Coffee which is prepared from the grinded roasted seeds of harvested coffee cherries, is one of the most consumed beverage and traded commodity, globally. To manually monitor the coffee field regularly, and inform about plant and soil health, as well as estimate yield and harvesting time, is labor-intensive, time-consuming and error-prone. Some recent studies have developed sensors for estimating coffee yield at the time of harvest, however a more inclusive and applicable technology to remotely monitor multiple parameters of the field and estimate coffee yield and quality even at pre-harvest stage, was missing. Following precision agriculture approach, we employed machine learning algorithm YOLO, for image processing of coffee plant. In this study, the latest version of the state-of-the-art algorithm YOLOv7 was trained with 324 annotated images followed by its evaluation with 82 unannotated images as test data. Next, as an innovative approach for annotating the training data, we trained K-means models which led to machine-generated color classes of coffee fruit and could thus characterize the informed objects in the image. Finally, we attempted to develop an AI-based handy mobile application which would not only efficiently predict harvest time, estimate coffee yield and quality, but also inform about plant health. Resultantly, the developed model efficiently analyzed the test data with a mean average precision of 0.89. Strikingly, our innovative semi-supervised method with an mean average precision of 0.77 for multi-class mode surpassed the supervised method with mean average precision of only 0.60, leading to faster and more accurate annotation. The mobile application we designed based on the developed code, was named CoffeApp, which possesses multiple features of analyzing fruit from the image taken by phone camera with in field and can thus track fruit ripening in real time.

Graph self-supervised learning (SSL), including contrastive and generative approaches, offers great potential to address the fundamental challenge of label scarcity in real-world graph data. Among both sets of graph SSL techniques, the masked graph autoencoders (e.g., GraphMAE)--one type of generative method--have recently produced promising results. The idea behind this is to reconstruct the node features (or structures)--that are randomly masked from the input--with the autoencoder architecture. However, the performance of masked feature reconstruction naturally relies on the discriminability of the input features and is usually vulnerable to disturbance in the features. In this paper, we present a masked self-supervised learning framework GraphMAE2 with the goal of overcoming this issue. The idea is to impose regularization on feature reconstruction for graph SSL. Specifically, we design the strategies of multi-view random re-mask decoding and latent representation prediction to regularize the feature reconstruction. The multi-view random re-mask decoding is to introduce randomness into reconstruction in the feature space, while the latent representation prediction is to enforce the reconstruction in the embedding space. Extensive experiments show that GraphMAE2 can consistently generate top results on various public datasets, including at least 2.45% improvements over state-of-the-art baselines on ogbn-Papers100M with 111M nodes and 1.6B edges.

Data assimilation is often viewed as a framework for correcting short-term error growth in dynamical climate model forecasts. When viewed on the time scales of climate however, these short-term corrections, or analysis increments, can closely mirror the systematic bias patterns of the dynamical model. In this study, we use convolutional neural networks (CNNs) to learn a mapping from model state variables to analysis increments, in order to showcase the feasibility of a data-driven model parameterization which can predict state-dependent model errors. We undertake this problem using an ice-ocean data assimilation system within the Seamless system for Prediction and EArth system Research (SPEAR) model, developed at the Geophysical Fluid Dynamics Laboratory, which assimilates satellite observations of sea ice concentration every 5 days between 1982--2017. The CNN then takes inputs of data assimilation forecast states and tendencies, and makes predictions of the corresponding sea ice concentration increments. Specifically, the inputs are states and tendencies of sea ice concentration, sea-surface temperature, ice velocities, ice thickness, net shortwave radiation, ice-surface skin temperature, sea-surface salinity, as well as a land-sea mask. We find the CNN is able to make skillful predictions of the increments in both the Arctic and Antarctic and across all seasons, with skill that consistently exceeds that of a climatological increment prediction. This suggests that the CNN could be used to reduce sea ice biases in free-running SPEAR simulations, either as a sea ice parameterization or an online bias correction tool for numerical sea ice forecasts.

Early detection of volcanic eruptions is of major importance for protecting human life. Ground deformation and changes in seismicity, geochemistry, petrology, and gravimetry are used to assess volcanic activity before eruptions. Studies on Mt. Etna (Italy) have demonstrated that vegetation can be affected by pre-eruptive activity before the onset of eruptions. During two consecutive years before Mt. Etna's 2002/2003 flank eruption, enhanced vegetation index (NDVI) values were detected along a distinct line which later developed into an eruptive fissure. However, the mechanisms by which volcanic activity can lead to changes in pre-eruption tree growth processes are still not well understood. We analysed ${\delta}^{13}$C, ${\delta}^{18}$O and $^{14}$C in the rings of the survived trees growing near to the line where the pre-eruptive increase in NDVI was observed in order to evaluate whether the uptake of water vapour or fossil volcanic CO2 could have contributed to the enhanced NDVI. We found a dramatic decrease in ${\delta}^{18}$O in tree rings formed before 2002/2003 in trees close to the eruption fissure, suggesting uptake of volcanic water by trees during pre-eruptive magma degassing. Moist conditions caused by outgassing of ascending magma may also have led to an observed reduction in tree-ring ${\delta}^{13}$C following the eruption. Furthermore, only ambiguous evidence for tree uptake of degassed CO2 was found. Our results suggest that additional soil water condensed from degassed water vapour may have promoted photosynthesis, explaining local increases in NDVI before the 2002/2003 Mt. Etna flank eruption. Tree-ring oxygen stable isotopes might be used as indicators of past volcanic eruptions.

The era of transfer learning has revolutionized the fields of Computer Vision and Natural Language Processing, bringing powerful pretrained models with exceptional performance across a variety of tasks. Specifically, Natural Language Processing tasks have been dominated by transformer-based language models. In Natural Language Inference and Natural Language Generation tasks, the BERT model and its variants, as well as the GPT model and its successors, demonstrated exemplary performance. However, the majority of these models are pretrained and assessed primarily for the English language or on a multilingual corpus. In this paper, we introduce GreekBART, the first Seq2Seq model based on BART-base architecture and pretrained on a large-scale Greek corpus. We evaluate and compare GreekBART against BART-random, Greek-BERT, and XLM-R on a variety of discriminative tasks. In addition, we examine its performance on two NLG tasks from GreekSUM, a newly introduced summarization dataset for the Greek language. The model, the code, and the new summarization dataset will be publicly available.

Change detection in remote sensing imagery is essential for a variety of applications such as urban planning, disaster management, and climate research. However, existing methods for identifying semantically changed areas overlook the availability of semantic information in the form of existing maps describing features of the earth's surface. In this paper, we leverage this information for change detection in bi-temporal images. We show that the simple integration of the additional information via concatenation of latent representations suffices to significantly outperform state-of-the-art change detection methods. Motivated by this observation, we propose the new task of *Conditional Change Detection*, where pre-change semantic information is used as input next to bi-temporal images. To fully exploit the extra information, we propose *MapFormer*, a novel architecture based on a multi-modal feature fusion module that allows for feature processing conditioned on the available semantic information. We further employ a supervised, cross-modal contrastive loss to guide the learning of visual representations. Our approach outperforms existing change detection methods by an absolute 11.7\% and 18.4\% in terms of binary change IoU on DynamicEarthNet and HRSCD, respectively. Furthermore, we demonstrate the robustness of our approach to the quality of the pre-change semantic information and the absence pre-change imagery. The code is available at https://github.com/mxbh/mapformer.

