Dynamic information about barriers impeding traffic flow is becoming an essential part of transportation analysis in the context of disruptive events. The automated detection of barriers from individual or vehicular tracking data previously relied on various statistical models of traffic counts and geometric patterns of mobility paths. To account for the variance of traffic flow in peripheral parts of a network, often due to limited data or device coverage, we propose a barrier detection method using trajectory data based on the randomized shortest path model, a probabilistic model for trip paths. By overlaying the spatial distribution of multiple trips, we model traffic counts on each edge as a Poisson process and infer potential barriers from edges with unexpectedly low counts. Therefore, the proposed method incorporates a spatial criterion of path distribution and a statistical criterion of traffic counts for barrier detection. We validated the proposed method on both synthetic data and real-world data, by comparing it with methods using statistical and temporal criteria. Our results allow us to discuss the bias of randomized shortest path model for finite-size networks, the advantages of spatial and temporal criteria for this task, and the capability limitations of the proposed method.

Reservoir operation and water division have significant altered the downstream hydrologic regime and ecological environment. The ecological problems, such as algal blooms, invasive alien species, and fish reproduction, in the middle and lower reaches of Han River basin in China is becoming prominent in recent years. The objective of this study is to propose a multi-objective optimal operation scheme that simultaneously considers the hydrologic alternation and ecological water demand. Based on the Indicators of Hydrologic Alteration (IHA), the Revised Range of Variability Approach (RRVA) was used to estimate the change of hydrologic regime. Two new ecological indexes, i.e., Water Quantity Level (WQL) and Hydrologic Alteration (HA), were also established. The ecological water demand in the downstream was determined and set as the constrain of outflow discharge in the reservoir operation. Four objectives (water supply, power generation, WQL and HA) operation model of the Danjiangkou Reservoir was established and solved by the Non-dominated Sorting Genetic Algorithm II, in which the Gaussian Radial Basis Functions (GRBFs) were constructed to fit the reservoir operation rules. The results revealed that overall degree of hydrological alteration in the downstream is 62.38

Microbes play a major role in soil biogeochemistry, yet it remains unclear how microbial physiology, particularly microbial nitrogen (N) use efficiency (NUE), regulates the soil C/N ratio to maintain long-term stoichiometric stability. Here, we refined the definition and quantification of microbial NUE in the latest C-N coupled Microbial-ENzyme Decomposition (MEND) model and validated it against field experimental data from subtropical broadleaf and pine forests. Our results show that microbial anabolism, reflected in high C use efficiency (CUE) and NUE, contributes to the stabilization of the soil C/N ratio. Long-term simulations, based on a calibration strategy combining experimental calibration with independent post-experiment evaluation, revealed insignificant difference in microbial CUE between broadleaf and pine forests, but a notable higher NUE in the broadleaf forest (0.64) than that in the pine forest (0.38). Soil C/N ratio decreased as CUE and NUE increased, with a stronger association for NUE. Additionally, following a 33

Recognizing the action of plastic bag taking from CCTV video footage represents a highly specialized and niche challenge within the broader domain of action video classification. To address this challenge, our paper introduces a novel benchmark video dataset specifically curated for the task of identifying the action of grabbing a plastic bag. Additionally, we propose and evaluate three distinct baseline approaches. The first approach employs a combination of handcrafted feature extraction techniques and a sequential classification model to analyze motion and object-related features. The second approach leverages a multiple-frame convolutional neural network (CNN) to exploit temporal and spatial patterns in the video data. The third approach explores a 3D CNN-based deep learning model, which is capable of processing video data as volumetric inputs. To assess the performance of these methods, we conduct a comprehensive comparative study, demonstrating the strengths and limitations of each approach within this specialized domain.

