Long-term Temporal Dynamics Research Articles

Long-term dynamics of grasslands and livestock in Norwegian cultural landscapes: implications for a sustainable transition of rural livelihoods

ContextAbandonment of cultural landscape practices has had a notable impact on grasslands and domestic livestock that depend on them. This affects the prerequisites for sustainable transitions of cultural landscapes, which combine traditional livelihoods and novel ones like tourism.ObjectivesThe aims of this study were to explore (1) the long-term temporal dynamics of grassland and livestock in a regional gradient from coastal to inland landscapes in Norway and (2) the temporal development of tourism types among regions.MethodsUsing three regions as case studies, with 62 municipalities, we analyzed (1) the temporal dynamics of grassland and livestock using agricultural census data from 1918 to 1999 and (2) two tourism types. Kruskal–Wallis and ANOVA were used to examine whether the relative changes in grassland areas and livestock units, respectively, differed significantly among regions. A PCA was conducted to explore relationships between grassland and livestock types. The proportions of tourist categories were compared.ResultsThe grassland area and thematic resolution of census data declined over time. Grassland areas correlated with domestic livestock units. Multivariate analysis explained 68% of the variation in grassland and livestock types among municipalities. There was a notable increase in the number of tourists, with summer tourism dominating where the cultural landscape was well conserved.ConclusionsOur study underscores the importance of integrating regional historical trajectories for the conservation and use of valuable cultural landscapes, thereby providing sustainable transitions that combine traditional livelihoods and new types as tourism. Additional in-depth studies are needed to understand the detailed drivers of these changes.

Long-term temporal dynamics of total and potentially active microbiota affect the biogas quality from the anaerobic digestion of vinasse in a pilot-scale hybrid anaerobic reactor

The microbiota was evaluated through a time series of the 16S rrna gene in a pilot-scale Hybrid Anaerobic Reactor treating vinasse. The microbial composition, evenness and diversity differed between the DNA and cDNA, the latter showing a better response for %CH4 along monitoring of the microbiota. Moreover, the microbial composition and richness were distinct between the reactor's compartments, possibly contributing to the COD removal in the sludge blanket. The abundant Methanospirillum, Methanomassiliicoccaceae, Bacteroidetes-vadinHA17, Paludibacteraceae-H1, Prolixibacteraceae, Ruminococcaceae, Syntrophobacter and Desulfovibrio contributed to the %CH4, as well as the redundancy of fermentation and hydrogenotrophic and methylotrophic methanogenesis, and the evenness and diversity of potentially active microbiota and metabolisms. The microbial changes after the middle of reactor's operation required new adaptations and impacted the biogas quality, but did not affect COD removal. This study brings important knowledge about the microbial temporal dynamics, plasticity and contribution to the anaerobic digestion in pilot-scale reactors.

Edge-Enhanced TempoFuseNet: A Two-Stream Framework for Intelligent Multiclass Video Anomaly Recognition in 5G and IoT Environments

Surveillance video analytics encounters unprecedented challenges in 5G and IoT environments, including complex intra-class variations, short-term and long-term temporal dynamics, and variable video quality. This study introduces Edge-Enhanced TempoFuseNet, a cutting-edge framework that strategically reduces spatial resolution to allow the processing of low-resolution images. A dual upscaling methodology based on bicubic interpolation and an encoder–bank–decoder configuration is used for anomaly classification. The two-stream architecture combines the power of a pre-trained Convolutional Neural Network (CNN) for spatial feature extraction from RGB imagery in the spatial stream, while the temporal stream focuses on learning short-term temporal characteristics, reducing the computational burden of optical flow. To analyze long-term temporal patterns, the extracted features from both streams are combined and routed through a Gated Recurrent Unit (GRU) layer. The proposed framework (TempoFuseNet) outperforms the encoder–bank–decoder model in terms of performance metrics, achieving a multiclass macro average accuracy of 92.28%, an F1-score of 69.29%, and a false positive rate of 4.41%. This study presents a significant advancement in the field of video anomaly recognition and provides a comprehensive solution to the complex challenges posed by real-world surveillance scenarios in the context of 5G and IoT.

