Convolutional neural networks (CNNs) and Transformer architectures are among the most popular techniques for hyperspectral image super-resolution (HSI-SR). However, both architectures suffer from inherent deficiencies: CNNs are constrained to a limited receptive field, which hinders their ability to capture a wider range of spatial contexts, while Transformers are computationally intensive, making them expensive to train and deploy in large-scale scenes. Furthermore, existing methods tend to employ end-to-end spectral and spatial modeling paradigms for the HSI-SR task while neglecting the natural spectral patterns of high-dimensional HSIs. To overcome the above problems, this paper proposes a new progressive local–global state-space model for HSI-SR, named PLGMamba, which fully takes into account feature extraction and model design from a local to a global perspective. Initially, a progressive learning paradigm to extract spectral sequences (from local to global) is applied to the low-resolution HSI to ensure spatial and spectral fidelity. Then, a residual attention Mamba (RatMamba) is developed to capture the local–global spectral–spatial features of the reconstructed HSI, which integrates spectral–spatial local perception and global modeling into a unified residual learning framework. Finally, a residual Mamba (ResMamba) is designed to fuse progressive local spectral–spatial features and obtain the final high-resolution HSI. Experiments on 3 benchmark HSIs and 1 Gaofen-5 HSI demonstrated that the proposed PLGMamba outperforms classical and state-of-the-art HSI-SR approaches including the mainstream CNN, Transformer, and Mamba. The proposed method is available online: https://github.com/chengle-zhou/PLGMamba .

Recently, large vision-language models (VLMs) have made marked strides in vision-language capabilities through visual instruction tuning, showing great promise in the field of remote sensing image interpretation. However, existing remote sensing vision-language models (RSVLMs) often fall short in capturing the complex characteristics of remote sensing scenes. While some recent RSVLMs, such as EarthGPT and LHRSBot, have started to incorporate multiscale features, they typically adopt relatively shallow fusion strategies and lack tightly integrated mechanisms for fine-grained vision-language alignment. In this paper, we present Aquila, an advanced vision-language foundation model designed to enable richer visual feature representation and more precise vision-language feature alignment for remote sensing images. Our approach introduces a learnable hierarchical spatial feature integration (SFI) module that supports high-resolution image inputs and aggregates multi-scale visual features, allowing for the detailed representation of complex visual information. Additionally, the SFI module is repeatedly integrated into the layers of the large language model (LLM) to achieve deep vision-language feature alignment, without compromising the model’s performance in natural language processing tasks. These innovations significantly improve the model’s ability to learn from image–text data. We validate the effectiveness of Aquila through extensive quantitative experiments and qualitative analyses, demonstrating its superior performance.

Land cover mapping plays a critical role in monitoring land system changes. Despite advancements in remote sensing technologies, traditional satellite-based approaches are often constrained by cloud cover, coarse temporal resolution, and limitations in capturing fine-scale landscape dynamics, leading to gaps in continuous and real-time monitoring. Near-surface cameras offer a solution by providing high-frequency, ground-level observations that bridge temporal gaps and enhance spatial detail. Therefore, this study makes substantial contributions by pioneering the integration of near-surface camera observations with satellite imagery, addressing key challenges in imaging perspective differences and limited coverage of ground-based observations for enhanced land cover monitoring at a 30-m/10-m scale. A key innovation lies in leveraging near-surface cameras to reconstruct dense satellite data time series and capture daily land cover dynamics, addressing critical temporal gaps in traditional satellite-based approaches. The research further advances the field by implementing state-of-the-art deep learning techniques, particularly the Segment Anything Model (SAM), to achieve precise parcel-level delineation and reduce classification noise at a high-resolution (meter- and submeter level) scale. Furthermore, the framework’s ability to synthesize multimodal data sources (near-surface cameras, Sentinel-1/2, and high-resolution imagery) represents a methodological breakthrough in space and surface sensor integration for real-time land cover change detection, enabling time-sensitive applications and early warning systems for land system changes.