Although poor housekeeping leads to construction accidents, there is limited technological research on it. Existing methods for detecting poor housekeeping face many challenges, including limited explanations, lack of locating of poor housekeeping and annotated datasets. To address these challenges, this paper proposes the Housekeeping Change Detection Network (HCDN), integrating a feature fusion module and a large vision model. This paper introduces the approach to establish a change detection dataset (Housekeeping-CCD) focused on construction housekeeping, along with a housekeeping segmentation dataset. Experimental results of our Housekeeping-CCD dataset demonstrate that HCDN outperforms existing state-of-the-art (SOTA) methods, achieving average accuracy (89.32 %), mean IoU (76.97 %), and mean F-score (86.67 %). The contributions include significant performance improvements compared to existing methods, providing an effective tool for enhancing construction housekeeping and safety.

Image compression is of great significance for improving the efficiency of image transmission and storage. In this work, we proposed a scalable in-memory WalshHadamard Transform (WHT) using self-selective memristors for image compression. With the self-selective memristors and scalable in-memory WHT method, arbitrarily-size WHT computing can be realized with a fixed-size memristor array, and image compression is achieved with a small result error that the Peak Signal-to-Noise Ratio (PSNR) of the compressed result is 32 dB. In addition, non-ideal characteristics in devices and arrays are analyzed and compared with Discrete Cosine Transform (DCT), which shows that in-memory WHT has higher robustness.

The attention mechanism is essential to convolutional neural network (CNN) vision backbones used for sensing and imaging systems. Conventional attention modules are designed heuristically, relying heavily on empirical tuning. To tackle the challenge of designing attention mechanisms, this paper proposes a novel probabilistic attention mechanism. The key idea is to estimate the probabilistic distribution of activation maps within CNNs and construct probabilistic attention maps based on the correlation between attention weights and the estimated probabilistic distribution. The proposed approach consists of two main components: (i) the calculation of the probabilistic attention map and (ii) its integration into existing CNN architectures. In the first stage, the activation values generated at each CNN layer are modeled by using a Laplace distribution, which assigns probability values to each activation, representing its relative importance. Next, the probabilistic attention map is applied to the feature maps via element-wise multiplication and is seamlessly integrated as a plug-and-play module into existing CNN architectures. The experimental results show that the proposed probabilistic attention mechanism effectively boosts image classification accuracy performance across various CNN backbone models, outperforming both baseline and other attention mechanisms.

Curvilinear Structure segmentation has numerous applications in various fields, including providing a better understanding of defects such as cracks on roads or walls, thereby assessing the structural safety of buildings. Numerous curve segmentation methods based on Convolutional Neural Networks(CNN) have been developed in recent years. However, preserving feature integrity remains a challenge. To address this issue, this paper introduces a double dynamic network, called DD-Net, which consists of a dynamic structure for training and dynamic masking for inference. Firstly, to enhance the nonlinearity of the model, a paradigm to auto-adjust the CNN structure via a trainable module is formed for better fine-grained feature extraction. This dynamic structure (DS) paradigm enables a trade-off to extract or discard the feature data in the model training. Due to the data heterogeneity between the training and inference data, models may contain some useless nodes that are less effective during inference. The dynamic masking (DM) is proposed to omit the useless nodes based on the score difference between the train and inference feature statistics, thereby reducing redundant computations. To further improve the model's performance on thin-curve segmentation and to preserve feature integrity, we introduce a non-curve suppression (NCS) module. This module focuses on background information while considering foreground prediction to address noisy conditions. The experimental results show that our DD-Net achieves promising results on three benchmark datasets and outperforms state-of-the-art curve segmentation models.

