Search Results (1,761)

In video scene graph prediction, the aim is to capture structured object interactions that occur over time in dynamic visual content. While recent spatiotemporal attention-based models have improved performance, they often suffer from high computational costs and limited structural consistency across long sequences. Therefore, we developed a Fourier transformer mixer network (FTM-Net), a modular, frequency-aware architecture that integrates spatial and temporal modeling via spectral operations. It incorporates a resolution-invariant Fourier Mixer for global spatial encoding and a Fast Fourier Transform (FFT)-Net-based temporal encoder that efficiently represents long-range dependencies with less complexity. To improve structural integrity, we introduce a spectral consistency loss function that synchronizes high-frequency relational patterns between frames. Experiments conducted utilizing the Action Genome dataset demonstrate that FTM-Net surpasses previous methodologies in terms of both Recall@K and mean Recall@K while markedly decreasing parameter count and inference duration, providing an efficient, interpretable, and generalizable approach for structured video comprehension. Full article

Cyclone Global Navigation Satellite System (CYGNSS) has emerged as an effective technique for inland water body detection due to its high sensitivity to inland waters. However, existing methods for inland water body detection using CYGNSS are limited by the difficulty in balancing high spatiotemporal resolution with strong generalization capability. Moreover, the limited spatial redundancy in short-term CYGNSS data restricts its capacity for high-precision inland water detection on its own. To address these issues, this study proposed a novel dual-branch model, termed STRUE. The model integrated a Swin Transformer and ResNet within a U-Net-enhanced student-teacher framework. This framework was developed through the fusion of multi-source data, including CYGNSS, SMAP, FABDEM, MODIS, and GSWE. The results showed that, for inland water body detection, the model attained a spatial resolution of 0.01° and a temporal resolution of 7 days. In terms of performance, it achieved an F1-score (F1) of 0.914, a mean Intersection over Union (mIoU) of 0.880, a Matthews Correlation Coefficient (MCC) of 0.873, and a Recall (R) of 0.963. Additionally, compared with traditional methods and models, the proposed model demonstrated a better performance in spatial continuity, structural integrity, and detail recovery, while mitigating common limitations such as cloud obscuration, spatial incoherence, and overestimation artifacts. These results further enhance the capacity of spaceborne GNSS-R for inland water body detection. Full article

Accurate precipitation mapping is essential for effective disaster management; however, individual radar, satellite, and numerical weather prediction products often struggle in the topographically complex terrain of South Korea. This study proposes a high-resolution (~500 m) daily precipitation fusion framework that integrates Korea Meteorological Administration (KMA) radar, Global Precipitation Measurement (GPM) Integrated Multi-Satellite Retrievals for GPM (IMERG), and Local Data Assimilation and Prediction System (LDAPS) data. The framework employs a Random Forest model augmented with a monthly Empirical Cumulative Distribution Function (ECDF) correction. Auxiliary predictors are incorporated to enhance physical interpretability and stability, including terrain attributes to represent orographic effects, land-cover information to account for surface-related modulation of precipitation, and seasonal cyclic signals to capture regime-dependent variability. These predictors complement dynamic precipitation inputs and enable the model to effectively capture nonlinear spatiotemporal patterns, resulting in improved performance relative to individual radar, IMERG, and LDAPS products. Evaluation against Automated Synoptic Observing System (ASOS) observations yielded a correlation coefficient of 0.935 and a mean absolute error of 3.304 mm day⁻¹ in a Leave-One-Year-Out (LOYO) validation for 2024. Regional analyses further indicate substantial performance gains in complex mountainous areas, including the Yeongdong–Yeongseo region, where the proposed framework markedly reduces estimation errors under challenging winter conditions. Overall, the results demonstrate the potential of the proposed fusion framework to provide robust, high-resolution precipitation estimates in regions characterized by strong topographic and seasonal heterogeneity, supporting applications related to hazard analysis and hydrometeorological assessment. Full article