Utilizing differential atmospheric forces in the Very Low Earth Orbits (VLEO) regime for the control of the relative motion within a satellite formation is a promising option as any thrusting device has tremendous effects on the mission capacity due to the limited weight and size restrictions of small satellites. One possible approach to increase the available control forces is to reduce the mass of the respective satellites as well as to increase the available surface area. However, satellites of these characteristics suffer from rapid orbital decay and consequently have a reduced service lifetime. Therefore, achieving higher control forces is in contradiction to achieving a minimum orbital decay of the satellites, which currently represents one of the biggest challenges in the VLEO regime. In this work, the geometry of a given reference satellite, a 3UCubeSat, is optimized under the consideration of different surface material properties for differential lift and drag control applications while simultaneously ensuring a sustained VLEO operation. Notably, not only the consideration of sustainability but also the optimization with regard to differential lift is new in literature. It was shown that the advantageous geometries strongly depend on the type of gas-surface interaction and thus, two different final designs, one for each extreme type, are presented. In both cases, improvements in all relevant parameters could be achieved solely via geometry adaptions.

Despite the recent advances in deep neural networks, standard convolutional kernels limit the applications of these networks to the Euclidean domain only. Considering the geodesic nature of the measurement of the earth's surface, remote sensing is one such area that can benefit from non-Euclidean and spherical domains. For this purpose, we propose a novel Graph Neural Network architecture for spatial and spatio-temporal classification using satellite imagery. We propose a hybrid attention method to learn the relative importance of irregular neighbors in remote sensing data. Instead of classifying each pixel, we propose a method based on Simple Linear Iterative Clustering (SLIC) image segmentation and Graph Attention GAT. The superpixels obtained from SLIC become the nodes of our Graph Convolution Network (GCN). We then construct a region adjacency graph (RAG) where each superpixel is connected to every other adjacent superpixel in the image, enabling information to propagate globally. Finally, we propose a Spatially driven Attention Graph Neural Network (SAG-NN) to classify each RAG. We also propose an extension to our SAG-NN for spatio-temporal data. Unlike regular grids of pixels in images, superpixels are irregular in nature and cannot be used to create spatio-temporal graphs. We introduce temporal bias by combining unconnected RAGs from each image into one supergraph. This is achieved by introducing block adjacency matrices resulting in novel Spatio-Temporal driven Attention Graph Neural Network with Block Adjacency matrix (STAG-NN-BA). We evaluate our proposed methods on two remote sensing datasets namely Asia14 and C2D2. In comparison with both non-graph and graph-based approaches our SAG-NN and STAG-NN-BA achieved superior accuracy on all the datasets while incurring less computation cost. The code and dataset will be made public via our GitHub repository.

Mutually beneficial interactions between plant and pollinators play an essential role in the biodiversity, stability of the ecosystem and crop production. Despite their immense importance, rapid decline events of pollinators are common worldwide in past decades. Excessive use of chemical pesticides is one of the most important threat to pollination in the current era of anthropogenic changes. Pesticides are applied to the plants to increase their growth by killing harmful pests and pollinators accumulates toxic pesticides from the interacting plants directly from the nectar and pollen. This has a significant adverse effect on the pollinator growth and the mutualism which in turn can cause an abrupt collapse of the community however predicting the fate of such community dynamics remains a blur under the alarming rise in the dependency of chemical pesticides. We mathematically modeled the influence of pesticides in a multispecies mutualistic community and used 105 real plant-pollinator networks sampled worldwide as well as simulated networks, to assess its detrimental effect on the plant-pollinator mutualistic networks. Our results indicate that the persistence of the community is strongly influenced by the level of pesticide and catastrophic and irreversible community collapse may occur due to pesticide. Furthermore, a species rich, highly nested community with low connectance and modularity has greater potential to function under the influence of pesticide. We finally proposed a realistic intervention strategy which involves the management of the pesticide level of one targeted plant from the community. We show that our intervention strategy can significantly delay the collapse of the community. Overall our study can be considered as the first attempt to understand the consequences of the chemical pesticide on a plant-pollinator mutualistic community.

Identifying words that impact a task's performance more than others is a challenge in natural language processing. Transformers models have recently addressed this issue by incorporating an attention mechanism that assigns greater attention (i.e., relevance) scores to some words than others. Because of the attention mechanism's high computational cost, transformer models usually have an input-length limitation caused by hardware constraints. This limitation applies to many transformers, including the well-known bidirectional encoder representations of the transformer (BERT) model. In this paper, we examined BERT's attention assignment mechanism, focusing on two questions: (1) How can attention be employed to reduce input length? (2) How can attention be used as a control mechanism for conditional text generation? We investigated these questions in the context of a text classification task. We discovered that BERT's early layers assign more critical attention scores for text classification tasks compared to later layers. We demonstrated that the first layer's attention sums could be used to filter tokens in a given sequence, considerably decreasing the input length while maintaining good test accuracy. We also applied filtering, which uses a compute-efficient semantic similarities algorithm, and discovered that retaining approximately 6\% of the original sequence is sufficient to obtain 86.5\% accuracy. Finally, we showed that we could generate data in a stable manner and indistinguishable from the original one by only using a small percentage (10\%) of the tokens with high attention scores according to BERT's first layer.

Since large number of high-quality remote sensing images are readily accessible, exploiting the corpus of images with less manual annotation draws increasing attention. Self-supervised models acquire general feature representations by formulating a pretext task that generates pseudo-labels for massive unlabeled data to provide supervision for training. While prior studies have explored multiple self-supervised learning techniques in remote sensing domain, pretext tasks based on local-global view alignment remain underexplored, despite achieving state-of-the-art results on natural imagery. Inspired by DINO, which employs an effective representation learning structure with knowledge distillation based on global-local view alignment, we formulate two pretext tasks for self-supervised learning on remote sensing imagery (SSLRS). Using these tasks, we explore the effectiveness of positive temporal contrast as well as multi-sized views on SSLRS. We extend DINO and propose DINO-MC which uses local views of various sized crops instead of a single fixed size in order to alleviate the limited variation in object size observed in remote sensing imagery. Our experiments demonstrate that even when pre-trained on only 10% of the dataset, DINO-MC performs on par or better than existing state-of-the-art SSLRS methods on multiple remote sensing tasks, while using less computational resources. All codes, models, and results are released at https://github.com/WennyXY/DINO-MC.

