Seawater pumped energy storage (SPES) offers a promising solution to the intermittency of offshore wind and photovoltaic power in China's coastal regions. However, a targeted planning approach is lacking. We present a novel multi-stage, data-driven potential model that leverages the unique advantages of coastal depression topography to improve reservoir depth estimation and more accurately quantify energy storage potential. First, technically feasible locations are identified based on environmental and regulatory constraints. Next, costreduction criteria are applied to screen these candidate sites. Finally, a multi-objective optimization framework simultaneously minimizes total construction cost, number of sites, and spatial distribution disparities to select an optimal combination of sites. The results identify approximately 20,000 potential coastal sites with a combined storage capacity of 4379 GWh. Among them, 87 sites exhibit average construction costs of 949/kW, approaching those of conventional freshwater pumped storage systems. Considering the low excavation costs of depressions, the actual construction costs may be lower than the estimated values. Under projected 2050 electricity-storage requirements, only 15 well-distributed site combinations with a total installed capacity of 19.2 GW can satisfy national demand while remaining cost-competitive with alternative storage technologies. This study provides a scalable, data-driven framework to support coordinated planning of SPES infrastructure alongside renewable energy development.

Groundwater levels in large agricultural irrigation districts generally show strong spatiotemporal variability due to heterogeneous hydrological conditions, geological formations, and human activities. Such variability complicates groundwater management and underscores the need for high-resolution, efficient prediction of groundwater levels. To enhance computational efficiency without compromising the accuracy of MODFLOW, this study proposes a novel surrogate modeling framework, SRR-LSTM, for predicting groundwater levels at a 1-km grid scale. The core innovation of this framework lies in its grid clustering strategy. It couples K-means and LSTM to cluster grids with similar physical features, hydrological features, and groundwater level dynamics, thereby enhancing prediction accuracy. A case study in the Taobei Irrigation District, Northeast China, shows that SRR-LSTM achieves an approximately 80% improvement in computational efficiency compared with the physics-based model. Simultaneously, the proposed framework attains a Nash–Sutcliffe Efficiency (NSE) greater than 0.9 for 96% of the grids. This performance surpasses that of the three baseline schemes, which reach NSE values above 0.9 in 11% to 49% of the grids. Furthermore, SHAP is employed to reveal the spatial heterogeneity of input variable contributions and to quantify the combined effects of streamflow and human activities on groundwater dynamics.

This study introduces a Python-based framework for constructing differentiable hydrological models with a modular design to streamline research workflows. The framework integrates five key modules: hydrodataset and hydrodatasource for data preprocessing, hydromodel and torchhydro for traditional and differentiable modeling, and HydroDHM for orchestrating integrated workflows. The data modules automate preparation of diverse datasets, including open-access and proprietary resources. Hydromodel supports process-based model calibration and evaluation, while torchhydro enables neural network integration for differentiable models. HydroDHM coordinates these components through a unified interface for configuring and executing end-to-end modeling pipelines. Case studies in CAMELS basins demonstrate that differentiable models achieve comparable streamflow simulation performance to traditional approaches. By decoupling data handling from model development and providing uv-installable (and pip-compatible) modules, the framework ensures reproducibility, scalability, and adaptability across diverse hydrological contexts.

To address the low computational efficiency in solving complex one-dimensional water network models, this study proposes a water-level prediction correction method based on the Lattice Boltzmann Method (LBM), called WLPC-LBM. This method establishes a numerical method for analyzing unsteady, gradually varied flow in one-dimensional water networks, effectively handling the coupling boundary condition between river segments. Firstly, the LBM is utilized to handle the Saint–Venant equations (SVE). LBM only requires iteratively updating the distribution function to solve the linear system, thereby avoiding the complex solution process for nonlinear partial differential equations. Secondly, the one-dimensional water network hydrodynamic model is solved explicitly using the water-level prediction-correction method at the mesoscopic level, without segment numbering and using a global connection matrix. The WLPC-LBM is validated by two hypothetical water networks and a real-world case of the Yingna River. The study found that WLPC-LBM significantly improves the computational accuracy and efficiency of hydrodynamic simulations, demonstrating strong potential for practical applications.