To host the 2022 Winter Olympics, Beijing and Zhangjiakou implemented extensive ecological restoration projects, improving the ecological quality of the region. However, detailed evidence of long-term spatiotemporal dynamics in vegetation productivity remains limited. This study employed the Carnegie–Ames–Stanford Approach (CASA) to estimate the vegetation Net Primary Productivity (NPP) in the Beijing–Zhangjiakou region from 2004 to 2023, utilizing 250 m monthly NDVI data. The 30 m resolution China Land Cover Dataset (CLCD) was incorporated to mask non-vegetated pixels and refine the vegetation mask, reducing mixed-pixel effects. Spatiotemporal variations, seasonal change-point detection, interannual stability, and trend persistence were analyzed across administrative regions and land cover types. Results indicate pronounced spatial heterogeneity in NPP, with persistently high values in forest-dominated western and northern Beijing and northeastern Zhangjiakou, and lower values concentrated in Beijing’s built-up and cropland-dominated southeastern plain. Pixel-level boxplots suggest stronger intra-regional variability in Beijing than in Zhangjiakou. Across landcover types, forests generally maintain the highest NPP, while grasslands are relatively lower. Boxplots further show that shrubs exhibit the highest variability, with all types showing right-skewed distributions. Annual mean NPP increased significantly for the entire region, Beijing, and Zhangjiakou, with interannual increase rates of 3.57, 1.56, and 4.53 gC·m⁻²·yr⁻², respectively; the lowest values occurred in 2007 and the highest in 2022. Trend maps and category statistics consistently suggest that positive trends dominate most of the region and expanded slightly during 2014–2023. BEAST analysis suggests a stable seasonal NPP cycle with no significant seasonal change points. CV-based assessment indicates generally high to extremely high stability, whereas low-stability zones are mainly associated with urban expansion areas, surrounding croplands, and parts of Zhangjiakou grasslands. Hurst results suggest that persistently increasing trends cover more than 90% of the study area, while persistently decreasing trends account for about 5.25% and are primarily linked to Beijing’s expansion zones. Full article

As a core city in central China and a key node of the Changsha–Zhuzhou–Xiangtan (CZT) Metropolitan Area, Changsha has experienced profound territorial space restructuring amid rapid urbanization and high-quality development. This study focuses on the spatiotemporal evolution characteristics, driving mechanisms, and future optimization paths of production–living–ecological space (PLES) in Changsha, using three key time nodes: 2010, 2020, and 2025. Based on updated land use data (30 m spatial resolution), socioeconomic statistics, and the latest territorial spatial planning policies, we integrated multiple research methods including the land use transfer matrix, dynamic degree model, Logistic regression, and FLUS (Future Land Use Simulation) model. The results reveal the evolutionary law of PLES space from “rapid expansion” (2010–2020) to “quality improvement” (2020–2025) in Changsha and simulate the 2035 PLES layout under three scenarios (natural development, cultivated land protection, and ecological protection) incorporating rigid policy constraints such as urban development boundaries and ecological conservation red lines. This research provides updated scientific support for the coordinated and sustainable development of territorial space in new first-tier cities and metropolitan area cores. Full article

Timely access to soil moisture conditions in farmland crops is the foundation and key to achieving precise irrigation. Due to their high spatiotemporal resolution, unmanned aerial vehicle (UAV) remote sensing has become an important method for monitoring soil moisture. This study addresses soil moisture retrieval in alfalfa fields across different growth stages. Based on UAV multispectral images, a multi-source feature set was constructed by integrating spectral and texture features. The performance of three machine learning models—random forest regression (RFR), K-nearest neighbors regression (KNN), and XG-Boost—as well as two ensemble learning models, Voting and Stacking, was systematically compared. The results indicate the following: (1) The integrated learning models generally outperform individual machine learning models, with the Voting model performing best across all growth stages, achieving a maximum R² of 0.874 and an RMSE of 0.005; among the machine learning models, the optimal model varies with growth stage, with XG-Boost being the best during the branching and early flowering stages (maximum R² of 0.836), while RFR performs better during the budding stage (R² of 0.790). (2) The fusion of multi-source features significantly improved inversion accuracy. Taking the Voting model as an example, the accuracy of the fused features (R² = 0.874) increased by 0.065 compared to using single-texture features (R² = 0.809), and the RMSE decreased from 0.012 to 0.005. (3) In terms of inversion depth, the optimal inversion depth for the branching stage and budding stage is 40–60 cm, while the optimal depth for the early flowering stage is 20–40 cm. In summary, the method that integrates multi-source feature fusion and ensemble learning significantly improves the accuracy and stability of alfalfa soil moisture inversion, providing an effective technical approach for precise water management of artificial grasslands in arid regions. Full article