Spatial-temporal deep learning model based on Similarity Principle for dock shared bicycles ridership prediction

Demand prediction plays a critical role in traffic research. The key challenge of traffic demand prediction lies in modeling the complex spatial dependencies and temporal dynamics. However, there is no mature and widely accepted concept to support the solution of the above challenge. Essentially, a prediction model combined with similar objects in temporal and spatial dimensions could obtain better performance. This paper proposes a concept called the Similarity-based Principle (SP), which is applied to improve the prediction performance of deep learning models in complex traffic scenarios. For the temporal components, the long-term temporal dynamics in contemporaneous historical data for ridership are extracted by the Stacked Autoencoder (SAE) method. For the spatial components, the activity-based spatial geographic information (ABG-information) is used to capture the spatial correlation of the traffic network, which is reflected in the daily activities of humans. Specifically, the SP is applied to a Spatio-temporal Graph Convolutional Neural Network (STGCNN) model. In the case study, the Similarity-based Principle Spatio-temporal Graph Convolutional Neural Network (SP-STGCNN) model predicts demand for bicycle sharing in San Francisco. The results show that the SP effectively improves the model's performance. The prediction accuracy is enhanced by up to 10.34% compared with STGCNN. For spatial relationships, the model using the geographic information attribute performs better than that using the road information attribute and the distance attribute. It is proved that the construction of the Spatio-temporal model-based similarity principle can improve the performance.

Skeleton-Based Action Segmentation With Multi-Stage Spatial-Temporal Graph Convolutional Neural Networks

The ability to identify and temporally segment fine-grained actions in motion capture sequences is crucial for applications in human movement analysis. Motion capture is typically performed with optical or inertial measurement systems, which encode human movement as a time series of human joint locations and orientations or their higher-order representations. State-of-the-art action segmentation approaches use multiple stages of temporal convolutions. The main idea is to generate an initial prediction with several layers of temporal convolutions and refine these predictions over multiple stages, also with temporal convolutions. Although these approaches capture long-term temporal patterns, the initial predictions do not adequately consider the spatial hierarchy among the human joints. To address this limitation, we recently introduced multi-stage spatial-temporal graph convolutional neural networks (MS-GCN). Our framework replaces the initial stage of temporal convolutions with spatial graph convolutions and dilated temporal convolutions, which better exploit the spatial configuration of the joints and their long-term temporal dynamics. Our framework was compared to four strong baselines on five tasks. Experimental results demonstrate that our framework is a strong baseline for skeleton-based action segmentation.

Unprecedentedly high soil carbon stocks and their spatial variability in a seasonally dry Atlantic Forest in Brazil

There is a significant gap in the literature concerning the potential of tropical soils to stock carbon in the Brazilian Atlantic Forest biome. Furthermore, the key drivers responsible for the variability in carbon storage in tropical soils remain unknown. Therefore, experimental studies are crucial to highlight the importance of Atlantic Forests in storing organic matter, considering that forest soils are one of the most important carbon reservoirs on Earth. This study aimed: (i) to determine soil carbon stocks at the depths of 0–5 cm, 5–10 cm, 10–20 cm, 20–30 cm, 30–40 cm, 40–60 cm, and 60–100 cm in a forest fragment located in the Atlantic Forest hotspot; (ii) to describe the temporal variability of soil carbon stocks; and (iii) to identify the main biotic and abiotic drivers that influence their temporal and spatial variations. Soil carbon stocks at a 1-meter depth ranged from 201 to 396 Mg C/ha, with a spatial variability of 22.3 %. The highest accumulation of soil carbon in the 1-meter depth was in December (290.8 Mg C/ha), while the lowest was in August (246.1 Mg C/ha), highlighting the soil carbon stocks’ seasonality in the Atlantic Forest. Concerning the vertical distribution, ∼50 % of the soil carbon stocks were found in the 0–30 cm layer. The main drivers of carbon spatial distribution varied across different soil layers: (i) litterfall and long-term spatiotemporal patterns of throughfall were the primary drivers in the shallower layers (0–5 cm and 5–10 cm), while (ii) variability in tree sizes and long-term temporal dynamics of soil moisture were the main drivers in the deeper layers. Oxisols’ characteristics were crucial for the unprecedently high carbon stocks observed in the deep layers (67.7 Mg C/ha). Thus, the Atlantic Forest-Oxisol association is a prominent strategy to improve carbon stocks in Brazilian tropical soils.

