Survey of Multimodal Geospatial Foundation Models: Techniques, Applications, and Challenges

TL;DR

This survey advances a modality-centric view of geospatial foundation models, tracing their evolution from unimodal RS models to multimodal visual and vision-language GFMs. It articulates core training paradigms, backbones, and adaptation strategies, and analyzes modality-specific design principles for aligning heterogeneous RS data. The paper systematically reviews modalities, benchmarks, and tasks, highlighting cross-modal fusion (RGB, MS, SAR/InSAR, HS, LiDAR) and VL-GFM capabilities, while presenting real-world applications and datasets. It identifies key challenges in domain generalization, interpretability, computational costs, and privacy, and outlines future directions toward lightweight, open, and responsible universal geospatial AI. Overall, the work provides a comprehensive, modality-informed framework for evaluating and advancing multimodal geospatial intelligence with practical societal impact.

Abstract

Foundation models have transformed natural language processing and computer vision, and their impact is now reshaping remote sensing image analysis. With powerful generalization and transfer learning capabilities, they align naturally with the multimodal, multi-resolution, and multi-temporal characteristics of remote sensing data. To address unique challenges in the field, multimodal geospatial foundation models (GFMs) have emerged as a dedicated research frontier. This survey delivers a comprehensive review of multimodal GFMs from a modality-driven perspective, covering five core visual and vision-language modalities. We examine how differences in imaging physics and data representation shape interaction design, and we analyze key techniques for alignment, integration, and knowledge transfer to tackle modality heterogeneity, distribution shifts, and semantic gaps. Advances in training paradigms, architectures, and task-specific adaptation strategies are systematically assessed alongside a wealth of emerging benchmarks. Representative multimodal visual and vision-language GFMs are evaluated across ten downstream tasks, with insights into their architectures, performance, and application scenarios. Real-world case studies, spanning land cover mapping, agricultural monitoring, disaster response, climate studies, and geospatial intelligence, demonstrate the practical potential of GFMs. Finally, we outline pressing challenges in domain generalization, interpretability, efficiency, and privacy, and chart promising avenues for future research.

Figures (6)

• Figure 1: Review of representative geospatial foundation models.
• Figure 2: Graphical illustration of training paradigms. (a) Masked modeling, (b) Contrastive learning, (c) Generative learning.
• Figure 3: Model architectures across backbones. (a) CNN, (b) Transformer, (c) Hybrid model.
• Figure 4: Network architectures and task-specific adaptation strategies for multimodal geospatial foundation models.
• Figure 5: Benchmarks and evaluation framework for multimodal geospatial foundation models.