NASA's Earth Surface Mineral Dust Source Investigation (EMIT) mission seeks to use spaceborne imaging spectroscopy (hyperspectral imaging) to map the mineralogy of arid dust source regions. Here we apply recent developments in Joint Characterization (JC) and the spectral Mixture Residual (MR) to explore the information content of data from this novel mission. Specifically, for a mosaic of 20 spectrally diverse scenes we find: 1) a generalized three-endmember (Substrate, Vegetation, Dark; SVD) spectral mixture model is capable of capturing the preponderance (99% in 3 dimensions) of spectral variance with low misfit (99% of pixels with RMSE < 3.7%); 2) manifold learning (UMAP) is capable of identifying spatially coherent, physically interpretable clustering relationships in the spectral feature space; 3) UMAP yields results that are at least as informative when applied to the MR as when applied to raw reflectance; 4) SVD fraction information usefully contextualizes UMAP clustering relationships, and vice-versa (JC); and 5) when EMIT data are convolved to spectral response functions of multispectral instruments (Sentinel-2, Landsat 8/9, Planet SuperDove), SVD fractions correlate strongly across sensors but UMAP clustering relationships for the EMIT hyperspectral feature space are far more informative than for simulated multispectral sensors. Implications are discussed for both the utility of EMIT data in the near-term, and for the potential of high SNR spaceborne imaging spectroscopy more generally, to transform the future of optical remote sensing in the years and decades to come.

Monitoring the rock-physics properties of the subsurface is of great importance for reservoir management. For either oil and gas applications or CO2 storage, seismic data are a valuable source of information for tracking changes in elastic properties which can be related to fluids saturation and pressure changes within the reservoir. Changes in elastic properties can be estimated with time-lapse full-waveform inversion. Monitoring rock-physics properties, such as saturation, with time-lapse full-waveform inversion is usually a two-step process: first, elastic properties are estimated with full-waveform inversion, then, the rock-physics properties are estimated with rock-physics inversion. However, multiparameter time-lapse full-waveform inversion is prone to crosstalk between parameter classes across different vintages. This leads to leakage from one parameter class to another, which, in turn, can introduce large errors in the estimated rock-physics parameters. To avoid inaccuracies caused by crosstalk and the two-step inversion strategy, we reformulate time-lapse full-waveform inversion to estimate directly the changes in the rock-physics properties. Using Gassmann's model, we adopt a new parameterization containing porosity, clay content, and water saturation. In the context of reservoir monitoring, changes are assumed to be induced by fluid substitution only. The porosity and clay content can thus be kept constant during time-lapse inversion. We compare this parameterization with the usual density-velocity parameterization for different benchmark models. Results indicate that the proposed parameterization eliminates crosstalk between parameters of different vintages, leading to more accurate estimation of saturation changes. We also show that using the parameterization based on porosity, clay content, and water saturation, the elastic changes can be monitored more accurately.

A key challenge for much of the machine learning work on remote sensing and earth observation data is the difficulty in acquiring large amounts of accurately labeled data. This is particularly true for semantic segmentation tasks, which are much less common in the remote sensing domain because of the incredible difficulty in collecting precise, accurate, pixel-level annotations at scale. Recent efforts have addressed these challenges both through the creation of supervised datasets as well as the application of self-supervised methods. We continue these efforts on both fronts. First, we generate and release an improved version of the Agriculture-Vision dataset (Chiu et al., 2020b) to include raw, full-field imagery for greater experimental flexibility. Second, we extend this dataset with the release of 3600 large, high-resolution (10cm/pixel), full-field, red-green-blue and near-infrared images for pre-training. Third, we incorporate the Pixel-to-Propagation Module Xie et al. (2021b) originally built on the SimCLR framework into the framework of MoCo-V2 Chen et al.(2020b). Finally, we demonstrate the usefulness of this data by benchmarking different contrastive learning approaches on both downstream classification and semantic segmentation tasks. We explore both CNN and Swin Transformer Liu et al. (2021a) architectures within different frameworks based on MoCo-V2. Together, these approaches enable us to better detect key agricultural patterns of interest across a field from aerial imagery so that farmers may be alerted to problematic areas in a timely fashion to inform their management decisions. Furthermore, the release of these datasets will support numerous avenues of research for computer vision in remote sensing for agriculture.

Diabetes Mellitus is a metabolic disorder which may result in severe and potentially fatal complications if not well-treated and monitored. In this study, a quantitative analysis of the data collected using CGM (Continuous Glucose Monitoring) devices from eight subjects with type 2 diabetes in good metabolic control at the University Polyclinic Agostino Gemelli, Catholic University of the Sacred Heart, was carried out. In particular, a system of ordinary differential equations whose state variables are affected by a sequence of stochastic perturbations was proposed and used to extract more informative inferences from the patients' data. For this work, Matlab and R programs were used to find the most appropriate values of the parameters (according to the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC)) for each patient. Fitting was carried out by Particle Swarm Optimization to minimize the ordinary least squares error between the observed CGM data and the data from the ODE model. Goodness of fit tests were made in order to assess which probability distribution was best suitable for representing the waiting times computed from the model parameters. Finally, both parametric and non-parametric density estimation of the frequency histograms associated with the variability of the glucose elimination rate from blood were conducted and their representative parameters assessed from the data. The results show that the chosen models succeed in capturing most of the glucose fluctuations for almost every patient.

Automatically extracting roads from satellite imagery is a fundamental yet challenging computer vision task in the field of remote sensing. Pixel-wise semantic segmentation-based approaches and graph-based approaches are two prevailing schemes. However, prior works show the imperfections that semantic segmentation-based approaches yield road graphs with low connectivity, while graph-based methods with iterative exploring paradigms and smaller receptive fields focus more on local information and are also time-consuming. In this paper, we propose a new scheme for multi-task satellite imagery road extraction, Patch-wise Road Keypoints Detection (PaRK-Detect). Building on top of D-LinkNet architecture and adopting the structure of keypoint detection, our framework predicts the position of patch-wise road keypoints and the adjacent relationships between them to construct road graphs in a single pass. Meanwhile, the multi-task framework also performs pixel-wise semantic segmentation and generates road segmentation masks. We evaluate our approach against the existing state-of-the-art methods on DeepGlobe, Massachusetts Roads, and RoadTracer datasets and achieve competitive or better results. We also demonstrate a considerable outperformance in terms of inference speed.

Economic issues, such as inflation, energy costs, taxes, and interest rates, are a constant presence in our daily lives and have been exacerbated by global events such as pandemics, environmental disasters, and wars. A sustained history of financial crises reveals significant weaknesses and vulnerabilities in the foundations of modern economies. Another significant issue currently is people quitting their jobs in large numbers. Moreover, many organizations have a diverse workforce comprising multiple generations posing new challenges. Transformative approaches in economics and labour markets are needed to protect our societies, economies, and planet. In this work, we use big data and machine learning methods to discover multi-perspective parameters for multi-generational labour markets. The parameters for the academic perspective are discovered using 35,000 article abstracts from the Web of Science for the period 1958-2022 and for the professionals' perspective using 57,000 LinkedIn posts from 2022. We discover a total of 28 parameters and categorised them into 5 macro-parameters, Learning & Skills, Employment Sectors, Consumer Industries, Learning & Employment Issues, and Generations-specific Issues. A complete machine learning software tool is developed for data-driven parameter discovery. A variety of quantitative and visualisation methods are applied and multiple taxonomies are extracted to explore multi-generational labour markets. A knowledge structure and literature review of multi-generational labour markets using over 100 research articles is provided. It is expected that this work will enhance the theory and practice of AI-based methods for knowledge discovery and system parameter discovery to develop autonomous capabilities and systems and promote novel approaches to labour economics and markets, leading to the development of sustainable societies and economies.