Joint operation of reservoirs using forecast information has been demonstrated to enhance floodwater utilization efficiency. However, forecasting uncertainties may lead to flood risk that may propagate across the reservoir systems. Currently, no systematic framework exists to quantitatively assess how forecast uncertainties influence flood risks. This study established a risk-controlled framework to determine the upper bound of pre-fill volume (UBFV) of the reservoirs, ensuring no additional flood risk is introduced. In this model, forecast uncertainty is explicitly linked to flood risk through probabilistic constraints, and the UBFV for reservoir clusters is derived under risk-neutral conditions. Furthermore, inter-reservoir UBFV relationships within cascades are quantified, and the impacts of forecast uncertainty are comprehensively analyzed across three key dimensions: spatial rainfall distribution, forecast lead time, and allowable flood high water level. The Jinxia Reservoir Group and the Three Gorges Cascade Reservoir System are examined as a case study. Results show that: (1) The maximum UBFV is determined to be 52 x 108 m3 and 51 x 108 m3 for the Jinxia and Three Gorges systems, respectively, when utilizing 1-7-day rainfall forecasts. (2) UBFV initially increases and then decreases with longer forecast periods, reaching maximum value at 6-7 days in Jinxia and 3-5 days in Three Gorges. Higher forecast accuracy reduces UBFV. (3) The higher the allowable flood high water level, the higher the UBFV of the reservoir.

In data–scarce basins, hydrological models often fail to capture the peak–dominated, nonstationary behaviour of rainfall–runoff events, limiting cross–basin generalization. To reconcile physical interpretability with deep learning capacity, we develop a differentiable Xin’anjiang model (dXAJ) that couples the conceptual XAJ structure with an LSTM network via differentiable programming, and further integrates transfer learning to enhance regional adaptability. Using 663 flood events from 18 diverse basins in Anhui Province, China, we evaluate dXAJ against the XAJ and LSTM under three tiers: data–rich training, zero–shot ungauged transfer, and fine–tuning with limited local data. Regionally trained dXAJ achieves a median KGE of 0.785 on the test set, rising to 0.811 for large floods, confirming robust temporal generalization. In zero–shot transfer, the multi–source pre–trained dXAJ attains a median KGE of 0.694 and a median absolute peak error of 21.4%, with log–regression R2 > 0.90 for peak flows, demonstrating strong spatial transferability. With transfer learning, fine–tuning using only 33% of local events yields a KGE of 0.785 and reduces peak error to 15%, matching full–data performance, indicating that the primary benefit lies in lowering data demand rather than raising the upper bound. Parameter analysis further shows that transfer–learning corrections align with the natural drift induced by expanded data (84% directional consistency, correlation 0.64), confirming a physically consistent parameter space. These findings demonstrate that differentiable hydrological modelling integrated with transfer learning provides a practical, interpretable framework for improving flood prediction and cross–basin generalization in data–scarce regions.

As a typical high-dimensional, multiconstraint optimization problem, the joint flood control operation of reservoir groups faces challenges such as the curse of dimensionality and the coupling of multiple constraints. This paper proposes a joint flood control peak-shaving optimization operation method for reservoir groups based on the identification of critical reservoirs. The method begins by optimizing the selection of reservoirs participating in the joint operation process. To enhance computational efficiency and interpretability, critical reservoirs are identified preliminarily using hydrological and structural indicators, providing a structure-aware dimensionality reduction strategy that complements existing systemwide optimization approaches. This method was applied to a seven-reservoir system in the Dawen River basin in China. It achieved flood control and peak-shifting effects comparable to those of full-system optimization, while reducing the number of reservoirs by 40%. The dimensionality and decision-making complexity arise from the interdependencies between reservoirs and operational constraints. Although the system in this study involved only seven reservoirs, the primary goal was to demonstrate the feasibility of the proposed dimensionality-reduction method. This approach is scalable and can be applied to larger, more complex systems, in which its benefits would be more pronounced. The study was implemented in the Dawen River basin of Shandong Province, China, and provides a technical reference for similar research.

