, frameworks, and real‑world case studies that demonstrate how AI‑driven data engineering accelerates scientific discovery, strengthens data interoperability, and supports sustainability across Earth and Space Science disciplines. Contributions that advance FAIR and Open Science principles are especially encouraged.

Work building upon discussions or early results presented at the DARES’25 ECAI Workshop is also welcome.

SDG 9: Industry, Innovation & Infrastructure