Aerodynamic collision avoidance manoeuvres provide an opportunity for satellites in Low Earth Orbits to reduce the risk during close encounters. With rising numbers of satellites and objects in orbit, satellites experience close encounters more frequently. Especially those satellites without thrusting capabilities face the problem of not being able to performimpulsive evasive manoeuvres. For satellites in Low Earth Orbits, though, perturbing forces due to aerodynamic drag may be used to influence their trajectories, thus offering a possibility to avoid collisions. This work introduces a tool for the analysis of aerodynamic collision avoidance manoeuvres. Current space-weather data are employed to estimate the density the satellite encounters. Achievable in-track separation distances following a variation of the ballistic coefficient through a change in attitude are then derived by evaluating an analytical equation from literature. Considering additional constraints for the attitude, e.g., charging phases, and uncertainties in the used parameters, the influence of a manoeuvre on the conjunction geometry and the collision probability is examined. The university satellite Flying Laptop of the University of Stuttgart is used as an exemplary satellite for analysis, which show the general effectiveness of evasive manoeuvres employing aerodynamic drag. First manoeuvring strategies can be deducted and the influence of parameter uncertainties is assessed.

A fundamental challenge of bipartite graph representation learning is how to extract informative node embeddings. Self-Supervised Learning (SSL) is a promising paradigm to address this challenge. Most recent bipartite graph SSL methods are based on contrastive learning which learns embeddings by discriminating positive and negative node pairs. Contrastive learning usually requires a large number of negative node pairs, which could lead to computational burden and semantic errors. In this paper, we introduce a novel synergistic representation learning model (STERLING) to learn node embeddings without negative node pairs. STERLING preserves the unique local and global synergies in bipartite graphs. The local synergies are captured by maximizing the similarity of the inter-type and intra-type positive node pairs, and the global synergies are captured by maximizing the mutual information of co-clusters. Theoretical analysis demonstrates that STERLING could improve the connectivity between different node types in the embedding space. Extensive empirical evaluation on various benchmark datasets and tasks demonstrates the effectiveness of STERLING for extracting node embeddings.

An analytical expression is derived for the thermal response observed during spontaneous imbibition of water into a dry core of zeolitic tuff. Sample tortuosity, thermal conductivity, and thermal source strength are estimated from fitting an analytical solution to temperature observations during a single laboratory test. The closed-form analytical solution is derived using Green's functions for heat conduction in the limit of "slow" water movement; that is, when advection of thermal energy with the wetting front is negligible. The solution has four free fitting parameters and is efficient for parameter estimation. Laboratory imbibition data used to constrain the model include a time series of the mass of water imbibed, visual location of the wetting front through time, and temperature time series at six locations. The thermal front reached the end of the core hours before the visible wetting front. Thus, the predominant form of heating during imbibition in this zeolitic tuff is due to vapor adsorption in dry zeolitic rock ahead of the wetting front. The separation of the wetting front and thermal front in this zeolitic tuff is significant, compared to wetting front behavior of most materials reported in the literature. This work is the first interpretation of a thermal imbibition response to estimate transport (tortuosity) and thermal properties (including thermal conductivity) from a single laboratory test.

Knowledge on changes in glacier calving front positions is important for assessing the status of glaciers. Remote sensing imagery provides the ideal database for monitoring calving front positions, however, it is not feasible to perform this task manually for all calving glaciers globally due to time-constraints. Deep learning-based methods have shown great potential for glacier calving front delineation from optical and radar satellite imagery. The calving front is represented as a single thin line between the ocean and the glacier, which makes the task vulnerable to inaccurate predictions. The limited availability of annotated glacier imagery leads to a lack of data diversity (not all possible combinations of different weather conditions, terminus shapes, sensors, etc. are present in the data), which exacerbates the difficulty of accurate segmentation. In this paper, we propose Attention-Multi-hooking-Deep-supervision HookNet (AMD-HookNet), a novel glacier calving front segmentation framework for synthetic aperture radar (SAR) images. The proposed method aims to enhance the feature representation capability through multiple information interactions between low-resolution and high-resolution inputs based on a two-branch U-Net. The attention mechanism, integrated into the two branch U-Net, aims to interact between the corresponding coarse and fine-grained feature maps. This allows the network to automatically adjust feature relationships, resulting in accurate pixel-classification predictions. Extensive experiments and comparisons on the challenging glacier segmentation benchmark dataset CaFFe show that our AMD-HookNet achieves a mean distance error of 438 m to the ground truth outperforming the current state of the art by 42%, which validates its effectiveness.

Self-supervised learning (SSL) aims to learn intrinsic features without labels. Despite the diverse architectures of SSL methods, the projection head always plays an important role in improving the performance of the downstream task. In this work, we systematically investigate the role of the projection head in SSL. Specifically, the projection head targets the uniformity part of SSL, which pushes the dissimilar samples away from each other, thus enabling the encoder to focus on extracting semantic features. Based on this understanding, we propose a Representation Evaluation Design (RED) in SSL models in which a shortcut connection between the representation and the projection vectors is built. Extensive experiments with different architectures, including SimCLR, MoCo-V2, and SimSiam, on various datasets, demonstrate that the representation evaluation design can consistently improve the baseline models in the downstream tasks. The learned representation from the RED-SSL models shows superior robustness to unseen augmentations and out-of-distribution data.

The precipitation of charged particles from the magnetosphere into the ionosphere is one of the crucial coupling mechanisms between these two regions of geospace and is associated with multiple space weather effects, such as global navigation satellite system signal disruption and geomagnetically induced currents at ground level. While precipitating particle fluxes have been measured by numerous spacecraft missions over the past decades, it often remains difficult to obtain global precipitation patterns with a good time resolution during a substorm. Numerical simulations can help to bridge this gap and improve the understanding of mechanisms leading to particle precipitation at high latitudes through the global view they offer on the near-Earth space system. We present the first results on auroral (0.5-50 keV) proton precipitation within a 3-dimensional simulation of the Vlasiator hybrid-Vlasov model. The run is driven by southward interplanetary magnetic field conditions with constant solar wind parameters. We find that, on the dayside, cusp proton precipitation exhibits the expected energy-latitude dispersion and takes place in the form of successive bursts associated with the transit of flux transfer events formed through dayside magnetopause reconnection. On the nightside, the precipitation takes place within the expected range of geomagnetic latitudes, and it appears clearly that the precipitating particle injection is taking place within a narrow magnetic local time span, associated with fast Earthward plasma flows in the near-Earth magnetotail. Finally, the simulated precipitating fluxes are compared to observations from Defense Meteorological Satellite Program spacecraft during driving conditions similar to those in the simulation and are found to be in good agreement with the measurements.