Objective Floating-object detectors deployed in complex water environments must recognize small, weak-texture, and dynamically deformed targets under changes in camera location, weather, surface ripples, illumination, and occlusion. Existing deep detectors often depend on large fully annotated datasets, incur excessive computational and storage costs, and lose accuracy in unseen water scenes. Conventional domain generalization also requires multiple completely labelled source domains, whereas single-labelled domain generalization reduces annotation work but is vulnerable to source-specific bias. A double-labelled domain generalization (DLDG) method was therefore developed to balance detection accuracy, real-time efficiency, annotation cost, and generalization. The method used only two labelled source domains and several unlabelled source domains while explicitly correcting feature-extraction, classification, and localization biases.Methods A lightweight multiscale detector was constructed by coupling MobileNetV3, a Dynamic Feature Pyramid Network (DyFPN), and a Single Shot MultiBox Detector (SSD). The VGG16 backbone of SSD was replaced by MobileNetV3, and feature maps from different stages were delivered to DyFPN. Dynamic modules containing a gate, an Inception unit, and a skip connection were inserted into the lateral connections. A gating signal and Gumbel-Softmax one-hot decision determined whether each lateral convolution was executed. Useful cross-scale information was therefore selected according to the input, while redundant reflections, ripples, shoreline textures, and computation were reduced. Depthwise separable convolutions were also applied to the classification and box-regression layers. DLDG training was organized as a progressive "feature generation-category discrimination-spatial localization" process. Two labelled domains were first used to initialize the feature generator, classifier, and locator through cross-entropy and Smooth L1 losses. The remaining unlabelled domains were then used to filter three types of bias. For feature-extraction bias, cluster centers were calculated from SoftMax outputs and bounding-box regression results. Pseudo-labels were assigned according to cosine distance, and information maximization improved cluster separability and prevented concentration in a few classes. Both classification and localization information constrained pseudo-label generation, after which the feature extractor was updated using classification and localization losses from unlabelled domains. For classification bias, a conditional feature-projection network mapped unlabelled features into the discriminative space jointly defined by the two labelled domains. Consistency between true labels and pseudo-labels constrained projection, and domain-similarity attention reweighted common features across domains. For localization bias, foreground-anchor regression was trained with pseudo-labels. An adversarial regressor maximized its prediction discrepancy from the main regressor on source samples, and generalized intersection over union measured box differences. The three stages shared features and pseudo-labels and were optimized sequentially under a unified objective. Experiments were conducted using fixed-camera images from the modern water-conservancy demonstration area in Deqing County, Zhejiang Province, China. Sixty representative video sequences were sampled. After redundant frames were removed, 526 water-hyacinth images, 340 floating-weed images, and 300 plastic-bottle images were retained; training-set augmentation produced 11,623 samples. The 1,920 × 1,080-pixel images were divided into five scene domains: normal water surfaces, weather variation, wave interference, illumination variation, and target-level occlusion. The dataset was split 9:1 for training and validation, and 30 labelled-source/target combinations were generated. Experiments used PyTorch 1.10, Ubuntu 18.04, an Intel i7 processor, and an NVIDIA RTX 3080 graphics card. Stochastic gradient descent was configured with a momentum of 0.9, an initial learning rate of 0.01, a batch size of 16, and weight decay of 0.001. Performance was evaluated using mean average precision (mAP), F1 score, frames per second (FPS), floating-point operations, parameter count, and model size.Results and Discussions For double-labelled domain generalization, DLDG achieved 70.33% mAP, 71.02% F1, and 22.38 FPS on the GPU. It required 3.01 billion floating-point operations, 4.98 million parameters, and 21.24 MB of storage; CPU inference reached 7.98 FPS. The strongest compared method, domain-invariant feature enhancement domain adaptation, obtained 66.48% mAP, 66.85% F1, and 20.10 FPS. DLDG therefore improved mAP by 3.85 percentage points, F1 by 4.17 percentage points, and speed by 2.28 FPS, while using fewer parameters and less storage. Under conventional domain generalization with fully labelled source data, the method achieved 85.29% mAP, 86.28% F1, and 17.81 FPS, indicating that the progressive filtering mechanism was applicable under both limited-label and fully labelled settings. The network ablation showed that MobileNetV3-DyFPN reached 86.28% mAP, 87.33% F1, and 19.15 FPS, compared with 70.02% mAP, 71.28% F1, and 12.99 FPS for VGG16 with DyFPN. Thus, mAP increased by 16.26 percentage points and speed by 6.16 FPS. Bias-filtering ablations showed that the unfiltered model produced 62.29% mAP and 63.02% F1 at 24.37 FPS. Introducing all three filters increased mAP by 8.04 percentage points and F1 by 8.00 percentage points, while speed decreased by 1.99 FPS. Feature-extraction filtering produced the largest single-module improvement, increasing mAP by 2.43 percentage points. Localization filtering contributed more than classification filtering because adversarial regression and generalized intersection over union directly corrected box displacement. In wave-interference and low-illumination scenes, DLDG produced boxes more consistent with target boundaries, whereas FixMatch and open compound domain adaptation showed missed detections or localization shifts. Stable detections were also obtained for small plastic bottles and floating weeds whose edges were mixed with reflections and ripples.Conclusions Combining lightweight dynamic multiscale feature extraction with progressive filtering of feature, classification, and localization biases reduced dependence on extensive bounding-box annotation while maintaining real-time cross-domain detection. Two labelled domains provided model initialization, and additional unlabelled domains were exploited through classification-and-localization-constrained pseudo-labels, conditional feature projection, domain-similarity reweighting, and adversarial box regression. The method can support fixed-camera identification of floating-object accumulation, cleaning prioritization, and continuous inspection in heterogeneous water environments. The current domain definition was based mainly on visible water-surface states and object distributions because synchronized flow velocity, water level, wind speed, and rainfall were unavailable. Future integration of hydrodynamic and meteorological measurements, unmanned surface vehicles, and hydrological monitoring data could clarify the relationships among water conditions, floating-object transport, and detection performance.

