Operational disaster response goes beyond damage assessment, requiring responders to integrate multi-sensor signals, reason over road networks, populations and key facilities, plan evacuations, and produce actionable reports. However, prior work largely isolates remote-sensing perception or evaluates generic tool use, leaving the end-to-end workflows of emergency operations underexplored. In this paper, we introduce Disaster Operational Response Agent benchmark (DORA), the first agentic benchmark for end-to-end disaster response: 515 expert-authored tasks across 45 real-world disaster events spanning 10 types, paired with expert-verified, replayable gold trajectories totaling 3,500 tool-call steps. Tasks span five dimensions that cover the operational disaster-response pipeline: disaster perception, spatial relational analysis, rescue and evacuation planning, temporal evolution reasoning, and multi-modal report synthesis. Agents compose calls from a 108-tool MCP library over heterogeneous geospatial data: optical, SAR, and multi-spectral imagery across single-, bi-, and multi-temporal sequences (0.015-10m GSD), complemented by elevation and social vector layers. We comprehensively evaluate 13 frontier LLMs on our benchmark, revealing three persistent challenges: 1) disaster-domain grounding exposes unique failure modes (damage-semantic grounding, sensor-modality mismatch, and disaster-pipeline composition); 2) agents are doubly bottlenecked by tool selection and argument grounding, where gold tool-order hints improve accuracy by only 1.08-4.40

This article presents the scientific results of the 2025 Data Fusion Contest organized by the Image Analysis and Data Fusion Technical Committee, the University of Tokyo, RIKEN, and ETH Zurich. The focus of the contest was to develop innovative solutions for all-weather land-cover and building damage mapping using multimodal SAR and optical EO data at submeter resolution. The contest is organized into two distinct tracks. Track 1 focuses on land-cover mapping and Track 2 is about building damage mapping. The competition presented two primary technical challenges: the effective integration of multimodal data and the development of robust models capable of handling noisy labels. The contest saw significant global engagement, with Track 1 receiving 507 registrations and 3859 successful submissions, while Track 2 followed a similar trend with 423 registered teams and 5008 successful entries. This article provides the methodologies and results achieved by the first- and second-ranked teams from each track. To promote transparency and ensure the reproducibility of results, all participating teams in this year's contest have publicly released their code. Furthermore, the dataset used in this competition has been made publicly available to the community to support and encourage further research in the field. The dataset and related resources can be accessed via the official IEEE GRSS IADF TC website1.

Natural hazards cause hundreds of billions of dollars in losses annually, with impacts that disproportionately affect vulnerable populations. Earth observation (EO) provides comprehensive, repeatable measurements that are essential for assessing disaster impacts when ground-based systems fail. Over the past two decades, methods for post-disaster mapping impacts from EO data have evolved rapidly, from manual interpretation and spectral indices to deep learning and, most recently, foundation models. Yet this methodological diversity has outpaced systematic evaluation: most existing studies focus on individual hazards, single tasks, or specific method families, and lack reproducible cross-method comparison. In this paper, we provide a unified review and benchmark for EO-based disaster mapping. We define a taskdriven taxonomy distinguishing disaster extent mapping from infrastructure damage mapping, and systematically review methods across both tasks, spanning traditional, deep learning, and foundation model paradigms. We then benchmark over 30 methods under a standardized protocol across 13 datasets covering diverse hazards, sensors, and operational settings. Our studies reveal that architectural complexity yields diminishing returns for extent mapping, that change detection architectures benefit damage mapping but not extent mapping, and that foundation models do not yet consistently outperform task-specific baselines, particularly on non-RGB modalities. Based on these findings, we identify key priorities for advancing operational disaster mapping. Code and benchmark resources are publicly available at https://github.com/ChenHongruixuan/AnyDisasterMapping

