The Sustainable Development Goal Satellite 1 (SDGSAT-1) is the world’s first science satellite dedicated to serving the United Nations 2030 Agenda for Sustainable Development. The Multispectral Imager for Inshore (MII) onboard SDGSAT-1 provide an alternative data source to effectively monitor the coastal water environments. In this study, the performances of MII in red tide detection were evaluated along the Guangdong middle coast. First, the radiometric performances of MII were estimated. The in-orbit signal-to-noise ratios (SNRs) of the blue and green bands were in the range of 270–360 while those of the red and near-infrared red bands were in the range of 100–140. The spectral resolution of MII is sufficient to generate atmospheric correction (AC) and red tide indexes calculation based on its own spectral bands. Two types of red tide indexes were compared. The results showed that the hue angle red tide index (HA-RI) better distinguished red tide and turbid water pixels. Two small red tides that occurred in Daya Bay and Mirs Bay were detected by HA-RI with areas of 13.1 and 5.7 km2, respectively. As the spatial resolution of MII has been improved to 10 m, its ability to detect small-scale red tides has been significantly enhanced, and more detailed spatial characteristics can be observed. In addition, multisource data are required to compensate for the weakness of SDGSAT-1 in temporal resolution for red tide tracing. The design of a three-satellite constellation will further enhance the monitoring capacities of SDGSAT series.

Shallow waters hold ecological and economic importance. Traditional reflectance forward models, such as Hydrolight and Monte Carlo (MC), are difficult to access or time-consuming. Two-stream models show great performance in radiative transfer simulation, but most of them are applied in other mediums, or ignore the asymmetry scattering characteristic of shallow water. In this paper, we propose a general multilayer analytical radiative transfer-based (GMART) model considering the realistic volume scattering function of water body, which is theoretically applicable for various shallow waters. GMART results are validated against Hydrolight and MC. The mean absolute percentage error (MAPE) of GMART and Hydrolight remote sensing reflectance (Rrs) among different combinations of the bottom depth, solar zenith angle and wind speed spans in ranges of 5.37%-28.86%, 3.37%-39.28%,and 4.91%-26.73%, with root mean square error (RMSE) values from 1.73×10-4-4.49×10-3, 2.68×10-4-1.85×10-2and 1.59×10-4-2.92×10-3sr-1, respectively, for the coral, sand, and seagrass bottom. Comparisons of the bidirectional reflectance function with Hydrolight and MC indicate the capacity of GMART to compute bidirectional reflectance. Additionally, a comparison with field measurements is carried out. The minimum MAPE and RMSE between the modeled and measured normalizedRrsamong several stations in the Sanya coastal area are 13.99%, and 0.061, respectively. The mean MAPE and RMSE among 6 stations are 21.38% and 0.083, respectively. The field validation underscores the feasibility of the model in shallow water radiative transfer. Compared with MC, the computational efficiency of GMART is much higher. The new model is shown to be an alternative and efficient tool capable of addressing shallow water-related challenges across diverse environmental conditions.

Efficient and resilient traffic management relies on accurate prediction of air traffic states. However, the complex spatial-temporal dependencies of air traffic networks make this task challenging. To address this issue, we propose a novel deep learning framework, named Physics-Informed Graph Attention Transformer (), which leverages real-world data and knowledge to predict essential air traffic state parameters. Our approach utilizes fine-grained traffic state detection to extract critical features from aviation databases. The model employs GAT-based spatial learning blocks with temporal Transformers to capture the dynamic spatial-temporal dependencies of data. A dynamic graph generator layer is also utilized to update the airport network’s topological structure adaptively, strengthening the model prediction’s effectiveness. Furthermore, the fluid queuing-theoretic PDEs are incorporated into the loss function, enhancing the model’s interpretability and reliability. Our framework is evaluated on real-world air traffic datasets from 36 major airport hubs within the US. Experimental results demonstrate that our proposed framework efficiently makes accurate predictions and outperforms eight baselines. In conclusion, our proposed framework has the potential to be applied in real-time decision-making systems for air traffic management and provides promising directions for future research. The code for our project is available at: https://github.com/ymlasu/para-atm-collection/tree/master/air-traffic-prediction/PIGAT.