Deep supervised learning algorithms typically require a large volume of labeled data to achieve satisfactory performance. However, the process of collecting and labeling such data can be expensive and time-consuming. Self-supervised learning (SSL), a subset of unsupervised learning, aims to learn discriminative features from unlabeled data without relying on human-annotated labels. SSL has garnered significant attention recently, leading to the development of numerous related algorithms. However, there is a dearth of comprehensive studies that elucidate the connections and evolution of different SSL variants. This paper presents a review of diverse SSL methods, encompassing algorithmic aspects, application domains, three key trends, and open research questions. Firstly, we provide a detailed introduction to the motivations behind most SSL algorithms and compare their commonalities and differences. Secondly, we explore representative applications of SSL in domains such as image processing, computer vision, and natural language processing. Lastly, we discuss the three primary trends observed in SSL research and highlight the open questions that remain. A curated collection of valuable resources can be accessed at https://github.com/guijiejie/SSL.

We present a technique to automatically classify the wave type of seismic phases that are recorded on a single six-component recording station (measuring both three components of translational and rotational ground motion) at the earth's surface. We make use of the fact that each wave type leaves a unique 'fingerprint' in the six-component motion of the sensor. This fingerprint can be extracted by performing an eigenanalysis of the data covariance matrix, similar to conventional three-component polarization analysis. To assign a wave type to the fingerprint extracted from the data, we compare it to analytically derived six-component polarization models that are valid for pure-state plane wave arrivals. For efficient classification, we make use of the supervised machine learning method of support vector machines that is trained using data-independent, analytically-derived six-component polarization models. This enables the rapid classification of seismic phases in a fully automated fashion, even for large data volumes, such as encountered in land-seismic exploration or ambient noise seismology. Once the wave-type is known, additional wave parameters (velocity, directionality, and ellipticity) can be directly extracted from the six-component polarization states without the need to resort to expensive optimization algorithms. We illustrate the benefits of our approach on various real and synthetic data examples for applications such as automated phase picking, aliased ground-roll suppression in land-seismic exploration, and the rapid close-to real time extraction of surface wave dispersion curves from single-station recordings of ambient noise. Additionally, we argue that an initial step of wave type classification is necessary in order to successfully apply the common technique of extracting phase velocities from combined measurements of rotational and translational motion.

The State of Tripura lies in northeast India which is considered to be one of the most seismically active regions of the world. In the present study, a realistic Probabilistic Seismic Hazard Assessment (PSHA) of Tripura State based on improved seismogenic sources considering layered polygonal sources corresponding to hypo-central depth range of 0 to 25 km, 25 to 70 km and 70 to 180 km, respectively and data driven selection of suitable Ground Motion Prediction Equations (GMPEs) in a logic tree framework is presented. Analyses have been carried out by formulating a layered seismogenic source zonation together with smooth-gridded seismicity. Using the limited accelerogram records available, most suitable GMPEs have been selected after performing a thorough quantitative assessment and thus the uncertainty in selecting appropriate GMPEs in PSHA has been addressed by combining them with proper weight factor. The computations of seismic hazard are carried out in a higher resolution of grid interval of 0.05 $\degree$ X 0.05 $\degree$. The probabilistic seismic hazard distribution in terms of Peak Ground Acceleration (PGA) and 5% damped Pseudo Spectral Acceleration (PSA) at different time periods for 10% and 2% probability of exceedance in 50 years at engineering bedrock level have been presented. The final results show significant improvements over the previous studies, which is reflecetd in the local variation of the hazard maps. The design response spectra at engineering bedrock level can be computed for any location in the study region from the hazard distributions. The results will be useful for earthquake resistant design and construction of structures in this region.

Large pretrained language models have shown surprising in-context learning (ICL) ability. With a few demonstration input-label pairs, they can predict the label for an unseen input without parameter updates. Despite the great success in performance, its working mechanism still remains an open question. In this paper, we explain language models as meta-optimizers and understand in-context learning as implicit finetuning. Theoretically, we figure out that Transformer attention has a dual form of gradient descent. On top of it, we understand ICL as follows: GPT first produces meta-gradients according to the demonstration examples, and then these meta-gradients are applied to the original GPT to build an ICL model. We comprehensively compare the behaviors of in-context learning and explicit finetuning on real tasks to provide empirical evidence that supports our understanding. Experimental results show that in-context learning behaves similarly to explicit finetuning from multiple perspectives. Inspired by the dual form between Transformer attention and gradient descent, we design a momentum-based attention by analogy with gradient descent with momentum. The improved performance over vanilla attention further supports our understanding from another perspective, and more importantly, shows the potential to utilize our understanding for future model design. The code is available at \url{https://aka.ms/icl}.

Recent advances in artificial intelligence (AI) have significantly intensified research in the geoscience and remote sensing (RS) field. AI algorithms, especially deep learning-based ones, have been developed and applied widely to RS data analysis. The successful application of AI covers almost all aspects of Earth observation (EO) missions, from low-level vision tasks like super-resolution, denoising and inpainting, to high-level vision tasks like scene classification, object detection and semantic segmentation. While AI techniques enable researchers to observe and understand the Earth more accurately, the vulnerability and uncertainty of AI models deserve further attention, considering that many geoscience and RS tasks are highly safety-critical. This paper reviews the current development of AI security in the geoscience and RS field, covering the following five important aspects: adversarial attack, backdoor attack, federated learning, uncertainty and explainability. Moreover, the potential opportunities and trends are discussed to provide insights for future research. To the best of the authors' knowledge, this paper is the first attempt to provide a systematic review of AI security-related research in the geoscience and RS community. Available code and datasets are also listed in the paper to move this vibrant field of research forward.

We present a new method to invert variable stress changes of fractures from InSAR ground displacements. Fractures can be either faults or magma intrusions, embeded in a 3D heterogeneous crust with prominent topographies. The method is based on a fictituous domain approach using a finite element discretization of XFEM type. A cost function involves the misfit between the solution of the physical problem and the observed data together with the smoothing terms. Regularization parameters are determined by using L-curves. The method is then reformulated to be applied to InSAR data (masked and noisy), projected in Earth-Satellite directions. Synthetic tests confirm the efficiency and effectiveness of our method.

Insects as pollinators play a crucial role in ecosystem management and world food production. However, insect populations are declining, calling for efficient methods of insect monitoring. Existing methods analyze video or time-lapse images of insects in nature, but the analysis is challenging since insects are small objects in complex and dynamic scenes of natural vegetation. In this work, we provide a dataset of primary honeybees visiting three different plant species during two months of the summer period. The dataset consists of 107,387 annotated time-lapse images from multiple cameras, including 9,423 annotated insects. We present a method pipeline for detecting insects in time-lapse RGB images. The pipeline consists of a two-step process. Firstly, the time-lapse RGB images are preprocessed to enhance insects in the images. This Motion-Informed-Enhancement technique uses motion and colors to enhance insects in images. Secondly, the enhanced images are subsequently fed into a Convolutional Neural network (CNN) object detector. The method improves the deep learning object detectors You Only Look Once (YOLO) and Faster Region-based CNN (Faster R-CNN). Using Motion-Informed-Enhancement, the YOLO-detector improves the average micro F1-score from 0.49 to 0.71, and the Faster R-CNN-detector improves the average micro F1-score from 0.32 to 0.56 on the dataset. Our dataset and proposed method provide a step forward to automate the time-lapse camera monitoring of flying insects. The dataset is published on: https://vision.eng.au.dk/mie/