The top-of-atmosphere (TOA) Earth-reflected shortwave radiation flux is a crucial component of the Earth’s radiation budget (ERB). The Earth polychromatic imaging camera onboard the Deep Space Climate Observatory (DSCOVR/EPIC) provides a unique Earth observation perspective from the Sun–Earth Lagrange point 1. Traditionally, angular distribution models (ADMs) are required to account for the radiation anisotropy. However, no ADMs are available intended for DSCOVR/EPIC, while the development and application of ADMs involved complicated procedures. With an attempt to simplify and improve the derivation of TOA shortwave flux, this study proposed a machine learning-based approach to interpret the DSCOVR/EPIC data and evaluate its application potential in monitoring the global shortwave flux. With the Clouds and the Earth’s Radiant Energy System (CERES) products as the benchmark, 36 neural network models were developed for each scene type and air condition to estimate the TOA shortwave flux from DSCOVR/EPIC, and a model was then developed to relate the EPIC-derived flux to global TOA shortwave flux. Results showed that the neural network models worked well in estimating the TOA shortwave flux, which produced consistent spatial distributions with CERES products across scene level, daily, and monthly scales. With the developed models, the EPIC-derived flux could account for 97% variations of global TOA shortwave flux at the daily scale, which were better than the EPIC-L2 albedo product. Overall, this study demonstrated the promising potential of deep space-based Earth observation for ERB monitoring, and the proposed method holds potentials for geostationary satellites and Moon-based Earth observations.

Tree height is a key indicator in forest ecology, reflecting tree growth status and ecosystem structure. Traditional methods of tree height measurement rely on ground-based measurements, which are limited by cost and time. In recent years, the development of machine learning and multi-source remotely sensed technologies has provided new ways to measure tree height. In this study, we utilized light detection and ranging and satellite data to extract spectral, vegetation, texture, polarization, terrain, and season features. By integrating these features with machine learning, deep learning, and optimization methods, we dynamically estimated tree heights in Shenzhen during summer and winter from 2018 to 2023 and validated seasonal and regional scalability. It was found that (a) the seasonal tree height neural network demonstrated the highest prediction accuracy in tree height estimation ( R 2 = 0.72, mean absolute error = 1.89 m), and the optimization process of Shapley additive explanations reduced 23 features, which improved the prediction accuracy ( R 2 = 0.80, mean absolute error = 1.58 m) and saved computational resources; (b) the seasonal tree height neural network has a strong generalizability for estimating tree height across seasons and regions; and (c) during 2018 to 2023, tree heights in Shenzhen were mainly concentrated in 6 to 14 m, and the spatial distribution has a strong autocorrelation. Tree canopy heights in winter are generally lower than those in summer, and the tree growth rate shows spatial heterogeneity. Overall, this study uncovers the intricate interplay between tree growth and seasonal variations in its traits throughout the urbanization process in Shenzhen. It offers valuable data support and a theoretical foundation for urban greening management and ecological protection.

Forest canopy height reflects the vertical structure of forests and gives indications on the growth capacity of trees and the level of forest biomass. Despite an increasing availability of global canopy height maps, there is a lack of maps reflecting temporal dynamics, which is required to understand forest carbon sink capacity. This study uses publicly available global maps and Landsat data to construct a long time series of canopy height maps in southern China (1986 to 2019). Our predictions are in the same range as National Forest Inventory data, both spatially and temporally. The dataset shows clear signs of tree growth, with small trees dominating the first period, which grow taller over the 3 decades. In numbers, the mean canopy height of the forests in southern China increased from 6.4 ± 3.99 m in 1986 to 10.3 ± 5.54 m in 2019, which is an increase of 61.25%. Secondary forests have an overall higher canopy height (17.8 ± 2.12 m) as compared to plantations (14.6 ± 2.05 m), but plantations grow faster (0.20 ± 0.08 m/year) as compared to secondary forests (0.13 ± 0.08 m/year). A driver analysis shows that management is the dominating factor for the canopy height increase, while natural factors play a minor role. Our study reaffirms the human made surge in forest areas in China, quantifies their canopy height dynamics, and demonstrates that spectral data can indeed be used to track canopy height changes at scale.