Conventional flood-season regulation typically keeps reservoir water levels at or below the planned flood limited water level (FLWL). While this practice safeguards flood-control safety, it can leave flood-control storage unused in small-flood years or when short-term forecasts are available, limiting further gains in reservoir benefits. To address this issue, we propose an FLWL optimization and control framework for the lower Jinsha River cascade-Wudongde (WDD), Baihetan (BHT), Xiluodu (XLD), and Xiangjiaba (XJB) reservoirs (Jinxia Reservoir Group). The framework integrates dynamic control of flood-season operating water levels, formulation and allocation of total pre-storage volume, and risk–benefit analysis. It includes three modules: (1) estimating the cascade-wide total pre-storage volume under multiple constraints using a pre-release capacity constraint method; (2) allocating this volume among alternative schemes to derive reservoir-specific dynamic FLWL control domains; (3) assessing maximum and expected risk rates together with hydropower benefits to identify the optimal pre-storage allocation plan. Application results show a total pre-storage volume of 2.16 billion m³ (13.9

Aging water supply infrastructure requires operational strategies that balance hydraulic resilience, pressure-related failure risk, and energy efficiency. Traditional reactive strategies often fail to address pipe deterioration evolution in water distribution systems (WDSs). To transition from passive maintenance to active intervention, this study proposes an operation optimization framework based on risk threshold in WDSs. First, the hierarchical pipe grouping strategy is employed to overcome the challenge of imbalanced failure data, enabling a robust LightGBM ensemble model to accurately predict pipe failure probabilities. Second, a Bayesian probability-ratio method is used to derive group- and material-specific RAPs as operational upper-pressure thresholds. Third, the multi-objective optimization model is developed, in which System resilience, energy efficiency, leakage and water age are taken into account. The RAPs are embedded as constraints in the optimization model, solved by NSGA-III. SHLG water distribution system with massive industrial demand is used as case study. Compared with the current operation, the preventive strategy reduced pump energy consumption by 22.0%, system leakage by 6.3%, and network-wide average maximum pressure by 8.7%, the number of high-risk pipe segments decreased from 1,780 to 586 (67.1%), while system resilience, I_AHR increased from 0.68 to 0.82. This integrated framework provides precise, interpretable pipe failure predictions and operationalizes them into RAP threshold, offering a scientifically rigorous solution to ensure the sustainable operation and maintain the pressure-related risk control of urban water systems.