River algal blooms have become a global environmental problem due to their large impact range and environmental hazards. However, the complex mechanisms underlying these blooms make prediction and prevention challenging. Here, we employed empirical dynamic modeling (EDM) and machine learning to reveal the causes and predict diatom blooms from 2003 to 2017 in the Han River of China. The diatom cell density ranged from 0.1 to 5.1 x 10(7) cells L-1, whereas algal blooms often lasted for 10 days with density exceeding 10(7) cells L-1. The EDM results elucidated that, under consistent high nutrient concentrations, algal blooms were primarily regulated by eight environmental factors: water temperature in the Han River; water levels, flow velocities, and streamflow discharges in the Han River and the Yangtze River; and water level variation in the Han River. The poor performance (coefficient of determination R-2 < 0) of the multiple linear regression, EDM, and random forest models indicated the challenge of predicting algae density. Therefore, we used machine learning classification models to predict algal blooms occurrences. With the resampling techniques to account for imbalanced data, machine learning models achieved perfect classification prediction (Kappa value = 1) of the 13 algae-bloom events with a 10-day lead time during a 15-year period, providing an important reference for preemptive warning of riverine algal blooms. Plain Language Summary River algal blooms have become a global environmental problem due to their large impact range and environmental hazards. The Han River is one of the most frequent areas of river algal blooms in China, but there is no clear conclusion on the causes of algal blooms formation. In this study, we combined two methods to reveal the causes and predict the occurrence of algal blooms in the Han River. Our results show that the prediction of algal blooms in the Han River is feasible. These blooms are primarily affected by the water temperature and seven other hydrological variables in the Han and Yangtze River. This study provides a starting point for the preemptive warning of riverine algal blooms.

We introduced StyleEntity, a zero- shot image manipulation model that utilizes named entities as proxies during its training phase. This strategy enables our model to manipulate images using unseen textual descriptions during inference, all within a single training phase. Additionally, we proposed an inference technique termed Prompt Ensemble Latent Averaging (PELA). PELA averages the manipulation directions derived from various named entities during inference, effectively eliminating the noise directions, thus achieving stable manipulation. In our experiments, StyleEntity exhibited superior performance in a zero-shot setting compared to other methods. The code, model weights, and datasets are available at https://github.com/feng-zhida/StyleEntity.

Data transformation of the reference soil organic matter (SOM) decomposition rates (kref), often derived as turnover times or in alternative formats, is commonly used to develop ecological models for projecting the persistence of SOM. However, the effects of reciprocal or logarithmic transformation of kref on model performance and edaphic-climatic patterns remain uncertain. Here, we convert published kref values into reciprocal or logarithmic formats and establish machine learning models between the transformed kref and edaphic-climatic predictors. We show that models trained with the transformed kref exhibit a 11.6%-68.4% reduction in performance upon re-conversion to kref compared to those trained with the original kref. The variable importance analysis identifies distinct key predictors governing the original kref and its transformed counterparts. This suggests that data transformation alters the relative significance of predictors without necessarily improving kref prediction performance. Consequently, our study underscores the importance of directly focusing on the original values rather than alternative representations when dissecting a given variable's patterns and mechanisms in ecological modeling. Data transformation caused reduced predictive performance upon re-conversion to kref compared to the original kref. Models trained with data transformation exhibited divergent edaphic-climatic controls for different formatted kref. This study underscores the importance of directly focusing on original values rather than alternative representations. The first-order model is commonly used to simulate the dynamics of soil organic matter decomposition. The reference decomposition rates (kref) in the first-order model are usually transformed into reciprocal or logarithmic formats, which may affect model performance as well as the regulatory patterns of edaphic-climatic factors. In this study, we investigated the effects of data transformation on predictive performance and variable importance. We found that models trained with transformed kref exhibited reduced performance upon re-conversion to kref, compared to those trained with the original kref. In addition, data transformation altered the distribution and range of the dependent variable, leading to divergent controlling roles of explanatory variables pertaining to models trained with different formatted kref. Our study underscores the need to apply models directly to the original variables rather than their transformed representations.

As warmer temperatures enhance atmospheric moisture, hydrological droughts tend to intensify in most regions of the globe. Consequently, younger generations are expected to face a more severe risk of hydrological drought during their lifetimes, emphasizing the critical issue of intergenerational inequity due to climate change. To quantify exposure to hydrological drought across generations, we constructed a cascade model chain for drought simulation using hybrid terrestrial models, based on 5 GCM outputs under SSP5-85, five hydrological models and a deep learning model. We then projected future univariate and bivariate hydrological drought evolution in 4091 river basins, and quantified lifetime exposure to drought for the age groups born in 2020 and 1960. Drought severity and duration are projected to increase substantially in the Eastern America, Southern Brazil and Western Europe, over 79% of basins. Extreme droughts far beyond historical records are expected to become more frequent and impact Western Europe in particular. Of note, the exposure of the different age groups to hydrological drought shows a notable disequilibrium. Exposure of people born in 2020 to hydrological drought hazards is projected to increase by 12% over the late 21st century compared to those born in 1960, indicating that the acceleration of climate change is expected to increase the lifetime risk of future generations. The exposure factor of the newborns is 1.4 times higher than that of 80 years of age under warming condition. Our findings underscore that future drought conditions under extreme warming pose a significant threat to the living conditions of younger generations.

