Nowadays, particulate matter, especially that with small dimension as PM10, PM2.5 and PM1, is the air quality indicator most commonly associated with a number of adverse health effects. In this paper it is analyzed the impact that a natural event, such as the transport of Saharan dust, can have on increasing the particulate matter concentration in Sicily.Consulting the data of daily PM10 concentration, acquired by air quality monitoring network belonging to Agenzia Regionale Protezione Ambiente (Environmental Protection Regional Agency), it was possible to analyze the trend from 2013 to 2015. The days, in which the limit value was exceeded, were subjected to combined analysis. It was based on three models: interpretations of the air masses back trajectories, using the atmospheric model HYSPLIT (HYbrid Single Particle Lagrangian Integrated trajectory); on the calculation of the concentration on the ground and at high altitude particulate applying DREAM model (Dust REgional atmospheric model) and on the calculation of the concentration of mineral aerosols according to the atmospheric optical thickness (AOT) applying NAAPS model (Navy Aerosol Analysis and Prediction System).The daily limit value exceedances were attributed to the transport of Saharan dust events exclusively when the three models were in agreement with each other. Identifying the natural events, it was possible to quantify the contribution of the Saharan dust and consequently the reduction of the exceedances number.

Deep learning-based object detection and instance segmentation have achieved unprecedented progress. In this paper, we propose Complete-IoU (CIoU) loss and Cluster-NMS for enhancing geometric factors in both bounding box regression and Non-Maximum Suppression (NMS), leading to notable gains of average precision (AP) and average recall (AR), without the sacrifice of inference efficiency. In particular, we consider three geometric factors, i.e., overlap area, normalized central point distance and aspect ratio, which are crucial for measuring bounding box regression in object detection and instance segmentation. The three geometric factors are then incorporated into CIoU loss for better distinguishing difficult regression cases. The training of deep models using CIoU loss results in consistent AP and AR improvements in comparison to widely adopted $\ell_n$-norm loss and IoU-based loss. Furthermore, we propose Cluster-NMS, where NMS during inference is done by implicitly clustering detected boxes and usually requires less iterations. Cluster-NMS is very efficient due to its pure GPU implementation, and geometric factors can be incorporated to improve both AP and AR. In the experiments, CIoU loss and Cluster-NMS have been applied to state-of-the-art instance segmentation (e.g., YOLACT and BlendMask-RT), and object detection (e.g., YOLO v3, SSD and Faster R-CNN) models. Taking YOLACT on MS COCO as an example, our method achieves performance gains as +1.7 AP and +6.2 AR$_{100}$ for object detection, and +0.9 AP and +3.5 AR$_{100}$ for instance segmentation, with 27.1 FPS on one NVIDIA GTX 1080Ti GPU. All the source code and trained models are available at https://github.com/Zzh-tju/CIoU

The rampant coronavirus disease 2019 (COVID-19) has brought global crisis with its deadly spread to more than 180 countries, and about 3,519,901 confirmed cases along with 247,630 deaths globally as on May 4, 2020. The absence of any active therapeutic agents and the lack of immunity against COVID-19 increases the vulnerability of the population. Since there are no vaccines available, social distancing is the only feasible approach to fight against this pandemic. Motivated by this notion, this article proposes a deep learning based framework for automating the task of monitoring social distancing using surveillance video. The proposed framework utilizes the YOLO v3 object detection model to segregate humans from the background and Deepsort approach to track the identified people with the help of bounding boxes and assigned IDs. The results of the YOLO v3 model are further compared with other popular state-of-the-art models, e.g. faster region-based CNN (convolution neural network) and single shot detector (SSD) in terms of mean average precision (mAP), frames per second (FPS) and loss values defined by object classification and localization. Later, the pairwise vectorized L2 norm is computed based on the three-dimensional feature space obtained by using the centroid coordinates and dimensions of the bounding box. The violation index term is proposed to quantize the non adoption of social distancing protocol. From the experimental analysis, it is observed that the YOLO v3 with Deepsort tracking scheme displayed best results with balanced mAP and FPS score to monitor the social distancing in real-time.

Remote sensing image scene classification, which aims at labeling remote sensing images with a set of semantic categories based on their contents, has broad applications in a range of fields. Propelled by the powerful feature learning capabilities of deep neural networks, remote sensing image scene classification driven by deep learning has drawn remarkable attention and achieved significant breakthroughs. However, to the best of our knowledge, a comprehensive review of recent achievements regarding deep learning for scene classification of remote sensing images is still lacking. Considering the rapid evolution of this field, this paper provides a systematic survey of deep learning methods for remote sensing image scene classification by covering more than 160 papers. To be specific, we discuss the main challenges of remote sensing image scene classification and survey (1) Autoencoder-based remote sensing image scene classification methods, (2) Convolutional Neural Network-based remote sensing image scene classification methods, and (3) Generative Adversarial Network-based remote sensing image scene classification methods. In addition, we introduce the benchmarks used for remote sensing image scene classification and summarize the performance of more than two dozen of representative algorithms on three commonly-used benchmark data sets. Finally, we discuss the promising opportunities for further research.