Intelligent detection and tracking of surface objects is becoming increasingly important in water environment management. However, it remains a challenging problem in practical applications due to the complex environment, target scale and single-camera field of view issues. In this study, we propose a detection and tracking of floating targets method based on a multi-camera joint spatial strategy, which achieves intelligent monitoring of floating targets through interconnected surveillance cameras. Specifically, this study improves the network architecture of the Single Shot Multibox Detector (SSD) through the integration of a lightweight backbone network and feature pyramid network to improve the robustness of floating object detection. Then, a fast histogram of oriented gradient (FHOG) and a pyramid scale estimation strategy are introduced into the kernel correlation filter algorithm, and an improved image-matching algorithm is proposed to achieve accurate tracking and matching. Finally, a joint multi-camera relational optimization strategy is proposed to achieve continuous and accurate tracking of floating targets based on the single-camera parameter initialization. The proposed method is trained and compared with the state-of-the-art methods based on multiple scenarios. The comprehensive experimental results show that the proposed method can effectively cope with continuous detection and tracking in different scenarios, with IDF1, IDP and IDR reaching 87.23%, 89.37% and 84.37% respectively, and the speed reaching 29.83f/s. This work expands the intelligent detection and tracking of floating objects on the water surface to support integrated water environment management.

The release rules for large reservoirs are generally graded and judgment-based, allowing for some flexibility in managing uncertainty in forecast information. Ignoring this aspect can introduce bias in flood water resource utilization and flood risk analysis. This study proposes a dynamic control flood limit method for reservoirs, accounting for the uncertainty in forecast errors, to optimize the use of forecast data. Monte Carlo simulations are employed to model and assess uncertainty in the sample, and the error domain is systematically quantified to evaluate the probability that the observed inflow falls within the forecast error range. This determination is pivotal in ascertaining the utilization of forecast data, thereby facilitating the integration of multiple forecast lead times. The proposed method is applied to the Three Gorges Reservoir and compared with a traditional approach using single forecast data. Results show that: (1) the proposed method has been demonstrated to facilitate real-time dynamic decision-making, thereby enabling the selection of forecast data that exhibits both acceptable risk and optimal benefit. (2) during the rising flood stage of 20,190,724, this method has been shown to enhance power generation by 2.83 × 108 kWh while concomitantly reducing flood risk in comparison with conventional methods. Flood risks upstream and downstream are reduced by 0.415 and 0.005, respectively, without compromising the benefits of power generation. This methodology provides a reference for flood water resource management in basin reservoirs, providing decision-makers with a strategy that best aligns with their preferences.

In actual reservoir operations, water levels are often below the flood-limited water level, providing additional flood control storage and enabling the reservoirs to store more floodwater. This allows other reservoirs to reduce flood control storage through storage substitution, raising their water levels to enhance conservation benefits. The key question is how much the water level can raise, i.e., what is the substitution relationship of flood control storages. To address this issue, this study develops a flood control storage substitution model to utilize additional flood control storage without increasing the system's flood control risk. The cascade hydropower system of the Wudongde, Baihetan, Xiluodu, Xiangjiaba, and Three Gorges Project on the Yangtze River, China, is taken as a case study. Results indicate that the water level of the Three Gorges Project can be raised to 153 m when the operating water level of XLD and BHT are at dead water levels. The substituted ratio is less than 1.0 due to the discharge of the Three Gorges Project is reduced by substitution. The substituted ratio and upper bound of the substituted flood control storage are related with the volume of additional flood control storage, flood magnitudes and hydrographs, and the flood composition. In real-time operations, the storage substitution relationship can be used to adjust the operating water levels and increase power generation without increasing flood risk.