We describe the new global land water storage data set GLWS2.0, which contains total water storage anomalies (TWSA) over the global land except for Greenland and Antarctica with a spatial resolution of 0.5{\deg}, covering the time frame 2003 to 2019 without gaps, and including uncertainty quantification. GLWS2.0 was derived by assimilating monthly GRACE/-FO mass change maps into the WaterGAP global hydrology model via the Ensemble Kalman filter, taking data and model uncertainty into account. TWSA in GLWS2.0 is then accumulated over several hydrological storage variables. In this article, we describe the methods and data sets that went into GLWS2.0, how it compares to GRACE/-FO data in terms of representing TWSA trends, seasonal signals, and extremes, as well as its validation via comparing to GNSS-derived vertical loading and its comparison with the NASA Catchment Land Surface Model GRACE Data Assimilation (CLSM-DA). We find that, in the global average over more than 1000 stations, GLWS2.0 fits better than GRACE/-FO to GNSS observations of vertical loading at short-term, seasonal, and long-term temporal bands. While some differences exist, overall GLWS2.0 agrees quite well with CLSM-DA in terms of TWSA trends and annual amplitudes and phases.

In recent years, the field of intelligent transportation systems (ITS) has achieved remarkable success, which is mainly due to the large amount of available annotation data. However, obtaining these annotated data has to afford expensive costs in reality. Therefore, a more realistic strategy is to leverage semi-supervised learning (SSL) with a small amount of labeled data and a large amount of unlabeled data. Typically, semantic consistency regularization and the two-stage learning methods of decoupling feature extraction and classification have been proven effective. Nevertheless, representation learning only limited to semantic consistency regularization may not guarantee the separation or discriminability of representations of samples with different semantics; due to the inherent limitations of the two-stage learning methods, the extracted features may not match the specific downstream tasks. In order to deal with the above drawbacks, this paper proposes an end-to-end deep semi-supervised learning double contrast of semantic and feature, which extracts effective tasks specific discriminative features by contrasting the semantics/features of positive and negative augmented samples pairs. Moreover, we leverage information theory to explain the rationality of double contrast of semantics and features and slack mutual information to contrastive loss in a simpler way. Finally, the effectiveness of our method is verified in benchmark datasets.

In the U.S., corn is the most produced crop and has been an essential part of the American diet. To meet the demand for supply chain management and regional food security, accurate and timely large-scale corn yield prediction is attracting more attention in precision agriculture. Recently, remote sensing technology and machine learning methods have been widely explored for crop yield prediction. Currently, most county-level yield prediction models use county-level mean variables for prediction, ignoring much detailed information. Moreover, inconsistent spatial resolution between crop area and satellite sensors results in mixed pixels, which may decrease the prediction accuracy. Only a few works have addressed the mixed pixels problem in large-scale crop yield prediction. To address the information loss and mixed pixels problem, we developed a variational autoencoder (VAE) based multiple instance regression (MIR) model for large-scaled corn yield prediction. We use all unlabeled data to train a VAE and the well-trained VAE for anomaly detection. As a preprocess method, anomaly detection can help MIR find a better representation of every bag than traditional MIR methods, thus better performing in large-scale corn yield prediction. Our experiments showed that variational autoencoder based multiple instance regression (VAEMIR) outperformed all baseline methods in large-scale corn yield prediction. Though a suitable meta parameter is required, VAEMIR shows excellent potential in feature learning and extraction for large-scale corn yield prediction.

Producing labels for unlabeled data is error-prone, making semi-supervised learning (SSL) troublesome. Often, little is known about when and why an algorithm fails to outperform a supervised baseline. Using benchmark datasets, we craft five common real-world SSL data scenarios: few-label, open-set, noisy-label, and class distribution imbalance/misalignment in the labeled and unlabeled sets. We propose a novel algorithm called Contrastive Credibility Propagation (CCP) for deep SSL via iterative transductive pseudo-label refinement. CCP unifies semi-supervised learning and noisy label learning for the goal of reliably outperforming a supervised baseline in any data scenario. Compared to prior methods which focus on a subset of scenarios, CCP uniquely outperforms the supervised baseline in all scenarios, supporting practitioners when the qualities of labeled or unlabeled data are unknown.

Seismic phase picking and magnitude estimation are essential components of real time earthquake monitoring and earthquake early warning systems. Reliable phase picking enables the timely detection of seismic wave arrivals, facilitating rapid earthquake characterization and early warning alerts. Accurate magnitude estimation provides crucial information about the size of an earthquake and potential impact. Together, these steps contribute to effective earthquake monitoring, enhancing our ability to implement appropriate response measures in seismically active regions and mitigate risks. In this study, we explore the potential of deep learning in real time earthquake monitoring. To that aim, we begin by introducing DynaPicker which leverages dynamic convolutional neural networks to detect seismic body wave phases. Subsequently, DynaPicker is employed for seismic phase picking on continuous seismic recordings. To showcase the efficacy of Dynapicker, several open source seismic datasets including window format data and continuous seismic data are used for seismic phase identification, and arrival time picking. Additionally,the robustness of DynaPicker in classifying seismic phases was tested on the low magnitude seismic data polluted by noise. Finally, the phase arrival time information is integrated into a previously published deep learning model for magnitude estimation. This workflow is then applied and tested on the continuous recording of the aftershock sequences following the Turkey earthquake to detect the earthquakes, seismic phase picking and estimate the magnitude of the corresponding event. The results obtained in this case study exhibit a high level of reliability in detecting the earthquakes and estimating the magnitude of aftershocks following the Turkey earthquake.

With climate change, we are expecting more frequent extreme weather events in many regions worldwide. These events can trigger disruptive, deadly natural hazards, which catch the attention of the media and raise awareness in citizens and policymakers. Floods, wildfires, landslides are the object of a great deal of research. Yet, they remain difficult to predict and handle. Climate change also means warmer temperatures, especially on land. Glaciers melt, permafrost thaws. Everywhere, the ground gets warmer down to increasing depths. At first sight, not a big deal. What can a few extra degrees do? Microbes and fungi becomes more active, chemical equilibria shift. Silent changes are left unnoticed, even by scientists. In landslide studies, the stability of slopes is a balance of weights, water pressures, and mechanical strengths. Above freezing, temperature is left out, yet it should not be. The strength of clays, which frequently are abundant in soils, is sensitive to temperature. With a simple numerical exercise, we show to what extent temperature can condition and even undermine the stability of clay-bearing slopes, across the seasons and under global warming. A silent change that could make a non-extreme weather event have much more severe consequences.

