
Accurate monitoring of methane emissions is essential for mitigating global warming. Due to high spectral response, hyperspectral remote sensing technology plays an important role in precise detection and concentration estimation of methane emissions. However, existing detection and quantification approaches fail to consider discrepancies in hardware imaging performance and real emission scenarios, posing challenges to achieve effective plume-background separation and stable inversion. To this end, we propose a multi-satellite cooperative hyperspectral intelligent method for methane emission monitoring. This method introduces robust matched filter retrieval and cross-sensor detection mechanisms to obtain complete methane plumes, thereby avoiding sensor-specific sensitivity and environment interference. Furthermore, the methane emission flux and its uncertainties are simultaneously estimated to generate standardized methane dataset products. We expect this technique can provide new investigation and development directions for the experienced researchers who understand hyperspectral methane monitoring to some extent. It further provides high-resolution reliable data foundation for estimating global methane emissions based on multi-satellite hyperspectral platforms, reducing the uncertainty of existing emission inventories and hotspot event tracking. The code and data will be available at https://github.com/hanzhu97702/MSHMM, contributing to the remote sensing community.
Building height (BH) is a fundamental descriptor of urban form and a key variable for understanding urban growth, energy consumption, and carbon emissions. The increasing availability of satellite observations has substantially advanced remote sensing-based BH estimation (BHE), leading to diverse methodological developments over the past two decades. This article provides a systematic review of current BHE research. We categorize existing BHE methods into geometric, physical-statistical, and artificial intelligence (AI)-based methods and compare their principles, advantages, limitations, and applicable scenarios. We identify a methodological shift from geometry-based approaches to data-driven paradigms, with deep learning (DL) becoming increasingly dominant and large multimodal models (LMMs) emerging as a new frontier. Based on the reviewed literature, we synthesize BHE accuracy across different estimation units. We further summarize 17 representative large-scale BH datasets by estimation method, accuracy, spatial resolution, temporal coverage, and height detection capability. These datasets reveal two major trends: finer spatial granularity, from grids to individual buildings, and expanded temporal coverage, from single-year to time-series mapping. Finally, we discuss key challenges in the BHE domain: limited model generalization, height estimation challenges in mixed and occluded built-up areas, and the lack of uncertainty quantification in most existing studies and BH products. We also outline promising future directions to advance robust, generalizable, and uncertainty-aware BHE and highlight urban agents as a promising frontier for autonomous BHE and intelligent urban analysis. Overall, this article provides a structured overview of the BHE domain and serves as a reference for future BHE research.
Hyperspectral image (HSI) fusion aims to reconstruct high-spatial-resolution (HR) HSIs from low-resolution HSI observations and complementary high-resolution guidance. Although these tasks share the goal of enhancing spatial details while preserving spectral fidelity, they differ in observation configuration and native spectral layout, making unified modeling challenging. Most existing methods remain reconstruction driven and open loop, with semantic interpretation rarely involved in fusion. This article proposes an observation-conditioned semantic closed-loop (OCSC) framework for unified HSI fusion in Earth observation (EO) by treating fusion tasks as configuration-dependent inverse problems. A compact observation descriptor represents task- and sensor-related differences, while sensor-specific spectral adapters connect native HSI spaces with a shared latent spectral space. This enables shared reconstruction while preserving native spectral identity. In this space, fusion is performed on a spectral manifold, where scene-adaptive spectral bases describe spectral structure and high-resolution coefficient maps encode spatial variations. Guidance details selectively refine the coefficient maps according to estimated local reliability and spectral compatibility. A dual-closed-loop mechanism maintains consistency with the input observations in the inner loop and provides confidence-weighted semantic feedback in the outer loop. Experiments on seven public benchmarks show advantages in fusion fidelity and downstream interpretation. Compared with the best competing results on the same datasets and metrics, the proposed framework reduces the spectral angle mapper (SAM) by an average of 2.75% on five reference-based fusion benchmarks and improves the downstream segmentation mean intersection over union (mIoU) by 3.57%. Experiments on a spatially disjoint dataset constructed from a Dongfang Huiyan Hyperspectral-01 (DFHY-HS01) on-orbit scene further assess the framework under real satellite observations. The code, evaluation protocols, and DFHY-HS01 satellite dataset are publicly available at https://github.com/JUSTM0VE0N/RSMIF_Project.
Soil degradation (SD) threatens ecosystem services, creating an urgent need for scalable and consistent monitoring. Remote sensing offers unique potential for repeated and multiscale observation across heterogeneous landscapes. However, all-type SD monitoring remains challenging because physical, chemical, and biological degradation processes differ greatly in observability. This article reviews remote sensing of SD types and organizes existing studies around two evidence streams: direct indicators and indirect proxies. Direct indicators [e.g., soil mineral composition, soil organic matter (SOM), soil surface roughness, and soil moisture] provide process-related evidence but are constrained by point-to-pixel scale mismatch and limited transferability. Indirect proxies (i.e., vegetation condition and land-use/land-cover change) enable broad-area screening but are vulnerable to environmental and anthropogenic confounders. Consequently, future research should focus on direct indicators, indirect proxies, and all-type SD monitoring, with emphasis on spatial superresolution through multisource/multitemporal fusion, an attribution-oriented framework, and foundation models for unified SD monitoring.
Change detection (CD) is an essential task in the field of remote sensing (RS), supporting applications such as urban management, disaster monitoring, resource surveys, and environmental assessment. CD involves cross-temporal feature alignment, consistency modeling, change discrimination, and semantic reasoning while also requiring robust and generalizable representations across multisource and multiscale data. With the rapid advancement of foundation models (FMs), large-scale pretrained models have demonstrated significant advantages in representation capability, cross-scene transferability, and data efficiency. These advances provide new opportunities for advancing research in CD tasks. This survey focuses on the emerging field of FMs for CD (FM4CD). This article outlines the background and motivation for incorporating FMs into CD tasks. Then, we systematically overview representative CD methods from the perspectives of FMs, summarizing their characteristics and advantages in change modeling and semantic understanding. Subsequently, this article analyzes key factors in FMs for CD tasks, including dataset scale, adaptation strategies, and generalization capabilities. This article aims to provide researchers with a systematic analysis to advance the development of FMs for RSCD, thereby driving the expansion of CD technology into complex and diverse application scenarios. We will continuously track and update related works at https://github.com/Jscript10/Awesome-FM4CD.