Aftershock occurrence is characterized by scaling behaviors with quite universal exponents. At the same time, deviations from universality have been proposed as a tool to discriminate aftershocks from foreshocks. Here we show that the change in rheological behavior of the crust, from velocity weakening to velocity strengthening, represents a viable mechanism to explain statistical features of both aftershocks and foreshocks. More precisely, we present a model of the seismic fault described as a velocity weakening elastic layer coupled to a velocity strengthening visco-elastic layer. We show that the statistical properties of aftershocks in instrumental catalogs are recovered at a quantitative level, quite independently of the value of model parameters. We also find that large earthquakes are often anticipated by a preparatory phase characterized by the occurrence of foreshocks. Their magnitude distribution is significantly flatter than the aftershock one, in agreement with recent results for forecasting tools based on foreshocks.

Detecting the user's intent and finding the corresponding slots among the utterance's words are important tasks in natural language understanding. Their interconnected nature makes their joint modeling a standard part of training such models. Moreover, data scarceness and specialized vocabularies pose additional challenges. Recently, the advances in pre-trained language models, namely contextualized models such as ELMo and BERT have revolutionized the field by tapping the potential of training very large models with just a few steps of fine-tuning on a task-specific dataset. Here, we leverage such models, namely BERT and RoBERTa, and we design a novel architecture on top of them. Moreover, we propose an intent pooling attention mechanism, and we reinforce the slot filling task by fusing intent distributions, word features, and token representations. The experimental results on standard datasets show that our model outperforms both the current non-BERT state of the art as well as some stronger BERT-based baselines.

The great success of Transformer-based models benefits from the powerful multi-head self-attention mechanism, which learns token dependencies and encodes contextual information from the input. Prior work strives to attribute model decisions to individual input features with different saliency measures, but they fail to explain how these input features interact with each other to reach predictions. In this paper, we propose a self-attention attribution method to interpret the information interactions inside Transformer. We take BERT as an example to conduct extensive studies. Firstly, we apply self-attention attribution to identify the important attention heads, while others can be pruned with marginal performance degradation. Furthermore, we extract the most salient dependencies in each layer to construct an attribution tree, which reveals the hierarchical interactions inside Transformer. Finally, we show that the attribution results can be used as adversarial patterns to implement non-targeted attacks towards BERT.

Attention is a key component of Transformers, which have recently achieved considerable success in natural language processing. Hence, attention is being extensively studied to investigate various linguistic capabilities of Transformers, focusing on analyzing the parallels between attention weights and specific linguistic phenomena. This paper shows that attention weights alone are only one of the two factors that determine the output of attention and proposes a norm-based analysis that incorporates the second factor, the norm of the transformed input vectors. The findings of our norm-based analyses of BERT and a Transformer-based neural machine translation system include the following: (i) contrary to previous studies, BERT pays poor attention to special tokens, and (ii) reasonable word alignment can be extracted from attention mechanisms of Transformer. These findings provide insights into the inner workings of Transformers.

It is well known that featuremap attention and multi-path representation are important for visual recognition. In this paper, we present a modularized architecture, which applies the channel-wise attention on different network branches to leverage their success in capturing cross-feature interactions and learning diverse representations. Our design results in a simple and unified computation block, which can be parameterized using only a few variables. Our model, named ResNeSt, outperforms EfficientNet in accuracy and latency trade-off on image classification. In addition, ResNeSt has achieved superior transfer learning results on several public benchmarks serving as the backbone, and has been adopted by the winning entries of COCO-LVIS challenge. The source code for complete system and pretrained models are publicly available.

Precision agriculture has become a key factor for increasing crop yields by providing essential information to decision makers. In this work, we present a deep learning method for simultaneous segmentation and counting of cranberries to aid in yield estimation and sun exposure predictions. Notably, supervision is done using low cost center point annotations. The approach, named Triple-S Network, incorporates a three-part loss with shape priors to promote better fitting to objects of known shape typical in agricultural scenes. Our results improve overall segmentation performance by more than 6.74% and counting results by 22.91% when compared to state-of-the-art. To train and evaluate the network, we have collected the CRanberry Aerial Imagery Dataset (CRAID), the largest dataset of aerial drone imagery from cranberry fields. This dataset will be made publicly available.

A question of global concern regarding the sustainable future of humankind stems from the effect due to aerosols on the global climate. The quantification of atmospheric aerosols and their relationship to climatic impacts are key to understanding the dynamics of climate forcing and to improve our knowledge about climate change. Due to its response to precipitation, temperature, topography and human activity, one of the most dynamical atmospheric regions is the atmospheric boundary layer (ABL): ABL aerosols have a sizable impact on the evolution of the radiative forcing of climate change, human health, food security, and, ultimately, on the local and global economy. The identification of ABL pattern behaviour requires constant monitoring and the application of instrumental and computational methods for its detection and analysis. Here, we show a new method for the retrieval of ABL top arising from light detection and ranging (LiDAR) signals, by training a convolutional neural network in a supervised manner; forcing it to learn how to retrieve such a dynamical parameter on real, non-ideal conditions and in a fully automated, unsupervised way. Our findings pave the way for a full integration of LiDAR elastic, inelastic, and depolarisation signal processing, and provide a novel approach for real-time quantitative sensing of aerosols.