A novel positional encoded attention-based Long short-term memory network for state of charge estimation of lithium-ion battery

Accurate estimation of the state of charge (SOC) in battery management systems is critical to the global electrification revolution in various fields. This article introduces a novel long short-term memory (LSTM) network enhanced with positional encoding and a time-step internal attention mechanism for accurate SOC estimation. The proposed method adeptly utilizes positional encoding to transform one-dimensional battery data sequences into a multi-dimensional space through a series of sinusoidal functions parameterized by a diverse frequency spectrum, capturing unique temporal characteristics and relationships inherent in battery data. The enhanced data is then processed by the time-step internal attention, which adeptly discerns interdependencies among input variables at each specific time interval. This dynamic adjustment captures evolving relationships, ensuring a transformed data representation. This enriched data is subsequently processed by the LSTM network, addressing long-term correlations and temporal dynamics. The model was trained and tested on a public dataset that includes the data from different working conditions and temperatures. Primary training followed a leave-one-out (LOO) approach focusing on working conditions, with further validation across temperature variations. Empirical analysis demonstrated that the proposed method surpassing baseline methods, attaining the optimal root mean square error and mean absolute error of 0.91 % and 0.68 % respectively.

Long-term mediation of a sexual risk-reduction intervention for South African adolescents.

Black adolescents in South Africa are disproportionately affected by HIV. A cluster-randomized controlled experiment examining the effects of a sexual risk-reduction intervention successfully reduced self-reported intercourse and unprotected intercourse. Based on long-term follow-up assessments, the present research examines theoretical constructs that could potentially mediate the intervention effects and how time and gender, respectively, moderated the mediation. The behavioral outcome was measured by asking whether participants had had any vaginal sex in the past 3 months. Mediation and moderated mediation were tested based on the 3-, 6-, 12-, 42-, and 54-month postintervention outcomes. Three variables through which the sexual risk-reduction intervention had a significant mediated effect on the behavioral outcome were identified: abstinence career opportunities outcome expectancy (α × β product = -0.086, 95% asymmetric confidence interval [ACI] [-0.126, -0.047]), expected parental approval of sexual intercourse (α × β product = -0.061, [-0.102, -0.025]), and self-efficacy to avoid sexual-risk situations (α × β product = -0.022, [-0.049, -0.001]). The moderated mediation analysis showed that gender moderated the intervention's effects on abstinence prevention outcome expectancy (B = -0.186, SEB = 0.079, p = .019), expected parental approval of sexual intercourse (B = 0.143, SEB = 0.058, p = .013), and self-efficacy to avoid sexual-risk situations (B = -0.293, SEB = 0.112, p = .009). The moderated mediation analysis also revealed that time moderated the effects of the intervention on abstinence career opportunities outcome expectancy (B = -0.293, SEB = 0.106, p = .006), self-efficacy to avoid sexual-risk situations (B = 0.335, SEB = 0.060, p < .001), and cultural myths regarding HIV transmission (B = 0.138, SEB = 0.042, p = .001); and the association between four theoretical constructs and the behavioral outcome: abstinence career opportunities outcome expectancy (B = -0.267, SEB = 0.104, p = .001), self-efficacy to refuse sex (B = -0.132, SEB = 0.043, p = .002), self-efficacy to avoid sexual-risk situations (B = -0.093, SEB = 0.055, p = .009), and HIV risk-reduction knowledge (B = -0.286, SEB = 0.134, p = .003). The present study identifies theoretical constructs that mediated the intervention effects on the sexual behavior among South African adolescents for an extended period of time. The findings also reveal gender differences in psychological mechanisms initiated by a sexual risk-reduction intervention and the long-term temporal dynamics of the intervention. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Video Transformers: A Survey.

Transformer models have shown great success handling long-range interactions, making them a promising tool for modeling video. However, they lack inductive biases and scale quadratically with input length. These limitations are further exacerbated when dealing with the high dimensionality introduced by the temporal dimension. While there are surveys analyzing the advances of Transformers for vision, none focus on an in-depth analysis of video-specific designs. In this survey, we analyze the main contributions and trends of works leveraging Transformers to model video. Specifically, we delve into how videos are handled at the input level first. Then, we study the architectural changes made to deal with video more efficiently, reduce redundancy, re-introduce useful inductive biases, and capture long-term temporal dynamics. In addition, we provide an overview of different training regimes and explore effective self-supervised learning strategies for video. Finally, we conduct a performance comparison on the most common benchmark for Video Transformers (i.e., action classification), finding them to outperform 3D ConvNets even with less computational complexity.

