Flood mapping from optical satellite imagery is often hindered by persistent cloud cover, while historical-water-occurrence-based gap-filling may fail when flood extent exceeds previous records. This study develops a Sentinel-1/Sentinel-2 framework for flood mapping in China, Spain, and Somalia. Sentinel-1-derived water maps provide cloud-independent dynamic references. A Markov chain first fills cloud-covered pixels whose adjacent satellite observations show consistent water or non-water states, and a spatiotemporal Markov random field then reconstructs the remaining uncertain pixels using spatial neighborhood coherence and temporally weighted observations. The gap-filled Sentinel-2 maps are further combined with Sentinel-1 maps to generate denser flood sequences. Across the three cases, the framework achieved 10 m resolution, mIoU values of 0.82–0.99, and effective mapping intervals of 1.6–3.6 days. Independent PlanetScope validation, component analysis, confidence intervals, and significance tests supported its reliability. The framework can provide timely flood-extent information for disaster response and flood-risk management.

Submerged air-jet scouring is promising for underwater excavation, yet the influence of bed slope remains poorly understood. This study experimentally investigates air-jet scouring on inclined sand beds with varying jet angles (θ), nozzle heights (H), and slope directions (uphill/downhill). Using four dimensionless parameters and K-means clustering, 18 tests are classified into three distinct scour patterns. Pattern I appears as a shallow smooth crater with negligible post-cessation backfilling regardless of slope. Pattern II is a moderate-depth V-shaped crater, with uphill conditions amplifying back-filling by more than five times. Pattern III is a deep platform crater where dense granular flow reshapes the scour profile; uphill configurations significantly lower the threshold for this regime, with back-filling reaching up to 63% of the dynamic scour depth. A dimensionless gravity–momentum ratio G is proposed to quantify slope–jet coupling, which correlates well with relative backfilling intensity Δ*. Results show that downhill conditions favor predictable scour geometry, while uphill slopes pose considerable risks of intense morphological reconstruction.

Selecting an efficient operational mode for air-jet seabed scouring requires understanding how jet expansion state governs both mechanism and performance. This study experimentally compares under-expanded and fully expanded air jets impinging on a non-cohesive sand bed in quiescent water. High-speed imaging reveals two distinct mechanisms: the expanded jet drives continuous viscous shear erosion (VSE), reaching dynamic equilibrium rapidly (0.48-3.9 s) via stable wall-bounded shear flow, whereas the under-expanded jet triggers violent bearing-capacity failure (BCF), requiring significantly longer times (5.9-7.0 s) due to intermittent explosive ejections. Under identical flow input, the expanded jet demonstrated unequivocally superior performance, achieving an 11-24% increase in maximum particle entrainment height, an expansion of cumulative entrainment area by up to 1.9 times, and a remarkable enhancement of the horizontal diffusion rate by a factor of 3.9 to 7.9. Morphologically, its scour profile closely matches the classical shear-driven model (R2 = 0.98), while the underexpanded jet yields a concave, non-classical profile (R2 = 0.63) with limited downstream transport. The jet expansion state, controlled by standoff distance relative to the Mach disk, thus governs the transition between efficient shear dominated and inefficient explosion dominated regimes, providing a quantitative basis for selecting the fully expanded mode in seabed trenching and similar marine engineering applications.

Understanding the seasonal lake dynamics is critical for water resource management and climate adaptation, yet intra-annual variability of Tibetan Plateau (TP) lakes remains poorly characterized. Here we present a monthly lake-surface area dataset covering 23,623 lakes (2000-2021), and propose a two-tiered classification framework identifying six distinct seasonal patterns. It reveals that semi-annual-cycle lakes predominate in endorheic regions while annual-cycle lakes concentrate in exorheic basins. Annual-cycle lakes are governed by single dominant factors and exhibit remarkable stability. In contrast, semi-annual-cycle lakes reflect coupled spring snowmelt and late-summer precipitation dynamics, showing high vulnerability to transitions. The Spring Peak (SP) pattern, whose shifts are attributable to intensified glacial melt and permafrost thaw, serves as a sensitive indicator of environmental changes. Seasonal complexity scales non-linearly with lake size. The post-2015 lake expansion coincided with rapid intensification of seasonal amplitude, indicating a fundamental hydrological transition that could threaten pastoral systems and water security across vulnerable endorheic regions.