Change detection in heterogeneous remote sensing images is crucial for disaster damage assessment. Recent methods use homogenous transformation, which transforms the heterogeneous optical and SAR remote sensing images into the same feature space, to achieve change detection. Such transformations mainly operate on the low-level feature space and may corrupt the semantic content, deteriorating the performance of change detection. To solve this problem, this paper presents a new homogeneous transformation model termed deep homogeneous feature fusion (DHFF) based on image style transfer (IST). Unlike the existing methods, the DHFF method segregates the semantic content and the style features in the heterogeneous images to perform homogeneous transformation. The separation of the semantic content and the style in homogeneous transformation prevents the corruption of image semantic content, especially in the regions of change. In this way, the detection performance is improved with accurate homogeneous transformation. Furthermore, we present a new iterative IST (IIST) strategy, where the cost function in each IST iteration measures and thus maximizes the feature homogeneity in additional new feature subspaces for change detection. After that, change detection is accomplished accurately on the original and the transformed images that are in the same feature space. Real remote sensing images acquired by SAR and optical satellites are utilized to evaluate the performance of the proposed method. The experiments demonstrate that the proposed DHFF method achieves significant improvement for change detection in heterogeneous optical and SAR remote sensing images, in terms of both accuracy rate and Kappa index.

This work describes experiments which probe the hidden representations of several BERT-style models for morphological content. The goal is to examine the extent to which discrete linguistic structure, in the form of morphological features and feature values, presents itself in the vector representations and attention distributions of pre-trained language models for five European languages. The experiments contained herein show that (i) Transformer architectures largely partition their embedding space into convex sub-regions highly correlated with morphological feature value, (ii) the contextualized nature of transformer embeddings allows models to distinguish ambiguous morphological forms in many, but not all cases, and (iii) very specific attention head/layer combinations appear to hone in on subject-verb agreement.

Based on new radiative transfer numerical evaluations, we reconsider an argument presented by Schack in 1972 that says that saturation of the absorption of infrared radiation by carbon dioxide in the atmosphere sets in as soon as the relative concentration of carbon dioxide exceeds a lower limit of approximately 300 ppm. We provide a concise brief and explicit representation of the greenhouse effect of the earth's atmosphere. We find an equilibrium climate sensitivity (temperature increase $\Delta T$ due to doubling of atmospheric $CO_2$ concentration) of $\Delta T \simeq 0.5 ^0C$. We elaborate on the consistency of these results on $\Delta T$ with results observationally obtained by satellite-based measurements of short-time radiation-flux versus surface-temperature changes.

In the context of natural-based wastewater treatment technologies (such as constructed wetlands - CW) the use of a low-cost, continuous-like biosensor tool for the assessment of operational conditions is of key importance for plant management optimization. The objective of the present study was to assess the potential use of constructed wetland microbial fuel cells (CW-MFC) as a domestic wastewater COD assessment tool. For the purpose of this work four lab-scale CW-MFCs were set up and fed with pre-settled domestic wastewater at different COD concentrations. Under laboratory conditions two different anodic materials were tested (graphite rods and gravel). Furthermore, a pilot-plant based experiment was also conducted to confirm the findings previously recorded for lab-scale experiments. Results showed that in spite of the low coulombic efficiencies recorded, either gravel or graphite-based anodes were suitable for the purposes of domestic wastewater COD assessment. Significant linear relationships could be established between inlet COD concentrations and CW-MFC Ecell whenever contact time was above 10 hours. Results also showed that the accuracy of the CW-MFC was greatly compromised after several weeks of operation. Pilot experiments showed that CW-MFC presents a good bio-indication response between week 3 and 7 of operation (equivalent to an accumulated organic loading between 100 and 200 g COD/m2, respectively). Main conclusion of this work is that of CW-MFC could be used as an "alarm-tool" for qualitative continuous influent water quality assessment rather than a precise COD assessment tool due to a loss of precision after several weeks of operation.

Time-lapse seismic monitoring aims at resolving changes in a producing reservoir from changes in the reflection response. When the changes in the reservoir are very small, the changes in the seismic response can become too small to be reliably detected. In theory, multiple reflections can be used to improve the detectability of traveltime changes: a wave that propagates several times down and up through a reservoir layer will undergo a larger time shift due to reservoir changes than a primary reflection. Since we are interested in monitoring very local changes (usually in a thin reservoir layer), it would be advantageous if we could identify the reservoir-related internal multiples in the complex reflection response of the entire subsurface. We introduce a Marchenko-based method to isolate these multiples from the complete reflection response and illustrate the potential of this method with numerical examples.

Recently, numerous handcrafted and searched networks have been applied for semantic segmentation. However, previous works intend to handle inputs with various scales in pre-defined static architectures, such as FCN, U-Net, and DeepLab series. This paper studies a conceptually new method to alleviate the scale variance in semantic representation, named dynamic routing. The proposed framework generates data-dependent routes, adapting to the scale distribution of each image. To this end, a differentiable gating function, called soft conditional gate, is proposed to select scale transform paths on the fly. In addition, the computational cost can be further reduced in an end-to-end manner by giving budget constraints to the gating function. We further relax the network level routing space to support multi-path propagations and skip-connections in each forward, bringing substantial network capacity. To demonstrate the superiority of the dynamic property, we compare with several static architectures, which can be modeled as special cases in the routing space. Extensive experiments are conducted on Cityscapes and PASCAL VOC 2012 to illustrate the effectiveness of the dynamic framework. Code is available at https://github.com/yanwei-li/DynamicRouting.