Simulating conversations: A Markov chain model of a central speaker's mnemonic influence over a group of communicating listeners.

Through their selective rehearsal, Central Speakers can reshape collective memory in a group of listeners, both by increasing accessibility for mentioned items (shared practice effects) and by decreasing relative accessibility for related but unmentioned items (socially shared retrieval induced forgetting, i.e., SSRIF). Subsequent networked communication in the group can further modify these mnemonic influences. Extant empirical work has tended to examine such downstream influences on a Central Speaker's mnemonic influence following a relatively limited number of interactions - often only two or three conversations. We develop a set of Markov chain simulations to model the long-term dynamics of such conversational remembering across a variety of group types, based on reported empirical data. These models indicate that some previously reported effects will stabilize in the long-term collective memory following repeated rounds of conversation. Notably, both shared practice effects and SSRIF persist into future steady states. However, other projected future states differ from those described so far in the empirical literature, specifically: the amplification of shared practice effects in communicational versus solo remembering non-conversational groups, the relatively transient impact of social (dis)identification with a Central Speaker, and the sensitivity of communicating networks to much smaller mnemonic biases introduced by the Central Speaker than groups of individual rememberers. Together, these simulations contribute insights into the long-term temporal dynamics of collective memory by addressing questions difficult to tackle using extant laboratory methods, and provide concrete suggestions for future empirical work.

Model of toxic fibrosis in Wistar rats: morphological and molecular-genetic parameters of the transition point to cirrhosis

BACKGROUND: To date, when studying the cellular and molecular genetic mechanisms of liver fibrogenesis in experimental rat models, no attention is paid to the transition point of fibrosis to cirrhosis as a separate stage. As this pathology develops, changes in the cell phenotype and gene expression are dynamic in nature, so it is necessary to evaluate them in a long-term temporal dynamics.&#x0D; AIM: The aim of this work is to study the morphological and molecular genetic changes in the liver of Wistar rats during nodular parenchymal rearrangement.&#x0D; METHODS: Fibrosis and cirrhosis of the liver in male Wistar rats was induced with a freshly prepared solution of thioacetamide, which was administered intragastrically through a tube at a dose of 200 mg/kg of body weight 2 times a week for 13 weeks. The level of mRNA of the tweak (tnfsf12), fn14 (tnfrsf12a), ang, vegfa, cxcl12 (sdf), and mmp-9 genes in the liver was detected by real-time polymerase chain reaction. Immunohistochemical study was performed on paraffin sections. -SMA, FAP, CD68, CD206, CX3CR1, CD45 were used as markers. The area of interlobular veins and interlobular arteries was measured (m2). The number of sinusoidal capillaries and interlobular veins was counted.&#x0D; RESULTS: Based on the results obtained, it is possible to establish the transition point of fibrosis to cirrhosis as an independent separate stage of fibrogenesis. This transition was identified at stage F5, and the process itself from F4/F5 to F6.&#x0D; With the growth of fibrous tissue and nodular restructuring of the liver parenchyma, no progression of dystrophic processes and an increase in the zones of necrosis and necrobiosis of hepatocytes were noted. The number of -SMA+ and FAP+ cells in the period F4F5 did not change (p=0.2073 and p=0.3775, respectively). At the same time, significant F6 cirrhosis was accompanied by an increase in their number by 1.5 times (p 0.00001). Differences in the number of CD68+ cells were revealed only at the F4/F5 stage (2.0 times higher than the control, p 0.00001). The number of CD206+, CX3CR1+ and CD45+ cells remained the same. An increase in the number of interlobular veins (p 0.00001) and a decrease in sinusoidal capillaries (p 0.00001) were found compared to the control.&#x0D; The transition to cirrhosis was characterized by changes in the expression levels of tweak, fn14, ang, vegfa, cxcl12, and mmp-9 mRNAs, as well as the presence and strength of the relationship between them. Significant correlations were revealed between the target genes (r=0.50.84; p 0.01).&#x0D; CONCLUSION: The transition point of fibrosis to cirrhosis has special morphological and molecular genetic features that expand knowledge about pathomorphological changes in the liver.

