Multimodal Earth Observation Data Fusion Technology

Dear Colleagues,

Earth observation has entered an era of multi-sensor, multi-source, multi-scale, and multi-temporal sensing, driven by the proliferation of heterogeneous sensors—including optical, SAR, LiDAR, hyperspectral, and thermal imagers—deployed on satellite constellations (e.g., Sentinel, Landsat, and Gaofen), airborne platforms, unmanned aerial vehicles (UAVs), and ground-based sensor networks. The heterogeneity of these sensing modalities poses significant challenges for traditional unimodal analysis methods that rely on a single sensor. Multi-sensor data fusion has thus emerged as a transformative paradigm that integrates complementary information from heterogeneous sensors and data types—spanning sensor-level, data-level, and feature-level fusion—to achieve more robust, accurate, and comprehensive Earth observation analytics. Beyond the fusion of physical sensor measurements, the integration of remote sensing imagery with textual descriptions represents an emerging frontier—vision-language multimodal learning—that promises to unlock richer semantic understanding of Earth observation data.

Despite significant advancements, several critical gaps remain in the existing body of research. A notable limitation is the lack of unified frameworks for cross-sensor geometric and radiometric registration and calibration, which hinders the seamless integration of heterogeneous measurements. Furthermore, current research often treats individual modalities as isolated entities, failing to fully exploit their complementarity under missing or incomplete data conditions. Another challenge lies in the limited operational deployment of advanced technologies, such as deep learning, vision-language models, and edge computing, which, while promising, remain underutilized in operational multi-sensor fusion systems, and whose interpretability and uncertainty quantification further constrain adoption. Future research directions aim to address these gaps by developing standardized cross-sensor fusion frameworks, scale-aware and modality-robust learning architectures, and real-time onboard processing, while fostering interdisciplinary collaboration among sensor scientists, photogrammetrists, remote sensing experts, and computer scientists. In light of the above, this Topic aims to collect innovative original manuscripts on the theoretical, methodological, and applied aspects of multi-sensor and multimodal Earth observation data fusion, covering the full chain from sensor and imaging system design, through image and signal processing, to AI-driven fusion algorithms and their deployment on satellite, airborne, and UAV platforms. Review articles and meta-analysis papers on these topics are additionally welcome. Topics of interest include the following:

• Deep learning-based multi-sensor and multimodal fusion architectures (e.g., transformer, CNN, attention mechanisms, and graph neural networks);
• Sensor-level, data-level, and feature-level fusion methods for heterogeneous Earth observation sensor data;
• Multi-sensor system design, calibration, geometric and radiometric registration, and quality assessment for Earth observation;
• Cross-modal feature alignment, domain adaptation, and representation learning for heterogeneous remote sensing data;
• Missing modality imputation and robust fusion strategies under incomplete or noisy data conditions;
• SAR-optical, LiDAR-optical, and hyperspectral-multispectral sensor fusion for multi-sensor Earth observation and target detection;
• Vision-language models and image-text multimodal learning for remote sensing interpretation (e.g., visual grounding, referring expression comprehension);
• Interpretability, uncertainty quantification, and explainable AI in multimodal fusion models;
• Real-time and edge-computing-oriented multi-sensor fusion frameworks for operational Earth observation monitoring.

Dr. Weiwei Qin
Prof. Dr. Zhiqiang Tian
Dr. Jianhui Xu
Dr. Jing Geng
Dr. Bing He
Topic Editors