Studies agree on a significant global mean sea level rise in the 20th century and its recent 21st century acceleration in the satellite record. At regional scale, the evolution of sea level probability distributions is often assumed to be dominated by changes in the mean. However, a quantification of changes in distributional shapes in a changing climate is currently missing. To this end, we propose a novel framework quantifying significant changes in probability distributions from time series data. The framework first quantifies linear trends in quantiles through quantile regression. Quantile slopes are then projected onto a set of four $orthogonal$ polynomials quantifying how such changes can be explained by $independent$ shifts in the first four statistical moments. The framework proposed is theoretically founded, general and can be applied to any climate observable with close-to-linear changes in distributions. We focus on observations and a coupled climate model (GFDL-CM4). In the historical period, trends in coastal daily sea level have been driven mainly by changes in the mean and can therefore be explained by a shift of the distribution with no change in shape. In the modeled world, robust changes in higher order moments emerge with increasing CO2 concentration. Such changes are driven in part by ocean circulation alone and get amplified by sea level pressure fluctuations, with possible consequences for sea level extremes attribution studies.

Accelerating the numerical integration of partial differential equations by learned surrogate model is a promising area of inquiry in the field of air pollution modeling. Most previous efforts in this field have been made on learned chemical operators though machine-learned fluid dynamics has been a more blooming area in machine learning community. Here we show the first trial on accelerating advection operator in the domain of air quality model using a realistic wind velocity dataset. We designed a convolutional neural network-based solver giving coefficients to integrate the advection equation. We generated a training dataset using a 2nd order Van Leer type scheme with the 10-day east-west components of wind data on 39$^{\circ}$N within North America. The trained model with coarse-graining showed good accuracy overall, but instability occurred in a few cases. Our approach achieved up to 12.5$\times$ acceleration. The learned schemes also showed fair results in generalization tests.

Shear rupture and fault slip in crystalline rocks like granite produce large dilation, impacting the spatiotemporal evolution of fluid pressure in the crust during the seismic cycle. To explore how fluid pressure variations are coupled to rock deformation and fault slip, we conducted laboratory experiments under upper crustal conditions while monitoring acoustic emissions and in situ fluid pressure. Our results show two separate faulting stages: initial rupture propagation, associated with large dilatancy and stabilised by local fluid pressure drops, followed by sliding on the newly formed fault, promoted by local fluid pressure recharge from the fault walls. This latter stage had not been previously recognised and can be understood as fluid-induced afterslip, co-located with the main rupture patch. Upscaling our laboratory results to the natural scale, we expect that spontaneous fault zone recharge could be responsible for early afterslip in locally dilating regions of major crustal faults, independently from large-scale fluid flow patterns.

We present a near-real-time global gridded daily CO$_2$ emissions dataset (GRACED) throughout 2021. GRACED provides gridded CO$_2$ emissions at a 0.1degree*0.1degree spatial resolution and 1-day temporal resolution from cement production and fossil fuel combustion over seven sectors, including industry, power, residential consumption, ground transportation, international aviation, domestic aviation, and international shipping. GRACED is prepared from a near-real-time daily national CO$_2$ emissions estimates (Carbon Monitor), multi-source spatial activity data emissions and satellite NO$_2$ data for time variations of those spatial activity data. GRACED provides the most timely overview of emissions distribution changes, which enables more accurate and timely identification of when and where fossil CO$_2$ emissions have rebounded and decreased. Uncertainty analysis of GRACED gives a grid-level two-sigma uncertainty of value of 19.9% in 2021, indicating the reliability of GRACED was not sacrificed for the sake of higher spatiotemporal resolution that GRACED provides. Continuing to update GRACED in a timely manner could help policymakers monitor energy and climate policies' effectiveness and make adjustments quickly.

In the remote sensing community, Land Use Land Cover (LULC) classification with satellite imagery is a main focus of current research activities. Accurate and appropriate LULC classification, however, continues to be a challenging task. In this paper, we evaluate the performance of multi-temporal (monthly time series) compared to mono-temporal (single time step) satellite images for multi-label classification using supervised learning on the RapidAI4EO dataset. As a first step, we trained our CNN model on images at a single time step for multi-label classification, i.e. mono-temporal. We incorporated time-series images using a LSTM model to assess whether or not multi-temporal signals from satellites improves CLC classification. The results demonstrate an improvement of approximately 0.89% in classifying satellite imagery on 15 classes using a multi-temporal approach on monthly time series images compared to the mono-temporal approach. Using features from multi-temporal or mono-temporal images, this work is a step towards an efficient change detection and land monitoring approach.

Automated crop-type classification using Sentinel-2 satellite time series is essential to support agriculture monitoring. Recently, deep learning models based on transformer encoders became a promising approach for crop-type classification. Using explainable machine learning to reveal the inner workings of these models is an important step towards improving stakeholders' trust and efficient agriculture monitoring. In this paper, we introduce a novel explainability framework that aims to shed a light on the essential crop disambiguation patterns learned by a state-of-the-art transformer encoder model. More specifically, we process the attention weights of a trained transformer encoder to reveal the critical dates for crop disambiguation and use domain knowledge to uncover the phenological events that support the model performance. We also present a sensitivity analysis approach to understand better the attention capability for revealing crop-specific phenological events. We report compelling results showing that attention patterns strongly relate to key dates, and consequently, to the critical phenological events for crop-type classification. These findings might be relevant for improving stakeholder trust and optimizing agriculture monitoring processes. Additionally, our sensitivity analysis demonstrates the limitation of attention weights for identifying the important events in the crop phenology as we empirically show that the unveiled phenological events depend on the other crops in the data considered during training.

Recently, the availability of remote sensing imagery from aerial vehicles and satellites constantly improved. For an automated interpretation of such data, deep-learning-based object detectors achieve state-of-the-art performance. However, established object detectors require complete, precise, and correct bounding box annotations for training. In order to create the necessary training annotations for object detectors, imagery can be georeferenced and combined with data from other sources, such as points of interest localized by GPS sensors. Unfortunately, this combination often leads to poor object localization and missing annotations. Therefore, training object detectors with such data often results in insufficient detection performance. In this paper, we present a novel approach for training object detectors with extremely noisy and incomplete annotations. Our method is based on a teacher-student learning framework and a correction module accounting for imprecise and missing annotations. Thus, our method is easy to use and can be combined with arbitrary object detectors. We demonstrate that our approach improves standard detectors by 37.1\% $AP_{50}$ on a noisy real-world remote-sensing dataset. Furthermore, our method achieves great performance gains on two datasets with synthetic noise. Code is available at \url{https://github.com/mxbh/robust_object_detection}.

Recently, unsupervised adversarial training (AT) has been highlighted as a means of achieving robustness in models without any label information. Previous studies in unsupervised AT have mostly focused on implementing self-supervised learning (SSL) frameworks, which maximize the instance-wise classification loss to generate adversarial examples. However, we observe that simply maximizing the self-supervised training loss with an untargeted adversarial attack often results in generating ineffective adversaries that may not help improve the robustness of the trained model, especially for non-contrastive SSL frameworks without negative examples. To tackle this problem, we propose a novel positive mining for targeted adversarial attack to generate effective adversaries for adversarial SSL frameworks. Specifically, we introduce an algorithm that selects the most confusing yet similar target example for a given instance based on entropy and similarity, and subsequently perturbs the given instance towards the selected target. Our method demonstrates significant enhancements in robustness when applied to non-contrastive SSL frameworks, and less but consistent robustness improvements with contrastive SSL frameworks, on the benchmark datasets.