Exploring Spatio–Temporal Graph Convolution for Video-Based Human–Object Interaction Recognition

Video-based human-object interaction recognition is a challenging task since the state of objects as well as their correlations change constantly in the video. Existing methods mainly use 3DCNN or use separate components ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., GCN + RNN) to model the spatial correlation or the temporal correlation respectively, but ignore modeling spatio-temporal correlations simultaneously and long-term temporal dynamics of objects. In this paper, we propose a novel model, named Spatio-Temporal Interaction Graph Parsing Networks (STIGPN), for human-object interaction recognition in videos. STIGPN captures both spatial and temporal correlations simultaneously and thus can capture intra-frame and inter-frame dependencies efficiently and effectively. To model long-term temporal dynamics of objects, we introduce spatio-temporal feature enhancement, which can improve the detection of the salient human-object interaction pairs. We explore three types of spatio-temporal graph convolutions to simultaneously capture the spatio-temporal correlations and assess their effectiveness as the basic building block of STIGPN. Extensive experiments on CAD-120, Something-Else and Charades datasets show that our proposed solution leads to competitive results compared with the state-of-the-art methods. Code for STIGPN is available at: https://github.com/NingWang2049/STIGPN2.

Using the Google Earth Engine cloud-computing platform to assess the long-term spatial temporal dynamics of land use and land cover within the Letaba watershed, South Africa

Population growth and environmental shifts have elevated the pressure on land use and cover (LULC), necessitating vital management and adaptive strategies to preserve the balance between ecosystem services and human well-being in watersheds. It's pivotal to understand the implications of human-induced shifts from natural to human-dominated surroundings This study utilized Google Earth Engine (GEE) to analyze 31 years of LULC changes in Letaba watershed. Using GEE's random forest, LULC classes were mapped with 93% to 99% accuracy across four timeframes (1990, 2000, 2010, and 2021). Trends revealed declining water bodies (-4%), bare surfaces (-2%), natural forest (-3%), and grassland (-3%), while shrublands, plantations, and built-up areas increased at annual rates of 41%, 24%, and 47% respectively. This transformation reflects population-driven shifts, necessitating adaptive strategies. Given the importance of plantations for income, embracing climate-smart agriculture could ensure long-term food and environmental security, thus addressing the evolving dynamics in Letaba watershed.

DeepCredNet: Empowering Credibility Detection in Online Media Posts with Robust Deep Neural Networks

The rampant dissemination of misinformation and fake news through online media necessitates effective methods for detecting credibility. This study proposes a novel approach that employs a deep neural network (DNN) architecture, specifically Long Short-Term Memory (LSTM), to accurately assess the credibility of online media posts. The proposed LSTM-based DNN model leverages the inherent sequential nature of textual content and metadata associated with the posts. By capturing long-term dependencies and temporal dynamics, the model effectively learns intricate patterns and features crucial for credibility detection. A large labelled dataset comprising both credible and non-credible posts is employed to train the LSTM-based DNN. The input features encompass textual content, user information, and contextual details such as the post source and timing. The LSTM layers within the network enable the model to capture and retain relevant information over extended periods, enhancing its discriminative capabilities. Experimental evaluations validate the efficacy of the proposed approach, showcasing its ability to discern between credible and non-credible online media posts with high accuracy and robust performance. The real-time applicability of this method enables prompt credibility assessment, offering valuable support in combating misinformation and aiding users in making informed decisions while consuming online media.

MRE-Net: Multi-Rate Excitation Network for Deepfake Video Detection

The current social media is flooded with hyper realistic face-synthetic videos due to the explosion of DeepFake technology that has brought a serious impact on human society security, which calls for further exploring on deepfake video detection methods. Existing methods attempt to isolated capture spatial artifacts or extract the homogeneous temporal inconsistency to detect deepfake video, but little attention has been paid to the exploitation of dynamic spatial-temporal inconsistency. To mitigate this issue, in this paper, we propose a novel Multi-Rate Excitation Network (MRE-Net) to effectively excite dynamic spatial-temporal inconsistency from the perspective of multiple rates for deepfake video detection. The proposed MRE-Net is composed of two components: Bipartite Group Sampling (BGS) and multiple rate branches. The BGS draws the entire video into multiple bipartite groups with different rates to cover various face motion dynamic evolution. We further design multiple rate branches to capture both short-term and long-term spatial-temporal inconsistency from corresponding bipartite groups of BGS. Concretely, for the early stages of the multi-rate branches, Momentary Inconsistency Excitation (MIE) module is developed to encode the spatial artifacts and intra-group short-term temporal inconsistency. Meanwhile, for the last stages of the multi-rate branches, Longstanding Inconsistency Excitation (LIE) module is constructed to perceive inter-group long-term temporal dynamics. Extensive experiments and visualizations conducted on four popular datasets demonstrate the effectiveness of the proposed method against state-of-the-art deepfake detection methods.