Forests play a critical role in Earth’s ecosystem, yet monitoring their long-term, large-scale spatiotemporal dynamics remains a significant challenge. This study addresses this gap by developing an integrated framework to map annual forest distribution in Hunan, China, from 1999 to 2023 at a high resolution of 30 m. Our methodology combines multi-temporal satellite imagery (Landsat 5/7/8/9) with key environmental variables, including digital elevation models, temperature, and precipitation data. To efficiently reconstruct historical maps, training samples were automatically derived from a reliable 2023 forest product using a transferable logic, drastically reducing manual annotation effort. Comprehensive evaluations demonstrate the robustness of our approach: (1) Qualitative analyses reveal superior spatial detail and temporal consistency compared to existing global forest maps. (2) Rigorous quantitative validation based on ∼9000 reference samples confirms high and stable accuracy (∼92.4%) and recall (∼91.9%) over the 24-year period. (3) Furthermore, comparisons with government forestry statistics show strong agreement, validating the practical utility of the data. This work provides a valuable, accurate long-term dataset that forms a scientific basis for critical downstream applications such as ecological conservation planning, carbon stock assessment, and climate change research, thereby highlighting the transformative potential of multi-source data fusion and automated methods in advancing geospatial monitoring. Full article

This study presents a 31-year (1993–2023) wave hindcast using a high-resolution two-domain nested numerical wave model implemented with Simulating Waves Nearshore (SWAN). The spatiotemporal variability and long-term trends of two wave parameters (significant wave height H_s and spectral peak period T_peak) are systematically analyzed for the Yangtze River Delta (YRD) and its adjacent waters. Validation against in situ buoy measurements confirms that the SWAN model effectively reproduces the regional wave conditions. Results indicate that mean wave conditions are primarily modulated by the Asian monsoon, whereas extreme wave events are predominantly influenced by typhoons. This leads to pronounced differences in spatial patterns and seasonal variability between mean and maximum H_s values. In addition, the regional interannual variations of H_s and T_peak exhibit different degrees of correlation with the Niño 3.4 index, the Pacific Decadal Oscillation (PDO) index and the Western Pacific Subtropical High Ridge Position (WPSH) Index. Overall, both H_s and T_peak exhibit positive trends over the study period, and both positive trends shift remarkably between seasons. The positive trends in mean wave conditions are mild during spring and summer but more pronounced in autumn and winter. Statistically significant increases in seasonal mean H_s are identified in parts of the East China Sea (0.35 cm a⁻¹ in autumn) and the southern Yellow Sea (0.27 cm a⁻¹ in winter). Notably, not all trends are positive: the 90th percentiles of both H_s and T_peak during summer exhibit widespread declining trends, although they are not statistically significant. Full article

Water quality observations are vital for effectively managing coastal resources and influencing decisions from emergency beach closures to aquaculture leasing agreements. This study focuses on deriving two water quality parameters—Chlorophyll a (Chl-a) and suspended particulate matter (SPM)—through the high-resolution multispectral imager (MSI) onboard the Sentinel 2A&B satellites, specifically for the Ningde coastal region, which is a crucial aquaculture hub in China. Since more than 90% of the signals captured by satellites are affected by atmospheric interference, it is crucial to apply a process called “atmospheric correction” (AC) to isolate the water contribution, known as water leaving reflectance, from the radiance measured at the top of the atmosphere. Our research assesses five published AC models and various algorithms designed to accurately estimate Chl-a and SPM from water leaving reflectance. We determine the most effective combination by comparing these findings against in situ data gathered from eleven locations in the Ningde coastal region (POLYMER-SOLID with lowest metric RMSLE (0.29), and MAE (1.68) and POLYMER-MDN with the lowest metric RMSLE (0.59), and MAE (0.56)). Our study underscores the importance of selecting locally validated AC models and algorithms for generating water quality products, as this enhances the utility of remote sensing data in monitoring water quality. Moreover, we conduct a spatiotemporal analysis of the water quality parameters from 2016 to 2021, revealing significant interannual variability that underlines the need for continuous monitoring and robust data analysis in coastal management efforts. Full article