Rapid post-disaster response requires timely, building-level information on whether structures remain intact, are damaged, or are destroyed. Post-event optical imagery, however, may be unavailable because of cloud, smoke, or darkness. The Bright Challenge evaluated all-weather building damage mapping from a submeter-resolution pre-event optical image and a post-event SAR image. Participants were required to detect and delineate each building and assign exactly one of three mutually exclusive damage labels. The challenge extended the globally distributed \textsc{Bright} dataset with instance-level annotations for about 291,000 buildings across 16 disaster events spanning seven disaster types. The final phase was evaluated exclusively on two 2025 events absent from training: a wildfire event in California and a hurricane in Jamaica. A total of 157 participants made 1,289 submissions, and 46 teams entered the final phase. The two winning solutions achieved test mAPs of 0.182 and 0.181, approximately 8.7 times the public baseline of 0.021, but remained far below the best in-domain holdout score of 0.513. Across teams ranked in both phases, performance declined sharply and the rank order changed substantially. The two leading solutions independently favored modality-specific encoding, staged or late optical--SAR fusion, and an optical-dominant separation of building localization from damage recognition. The winning method additionally used scene-aware threshold adjustment and pseudo-label adaptation. These results identify cross-event generalization and stable severity discrimination as the principal remaining challenges. All data, annotations, baseline code, and winning solutions are publicly available at https://github.com/ChenHongruixuan/BRIGHT.

Local spectral features and global spatial context are essential for hyperspectral image (HSI) classification. However, existing methods based on convolutional neural networks (CNNs), graph convolutional networks (GCNs), and Transformers often rely on multibranch structures to separately extract and fuse local and global features, resulting in high computational complexity and redundant information that can negatively affect classification performance. To address these issues, we propose a two-stage graph convolutional mamba network (TGMN) that enables efficient modeling of local and global features through sequential intrasubgraph local feature extraction and intersubgraph global information learning. Specifically, in the first stage, we partition the HSI into superpixel regions and treat each superpixel as a subgraph, where a GCN is applied to aggregate spectral-spatial features within each subgraph. We further design a downsampled subgraph feature reconstruction (DSFR) module that dynamically selects key nodes to reduce redundancy, highlight critical features, and enhance model representation capability. In the second stage, the Mamba network models the global dependencies between subgraphs and introduces a region-relation aware absolute positional encoding (RAPE) module. This module encodes spatial positional information into embedded vectors by integrating the relative distance and direction between the geometric center of each superpixel and the image center, which are then deeply fused with the feature matrix to improve spatial relationship comprehension. The two-stage sequential structure ensures effective local and global feature extraction, avoiding the high computational complexity and redundancy issues commonly associated with multibranch models. Experiments on three benchmark datasets demonstrate its superiority, achieving classification accuracies of 98.54%, 98.30%, and 96.94% on the Indian Pines, Dioni, and Honghu datasets, respectively. Compared to state-of-the-art methods, TGMN achieves higher classification accuracy with significantly lower computational cost, demonstrating its efficiency and effectiveness for HSI classification.

Change Detection (CD) between images with different modalities is a fundamental capability for remote sensing. In this work, we pinpoint the commonalities between Multimodal Change Detection (MCD) and Multimodal Image Matching (MIM). Accordingly, we present a new unsupervised CD framework designed from the perspective of Image Matching (IM), called IM4CD. It unifies the IM and CD tasks into a single, coherent framework. In this framework, we abandon the prevalent strategy in MCD to compare per-pixel image features, since it is in practice quite difficult to design features that are truly invariant across modalities. Instead, we propose to compute similarity by local template matching and utilize the spatial offset of response peaks to represent change intensity between images with different modalities, and then to integrate it tightly with the co-registration of the two images, which anyway includes such a matching step. In this way, the same off-the-shelf descriptors used for MIM also support MCD. In other words, we first extract modality-independent features, then detect salient points to obtain initial pairs of corresponding Control Points (CP). When matching those points to accurately register the images, CP pairs located in unchanged areas show low residuals, whereas those in changed areas show high residuals. The CPs can then be connected into a Conditional Random Field (CRF), leveraging modality-independent structural relationships to estimate dense change maps. Experimental results show the effectiveness of our method, including robustness to registration errors, its compatibility with different image descriptors, and promising potential for challenging real-world disaster response scenarios.