We investigate multi-agent reinforcement learning for stochastic games with complex tasks, where the reward functions are non-Markovian. We utilize reward machines to incorporate high-level knowledge of complex tasks. We develop an algorithm called Q-learning with reward machines for stochastic games (QRM-SG), to learn the best-response strategy at Nash equilibrium for each agent. In QRM-SG, we define the Q-function at a Nash equilibrium in augmented state space. The augmented state space integrates the state of the stochastic game and the state of reward machines. Each agent learns the Q-functions of all agents in the system. We prove that Q-functions learned in QRM-SG converge to the Q-functions at a Nash equilibrium if the stage game at each time step during learning has a global optimum point or a saddle point, and the agents update Q-functions based on the best-response strategy at this point. We use the Lemke-Howson method to derive the best-response strategy given current Q-functions. The three case studies show that QRM-SG can learn the best-response strategies effectively. QRM-SG learns the best-response strategies after around 7500 episodes in Case Study I, 1000 episodes in Case Study II, and 1500 episodes in Case Study III, while baseline methods such as Nash Q-learning and MADDPG fail to converge to the Nash equilibrium in all three case studies.

Human-robot collaboration has been an important topic in the HRI communities. In this paper, we explore how robots can contribute to gas pipeline inspection work, and how they can support one of the most important elements of energy transportation infrastructure. To situate robots in the gas energy industry, we conducted a collaborative design study, where our co-designers were diverse stakeholders: from pipeline researchers to utility workers. The contribution of this paper is threefold: First, we explore gas pipeline work settings as a new context where robots can provide significant benefit, considering that public infrastructure is vast but understudied. Second, we collaboratively envisioned the design and use cases together with workers who are not often invited to human-robot collaboration research. Lastly, we address the importance of viewing humans in human-robot collaboration as “workers” whose roles and expertise are shaped within organizational dynamics. This study aims to shed light on the importance of a more nuanced understanding of work contexts and the positionality of robots within organizations.

Data augmentation is a widely used technique in machine learning to improve model performance. However, existing data augmentation techniques in natural language understanding (NLU) may not fully capture the complexity of natural language variations, and they can be challenging to apply to large datasets. This paper proposes the Random Position Noise (RPN) algorithm, a novel data augmentation technique that operates at the word vector level. RPN modifies the word embeddings of the original text by introducing noise based on the existing values of selected word vectors, allowing for more fine-grained modifications and better capturing natural language variations. Unlike traditional data augmentation methods, RPN does not require gradients in the computational graph during virtual sample updates, making it simpler to apply to large datasets. Experimental results demonstrate that RPN consistently outperforms existing data augmentation techniques across various NLU tasks, including sentiment analysis, natural language inference, and paraphrase detection. Moreover, RPN performs well in low-resource settings and is applicable to any model featuring a word embeddings layer. The proposed RPN algorithm is a promising approach for enhancing NLU performance and addressing the challenges associated with traditional data augmentation techniques in large-scale NLU tasks. Our experimental results demonstrated that the RPN algorithm achieved state-of-the-art performance in all seven NLU tasks, thereby highlighting its effectiveness and potential for real-world NLU applications.

Predicting separation errors in the daily tasks of air traffic controllers (ATCOs) is essential for the timely implementation of mitigation strategies before performance declines and the prevention of loss of separation and aircraft collisions. However, three challenges impede accurate separation errors forecasting: 1) compounding relationships between many human factors and control processes require sufficient operation process data to capture how separation errors occur and propagate within controller-in-the-loop processes; 2) previous human factor measurement approaches are disruptive to controllers’ daily operations because they use invasive sensors, such as electroencephalography (EEG) and electrocardiography (ECG), 3) errors accumulated in using the tasks and human behaviors for estimating system dynamics challenge accurate separation error predictions with sufficient leading time for proactive control actions. This study proposed a separation error prediction framework with a long leading time (>50 s) to address the above challenges, including 1) a multi-factorial model that characterizes the inter-relationships between task complexity, behavioral activity, cognitive load, and operational performance; 2) a multimodal data analytics approach to non-intrusively extract the task features (i.e., traffic density) from high-fidelity simulation systems and visual behavioral features (i.e., head pose, eyelid movements, and facial expressions) from ATCOs’ facial videos; 3) an encoder-decoder Long Short-Term Memory (LSTM) network to predict long-time-ahead separation errors by integrating multimodal features for reducing accumulated errors. A user study with six experienced ATCOs tested the proposed framework using the Phoenix Terminal Radar Approach Control (TRACON) simulator. The authors evaluated the model performance through two types of metrics: 1) point-level metrics, including precision, recall, and F1-score, and 2) sequence-level metrics, including alignment accuracy and sequence similarity. The results showed that 1) the model using the task and visual behavioral features significantly improved the prediction performance compared to the model using one single feature (eyelid movements), with an improvement of up to 26.95% in alignment accuracy for 10s-ahead prediction; 2) the model that combined task and visual behavioral features had a higher or comparable performance to models with different hybrid features, achieving an alignment accuracy of 82.38% for 50s-ahead error prediction; and (3) the proposed method outperformed three baseline models – Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and classic LSTM – by 8.21%, 3.47%, and 3.14% in alignment accuracy, respectively, for predicting 50s-ahead separation errors. These results suggest that the proposed model can effectively predict separation errors in air traffic control.