High-precision temperature prediction is crucial for dealing with extreme weather events under the background of global warming. However, due to the limitations of computing resources, numerical weather prediction models are difficult to directly provide high spatio-temporal resolution data that meets the specific application requirements of a certain region. This problem is particularly prominent in areas with complex terrain. The use of remote sensing data, especially high-resolution terrain data, provides key information for understanding and simulating the interaction between land and atmosphere in complex terrain, making the integration of remote sensing and NWP outputs to achieve high-precision meteorological element downscaling a core challenge. Aiming at the challenge of temperature scaling in complex terrain areas of Southwest China, this paper proposes a novel deep learning model—Terrain Adaptive U-Net Transformer (TAUT). This model takes the encoder–decoder structure of U-Net as the skeleton, deeply integrates the global attention mechanism of Swin Transformer and the local spatiotemporal feature extraction ability of three-dimensional convolution, and innovatively introduces the multi-branch terrain adaptive module (MBTA). The adaptive integration of terrain remote sensing data with various meteorological data, such as temperature fields and wind fields, has been achieved. Eventually, in the complex terrain area of Southwest China, a spatio-temporal high-resolution downscaling of 2 m temperature was realized (from 0.1° in space to 0.01°, and from 3 h intervals to 1 h intervals in time). The experimental results show that within the 48 h downscaling window period, the TAUT model outperforms the comparison models such as bilinear interpolation, SRCNN, U-Net, and EDVR in all evaluation metrics (MAE, RMSE, COR, ACC, PSNR, SSIM). The systematic ablation experiment verified the independent contributions and synergistic effects of the Swin Transformer module, the 3D convolution module, and the MBTA module in improving the performance of each model. In addition, the regional terrain verification shows that this model demonstrates good adaptability and stability under different terrain types (mountains, plateaus, basins). Especially in cases of high-temperature extreme weather, it can more precisely restore the temperature distribution details and spatial textures affected by the terrain, verifying the significant impact of terrain remote sensing data on the accuracy of temperature downscaling. The core contribution of this study lies in the successful construction of a hybrid architecture that can jointly leverage the local feature extraction advantages of CNN and the global context modeling capabilities of Transformer, and effectively integrate key terrain remote sensing data through dedicated modules. The TAUT model offers an effective deep learning solution for precise temperature prediction in complex terrain areas and also provides a referential framework for the integration of remote sensing data and numerical model data in deep learning models. Full article

In image super-resolution tasks, methods based on Generative Adversarial Networks (GANs), Transformer models, and diffusion models demonstrate robust global modeling capabilities and outstanding performance. However, their computational costs remain prohibitively high, limiting deployment on resource-constrained devices. Meanwhile, frequency-domain approaches based on convolutional neural networks (CNNs) capture complementary structural information but lack long-range dependencies, resulting in suboptimal perceptual image quality. To overcome these limitations, we propose a micro-Transformer-based architecture. This framework enriches high-frequency image information through wavelet transform-based frequency-domain features, integrates spatio-temporal and frequency-domain cross-feature fusion, and incorporates a discriminator constraint to achieve image super-resolution. Extensive experiments demonstrate that this approach achieves competitive PSNR/SSIM performance while maintaining reasonable computational complexity. Its visual quality and efficiency outperform most existing SR methods. Full article

The integration of complementary metal–oxide–semiconductor (CMOS) and micro-electromechanical systems (MEMSs) technologies for miniaturized biosensor fabrication enables unprecedented spatiotemporal resolution in monitoring the bioelectrical activity of the nervous system. Wafer-level CMOS technology incurs high costs, but multi-project wafer (MPW) runs mitigate this by allowing multiple users to share a single wafer. Still, monolithic CMOS biosensors require specialized surface materials or device geometries incompatible with standard CMOS processes. Performing MEMS post-processing on the few square millimeters available in MPW dies remains a significant challenge. In this paper, we present a MEMS post-processing workflow tailored for CMOS dies that supports both surface material modification and layout shaping for intracortical biosensing applications. To address lithographic limitations on small substrates, we optimized spray-coating photolithography methods that suppress edge effects and enable reliable patterning and lift-off of diverse materials. We fabricated a needle-like, 512-channel simultaneous neural recording active pixel sensor (SiNAPS) technology based neural probe designed for integration with optical fibers for optogenetic studies. To mitigate photoelectric effects induced by light stimulation, we incorporated a photoelectric shield through simple modifications to the photolithography mask. Optical bench testing demonstrated >96% light-shielding effectiveness at 3 mW of light power applied directly to the probe electrodes. In vivo experiments confirmed the probe’s capability for high-resolution electrophysiological measurements. Full article