The Tibetan Plateau (TP) is highly sensitive to climate change, yet existing Landsat-based surface water products suffer from pervasive data gaps caused by clouds, terrain shadows, and seasonal snow/ice, hindering fine-scale intra-annual hydrological analysis. Here we present a spatially complete, monthly 30-m resolution surface water dataset for the entire TP spanning 2000–2021 (TP MWH), constructed using a stepwise gap-filling (SGF) framework. The SGF method synergistically integrates Joint Research Centre (JRC)'s Yearly Water Classification History, Monthly Water Recurrence, and spatiotemporal neighborhood consistency, forming a four-step hierarchical judgment system that preserves the high precision of the original JRC classification while compensating for its inherent seasonal water omission bias. Validated against 12,141 stratified random samples covering 12 months, 12 hydrological basins, and 8 typical plateau surface features, the TP MWH dataset achieves an overall accuracy of 98.1%, a recall of 97.8%, a precision of 98.4%, an F1-score of 0.981, a Kappa coefficient of 0.961, and maintains stable high performance (>97.4% accuracy, >97.0% recall, >96.5% precision, >0.946 Kappa) throughout the year, outperforming mainstream JRC and GLAD products especially in winter freezing periods. Cross-comparison with the global monthly lake dataset confirms consistency in both interannual trends (Spearman r = 0.998) and intra-annual seasonal variability (Spearman r = 0.949). Leveraging this gap-free dataset, we reveal a three-stage interannual expansion of surface water and intensified intra-annual fluctuations after 2016, indicating growing hydrological instability of the TP. The 12 sub-basins exhibit four distinct heterogeneous evolution regimes, reflecting differentiated responses to monsoon, westerly circulation, and cryospheric meltwater changes. This dataset provides a reliable baseline for seasonal hydrological research and water resource management on the TP, and is publicly available at https://doi.org/10.5281/zenodo.13910635.

Accurately measuring flow fields across diverse scenarios is essential for understanding physical phenomena. Particle Image Velocimetry (PIV) allows non-contact surface flow acquisition, but obtaining reliable prototype scale data, including in complex waterways and marine settings, remains a major challenge. This study proposes GMFlow-PIV, a novel framework that adapts the global matching optical flow model (GMFlow) to PIV applications through the design of physically-informed divergence and vorticity loss functions. Training and testing on synthetic datasets, GMFlow-PIV outperforms the original model across all test cases and reduces the average endpoint error (AEE) by 28.7% and 38.2% compared to UnLiteFlowNet-PIV and RAFT256-PIV in surface-quasi-geostrophic (SQG) flows, respectively. Parametric analysis demonstrates enhanced performance in scenarios with ultra-low/high particle diameters (<1 pixel or >5 pixels), sparse seeding densities (<0.03 particles per pixel), and large displacements up to 18 pixels. When validated on experimental PIV datasets and field flows, GMFlow-PIV exhibits improved generalization capabilities. It reduces inference time by 86.5%, 69.6% and 84.1% compared to PIV, RAFT256-PIV, and RAFT32-PIV, respectively, while maintaining competitive accuracy. By balancing high computational efficiency with robust performance, GMFlow-PIV shows promising potential for application in complex real-world scenarios.

Accurate estimation of discharge through submerged spillways is critical for reservoir operation and structural maintenance. Traditional methods using empirical formulas or intrusive measurements fail to characterize high-speed pressurized flow hydrodynamics. We introduce a novel Eulerian monocular photogrammetry (EMP) method that non-intrusively quantifies discharge by integrating pixel-level jet surface geometry into a theoretical velocimetry framework. This framework further combines image-derived streamlines and cross-sectional flow analysis to establish a robust relationship between jet images and flow discharge. The methodology was validated through controlled laboratory experiments and a field application at Xiluodu dam. In laboratory tests, EMP-derived discharge estimates showed excellent agreement with direct measurements and demonstrated robustness against variations in camera positioning. Field application matched theoretical/numerical results with minor underestimation, consistent with similar projects. Bridging visual observation and quantitative analysis, EMP enhances spillway discharge estimation reliability and efficiency for modern hydraulic engineering.

MobileViT v2, a Transformer-CNN hybrid deep learning model, was restructured and successfully applied to predict urban pluvial flood maximum water depth for the first time. The hybrid model leverages the advantages of both Transformers and CNNs to enhance global and local modeling capabilities, thereby achieving high prediction accuracy even on unseen terrains. Furthermore, we proposed a data augmentation approach to satisfy the substantial data requirements of the Transformer blocks and mitigate overfitting caused by limited training data. Moreover, we utilized an attribution method to analyze the effectiveness and physical plausibility of both the hybrid model and the data augmentation, thereby enhancing the interpretability of the proposed framework. Results indicate that the hybrid model with data augmentation demonstrated a significant improvement in performance compared to the standard ViT and a modest enhancement compared to representative CNN baselines. These findings demonstrate the hybrid model’s superior generalization capability and greater potential for further improvements. Additionally, the results validate data augmentation as an easy-to-implement and widely applicable approach for enhancing prediction accuracy with limited data in flood prediction tasks. Finally, this study presents a promising method for enhancing prediction accuracy in flood prediction tasks and represents the successful application of hybrid models to this task across diverse unseen spatial terrain patches.