The self-potential (SP) method is a passive geophysical technique, which may offer insights about water and ionic fluxes in the vadose zone. The main obstacles presently prohibiting its routine use in quantitative vadose zone hydrology are the superposition of signals arising from various source mechanisms, difficult-to-predict electrode polarization effects that depend on electrode design and age, as well as water saturation, pore water chemistry, clay content, and temperature in the immediate vicinity of the electrodes. We present a unique long-term SP monitoring experiment focusing on the first four years of data acquired at different depths in the vadose zone within the HOBE hydrological observatory in Denmark. Using state-of-the-art SP theory combined with flow and transport simulations, we attempt to replicate the observed data and suggest reasons for observed discrepancies. The predictions are overall satisfactory during the first six months of monitoring after which both the patterns and magnitudes of the observed data change drastically. Our main observations are (1) that predicted SP magnitudes are strongly sensitive to how the effective excess charge scales with water saturation implying that continued research is needed to build more accurate models of electrokinetic phenomena in unsaturated conditions, (2) that significant changes in electrode polarization occur in the shallowest electrodes at time scales of a year, suggesting that electrode effects cannot be ignored and that explicit electrode modeling should be considered in future monitoring studies, and (3) that multi-rate mass transfer and reactive transport modeling are needed to better predict salinity and pore water conductivity. We hope to stimulate other researchers to test new SP modeling approaches and interpretation strategies against these data by making the SP and complimentary data time-series available.

Terra Seismic can predict most major earthquakes (M6.2 or greater) at least 2 - 5 months before they will strike. Global earthquake prediction is based on determinations of the stressed areas that will start to behave abnormally before major earthquakes. The size of the observed stressed areas roughly corresponds to estimates calculated from Dobrovolskys formula. To identify abnormalities and make predictions, Terra Seismic applies various methodologies, including satellite remote sensing methods and data from ground-based instruments. We currently process terabytes of information daily, and use more than 80 different multiparameter prediction systems. Alerts are issued if the abnormalities are confirmed by at least five different systems. We observed that geophysical patterns of earthquake development and stress accumulation are generally the same for all key seismic regions. Thus, the same earthquake prediction methodologies and systems can be applied successfully worldwide. Our technology has been used to retrospectively test data gathered since 1970 and it successfully detected about 90 percent of all significant quakes over the last 50 years.

Space weather indices are commonly used to drive operational forecasts of various geospace systems, including the thermosphere for mass density and satellite drag. The drivers serve as proxies for various processes that cause energy flow and deposition in the geospace system. Forecasts of neutral mass density is a major uncertainty in operational orbit prediction and collision avoidance for objects in low earth orbit (LEO). For the strongly driven system, accuracy of space weather driver forecasts is crucial for operations. The High Accuracy Satellite Drag Model (HASDM) currently employed by the United States Air Force in an operational environment is driven by four (4) solar and two (2) geomagnetic proxies. Space Environment Technologies (SET) is contracted by the space command to provide forecasts for the drivers. This work performs a comprehensive assessment for the performance of the driver forecast models. The goal is to provide a benchmark for future improvements of the forecast models. Using an archived data set spanning six (6) years and 15,000 forecasts across solar cycle 24, we quantify the temporal statistics of the model performance.

Change detection is a basic task of remote sensing image processing. The research objective is to identity the change information of interest and filter out the irrelevant change information as interference factors. Recently, the rise of deep learning has provided new tools for change detection, which have yielded impressive results. However, the available methods focus mainly on the difference information between multitemporal remote sensing images and lack robustness to pseudo-change information. To overcome the lack of resistance of current methods to pseudo-changes, in this paper, we propose a new method, namely, dual attentive fully convolutional Siamese networks (DASNet) for change detection in high-resolution images. Through the dual-attention mechanism, long-range dependencies are captured to obtain more discriminant feature representations to enhance the recognition performance of the model. Moreover, the imbalanced sample is a serious problem in change detection, i.e. unchanged samples are much more than changed samples, which is one of the main reasons resulting in pseudo-changes. We put forward the weighted double margin contrastive loss to address this problem by punishing the attention to unchanged feature pairs and increase attention to changed feature pairs. The experimental results of our method on the change detection dataset (CDD) and the building change detection dataset (BCDD) demonstrate that compared with other baseline methods, the proposed method realizes maximum improvements of 2.1\% and 3.6\%, respectively, in the F1 score. Our Pytorch implementation is available at https://github.com/lehaifeng/DASNet.