Groundwater exploitation has intensified land subsidence risks, yet spatiotemporal pattern of county-level surface deformation and their implications for potential building damage in resource-based arid urban agglomerations remain poorly understood. Here, long-term surface deformation across 32 counties within the HohhotBaotou-Ordos-Yulin (HBOY) urban agglomeration is derived by multi-temporal InSAR from ALOS-1/PALSAR and Sentinel-1A SAR imagery. A building-level potential damage assessment method is developed to bridge the gap between InSAR-derived deformation and building risk. It links and aggregates deformation points to building footprints and applies a standardized matrix to assign building damage intensity levels. Median subsidence velocities exceeding 3 mm/year were detected in three cities during 2007-2011 and increased to seven cities during 2018-2022. Hohhot was subjected to the most severe surface subsidence, with median velocities of -6.6 mm/year and -5.9 mm/year measured in the two periods. Surface subsidence in all cities has predominantly exhibited a uniform pattern, which is largely attributed to regional groundwater extraction. In addition, the heterogeneity of surface subsidence among cities is controlled by geological lithology, soil texture and the thickness of compressible layers. Despite widespread subsidence, most buildings in HBOY were classified as low damage intensity. The proportion of buildings with high or extremely-high damage intensity decreased from 3.03% to 2.27% between the two study phases, mainly associated with subsidence funnels and differential deformation linked to residual subsidence of high-rise buildings on compressible soils. These findings can provide critical support for targeted urban subsidence management and sustainable development.

Road marking lines can be extracted from aerial images using semantic segmentation (SS) models; however, in this work, a conditional generative adversarial network, RoadMark-cGAN, is proposed for direct extraction of these representations with image-to-image translation techniques. The generator features residual and attention blocks added in a functional bottleneck, while the discriminator features a modified PatchGAN, with an optimized encoder and an attention block added. The proposed model is improved in three versions (v2 to v4), in which dynamic dropout techniques and a novel "Morphological Boundary-Sensitive Class-Balanced" (MBSCB) loss are progressively added to better handle the high class imbalance present in the data. All models were trained on a novel "RoadMarking-binary" dataset (29,405 RGB orthoimage tiles of 256 x 256 pixels and their corresponding ground truth masks) to learn the distribution of road marking lines found on pavement. The metrical evaluation on the test set containing 2045 unseen images showed that the best proposed model achieved average improvements of 45.2% and 1.7% in the Intersection-over-Union (IoU) score for the positive, underrepresented class when compared to the best Pix2Pix and SS models, respectively, trained for the same task. Finally, a qualitative, visual comparison was conducted to assess the quality of the road marking predictions of the best models and their mapping performance.

Vision-language models (VLMs) have shown remarkable promise in Earth Vision, particularly in providing human-interpretable analysis of remote sensing imagery. While existing VLMs excel at general visual perception tasks, they often fall short in addressing the complex needs of geoscience, which requires comprehensive urban analysis across geographical, social, and economic dimensions. To bridge this gap, we expand VLM capabilities to tackle sustainable urban development challenges by integrating two complementary sources: remote sensing (RS) and street-view (SV) imagery. Specifically, we first design a multi-view vision-language dataset (CitySet), comprising 20,589 RS images, 1.1 million SV images, and 0.8 million question-answer pairs. CitySet facilitates geospatial object reasoning, social object analysis, urban economic assessment, and sustainable development report generation. Besides, we develop CityVLM to integrate macro-and micro-level semantics using geospatial and temporal modeling, while its language modeling component generates detailed urban reports. We extensively benchmarked 10 advanced VLMs on our dataset, revealing that state-of-the-art models struggle with urban analysis tasks, primarily due to domain gaps and limited multi-view data alignment capabilities. By addressing these issues, CityVLM achieves superior performance consistently across all tasks and advances automated urban analysis through practical applications like heat island effect monitoring, offering valuable tools for city planners and policymakers in their sustainability efforts.