Accurate mapping of land use and land cover (LULC) is crucial for various applications such as urban planning, environmental management, and sustainable development, particularly in rapidly growing urban areas. African cities such as Yaoundé, Cameroon, are particularly affected by this rapid and often uncontrolled urban growth with complex spatio-temporal dynamics. Effective modeling of LULC indicators in such areas requires robust algorithms for high-resolution images segmentation and classification, as well as reliable data with great spatio-temporal distributions. Among the most suitable data sources for these types of studies, Sentinel-2 image time series, thanks to their high spatial (10 m) and temporal (5 days) resolution, are a valuable source of data for this task. However, for an effective LULC modeling purpose in such dynamic areas, many challenges remain, including spectral confusion between certain classes, seasonal variability, and spatial heterogeneity. This study proposes a hybrid deep learning architecture combining U-Net and Convolutional Long Short-Term Memory (ConvLSTM) layers, allowing the spatial structures and temporal dynamics of the Sentinel-2 series to be exploited jointly. Applied to the Yaoundé region (Cameroon) over the period 2018–2025, the hybrid model significantly outperforms the U-Net and ConvLSTM models alone. It achieves a macro-average F1 score of 0.893, an accuracy of 0.912, and an average IoU of 0.811 on the test set. These segmentation performances reached up to 0.948, 0.953, and 0.910 for precision, F1-score, and IoU, respectively, on the built-up areas class. Moreover, despite its better performance, in terms of complexity, the figures confirm that the hybrid does not significantly penalize evaluation speed. These results demonstrate the relevance of jointly integrating space and time for robust LULC classification from multi-temporal satellite images. Full article

Geostationary atmospheric motion vectors (e.g., FY4A AMVs) are routine mid-upper atmospheric observations used in numerical weather prediction (NWP) models, yet their complex spatiotemporal errors and assimilation limitations, i.e., high-temporal/coarse-spatial data and large-scale-adjustment/direct-assimilation scheme, leave unclear impacts of AMVs assimilation on nowcasting forecasts. To this end, a Nudging-Forced–3DVar scheme (NFV) is designed within a multi-scale (i.e., 12, 4, and 1 km) regional NWP framework to exploit AMVs characteristics; ablation experiments for the Zhengzhou “7.20” rainstorm isolate Nudging and 3DVar impacts on assimilation and nowcasting. Results show the following: (1) large-scale Nudging and high-resolution 3DVar both improve mid-upper analyses, with the former ingesting more observations; (2) Nudging retains large-scale background updates but yields significant misses, whereas 3DVar intensifies rainfall extremes yet blurs fine structures; (3) NFV merges its strengths, modulating deep convection through upper-level systems and markedly improving rainfall spatiotemporal patterns. Therefore, NFV is recommended for the FY4A AMVs’ future numerical nowcasting, which provides useful guidance for the regional application of geostationary 3D winds. Full article

Accurate traffic flow forecasting is a fundamental component of Intelligent Transportation Systems and proactive urban mobility management. However, the inherent complexity of urban traffic flow, characterized by non-stationary dynamics and multi-scale temporal dependencies, poses significant modeling challenges. Existing spatio-temporal models often struggle to reconcile the discrepancy between static physical road constraints and highly dynamic, state-dependent spatial correlations, while their reliance on fixed temporal receptive fields limits the capacity to disentangle overlapping periodicities and stochastic fluctuations. To bridge these gaps, this study proposes a novel Multi-scale Graph-Decoupling Spatial–temporal Network (MS-GSTN). MS-GSTN leverages a Hierarchical Moving Average decomposition module to recursively partition raw traffic flow signals into constituent patterns across diverse temporal resolutions, ranging from systemic daily trends to high-frequency transients. Subsequently, a Tri-graph Spatio-temporal Fusion module synergistically models scale-specific dependencies by integrating an adaptive temporal graph, a static spatial graph, and a data-driven dynamic spatial graph within a unified architecture. Extensive experiments on four large-scale real-world benchmark datasets demonstrate that MS-GSTN consistently achieves superior forecasting accuracy compared to representative state-of-the-art models. Quantitatively, the proposed framework yields an overall reduction in Mean Absolute Error of up to 6.2% and maintains enhanced stability across multiple forecasting horizons. Visualization analysis further confirms that MS-GSTN effectively identifies scale-dependent spatial couplings, revealing that long-term traffic flow trends propagate through global network connectivity while short-term variations are governed by localized interactions. Full article