Forbush decreases (FDs), which are short-term drops in the flux of galactic cosmic rays, are caused by the shielding from strong and/or turbulent magnetic structures in the solar wind, especially interplanetary coronal mass ejections (ICMEs) and their associated shocks, as well as corotating interaction regions. Such events can be observed at Earth, for example, using neutron monitors, and also at many other locations in the solar system, such as on the surface of Mars with the Radiation Assessment Detector instrument onboard Mars Science Laboratory. They are often used as a proxy for detecting the arrival of ICMEs or corotating interaction regions, especially when sufficient in situ solar wind measurements are not available. We compare the properties of FDs observed at Earth and Mars, focusing on events produced by ICMEs. We find that FDs at both locations show a correlation between their total amplitude and the maximum hourly decrease, but with different proportionality factors. We explain this difference using theoretical modeling approaches and suggest that it is related to the size increase of ICMEs, and in particular their sheath regions, en route from Earth to Mars. From the FD data, we can derive the sheath broadening factor to be between about 1.5 and 1.9, agreeing with our theoretical considerations. This factor is also in line with previous measurements of the sheath evolution closer to the Sun.

In this paper we provide a comprehensive introduction to knowledge graphs, which have recently garnered significant attention from both industry and academia in scenarios that require exploiting diverse, dynamic, large-scale collections of data. After some opening remarks, we motivate and contrast various graph-based data models and query languages that are used for knowledge graphs. We discuss the roles of schema, identity, and context in knowledge graphs. We explain how knowledge can be represented and extracted using a combination of deductive and inductive techniques. We summarise methods for the creation, enrichment, quality assessment, refinement, and publication of knowledge graphs. We provide an overview of prominent open knowledge graphs and enterprise knowledge graphs, their applications, and how they use the aforementioned techniques. We conclude with high-level future research directions for knowledge graphs.

In non-uniform valleys, dam-break flood waves can be significantly affected by downstream variations in river width, slope and roughness. To model these effects, we derive new analytical solutions to the kinematic wave equation, applicable to rating curves in the power law form and hydrographs of generic shape as long as they produce a single shock at the wave front. New results are first obtained for uniform channels, using the Gauss-Green theorem applied to characteristic-bounded regions of the plane. The results are then extended to non-uniform valleys, using a change of variable that homogenizes river properties by rescaling the distance coordinate. The solutions are illustrated and validated for idealized cases, then applied to three historical dam failures: the 2008 breaching failure of the Tangjiashan landslide dam, the 1976 piping failure of Teton Dam, and the 1959 sudden failure of Malpasset Dam. In spite of the much reduced computational cost and data requirements, the results agree well with the field data and with more elaborate simulations. They also clarify how both river and hydrograph properties affect flood propagation and attenuation.

A tornado is an extreme weather condition that can cause enormous damage to human society. In this paper, we propose a low cost, environmentally friendly method to fight against tornadoes using clusters of connected balloons. Deployed over a sufficiently wide area, these large balloons may be able to reduce the wind speed of the tornado, block and disrupt the convective flow of air, and destroy the tornado.

Cloud-aerosol interactions remain a major obstacle to understanding climate and severe weather. Observations suggest that aerosols enhance tropical thunderstorm activity; past research, motivated by the importance of understanding aerosol impacts on clouds, has proposed several mechanisms that could explain that observed link. Here, we show that high-resolution atmospheric simulations can reproduce the observed link between aerosols and convection. However, we also show that previously proposed mechanisms are unable to explain the invigoration. Examining underlying processes reveals that, in our simulations, high aerosol concentrations increase environmental humidity by producing clouds that mix more condensed water into the surrounding air. In turn, higher humidity favors large-scale ascent and stronger convection. Our results provide a physical reason to expect invigorated thunderstorms in high-aerosol regions of the tropics.

Building footprint maps are vital to many remote sensing applications, such as 3D building modeling, urban planning, and disaster management. Due to the complexity of buildings, the accurate and reliable generation of the building footprint from remote sensing imagery is still a challenging task. In this work, an end-to-end building footprint generation approach that integrates convolution neural network (CNN) and graph model is proposed. CNN serves as the feature extractor, while the graph model can take spatial correlation into consideration. Moreover, we propose to implement the feature pairwise conditional random field (FPCRF) as a graph model to preserve sharp boundaries and fine-grained segmentation. Experiments are conducted on four different datasets: (1) Planetscope satellite imagery of the cities of Munich, Paris, Rome, and Zurich; (2) ISPRS benchmark data from the city of Potsdam, (3) Dstl Kaggle dataset; and (4) Inria Aerial Image Labeling data of Austin, Chicago, Kitsap County, Western Tyrol, and Vienna. It is found that the proposed end-to-end building footprint generation framework with the FPCRF as the graph model can further improve the accuracy of building footprint generation by using only CNN, which is the current state-of-the-art.

The analysis of satellite thermal images of the Earth's surface within the spectral range of 10.5-11.3 mkm has shown that over some linear structures of the Middle-Asian seismically active region there is observed a stable in time and space increasing intensity of the outgoing radiation flux as compared to contiguous blocks. A retrospective analysis of a continuous series of observations of the outgoing IR radiation flux has shown that in certain individual zones of some major tectonic dislocations there appear from time to time positive anomalies of IR radiation, for instance at the point of intersection of the Talasso-Ferghana and Tamdy-Tokrauss faults. These anomalies last from 2 to 10 days. The spontaneous anomalies are characterized by a pulsating variation of area. The space confinement and duration of these anomalies permit distinguishing theme noise anomalies caused by meteorological factors. The time of the appearance of these anomalies coincides with the activation of faults over which there has been detected an increase of the outgoing IR radiation flux. In 1984 the majority of crustal earthquakes, of a magnitude over 4, in the Tien Shan were accompanied by the appearance of a positive anomaly of the IR radiation at the point of the intersection of the faults. The area of anomalies was n*10000 km2. The most outstanding example of such activization is the Ghazli earthquake of 19.03.1984 M7.2. At the point of the intersection of the Tamdy-Tokrauss and Talasso- Ferghana faults there was detected on March 11 a positive anomaly of the outgoing IR radiation flux of exceptional intensity and enormous area (about 100 thousand km2). The subsequent earthquakes in the zone of the Tamdy-Tokrauss fault in the summer of 1984 of the magnitudes from 4.3 to 5.3 were also preceded by the appearance of a positive anomaly of the outgoing IR radiation at the point of intersection of the faults.