Open-vocabulary segmentation enables pixel-level recognition from an open set of textual categories, allowing generalization beyond fixed classes. Despite great potential in remote sensing, progress in this area remains largely limited to clear-sky optical data and struggles under cloudy or haze-contaminated conditions. We present MM-OVSeg, a multimodal Optical–SAR fusion framework for resilient open-vocabulary segmentation under adverse weather conditions. MM-OVSeg leverages the complementary strengths of the two modalities—optical imagery provides rich spectral semantics, while synthetic aperture radar (SAR) offers cloud-penetrating structural cues. To address the cross-modal domain gap and the limited dense prediction capability of current vision–language models, we propose two key designs: a cross-modal unification process for multi-sensor representation alignment, and a dual-encoder fusion module that integrates hierarchical features from multiple vision foundation models for text-aligned multimodal segmentation. Extensive experiments demonstrate that MM-OVSeg achieves superior robustness and generalization across diverse cloud conditions. All dataset and code will be publicly released.

Synthetic Aperture Radar (SAR) is a crucial remote sensing technology, enabling all-weather, day-and-night observation with strong surface penetration for precise and continuous environmental monitoring and analysis. However, SAR image interpretation remains challenging due to its complex physical imaging mechanisms and significant visual disparities from human perception. Recently, Vision-Language Models (VLMs) have demonstrated remarkable success in RGB image understanding, offering powerful open-vocabulary interpretation and flexible language interaction. However, their application to SAR images is severely constrained by the absence of SAR-specific knowledge in their training distributions, leading to suboptimal performance. To address this limitation, we introduce SARLANG-1M, a large-scale benchmark tailored for multimodal SAR image understanding, with a primary focus on integrating SAR with textual modality. SARLANG-1M comprises more than 1 million high-quality SAR image-text pairs collected from over 59 cities worldwide. It features hierarchical resolutions (ranging from 0.1 to 25 meters), fine-grained semantic descriptions (including both concise and detailed captions), diverse remote sensing categories (1,696 object types and 16 land cover classes), and multi-task question-answering pairs spanning seven applications and 1,012 question types. Extensive experiments on mainstream VLMs demonstrate that fine-tuning with SARLANG-1M significantly enhances their performance in SAR image interpretation, reaching performance comparable to human experts. The dataset and code will be made publicly available at https://github.com/Jimmyxichen/SARLANG-1M.

Over the past two decades, the annual Data Fusion Contest (DFC) has emerged as a platform that invites researchers worldwide to advance data fusion and image analysis methodologies. The contest focuses on the challenges of handling large-scale, multisensor, multimodal, and multitemporal data, fostering innovation in remote sensing research. The past editions have introduced novel and demanding problem settings, thereby creating benchmarks that have shaped progress in the field. The contest has been hosted annually since 2006 by the Image Analysis and Data Fusion Technical Committee (IADF TC) of the IEEE Geoscience and Remote Sensing Society (GRSS) [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21]. This year’s contest focus was all-weather land cover and building damage mapping.

Large vision-language models (VLMs) have made great achievements in Earth vision. However, complex disaster scenes with diverse disaster types, geographic regions, and satellite sensors have posed new challenges for VLM applications. To fill this gap, we curate a remote sensing vision-language dataset (DisasterM3) for global-scale disaster assessment and response. DisasterM3 includes 26,988 bi-temporal satellite images and 123k instruction pairs across 5 continents, with three characteristics: 1) Multi-hazard: DisasterM3 involves 36 historical disaster events with significant impacts, which are categorized into 10 common natural and man-made disasters. 2)Multi-sensor: Extreme weather during disasters often hinders optical sensor imaging, making it necessary to combine Synthetic Aperture Radar (SAR) imagery for post-disaster scenes. 3) Multi-task: Based on real-world scenarios, DisasterM3 includes 9 disaster-related visual perception and reasoning tasks, harnessing the full potential of VLM's reasoning ability with progressing from disaster-bearing body recognition to structural damage assessment and object relational reasoning, culminating in the generation of long-form disaster reports. We extensively evaluated 14 generic and remote sensing VLMs on our benchmark, revealing that state-of-the-art models struggle with the disaster tasks, largely due to the lack of a disaster-specific corpus, cross-sensor gap, and damage object counting insensitivity. Focusing on these issues, we fine-tune four VLMs using our dataset and achieve stable improvements across all tasks, with robust cross-sensor and cross-disaster generalization capabilities. The code and data are available at: https://github.com/Junjue-Wang/DisasterM3.