Mapping bathymetry with satellite-based imagery for optically shallow waters is important for the management of coastal areas. However, mapping ultra-high-spatial-resolution bathymetry (spatial-resolution < 5 m) for a wide range of coastal optically shallow waters is limited by high cost and the absence of in situ data. Therefore, in this study, a new downscaled bathymetric mapping approach (DBMA-RGB) was established using freely accessed Landsat-8 (spatial-resolution = 30 m) and cost-effective Google Earth Pro-exported ultra-high-spatial-resolution red-green-blue (RGB) imagery. The new approach introduced the scale-invariance assumption that the log-ratio model (LRM) which is calibrated at a high scale is valid at a low scale. To reduce the inversion errors from Landsat-8 imagery, a theoretical maximum detection depth (MDD) model was proposed for the optimization-based inversion method, which is the foundation of DBMA-RGB for working without in situ bathymetric data. The theoretical MDDs overall matched the true MDDs except for the abnormal values, with root mean square error and correlation coefficient values of 3.30 m and 0.68, respectively. Additionally, a set of correction coefficients was obtained to improve the accuracy of the water depth derived from Landsat-8 imagery. Finally, the DBMA-RGB was applied to the RGB imagery of sixteen coastal areas around the world, incorporating the bathymetric results from Landsat-8 imagery. The comparison results indicated that the DBMA-RGB yielded similar results to the LRM calibrated using in situ bathymetric data. In summary, DBMA-RGB demonstrated its feature in accurately mapping ultra-high-spatial-resolution bathymetry for a wide range of coastal optically shallow waters without in situ bathymetric data.

Optical satellite remote sensing (RS) is a time- and cost-effective approach for shallow-water bathymetry over large areas. The photogrammetric and physics-based methods, which do not requirein situdepth calibration data, are two implementations of RS-based bathymetry. This article compares these two bathymetric methods by using the same WorldView-2 imagery product of a study area in the Xisha Islands of the South China Sea. The focus is to investigate the difference between the resulting depths. The resulted accuracy of the photogrammetric method was slightly higher than that of the physics-based method based on the backward image, but both of them were significantly higher than that of the physics-based method based on the forward image. In the calm clear waters, the resulted depth of the photogrammetric method was closer to the ground truth than that of the physics-based method based on the backward image. In the shoal and adjacent waters where the water was turbid or larger waves were present, the photogrammetric method produced a highly noisy digital depth model (DDM), whereas the physics-based method derived a low-noise DDM from the backward image. In conclusion, the photogrammetric method outperforms the physics-based method in calm clear waters, but the physics-based method performs better than the photogrammetric method in shoal and adjacent waters that has turbid water or waves larger than the image spatial resolution. The photogrammetric DDM is noisier than the physics-based DDM. The accuracy of the physics-based method is also influenced by the non-optimal sun-target-sensor geometry of the images used.

A novel method for aircraft conflict resolution in air traffic management (ATM) using physics informed deep reinforcement learning (RL) is proposed. The motivation is to integrate prior physics understanding and model in the learning algorithm to facilitate the optimal policy searching and to present human-explainable results for display and decision-making. First, the information of intruders’ quantity, speeds, heading angles, and positions are integrated into an image using the solution space diagram (SSD), which is used in the ATM for conflict detection and mitigation. The SSD serves as the prior physics knowledge from the ATM domain which is the input features for learning. A convolution neural network is used with the SSD images for the deep reinforcement learning. Next, an actor-critic network is constructed to learn conflict resolution policy. Several numerical examples are used to illustrate the proposed methodology. Both discrete and continuous RL are explored using the proposed concept of physics informed learning. A detailed comparison and discussion of the proposed algorithm and classical RL-based conflict resolution is given. The proposed approach is able to handle arbitrary number of intruders and also shows faster convergence behavior due to the encoded prior physics understanding. In addition, the learned optimal policy is also beneficial for proper display to support decision-making. Several major conclusions and future work are presented based on the current investigation.