NOAA/AVHRR thermal images indicated the presence of positive thermal anomalies that are associated with the large linear structures and fault systems of the Earth's crust. The relation between thermal anomalies and seismic activity was established for Middle Asia on the basis of a 7-year series of thermal images. Thermal anomaly has been located near Beijing, on the border between the mountains and plain. The size of this anomaly is about 700 km long and 50 km wide. The anomaly appeared about 6-24 days before and continued about a week after an earthquake. The anomaly was sensitive to crust earthquakes with a magnitude more than 4.7 and for a distance of up to 500 km. The amplitude of this anomaly was about 3 C.

As photovoltaic power is expanding rapidly worldwide, it is imperative to assess its promise under future climate scenarios. While a great deal of research has been devoted to trends of mean solar radiation, less attention has been paid to its intermittent character, a key challenge when compounded with uncertainties related to climate variability. Using both satellite data and climate model outputs, here we characterize solar radiation intermittency to assess future photovoltaic reliability. We find that the relation between the future power supply and long-term trends of mean solar radiation is highly nonlinear, thus making power reliability more sensitive to the fluctuations of mean solar radiation in regions where insolation is the highest. Our results highlight how reliability analysis must account simultaneously for the mean and intermittency of solar inputs when assessing the impacts of climate change on photovoltaics.

Classical and more recently deep computer vision methods are optimized for visible spectrum images, commonly encoded in grayscale or RGB colorspaces acquired from smartphones or cameras. A more uncommon source of images exploited in the remote sensing field are satellite and aerial images. However, the development of pattern recognition approaches for these data is relatively recent, mainly due to the limited availability of this type of images, as until recently they were used exclusively for military purposes. Access to aerial imagery, including spectral information, has been increasing mainly due to the low cost of drones, cheapening of imaging satellite launch costs, and novel public datasets. Usually remote sensing applications employ computer vision techniques strictly modeled for classification tasks in closed set scenarios. However, real-world tasks rarely fit into closed set contexts, frequently presenting previously unknown classes, characterizing them as open set scenarios. Focusing on this problem, this is the first paper to study and develop semantic segmentation techniques for open set scenarios applied to remote sensing images. The main contributions of this paper are: 1) a discussion of related works in open set semantic segmentation, showing evidence that these techniques can be adapted for open set remote sensing tasks; 2) the development and evaluation of a novel approach for open set semantic segmentation. Our method yielded competitive results when compared to closed set methods for the same dataset.

Accurate estimation of water relative permeability has been of great interest in various research areas because of its broad applications in soil physics and hydrology as well as oil and gas production and recovery. Critical path analysis (CPA), a promising technique from statistical physics, is well known to be applicable to heterogeneous media with broad conductance or pore size distribution (PSD). By heterogeneity, we mean variations in the geometrical properties of pore space. In this study, we demonstrate that CPA is also applicable to packings of spheres of the same size, known as homogeneous porous media. More specifically, we apply CPA to model water relative permeability (krw) in mono-sized sphere packs whose PSDs are fairly narrow. We estimate the krw from (1) the PSD and (2) the PSD and saturation-dependent electrical conductivity ({\sigma}_r) for both drainage and imbibition processes. We show that the PSD of mono-sized sphere packs approximately follows the log-normal probability density function. Comparison with numerical simulations indicate that both the imbibition and drainage krw are estimated from the PSD and {\sigma}_r data more accurately than those from the PSD. We show that CPA can estimate krw in mono-sized sphere packs precisely.

Energetic particle instruments on board GPS satellites form a powerful global prompt proton sensor network (GPPSn) that provides an unprecedented opportunity to monitor and characterize solar energetic protons targeting the Earth. The medium-Earth-orbits of the GPS constellation have the unique advantage of allowing solar energetic protons to be simultaneously measured from multiple points in both open- and closed-field line regions. Examining two example intervals of solar proton events, we showcase in this study how GPS proton data are prepared, calibrated and utilized to reveal important features of solar protons, including their source, acceleration/scattering by interplanetary shocks, the relative position of Earth when impinged by these shocks, the shape of solar particle fronts, the access of solar protons inside the dynamic geomagnetic field, as well temporally-varying proton distributions in both energy and space. By comparing to Van Allen Probes data, GPS proton observations are further demonstrated not only to be useful for qualitatively monitoring the dynamics of solar protons, but also for quantitative scientific research including determining cutoff L-shells. Our results establish that this GPPSn can join forces with other existing solar proton monitors and contribute to observing, warning, understanding and ultimately forecasting the incoming solar energetic proton events.