A thorough understanding of the ecological impacts behind the hydrologic alteration is still insufficient and hinders the watershed management. Here, we used eco-flow indicators, multiple hydrological indicators, and fluvial biodiversity to investigate the ecological flow in different temporal scales. The case study in the Han River shows a decrease in high flows contributed to the decrease in eco-surplus and increase in eco-deficit in summer and autumn, while the decrease in eco-deficit can be attributed to the change of low flow in spring. An integrated hydrologic alteration was over 48% degree and was under moderate ecological risk degree in impact period I, while DHRAM scores showed the Huangzhuang station faced a high ecological risk degree in impact period II. The decrease (increase) in total seasonal eco-surplus (eco-deficit) was identified after alteration with the change in seasonal eco-flow indicators contributions. Shannon index showed a decreasing trend, indicating the degradation of fluvial biodiversity in the Han River basin. Eco-flow indicators such as eco-surplus and eco-deficit are in strong relationships with 32 hydrological indicators and can be accepted for ecohydrological alterations at multiple temporal scales. This study deepens the understanding of ecological responses to hydrologic alteration, which may provide references for water resources management and ecological security maintenance.

The development of geocapabilities has been a concern in geographical education, and the application of geospatial technologies such as geographical information systems (GIS) has been popularized in both geography and non-geography courses. However, there are still gaps in how GIS courses could be designed to promote students' geocapabilities. In order to facilitate GIS teaching and learning rooted in educational theory, we designed a course on story maps based on the First Principles of Instruction. We examine students' perceptions of learning experiences and geocapabilities through two rounds of course delivery and student survey. According to student feedback and instructor ratings, the designed course activities supported their learning of subject content and developed both of their geography-related and general skills. However, the objective for the development of geocapabilities was found to be too broad or vague for students. We discuss the application of First Principles of Instruction, the relevance of considering geocapabilities, and the implication of the study findings for GIS pedagogy.

River algal blooms pose a global ecological and environmental problem, resulting in serious consequences for watershed ecosystems and human health. However, current research has yet to fully identify factors influencing river algal blooms or accurately predict chlorophyll-a concentration, a key indicator, at various lead-times using high-frequency data. Here, we employed Empirical Dynamic Modeling (EDM) and machine learning techniques to forecast daily chlorophyll-a concentration in the Han River of China. The Convergent Cross-Mapping (CCM) analysis revealed the causal relationships between chlorophyll-a concentration and eight variables: pH, total nitrogen, water level in the Han River, water level in the Yangtze River, water temperature, permanganate index, sunlight duration and total phosphorus in the Han River. Subsequent multivariate EDM models with three lead-times (i.e., 1-day, 5-day, and 10-day) only showed acceptable performance in model training (R-2 = 0.19-0.73, MAE = 2.76-9.17 mu g/L), while exhibited poor predictive performance in model testing (R-2 = - 0.50-0, MAE = 9.64-13.09 mu g/L). However, Gradient Boosting Machine (GBM) and Random Forest (RF) models with three lead-times exhibited robust performance in both model training (R-2 >= 0.8, MAE < 4 mu g/L) and testing (R-2 > 0.6, MAE < 6 mu g/L), demonstrating that machine learning models were more suitable than multivariate EDM models for reliably predicting algal blooms. Our study contributes valuable tools for predicting daily chlorophyll-a concentration. The methods presented herein hold broad applicability and offer insights into predicting the entire process of river algal blooms based on daily monitoring data.