Tracing Time-Varying Characteristics of Meteorological Drought through Nonstationary Joint Deficit Index

Abstract Standardized precipitation index (SPI) is one of the frequently used meteorological drought indices. However, the time-varying characteristics observed in the historical precipitation data questions the reliability of SPI and motivated the development of nonstationary SPI. To overcome some of the limitations in the existing nonstationary drought indices, a new framework for drought index is proposed, incorporating the temporal dynamics in the precipitation. The proposed drought index is developed by coupling the joint deficit index with the extended time sliding window–based nonstationary modeling (TSW-NSM). The proposed nonstationary joint deficit index (NJDI) detects the signature of nonstationarity in the distribution parameter and models both long-term (i.e., trend) and short-term (i.e., step-change) temporal dynamics of distribution parameters. The efficacy of NJDI is demonstrated by employing it to identify the meteorological drought-prone areas over India. The changes observed in the distribution parameter of rainfall series reveal an increasing number of dry days in recent decades all over India, except the northeast. Comparison of NJDI and stationary joint deficit index (JDI) reveals that JDI overestimates drought when frequent severe dry events are clustered and underestimates when these events are scattered, which indicates that the traditional index is biased toward the lowest magnitude of precipitation while classifying the drought. Moreover, NJDI could closely capture historical droughts and their spatial variations, thereby reflecting the temporal dynamics of rainfall series and the changes in the pattern of dry events over India. NJDI proves to be a potentially reliable index for drought monitoring in a nonstationary climate. Significance Statement Drought is one of the most severe natural disasters and is expected to intensify under a warming climate. There has been progress on developing newer methodologies to characterize drought severity under the changing climate. However, some limitations remain in capturing the temporal changes of the precipitation, especially the nonstationarity in variance (rapid increases in extreme precipitation versus the average precipitation). We propose an extension to the time sliding window approach to capture the nonstationarity in variance and mean while incorporating short-term (step-change) and long-term (trend) temporal dynamics. We apply the methodology to a subcontinent-sized heterogeneous country of India with gridded rainfall dataset (0.25° × 0.25°, 1901–2013) from the India Meteorological Department. We find that the majority of grids exhibit negative and positive trends in shape and scale parameters, respectively, which ultimately leads to an increase in the number of drier events, whereas a contradictory pattern is exhibited in the northwest Indian region (decrease in the number of dry days). The proposed index gives a potential framework for drought monitoring under the changing climate and can be extended to develop a multivariate nonstationary joint deficit index by incorporating other hydrological variables (e.g., soil moisture, diurnal temperature range, streamflow).

Long-Short Term Spatio-Temporal Aggregation for Trajectory Prediction

Pedestrian trajectory prediction in crowd scenes plays a significant role in intelligent transportation systems. The main challenges are manifested in learning motion patterns and addressing future uncertainty. Typically, trajectory prediction is considered in two dimensions, including temporal dynamics modeling and social interactions capturing. For temporal dependencies, although existing models based on recurrent neural networks (RNNs) or convolutional neural networks (CNNs) achieve high performance on short-term prediction, they still suffer from limited scalability for long sequences. For social interactions, previous graph-based methods only consider fixed features but ignore dynamic interactions between pedestrians. Considering that the transformer network has a strong capability of capturing spatial and long-term temporal dynamics, we propose Long-Short Term Spatio-Temporal Aggregation (LSSTA) network for human trajectory prediction. First, a modern variant of graph neural networks, named spatial encoder, is presented to characterize spatial interactions between pedestrians. Second, LSSTA utilizes a transformer network to handle long-term temporal dependencies and aggregates the spatial and temporal features with a temporal convolution network (TCN). Thus, TCN is combined with the transformer to form a long-short term temporal dependency encoder. Additionally, multi-modal prediction is an efficient way to address future uncertainty. Existing auto-encoder modules are extended with static scene information and future ground truth for multi-modal trajectory prediction. Experimental results on complex scenes demonstrate the superior performance of our method in comparison to existing approaches.