Groundwater resource management faces significant challenges due to groundwater overdraft and waterlogging. Establishing thresholds of the water table depth (WTD) is crucial to ascertain whether WTDs align with ranges conducive to the health of social‐ecological systems. However, existing studies often overlook multiple protection targets, dominant targets across different seasons, and spatial variations of thresholds. The long‐term effects of WTDs exceeding threshold ranges of the WTD also need to be further explored. Here we propose a novel framework for calculating grid‐scale thresholds across seasons, incorporating multiple targets. This framework calculates frequency, duration, and magnitude metrics, offering an evaluation of multiple groundwater management targets over decades. We apply this framework to the lower Tao'er River Basin in China, revealing threshold depths for shallow water tables ranges of 1.16–2.05 m and 1.16–4.05 m during non‐growth and growth periods, while threshold depths for deep water tables ranges from 6.28–33.54 m and 1.96–30.72 m, respectively. Climate change scenarios demonstrate minimal frequency changes but significant deterioration in duration and magnitude compared to the historical scenario. Grids with duration of transgressions more than 12 months expand by 1–2 times, while grids exceeding thresholds of the WTD by 2 m increase by 37%–81% under climate change and intensified pumping scenarios. A 20% increase in groundwater pumping leads to an average rise of 151%, 224%, and 147% deterioration in frequency, duration, and magnitude. Furthermore, 1%–6% of grids face dual challenges of groundwater storage reduction and waterlogging. These findings can inform groundwater resource management under various potential futures.

Uncertainty in forecast information can introduce flood risk into the process of reservoir flood water resource utilization. However, reservoirs exhibit an accommodative transformation relationship to forecast uncertainty. A flood risk analysis model that ignores this relationship may lead to biased flood risk calculations, which in turn may result in incorrect scheduling decisions. In this study, flood risk in the process of reservoir flood water resource utilization is divided into two components: pre-release volume risk (PVR) and pre-release timing risk (PTR), based on an analysis of the underlying risk mechanism. A flood risk calculation method, based on the conditional value-at-risk (CVaR) theory, is proposed. The accommodative transformation relationship of forecast uncertainty in reservoirs is quantified through numerical simulation, and the factors influencing this relationship are explored. Finally, a forecast information utilization method based on the accommodative transformation relationship is presented. The proposed method is validated using the Three Gorges Reservoir as a case study. The main research findings are as follows: (1) Reservoirs have an accommodation space for forecast uncertainty in flood water resource utilization. Flood risk only becomes significant when this uncertainty exceeds the accommodation space. Neglecting this relationship may lead to an overestimation of flood risk, hindering optimal flood water resource utilization. (2) The faster the inflow rises, the larger the accommodation space for forecast uncertainty in the reservoir, but this also increases the flood risk. (3) The forecast utilization approach based on accommodation space can increase power generation by 2.25 x 108 kWh without raising flood risk during a typical flood rise.

Recent advancements in deep learning (DL) have significantly improved hydrological modeling by extracting generalities from large-sample datasets and enhancing predictive accuracy. However, DL models often rely heavily on large volumes of data, which are often unavailable or insufficient in many real-world hydrological applications. This challenge has prompted interest in integrating DL with physically based hydrological models (PBHMs). This study explores such integration using differentiable programming with the Xin’anjiang model. We introduce two advanced model variants: the differentiable Xin’anjiang model (dXAJ), which retains the Xin’anjiang model’s structure while incorporating Long Short-Term Memory (LSTM) networks for parameter learning, and the dXAJnn model, which replaces the traditional evapotranspiration module of dXAJ model with a neural network. Both models were evaluated against the evolutionary algorithm-calibrated XAJ model (eXAJ) across five basins in the Three Gorge region of China and eight basins from the CAMELS dataset under varying data-limited conditions. Our results showed that both dXAJ and dXAJnn models outperformed the eXAJ model in streamflow prediction accuracy as they have different optimization mechanism, demonstrating that the local optimization mechanism in differentiable models (DMs) tends to generalize better during validation than global optimization approaches in data-limited contexts. The DMs also provided reliable evapotranspiration estimates, even without using evapotranspiration data for calibration. Although the dXAJnn model offered greater flexibility, it did not consistently yield better results and exhibited a tendency toward overfitting in certain basins. The study also found that both models require a minimum of three years of training data (including a one-year warm-up period) to achieve acceptable predictive performance, with longer data records further preventing overfitting. These findings underscore the ability of DMs to effectively balance data-driven techniques and physical mechanisms, highlighting the importance of sufficient training data.