Spartina alterniflora has become the most problematic invasive species in China’s coastal regions due to its rapid growth, robust reproductive capacity, and extensive adaptability. It has importantly disrupted the structure and function of coastal wetland ecosystems, thereby posing a serious threat to the ecological security of these wetlands . China is currently engaged in a nationwide initiative to manage the invasive species S. alterniflora . An accurate and up-to-date understanding of the current distribution and dynamic changes of S. alterniflora is essential for formulating effective control measures. Remote sensing technology has enabled the rapid, large-scale monitoring of S. alterniflora . However, traditional remote sensing methods typically focus on single-period images of specific small- to medium-scale areas and depend heavily on a substantial number of training samples. Consequently, these methods exhibit weak model transferability and poor generalization capabilities, rendering them unsuitable for the fine-scale identification of S. alterniflora across extensive regions. This research proposed an S. alterniflora index (SAI) derived from Sentinel-2 imagery. The SAI was constructed using the Sentinel-2 Red and near-infrared (NIR) bands, formulated as (Red-NIR)/NIR, to accentuate the distinctions in greenness and moisture between S. alterniflora and other land cover types. This study surveyed 6 representative S. alterniflora distribution areas along the coastal regions of China. We compared the S. alterniflora extraction results using SAI with those obtained using common vegetation indices, sensitive bands, and classic machine learning-based methods. The results demonstrate that SAI surpasses other vegetation index and sensitive bands in extracting S. alterniflora , showing performance comparable to that of support vector machine. Furthermore, we applied this index to Landsat-8 images to test its performance on different datasets. We also validated its effectiveness for both native and invasive Spartina spp. habitats worldwide. Finally, we conducted S. alterniflora extraction across coastal regions of China, acquiring a 2020 dataset with a 10-m resolution. Comparative analysis with official statistics and existing datasets yielded favorable results. Therefore, the proposed method in this study shows promising potential for application in S. alterniflora monitoring, providing technical support for effective management and enhanced protection of coastal wetlands.

Agricultural film plays a vital role in enhancing land productivity. However, concerns have arisen regarding its impact on ecology and soil environment. Accurate and timely agricultural film maps are critical for supporting strategic agricultural planning and predicting environmental impacts. In this paper, we summarized the current status of agricultural film mapping, including plastic greenhouses (PGs) and plastic-mulched farmland (PMF), from the evolution of remote sensing data, sample sources, spectral-temporal-spatial features, and advantages and disadvantages of classification algorithms. The findings revealed that medium- and low-resolution images were used for large-scale PGs and PMF mapping, while high-resolution images were combined with deep learning to extract local deep information. The synergy between the spectral, temporal, and spatial features can definitely improve classification accuracy, especially through object-based classification methods. Deep learning has apparent advantages than traditional machine learning algorithms in extracting PGs details, rarely used for mapping PMF. There are some problems, i.e., the diversity of film types, the difference of coverage time, and the variation of spectral properties, which lead to the scarcity of large-scale PGs and PMF maps despite numerous efforts in agricultural film mapping. To advance the field, future directions should focus on combining multi-source data, collaborating spectral-temporal-spatial features to extract types, start-end dates, and durations of mulching, and expanding from local to national or global scales. The accurate and timely agricultural film maps are expected to support effective land management, rationalize human land use behavior, and inform policy formulation for environmental sustainability.

Plantations for producing profitable products play a critical role in supporting local livelihoods, which requires accurate monitoring to guide sufficient management activities. Additionally, recent expansion of plantations for sequestering atmospheric CO 2 motivated by China Certified Emission Reduction (CCER) guideline further underscore the urgent need for accurate and cost-effective methods to estimate plantation aboveground biomass (AGB). High-resolution canopy height maps (CHMs) are essential for capturing detailed plantation traits for AGB estimation, especially given the typically small scale of plantations. While airborne or unmanned aerial vehicle (UAV)-based lidar remains the gold standard for acquiring high-resolution CHMs, its high cost limits widespread use. With advancements in deep learning, predicting CHMs using remote sensing RGB data has emerged as a cost-effective alternative, although challenges remain in accurately extracting canopy height-related features. To address these challenges, we develop a novel model for high-resolution CHM generation based on large vision foundation model (LVFM). This model integrates a feature extractor, a self-supervised feature enhancement module to avoid spatial detail loss in feature extraction, and a height estimator to produce high-resolution CHMs. Tested in the Fangshan District of Beijing, China—a region characterized by small, fragmented plantation parcels—our model, utilizing high-resolution (1-m grid) RGB imagery from Google Earth, demonstrated superior performance compared to existing methods, including conventional convolutional neural networks (CNNs) and naive LVFM implementations. The model achieved a mean absolute error of 0.09 m, a root mean square error of 0.24 m, and a correlation coefficient of 0.78 when evaluated against lidar-based CHM observations in pixel-wise assessments. Our model also exhibits satisfactory performance when being generalized into nontraining regions. Additionally, the CHMs generated by our model enabled over 90% success in individual tree detection and showed high accuracy in AGB estimation and a reasonable performance in tracking plantations’ growth. Our approach offers a promising tool for evaluating carbon sequestration in plantations and natural forests covering a large region.