In this study, a novel spatiotemporal hydrodynamic prediction task framework, named single frame prediction, was developed. The framework could generate results based on boundary conditions and a single flood map from the last time step, relying on hydrodynamic principles rather than historical trends, and doesn't require the assistance of traditional hydrodynamic models. Moreover, a post-processing method based on physical laws was developed to refine the outputs of deep learning models at each time step, aiming to reduce accumulated errors in long-term predictions. The performance of a widely used convolutional neural network-based model, U-Net, was evaluated to assess the feasibility of single frame prediction and the impact of the proposed post-processing method. The experiments showed that single frame prediction could produce accurate flood maps, demonstrating the feasibility of the novel framework. Furthermore, the results indicated that the physics-based post-processing method could mitigate errors at each step, thereby enhancing prediction accuracy across entire flood event, showing strong effectiveness and applicability in flood prediction. Additionally, an ablation experiment was conducted to assess the effectiveness of each step in the method. The single frame prediction provided a more comprehensive and interpretable depiction of flood prediction processes with essential hydrodynamic variables, including water depth and unit discharge on all grid cells. The post-processing method significantly reduced the accumulated error in the later stages of single frame prediction to an acceptable range with an average root-mean-square error of 0.041 m for water depth and 0.003 m2/s for unit discharge, suggesting a new technique for long-term flood predictions.

A set of high-resolution, time-resolved particle image velocimetry measurements were conducted in an open channel, with closely arranged glass spheres of 6 mm used to rough the bed, at low to moderate Reynolds numbers ( Re-tau approximate to 600-2000) and intermediate to high relative submergences ( h/k(s) = 6.3-14.7, where h is the flow depth and k(s) is the equivalent roughness height). Analyses of the wall-attached motions (WAMs) in rough-wall open channel flows (OCFs) are performed using linear coherence spectra at various wall-normal positions, and results of two smooth-wall OCFs are also included for comparison ( Re-tau approximate to 500 and 900). The WAMs in rough-wall OCFs exhibit self-similar properties in the region of 0.2

River level predicting underpins the management of water resource projects, steers navigational activities in rivers, and protects the lives and properties of riverside communities, etc. Traditionally, hydrological-hydraulic coupled models have been at the forefront of simulating and predicting river levels, achieving notable success. Despite their utility, these models encounter limitations due to the exhaustive demand for various data types-often difficult to obtain-and the ambiguity in determining downstream boundary conditions for the hydraulic model. Responding to these limitations, this study utilizes Long Short-Term Memory (LSTM) model, a deep learning technique, to predict river levels using upstream discharges. Three approaches were used to further enhance the accuracy and reliability of our model. Firstly, we incorporated historical water level data at or downstream of the predicted station as input, secondly, we classified the datasets based on physical principles, and thirdly, we employed data augmentation techniques. These methods were evaluated within the JingjiangDongting river-lake system in China. It achieves high prediction accuracy of water level and can mitigate the impact of input inaccuracies. The incorporation of water level data as input and the Classification-Enhanced LSTM model that segregates the input data according to rising and recession trends of water level, significantly improve prediction accuracy under extreme water level conditions compared with other deep learning approaches. The proposed model uses easily accessible data to predict water levels, offering enhanced robustness and new strategies for improving prediction accuracy under extreme conditions. It is applicable for predicting water levels at any hydrological station along a river and can enhance the prediction accuracy of hydraulic models by proving a robust downstream boundary condition.