Timely building damage detection is crucial in disaster decision-making. Unlike existing studies using sub-meter resolution images, we employ stable, mid-high resolution (1-3 meters) remote sensing data to develop building flood damage classification methods for quick disaster response. A mid-high-resolution dataset of 973 Planet-sliced images from three floods in different countries is presented to facilitate building damage classification. Based on this, we propose a Cross Fusion Pyramid Swin Transformer (CFP-SwinT), which balances local and global representations in flood scenes. The model employs feature pyramids to classify multi-scale building groups, and integrates spatial-channel features from dual-time images via a cross-fusion block to jointly capture building groups (pre-disaster) and damage severity (post-disaster). Through extensive experiments, CFP-SwinT outperformed all other backbones and settings and achieved 80.2% accuracy on four-level classification, demonstrating its effective focus on multi-scale disaster information and dual-temporal changes. To enable rapid adaptation to different disaster-affected regions, we combined an unsupervised domain adaptation method with CFP-SwinT, allowing a model trained on one disaster to achieve 77.8% accuracy on unseen flood events. This approach avoids sample labeling, which can benefit decision-making efficiency during disasters. The code for the proposed method is available at https://github.com/WuChuYi/Cross-Fusion-Pyramid-Swin-Transformer.

In the context of Earth observation, the trade-off between spatial, spectral, and temporal resolution often limits the versatility of remote sensing images in many important applications. In response, this paper introduces a novel deep learning diffusion model, specifically tailored to improve the spatial resolution of the optical products acquired by the Sentinel-3 (S3) satellite. Our framework employs a diffusion probabilistic model, benefiting from the higher spatial resolution of the Sentinel-2 satellite during training via a new multi-modal loss formulation. This ensures consistency with the original S3 images while enhancing the spatial details. Two distinct conditional low-resolution encoders were experimented with, providing insights into their respective contributions to the diffusion process. The efficacy of the proposed model is demonstrated through extensive ablation studies and comparisons with state-of-the-art methods, using both synthetic and real S3 products. The findings indicate that our model successfully improves spatial resolution while maintaining the integrity of the spectral information, contributing to the field of remote sensing single-image super-resolution.

Disaster events occur around the world and cause significant damage to human life and property. Earth observation (EO) data enables rapid and comprehensive building damage assessment, an essential capability crucial in the aftermath of a disaster to reduce human casualties and inform disaster relief efforts. Recent research focuses on developing artificial intelligence (AI) models to accurately map unseen disaster events, mostly using optical EO data. These solutions based on optical data are limited to clear skies and daylight hours, preventing a prompt response to disasters. Integrating multimodal EO data, particularly combining optical and synthetic aperture radar (SAR) imagery, makes it possible to provide all-weather, day-and-night disaster responses. Despite this potential, the lack of suitable benchmark datasets has constrained the development of robust multimodal AI models. In this paper, we present a Building damage assessment dataset using veRy-hIGH-resoluTion optical and SAR imagery (Bright) to support AI-based all-weather disaster response. To the best of our knowledge, Bright is the first open-access, globally distributed, event-diverse multimodal dataset specifically curated to support AI-based disaster response. It covers five types of natural disasters and two types of human-made disasters across 14 regions worldwide, focusing on developing countries where external assistance is most needed. The dataset's optical and SAR images with spatial resolutions between 0.3 and 1 m provide detailed representations of individual buildings, making it ideal for precise damage assessment. We train seven advanced AI models on Bright to validate transferability and robustness. Beyond that, it also serves as a challenging benchmark for a variety of tasks in real-world disaster scenarios, including unsupervised domain adaptation, semi-supervised learning, unsupervised multimodal change detection, and unsupervised multimodal image matching. The experimental results serve as baselines to inspire future research and model development. The dataset (https://doi.org/10.5281/zenodo.14619797, Chen et al., 2025a), along with the code and pretrained models, is available at https://github.com/ChenHongruixuan/BRIGHT (last access: 7 November 2025) and will be updated as and when a new disaster data is available. Bright also serves as the official dataset for the 2025 IEEE GRSS Data Fusion Contest Track II. We hope that this effort will promote the development of AI-driven methods in support of people in disaster-affected areas.

Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.

The principle of the image fusion is to integrate complementary information of the heterogeneous images to obtain a fused image that is more in line with the visual effect of the human eyes. However, most decomposition methods cannot distinguish the textures and edges in an image, which is easy to produce the halo artifacts around edges. In this paper, we proposed a novel image decomposition strategy (co-occurrence analysis shearlet transform, CAST) to preprocess the input images depending on the co-occurrence statistic information to generate the base layer and detail layer components. In order to improve the sparseness of the base layer, the classified sparse dictionary in the measurement domain is introduced to enhance the robustness of incorrect registration. As for the detail layers, the adaptive dual-channel PCNN model is adopted as the fusion rule, in which the neurons are activated by the improved spatial frequency operator (ISF), and the model uses the sum of the improved weighted Laplacian energies (IWSWL) as the adaptive linking strength. Finally, the fused image can be generated by the inverse CAST. Based on the combination of the sparseness of the classified dictionary and the visual characteristics of PCNN model, the more valuable information of the source images can be fused, so that the final fused images conform to the human visual system. Qualitative and quantitative experimental results demonstrate the superiority of the proposed method over other typical fusion techniques on the publicly available datasets.

Climate change and human activities have significantly altered the dynamics of surface water area (SWA) in the Irtysh River Basin (IRB). While inter-annual trends in SWA can be detected using Landsat imagery, the characteristics of seasonal SWA changes under long-term scenarios remain uncertain due to reduced data availability caused by cloud cover. In this study, we propose a time-disaggregated water frequency (TWF) that is more suitable for seasonal surface water analysis and develop a cloud-filling algorithm utilizing a Random Forest approach. The results demonstrate that the TWF effectively represents seasonal surface water distribution and achieves high cloud-filling accuracy. Using this method, we reconstructed monthly cloud-filled SWA series for the IRB from 1985 to 2022 at a spatial resolution of 30 m with high accuracy (>94%). Analysis indicates that the multi-year average SWA of the IRB was 41,003 km(2), reflecting a decrease of 22%. The peak SWA occurs in spring (May), following the general trend of spring > summer > fall > winter. Surface water loss primarily occurs during summer and fall, particularly in the middle reaches of the Irtysh River Basin (35%). Time-series correlation analysis reveals that snowmelt, precipitation, and temperature are the most significant climatic factors affecting SWA in spring, summer, and fall.

High-resolution land cover mapping is vital for global challenges such as urban planning, environmental monitoring, disaster response, and sustainable development. Yet, building accurate large-scale datasets remains difficult due to complex terrain, multimodal sensor data, and varying atmospheric conditions. Synthetic aperture radar (SAR), with its cloud-penetrating, all-weather, and day-and-night imaging capabilities, provides a unique advantage, but progress has been hindered by the lack of tailored benchmark datasets. In this article, we introduce OpenEarthMap-SAR, a large-scale benchmark dataset for high-resolution land cover mapping using SAR imagery. OpenEarthMap-SAR includes 1.5 million segments across 5,033 aerial and satellite images (1,024 x 1,024 pixels) from 35 regions in Japan, France, and the United States. It features partially manually annotated and fully pseudo-labeled eight-class land cover labels at a ground sampling distance (GSD) of 0.15-0.5 m. We evaluate state-of-the-art semantic segmentation, unsupervised domain adaptation (UDA), and Image2Image/Label2Image translation methods and propose challenging benchmark settings to foster future development. OpenEarthMap-SAR also serves as the official dataset for track 1 of the IEEE Geoscience and Remote Sensing Society (GRSS) Data Fusion Contest, specifically for the task of semantic segmentation. The dataset and source codes with pretrained models have been made publicly available at https://zenodo.org/records/14622048 and https://github.com/cliffbb/OpenEarthMap-SAR.