Detection of floating objects in complicated aquatic environments has a wide range of applications, but it confronts significant hurdles due to the imbalance in detection accuracy and efficiency, and low domain generalization performance. To address these issues, this study proposes a novel floating object detection method based on double-labelled domain generalization. First, the Single Shot Multibox Detector (SSD) is improved by replacing the backbone network with a lightweight feature extraction network, and dynamic feature pyramid network is introduced to balance accuracy and efficiency. Then, this study initializes the improved SSD network based on the double-labelled data of the source domain, and filters the feature extractor bias, classification bias and location bias using pseudo-labelling and feature projection based on the un-labelled source domain data to minimize the bias to improve the domain generalization performance. The proposed method is trained on a self-constructed floating object dataset and is compared with state-of-the-art methods based on multiple scenarios. The results show that the proposed method achieves better performance in double-labelled domain generalization and conventional domain generalization tasks compared to other methods, achieving 70.33%, 22.38 f/s and 85.29% and 17.81 f/s in accuracy and speed respectively, satisfying the need for multi-scale floating object detection in complex environments and also alleviating the data labelling problem. This work effectively solves the problem of slow detection of floating objects due to the complex model structure and low generalization ability, and provides support for the rapid detection of floating objects in complex scenarios and the promotion of technology applications.

Many studies have evaluated the value of long-term inflow forecast for hydropower operation in high-precision watersheds and explored the impact of forecast uncertainty on hydropower benefits. However, few of them focuses on the low-accuracy regions and the specific impact of forecast errors on hydropower benefits. Taking Hunjiang cascaded system in China as an example, this study investigated the value of low accuracy long-term inflow forecast for hydropower operation, revealed the influence mechanism of forecast errors on hydropower benefits, and determined the critical threshold for accuracy of beneficial forecast. Results show that low accuracy forecast but higher than critical threshold can increase hydropower generation. Hunjiang has a low accuracy long-term forecast with qualified rate being 30-40 %, but annual hydropower generation can be increased by 14.86-29.58 million kWh. The influence mechanism of forecast errors on hydropower operation presents three typical scenarios, two of which are harmless. One is that the error tolerance of operation rules can avoid some decision-making errors; the other one is that when current misreporting is consistent with the inflow in a longer lead time, the misreporting may instead be beneficial. We conclude that low accuracy long-term forecast is valuable but higher accuracy brings higher benefits.

Floating materials seriously damage the landscape and ecosystem of rivers and visual surveillance has become an important technique for improving the water environment. However, it remains a challenging problem in practical applications due to small-scale targets and high scene complexity with many noise problems such as water wave disturbance, light and shadow change, and strong light emission. To address these issues, this study proposes a floating object detection and tracking method based on spatial–temporal information fusion. Specifically, this study improves the network architecture of the Single Shot Multibox Detector (SSD) by enhancing the high-resolution layers to adapt to the detection task of small floating targets. Then, an improved Kernel Correlation Filter (KCF) by introducing a fast histogram of oriented gradient (FHOG) and a pyramid scale estimation strategy is proposed to achieve the estimation of the position and size of floating objects. More significantly, a spatial–temporal information fusion strategy is applied to complement detection information with tracking information based on feature comparison. The proposed method is trained and compared with the state-of-the-art methods based on multiple scenarios. The results show that the proposed method has better performance than other methods in different scenarios, and achieves an average accuracy of more than 91% with a speed of 15.55 FPS, which prove that our method can well complete the detection and tracking task of floating objects. This work enriches the framework of “tracking by detection” and extends the application of floating object detection and tracking in surface vision.