This paper aims to present an online teaching pedagogic experience for the practice-oriented computer vision course during the COVID-19 pandemic. COVID-19 has been disruptive to the education system worldwide, particularly to the computer vision course that usually requires face-to-face lectures and project collaboration during the study. This paper addresses three fundamental questions in teaching computer vision courses: 1) how to design the course topic and adapt to the online teaching format?; 2) how to conduct hybrid project collaboration in a hybrid mode?; 3) how to conduct the course assessment efficiently online? More specifically, this paper presents the pedagogic experience, including learning objectives, course curriculum structure, teaching methodologies, as well as final holistic assessments. The presented approach is an effective way of teaching practical computer vision courses, as verified by feedback from students. These experiences can be insightful to other lecturers who need to design, develop and deliver similar courses in the post pandemic era.

全球变暖改变了气候系统的热力和动力环境,影响了陆地圈与大气圈的生物地球化学循环过程,对极端降水及陆地碳收支产生显著影响.现有研究发现极端降水一般呈Hook气候响应结构,但较少分析其形成机理,也未能量化降水对生态系统生产力的影响,不利于科学评估未来极端气候灾害及潜在生态风险.为了揭示全球极端降水的热力学驱动机理及生态水文效应,文章结合十余套卫星遥感、大气再分析、气候模式、陆面模式、机器学习重构和通量站观测数据集,评估了水-热-碳通量对极端降水的反馈效应,通过大气热力和动力机制的降水效应解释了Hook结构形成机理;基于ISIMIP3b框架下5个全球气候模式集合,预估了未来Hook结构迁移路径及其对极端降水的影响,最后结合CLM4.5陆面模式探讨了气候变化下降水的碳收支效应.研究发现,极端降水事件往往伴随着剧烈的水-热交换,降水与生态系统生产力及碳收支过程存在非线性响应关系;全球大多数地区极端降水强度上升、历时缩短,三维降水事件更趋集中;大气动力作用是形成Hook结构的关键因素,但该结构并不稳定,未来随全球变暖发生动态迁移,可能导致本世纪末极端降水强度增长10～40%;较充沛的降水有助于生态系统固碳,气候变化下碳收支对降水的响应特征较为稳定.

从幸福河提出的背景和建设目标出发,结合新阶段水利高质量发展的六条实施路径,构建适用于幸福河的评价指标体系包括:防洪、水资源、水环境、水生态、水节约、水文化共6个准则层及30个具体指标.根据需求层次理论对评价区域分级,将"幸福"程度分为1～5星级.使用主客观赋权方法,综合幸福河指数和可持续发展指标,对湖北省县(市)区汉江幸福河进行评价.结果表明:22个地区大部分为3星级,上中游幸福程度高于下游,影响幸福程度主要因素是水污染和水资源总量减少.评价结果符合客观规律,可为湖北省幸福河的评价考核提供技术支撑.

基于Budyko框架及径流变化情势指标同气象因子的拟合关系,拓展Budyko方程并得到微分方程.选择汉江上游安康和白河水文站的年均径流、汛期平均径流和非汛期平均径流资料系列,开展径流情势变化及归因研究.结果表明:所有径流指标均发生变异且明显减小;多元对数线性回归模型拟合的相关系数大于0.90,能够较好预估径流变化情势指标,并捕捉到径流变化情势指标同气象参数之间的非线性关系;基于Budyko假设的互补关系法性能优于全微分法,气候(流域下垫面)变化对安康站年均径流量、汛期平均径流量和非汛期平均径流量贡献的绝对值分别为35.89％(64.11％)、34.58％(65.42％)和71.12％(28.88％),对白河站年均径流量、汛期平均径流量和非汛期平均径流量贡献的绝对值分别为34.82％(65.18％)、26.29％(73.71％)和35.11％(64.89％).

Intelligent sensing systems have been fueled to make sense of visual sensory data to handle complex and difficult real-world sense-making challenges due to the rapid growth of computer vision and machine learning technologies [...]