We converted the recently developed BabyAI grid world platform to a sender/receiver setup in order to test the hypothesis that established deep reinforcement learning techniques are sufficient to incentivize the emergence of a grounded discrete communication protocol between generalized agents. This is in contrast to previous experiments that employed straight-through estimation or specialized inductive biases. Our results show that these can indeed be avoided, by instead providing proper environmental incentives. Moreover, they show that a longer interval between communications incentivized more abstract semantics. In some cases, the communicating agents adapted to new environments more quickly than a monolithic agent, showcasing the potential of emergent communication for transfer learning and generalization in general.

Semantic segmentation with deep learning has achieved great progress in classifying the pixels in the image. However, the local location information is usually ignored in the high-level feature extraction by the deep learning, which is important for image semantic segmentation. To avoid this problem, we propose a graph model initialized by a fully convolutional network (FCN) named Graph-FCN for image semantic segmentation. Firstly, the image grid data is extended to graph structure data by a convolutional network, which transforms the semantic segmentation problem into a graph node classification problem. Then we apply graph convolutional network to solve this graph node classification problem. As far as we know, it is the first time that we apply the graph convolutional network in image semantic segmentation. Our method achieves competitive performance in mean intersection over union (mIOU) on the VOC dataset(about 1.34% improvement), compared to the original FCN model.

The heat engine model of tropical cyclones describes a thermally direct overturning circulation. Outflowing air slowly subsides as radiative cooling to space balances adiabatic warming, a process that does not consume any work. However, we show here that the lateral spread of the outflow is limited by the environmental deformation radius, which at high latitudes can be rather small. In such cases, the outflowing air is radially constrained, which limits how far downward it can subside via radiative cooling alone. Some literature has invoked the possibility of `mechanical subsidence' or `forced descent' in the storm outflow region in the presence of high inertial stability, which would be a thermally indirect circulation. Mechanical subsidence in the subsiding branch of a tropical cyclone has not before been observed or characterized. A series of axisymmetric tropical cyclone simulations at different latitudes and domain sizes is conducted to study the impact of environmental inertial stability on storm dynamics. In higher latitude storms in large axisymmetric domains, the outflow acts as a wavemaker to excite an inertial wave at the environmental inertial (Coriolis) frequency. This inertial wave periodically ventilates the core of a high-latitude storm with its own low entropy exhaust air. The wave response is in contrast to the presumed forced descent model, and we hypothesize that this is because inertial stability provides less resistance than buoyant stability, even in highly inertially stable environments.

In this paper, we briefly introduce the basic questions in the measurements of solar magnetic fields and the possible error sources due to the approximation of the theory of radiation transfer of spectral lines in the solar atmosphere. We introduce some basic research progress in magnetic field measurement at Huairou Solar Observing Station of National Astronomical Observatories of the Chinese Academy of Sciences, especially concerning the non-potentiality in solar active regions, such as the magnetic shear, current and helicity. We also discuss some basic questions for the measurements of the magnetic fields and corresponding challenges for the future studies.

A correct localisation of tables in a document is instrumental for determining their structure and extracting their contents; therefore, table detection is a key step in table understanding. Nowadays, the most successful methods for table detection in document images employ deep learning algorithms; and, particularly, a technique known as fine-tuning. In this context, such a technique exports the knowledge acquired to detect objects in natural images to detect tables in document images. However, there is only a vague relation between natural and document images, and fine-tuning works better when there is a close relation between the source and target task. In this paper, we show that it is more beneficial to employ fine-tuning from a closer domain. To this aim, we train different object detection algorithms (namely, Mask R-CNN, RetinaNet, SSD and YOLO) using the TableBank dataset (a dataset of images of academic documents designed for table detection and recognition), and fine-tune them for several heterogeneous table detection datasets. Using this approach, we considerably improve the accuracy of the detection models fine-tuned from natural images (in mean a 17%, and, in the best case, up to a 60%).

Bounding box regression is the crucial step in object detection. In existing methods, while $\ell_n$-norm loss is widely adopted for bounding box regression, it is not tailored to the evaluation metric, i.e., Intersection over Union (IoU). Recently, IoU loss and generalized IoU (GIoU) loss have been proposed to benefit the IoU metric, but still suffer from the problems of slow convergence and inaccurate regression. In this paper, we propose a Distance-IoU (DIoU) loss by incorporating the normalized distance between the predicted box and the target box, which converges much faster in training than IoU and GIoU losses. Furthermore, this paper summarizes three geometric factors in bounding box regression, \ie, overlap area, central point distance and aspect ratio, based on which a Complete IoU (CIoU) loss is proposed, thereby leading to faster convergence and better performance. By incorporating DIoU and CIoU losses into state-of-the-art object detection algorithms, e.g., YOLO v3, SSD and Faster RCNN, we achieve notable performance gains in terms of not only IoU metric but also GIoU metric. Moreover, DIoU can be easily adopted into non-maximum suppression (NMS) to act as the criterion, further boosting performance improvement. The source code and trained models are available at https://github.com/Zzh-tju/DIoU.