Trends in nutrients in the Changjiang River

Better documentation and understanding of long-term temporal dynamics of nutrients in watersheds are necessary to support effective water quality management. We examined the hypothesis that the recent management of fertilizer use and pollution control in the Changjiang River Basin could govern the fluxes of nutrients from the river to the sea. Results based on historical data since 1962 and surveys in recent years show that concentrations of dissolved inorganic nitrogen (DIN) and phosphorus (DIP) in the mid- and downstream reaches were higher than those in the upper reaches due to intensive anthropogenic activities, while dissolved silicate (DSi) was distributed evenly from the up- to downstream reaches. Fluxes of DIN and DIP increased rapidly, and DSi declined during the 1962–1980 and 1980–2000. After the 2000s, concentrations and fluxes of DIN and DSi remained almost unchanged; those of DIP remained stable until the 2010s and slightly decreased afterward. The decline in fertilizer use explains 45 % of the variance in the decline of DIP flux, followed by pollution control, groundwater and water discharge. As a result, the molar ratio of DIN:DIP, DSi:DIP and ammonia:nitrate varied largely during 1962–2020, and the excess DIN relative to DIP and DSi lead to increased limitations of silicon and phosphorus. A turning point probably occurred for nutrient fluxes in the Changjiang River in the 2010s, with the pattern of DIN from continuous increase to stability and DIP from increase to decrease. This decline in phosphorus in the Changjiang River has many similarities with the rivers worldwide. The continued basin nutrient management is likely to have a major influence on river nutrient delivery and therefore may control coastal nutrient budget and ecosystem stability.

Observed and dark diversity dynamics over millennial time scales: fast life-history traits linked to expansion lags of plants in northern Europe.

Global change drivers (e.g. climate and land use) affect the species and functional traits observed in a local site but also its dark diversity-the set of species and traits locally suitable but absent. Dark diversity links regional and local scales and, over time, reveals taxa under expansion lags by depicting the potential biodiversity that remains suitable but is absent locally. Since global change effects on biodiversity are both spatially and temporally scale dependent, examining long-term temporal dynamics in observed and dark diversity would be relevant to assessing and foreseeing biodiversity change. Here, we used sedimentary pollen data to examine how both taxonomic and functional observed and dark diversity changed over the past 14 500 years in northern Europe. We found that taxonomic and functional observed and dark diversity increased over time, especially after the Late Glacial and during the Late Holocene. However, dark diversity dynamics revealed expansion lags related to species' functional characteristics (dispersal limitation and stress intolerance) and an extensive functional redundancy when compared to taxa in observed diversity. We highlight that assessing observed and dark diversity dynamics is a promising tool to examine biodiversity change across spatial scales, its possible causes, and functional consequences.

Short- and long-term effects of continuous compost amendment on soil microbiome community

Organic amendment, and especially the use of composts, is a well-accepted sustainable agricultural practice. Compost increases soil carbon and microbial biomass, changes enzymatic activity, and enriches soil carbon and nitrogen stocks. However, relatively little is known about the immediate and long-term temporal dynamics of agricultural soil microbial communities following repeated compost applications. Our study was conducted at two field sites: Newe Ya'ar (NY, Mediterranean climate) and Gilat (G, semi-arid climate), both managed organically over 4 years under either conventional fertilization (0, zero compost) or three levels of compost amendment (20, 40 and 60 m3/ha or 2, 4, 6 L/m2). Microbial community dynamics in the soils was examined by high- and low-time-resolution analyses. Annual community composition in compost-amended soils was significantly affected by compost amendment levels in G (first, second and third years) and in NY (third year). Repeated sampling at high resolution (9–10 times over 1 year) showed that at both sites, compost application initially induced a strong shift in microbial communities, lasting for up to 1 month, followed by a milder response. Compost application significantly elevated alpha diversity at both sites, but differed in the compost–dose correlation effect. We demonstrate higher abundance of taxa putatively involved in organic decomposition and characterized compost-related indicator taxa and a compost-derived core microbiome at both sites. Overall, this study describes temporal changes in the ecology of soil microbiomes in response to compost vs. conventional fertilization.