Ambitious satellite constellation projects by commercial entities and the ease of access to space in recent times have led to a dramatic proliferation of low-Earth space traffic. It jeopardizes space safety and long-term sustainability, necessitating better space traffic management (STM). Correct modeling of uncertainties in force models and orbital states, among other things, is an essential part of STM. For objects in the low-Earth orbit (LEO) region, the uncertainty in the orbital dynamics mainly emanate from limited knowledge of the atmospheric drag-related parameters and variables. In this paper, which extends the work by Paul et al. [2021], we develop a feed-forward deep neural network model for the prediction of the satellite drag coefficient for the full range of satellite attitude (i.e., satellite pitch $\in$ ($-90^0$, $+90^0$) and satellite yaw $\in$ ($0^0$, $+360^0$)). The model simultaneously predicts the mean and the standard deviation and is well-calibrated. We use numerically simulated physical drag coefficient data for training our neural network. The numerical simulations are carried out using the test particle Monte Carlo method using the diffuse reflection with incomplete accommodation gas-surface interaction model. Modeling is carried out for the well-known CHAllenging Minisatellite Payload (CHAMP) satellite. Finally, we use the Monte Carlo approach to propagate CHAMP over a three-day period under various modeling scenarios to investigate the distribution of radial, in-track, and cross-track orbital errors caused by drag coefficient uncertainty.

The development of low-level mesocyclones in supercell thunderstorms has often been explained via the development of storm-generated streamwise vorticity along a baroclinic gradient in the forward flank of supercells. However, the ambient streamwise vorticity of the environment (often quantified via storm-relative helicity), especially near the ground, is particularly skillful at discriminating between nontornadic and tornadic supercells. This study investigates whether the origins of the inflow air into supercell low-level mesocyclones, both horizontally and vertically, can help explain the dynamical role of environmental versus storm-generated vorticity in the development of low-level mesocyclone rotation. Simulations of supercells, initialized with wind profiles common to supercell environments observed in nature, show that the air bound for the low-level mesocyclone primarily originates from the undisturbed, ambient environment, rather than from along the forward flank, and from very close to the ground, often in the lowest 200 - 400 m of the atmosphere. Given that the near-ground environmental air comprises the bulk of the inflow into low-level mesocyclones, this likely explains the forecast skill of environmental streamwise vorticity in the lowest few hundred meters of the atmosphere. The low-level mesocyclone does not appear to require much augmentation from the development of additional horizontal vorticity in the forward flank. Instead, the dominant contributor to vertical vorticity within the low-level mesocyclone is from the environmental horizontal vorticity. This study hopefully clarifies the development of low-level mesocyclones in supercells.

Transformer-based pre-trained language models such as BERT have achieved remarkable results in Semantic Sentence Matching. However, existing models still suffer from insufficient ability to capture subtle differences. Minor noise like word addition, deletion, and modification of sentences may cause flipped predictions. To alleviate this problem, we propose a novel Dual Attention Enhanced BERT (DABERT) to enhance the ability of BERT to capture fine-grained differences in sentence pairs. DABERT comprises (1) Dual Attention module, which measures soft word matches by introducing a new dual channel alignment mechanism to model affinity and difference attention. (2) Adaptive Fusion module, this module uses attention to learn the aggregation of difference and affinity features, and generates a vector describing the matching details of sentence pairs. We conduct extensive experiments on well-studied semantic matching and robustness test datasets, and the experimental results show the effectiveness of our proposed method.

This study compares the skills of two numerical models having the same horizontal resolution but based on different principles in representing meso- and submesoscale features of ocean dynamics in the Lakshadweep Sea (North Indian Ocean). The first model, titled LD20-NEMO, is based on solving primitive equations using the NEMO (Nucleus for European Modelling of the Ocean) modelling engine. The second one, titled LD20-SDD, uses a newer Stochastic-Deterministic Downscaling method. Both models have 1/20o resolution and use the outputs from a Global Ocean Physics Analysis and Forecast model at 1/12o resolution available from Copernicus Marine Service (CMEMS). The LD20-NEMO uses only a 2D set of data from CMEMS as lateral boundary conditions. The LD20-SDD consumes the full 3D set of data from CMEMS and exploits the stochastic properties of these data to generate the downscaled field variables at higher resolution than the parent model. The skills of the three models, CMEMS, LD20-NEMO and LD20-SDD are assessed against remotely sensed and in-situ observations for the four-year period 2015-2018. All models show similar skills in reproducing temperature and salinity, however the SDD version performs slightly better than the NEMO version. This difference in resolution is particularly significant in simulation of vorticity and computation of the share of the sea occupied by highly non-linear processes. While the NEMO and SDD model show similar skill, the SDD model is more computationally efficient than the NEMO model by a large margin.

We consider the effects of leakage on the ability to realize a discrete time crystal (DTC) in a semiconductor quantum dot linear array being operated as a chain of singlet-triplet (ST) qubits. This system realizes an Ising model with an effective applied magnetic field, plus additional terms that can cause leakage out of the computational subspace. We demonstrate that, in the absence of these leakage terms, this model theoretically realizes a DTC phase over a broad parameter regime for six and eight qubits, with a broader parameter range for the eight-qubit case. We then reintroduce the leakage terms and find that the DTC phase disappears entirely over the same parameter range if the system is only subject to a uniform magnetic field, which does not suppress leakage. However, we find that the DTC phase can be restored if the system is instead subject to a magnetic field that alternates from qubit to qubit, which suppresses leakage. We thus show that leakage is a serious problem for the realization of a DTC phase in a chain of ST qubits, but is by no means insurmountable. Our work suggests that experiments manifesting small-system stable DTC should be feasible with currently existing quantum dot spin qubits.

Remote sensing benefits habitat conservation by making monitoring of large areas easier compared to field surveying especially if the remote sensed data can be automatically analyzed. An important aspect of monitoring is classifying and mapping habitat types present in the monitored area. Automatic classification is a difficult task, as classes have fine-grained differences and their distributions are long-tailed and unbalanced. Usually training data used for automatic land cover classification relies on fully annotated segmentation maps, annotated from remote sensed imagery to a fairly high-level taxonomy, i.e., classes such as forest, farmland, or urban area. A challenge with automatic habitat classification is that reliable data annotation requires field-surveys. Therefore, full segmentation maps are expensive to produce, and training data is often sparse, point-like, and limited to areas accessible by foot. Methods for utilizing these limited data more efficiently are needed. We address these problems by proposing a method for habitat classification and mapping, and apply this method to classify the entire northern Finnish Lapland area into Natura2000 classes. The method is characterized by using finely-grained, sparse, single-pixel annotations collected from the field, combined with large amounts of unannotated data to produce segmentation maps. Supervised, unsupervised and semi-supervised methods are compared, and the benefits of transfer learning from a larger out-of-domain dataset are demonstrated. We propose a \ac{CNN} biased towards center pixel classification ensembled with a random forest classifier, that produces higher quality classifications than the models themselves alone. We show that cropping augmentations, test-time augmentation and semi-supervised learning can help classification even further.