Collision-free path planning is a major challenge in managing unmanned aerial vehicles (UAVs) fleets, especially in uncertain environments. In this paper, we consider the design of UAV routing policies using multi-agent reinforcement learning, and propose a Multi-resolution, Multi-agent, Mean-field reinforcement learning algorithm, named3M-RL,for flight planning, where multiple vehicles need to avoid collisions with each other while moving towards their destinations. In the system we consider, each UAV makes decisions based on local observations, and does not communicate with other UAVs. The algorithm trains a routing policy using an Actor-Critic neural network with multi-resolution observations, including detailed local information and aggregated global information based on mean-field. The algorithm tackles the curse-of-dimensionality problem in multi-agent reinforcement learning and provides a scalable solution. We test our algorithm in different complex scenarios in both 2D and 3D space and our simulation results show that 3M-RL result in good routing policies.

Aircraft operations in the terminal area rely heavily on voice communications between pilots and air traffic controllers. This paper proposes a novel aircraft trajectory simulation framework by guiding the trajectory simulation following the voice command from controllers. Bayesian model selection is used for checking pilot compliances to controller commands with observed trajectories. This framework is named as Voice Communication-Augmented Simulation. The goal of the proposed study is to enable accurate trajectory predictions. The framework can act as a computer assistant for controllers to monitor pilot compliances and ensure safe operations. The proposed method is tested and validated with actual trajectory data from Sherlock Data Warehouse. The tests showed that the proposed framework can accurately simulate and monitor the flight level change of aircraft and update the approach procedure.

Sequential event prediction is a well-studied area and has been widely used in proactive management, recommender systems and healthcare. One major assumption of the existing sequential event prediction methods is that similar event sequence patterns in the historical record will repeat themselves, enabling us to predict future events. However, in reality, the assumption becomes less convincing when we are trying to predict rare or unique sequences. Furthermore, the representation of the event may be complex with hierarchical structures. In this paper, we aim to solve this issue by taking advantage of the multi-level or hierarchical representation of these rare events. We proposed to build a sequential Encoder-Decoder framework to predict the event sequences. More specifically, in the encoding layer, we built a hierarchical embedding representation for the events. In the decoding layer, we first predict the high-level events and the low-level events are generated according to a hierarchical graphical structure. We propose to link the encoding decoding layers with the temporal models for future event prediction. In this article, we further discussed applying the proposed model into the failure event prediction according to the aviation accident reports and have shown improved accuracy and model interpretability.

Chromophoric dissolved organic matter (CDOM) retrieval remains to be a challenging task in water color remote sensing research due to its highly spatial and temporal variability. In this paper, we present a novel CDOM retrieval algorithm that takes advantage of the sparse learning, which can simultaneously perform feature selection and parameter estimation. More specifically, by incorporating the band interaction terms into the original spectral matrix and let it be the basis matrix, then the inversion task can be converted to a classical sparse regression problem, namely LASSO, which can be efficiently solved by the coordinate descend algorithm. Experimental results conducted on both simulated and in-situ datasets have demonstrated the efficiency and superiority of the proposed method over some conventional empirical algorithms.

Imbalanced data widely exist in many high-impact applications. An example is in air traffic control, where among all three types of accident causes, historical accident reports with `personnel issues' are much more than the other two types (`aircraft issues' and `environmental issues') combined. Thus, the resulting data set of accident reports is highly imbalanced. On the other hand, this data set can be naturally modeled as a network, with each node representing an accident report, and each edge indicating the similarity of a pair of accident reports. Up until now, most existing work on imbalanced data analysis focused on the classification setting, and very little is devoted to learning the node representations for imbalanced networks. To bridge this gap, in this paper, we first propose Vertex-Diminished Random Walk (VDRW) for imbalanced network analysis. It is significantly different from the existing Vertex Reinforced Random Walk by discouraging the random particle to return to the nodes that have already been visited. This design is particularly suitable for imbalanced networks as the random particle is more likely to visit the nodes from the same class, which is a desired property for learning node representations. Furthermore, based on VDRW, we propose a semi-supervised network representation learning framework named ImVerde for imbalanced networks, where context sampling uses VDRW and the limited label information to create node-context pairs, and balanced-batch sampling adopts a simple under-sampling method to balance these pairs from different classes. Experimental results demonstrate that ImVerde based on VDRW outperforms state-of-the-art algorithms for learning network representations from imbalanced data.