Long-duration and time-resolved particle image velocimetry measurements were conducted in rough-wall open channel flows (OCFs), with the friction Reynolds number ranging from 642 to 2034. The primary objective is to investigate the impacts of various turbulent motions at different scales on the mean wall-shear stress ( $\langle \tau _w \rangle$ ). To achieve this aim, a physical decomposition of $\langle \tau _w \rangle$ was initially performed utilizing the double-averaged methodology proposed by Nikora et al. (2019 J. Fluid Mech. 872, 626-664). This method enabled the breakdown of $\langle \tau _w \rangle$ into three distinct constituents: viscous, turbulent and dispersive stress segments. The findings underscore the substantial roles that turbulent and dispersive stresses play, accounting for over 75 % and 9 % of $\langle \tau _w \rangle$ , respectively. Subsequently, a scale decomposition was further applied to analyse the contributions of coherent motions at different scales to $\langle \tau _w \rangle$ . Adopting typical cutoff streamwise wavelengths ( $\lambda _x = 3h$ and $10h$ ), the contribution of large-scale motions (LSMs) and very large-scale motions (VLSMs) to the overall wall-shear stress was quantified. It was revealed that turbulent motions with $\lambda _x \gt 3h$ and $\lambda _x \gt 10h$ contribute more than 40 % and 18 % of $\langle \tau _w \rangle$ , respectively. The scale decomposition of the wall-shear stress and the contribution from LSMs and VLSMs exhibit evident dependencies on the Reynolds number. The contribution of LSMs and VLSMs to $\langle \tau _w \rangle$ is lower in rough OCFs compared with those of smooth counterparts. Secondary currents induced by the rough wall are hypothesised to be responsible for the reduced strength of LSMs and VLSMs and decreases in their contribution to $\langle \tau _w \rangle$ .

Dunes are ubiquitous riverbed forms, yet how they modify the most recently documented energy-containing turbulent structures—very-large-scale motions (VLSMs)—remains unclear. To address this issue, high fidelity experiments were conducted in open-channel flows (OCFs) over isolated dunes with contrasting morphologies. Time-resolved, long-duration, and high-resolution velocity fields with extended streamwise coverage were obtained using an in-house multi-camera particle image velocimetry system, enabling detailed quantification of dune effects over long distances with fine spatial resolutions. Consistent with canonical OCFs, VLSMs are evident in both dune cases upstream of the lee face, where dune effects are minimal, as revealed by the bimodal features of the streamwise velocity spectrum, indicating two dominant energy-containing turbulent structures: large-scale motions and VLSMs. The scales and strength of these motions are comparable to those in canonical OCFs. In contrast, over the lee face and downstream side, pronounced dune morphology–dependent effects emerge. For low lee-slope dunes, VLSMs persist throughout the flow depth, whereas for steep dunes, their spectral signature is eliminated in the near-bed region and remains only near the free surface. Furthermore, the strength of VLSMs, measured by their contributions to streamwise turbulent kinetic energy (TKE) and Reynolds shear stress (RSS), is substantially reduced: from 60% to 30% for TKE and from 50% to 15% for RSS in the near-dune region. With increasing distance downstream, dune effects weaken, and VLSM strength gradually recovers.

The growing research domain of unsupervised learning methodologies for 3D-aware Generative Adversarial Networks (GANs) is particularly focused on utilizing extensive datasets containing unstructured single-view images. Recent advancements in 3D GANs have demonstrated their ability to enhance photorealistic synthesis and achieve multi-view consistency when generating radiance fields depicting human facial and body features. However, these methods have not yet effectively addressed human hands, primarily due to the increased complexity in learning the distribution of hand poses caused by diverse hand gestures and extensive self-occlusion. In this paper, our method represents a significant advancement in the domain of photorealistic 3D-aware image synthesis for articulated human hands. The key contribution of our model lies in the synthesis of high-quality 3D hand avatars, which incorporate intricate geometric details and capture detailed human hand features, such as skin wrinkles, in a more natural manner compared to previous approaches. We present a novel framework for effectively representing articulated human hands which integrates a heatmap-based feature generator, the innovative tri-plane feature volume representation, and a feature volume deformation method guided by a mesh. Our empirical findings demonstrate the superior performance of our methodology compared to preceding 3D and articulation-aware approaches in the generation of complex human hand representations. We validate the effectiveness of our model and the importance of each component via systematic ablation studies and demonstrate state-of-the-art 3D-aware synthesis with InterHand 2.6 M, achieving an FID of 7.53, which represents a $33 \%$ improvement compared to SOTA.

Rotated object detectors commonly encounter instability during the training process, primarily due to background noise and angular periodicity. Targets with elongated or non-convex shapes may introduce background noise during convolution, hindering accurately extracting features. Meanwhile, the periodicity of angles leads to predictions beyond the defined range, subsequently impeding the convergence. This letter introduces an Angle Adaptive Module (AAM) designed for the backbone, enhancing the ability of the model to accurately extract object features and dynamically select the optimal angle. Moreover, to mitigate the effect of angle periodicity, a method called Radian Regression Method (RRM) is proposed for predicting proper angles. It avoids directly regressing the value and instead produces the probability density distribution of the offset. We elaborately design numerous experiments to demonstrate the effectiveness of the proposed modules. As a result, the proposed method attains competitive results across various datasets, including DOTAv1.0, DOTAv1.5, DOTAv2.0, HRSC2016, and DIOR-R.

Satellite altimetry data has become essential for studying the dynamics of water bodies, especially in regions with limited or inaccessible data. Traditional low-resolution mode (LRM) satellites' accuracy cannot be guaranteed when it comes to assessing water levels in small- (< 200 m in width) and medium-sized (200-800 m in width) rivers. Synthetic aperture radar (SAR) altimeters, exemplified by Sentinel-3 A, have shown great potential for inland water altimetry. Nevertheless, developing algorithms to retrack the raw data remains an essential requirement in this context. This is attributed to the width of small-sized rivers, which is often narrower than the along-track resolution of both LRM and SAR altimeters. In addition, new altimeters may have long revisit cycles and different spatial coverage and cannot yield historical data necessary in some situations. To address these challenges, this study proposed a conditional threshold retracker (CTR). The CTR algorithm is well-designed and facilitates accurate water level monitoring. Moreover, we proposed an enhanced footprint filter (EFF), thus significantly bolstering the number of available cycles. Our findings demonstrate that the developed method substantially enhances the temporal and spatial resolution of both LRM and SAR altimetry satellites during water level monitoring in rivers of different climate types. The width of the thirteen selected rivers is on the order of 85-630 m. The CTR significantly improved the water level monitoring accuracy by 68 %- 78 %. Furthermore, the EFF increased the number of water level cycles by approximately 49 %-68 %. These findings have practical implications for obtaining accurate water level data, estimating river discharge and improving hydraulic model calibration.

Optical satellite imaging for surface water mapping often encounters significant challenges owing to persistent spatial data gaps caused by clouds, shadows, and sensor errors. This study presents a novel Stepwise Gap-Filling (SGF) method, designed to enhance the monthly surface water mapping and monitoring. The SGF method leverages temporal similarities and spatial correlations to reconstruct gap pixels originally classified as invalid observations. We validated this approach against historical high-resolution Google Earth images from 2887 sample points in the Siling Co Basin of the Tibetan Plateau. The results demonstrated substantial improvements in mapping accuracy, achieving an overall accuracy of 98.93%, a producer’ accuracy of 98.59%, and a user’ accuracy of 99.11%, markedly reducing the uncertainties in the original dataset. Importantly, the SGF method offers detailed insights into monthly surface water dynamics, which are closely aligned with annual trends. This study highlights the effectiveness of the SGF method for filling data gaps and its potential for widespread application in the monitoring and management of global water resources.

Object tracking has made significant progress in recent years. However, the state-of-the-art trackers are becoming increasingly heavy and expensive, making their deployment challenging in resource-constrained applications. In this study, we introduce MobileTrack, a visual object tracker that strikes a perfect balance between tracking accuracy and inference speed. Utilizing a novel coordinated perception-aware fusion module and a lightweight prediction head, our proposed methodology outperforms most Siamese trackers on various academic benchmarks in terms of both accuracy and efficiency. When deployed on resource-constrained embedded devices such as NVIDIA Jetson TX2, MobileTrack ensures real-time performance at a speed exceeding 33FPS, while LightTrack only operates at 18FPS. Therefore, MobileTrack holds significant potential to unlock a wide range of practical applications across various industries. MobileTrack is released at here.

Neural architecture search (NAS) has shown excellent performance. However, existing semantic segmentation models rely heavily on pre-training on Image-Net or COCO and mainly focus on the designing of decoders. Directly training the encoder–decoder architecture search models from scratch to SOTA for semantic segmentation requires even thousands GPU days, which greatly limits the application of NAS. To address this issue, we propose a novel neural architecture Search framework for Enhanced Decoder (SED). Utilizing the pre-trained hand-designing backbone and the searching space composed of light-weight cells, SED searches for a decoder which can perform high-quality segmentation. Furthermore, we attach switchable skip connection operations to search space, expanding the diversity of possible network structure. The parameters of backbone and operations selected in searching phrase are copied to retraining process. As a result, searching, pruning and retraining can be done in just 1 day. The experimental results show that the SED proposed in this paper only needs 1/4 of the parameters and calculation in contrast to hand-designing decoder, and obtains higher segmentation accuracy on Cityscapes. Transferring the same decoder architecture to other datasets, such as: Pascal VOC 2012, Camvid, ADE20K proves the robustness of SED.