Recent progress in NLP witnessed the development of large-scale pre-trained language models (GPT, BERT, XLNet, etc.) based on Transformer (Vaswani et al. 2017), and in a range of end tasks, such models have achieved state-of-the-art results, approaching human performance. This demonstrates the power of the stacked self-attention architecture when paired with a sufficient number of layers and a large amount of pre-training data. However, on tasks that require complex and long-distance reasoning where surface-level cues are not enough, there is still a large gap between the pre-trained models and human performance. Strubell et al. (2018) recently showed that it is possible to inject knowledge of syntactic structure into a model through supervised self-attention. We conjecture that a similar injection of semantic knowledge, in particular, coreference information, into an existing model would improve performance on such complex problems. On the LAMBADA (Paperno et al. 2016) task, we show that a model trained from scratch with coreference as auxiliary supervision for self-attention outperforms the largest GPT-2 model, setting the new state-of-the-art, while only containing a tiny fraction of parameters compared to GPT-2. We also conduct a thorough analysis of different variants of model architectures and supervision configurations, suggesting future directions on applying similar techniques to other problems.

Leveraging multilingual parallel texts to automatically generate paraphrases has drawn much attention as size of high-quality paraphrase corpus is limited. Round-trip translation, also known as the pivoting method, is a typical approach to this end. However, we notice that the pivoting process involves multiple machine translation models and is likely to incur semantic drift during the two-step translations. In this paper, inspired by the Transformer-based language models, we propose a simple and unified paraphrasing model, which is purely trained on multilingual parallel data and can conduct zero-shot paraphrase generation in one step. Compared with the pivoting approach, paraphrases generated by our model is more semantically similar to the input sentence. Moreover, since our model shares the same architecture as GPT (Radford et al., 2018), we are able to pre-train the model on large-scale unparallel corpus, which further improves the fluency of the output sentences. In addition, we introduce the mechanism of denoising auto-encoder (DAE) to improve diversity and robustness of the model. Experimental results show that our model surpasses the pivoting method in terms of relevance, diversity, fluency and efficiency.

We present a new model of time-dependent flow at low latitudes in the Earth's core between 2000 and 2018, derived from magnetic field measurements made on board the {\it Swarm} and CHAMP satellites and at ground magnetic observatories. The model, called {\it CoreFlo-LL.1}, consists of a steady background flow without imposed symmetry plus a time-dependent flow that is dominated by geostrophic and quasi-geostrophic components but also allows weak departures from equatorial symmetry. We find that the equatorial region beneath the core-mantle boundary is a place of vigorous, localised, fluid motions; time-dependent flow focused at low latitudes close to the core surface is able to reproduce rapid field variations observed at non-polar latitudes at and above Earth's surface. Magnetic field acceleration pulses are produced by alternating bursts of non-zonal azimuthal flow acceleration in this region. Such acceleration sign changes can occur within a year or less, and when the structures involved are of large spatial scale they can give rise to geomagnetic jerks at the Earth's surface.

In English semantic similarity tasks, classic word embedding-based approaches explicitly model pairwise "interactions" between the word representations of a sentence pair. Transformer-based pretrained language models disregard this notion, instead modeling pairwise word interactions globally and implicitly through their self-attention mechanism. In this paper, we hypothesize that introducing an explicit, constrained pairwise word interaction mechanism to pretrained language models improves their effectiveness on semantic similarity tasks. We validate our hypothesis using BERT on four tasks in semantic textual similarity and answer sentence selection. We demonstrate consistent improvements in quality by adding an explicit pairwise word interaction module to BERT.

The Loreau and Hector (2019) Comment on our paper (Pillai and Gouhier, 2019) failed to address the two core elements of our critique, both the circularity of the BEF research program, in general, and the mathematical flaws of the Loreau-Hector partitioning scheme, in particular. Loreau and Hector avoided dealing with the first part of our critique by arguing against a non-existent claim that all biodiversity effects could be reduced to coexistence, while the mathematical flaws in the Loreau-Hector partitioning method that we described in the second part of our critique were ignored altogether. Here, we address these misconceptions and demonstrate that all of the claims that were made in our original paper hold. We conclude that (i) BEF studies need to adopt baselines that account for coexistence in order to avoid overestimating the effects of biodiversity and (ii) the LH partitioning method should not be used unless the linearity of the abundance-ecosystem functioning relationship in monocultures can be verified for all species.

Mountain ecosystems are sensitive indicators of climate change. Long-term studies may be extremely useful in assessing the responses of high-elevation ecosystems to climate change and other anthropogenic drivers. Mountain research sites within the LTER (Long-Term Ecosystem Research) network are representative of various types of ecosystems and span a wide bioclimatic and elevational range. Here, we present a synthesis and a review of the main results from long-term ecological studies in mountain ecosystems at 20 LTER sites in Italy, Switzerland and Austria. We analyzed a set of key climate parameters, such as temperature and snow cover duration, in relation to vascular species composition, plant traits, abundance patterns, pedoclimate, nutrient dynamics in soils and water, phenology and composition of freshwater biota. The overall results highlight the rapid response of mountain ecosystems to climate change. As temperatures increased, vegetation cover in alpine and subalpine summits increased as well. Years with limited snow cover duration caused an increase in soil temperature and microbial biomass during the growing season. Effects on freshwater ecosystems were observed, in terms of increases in solutes, decreases in nitrates and changes in plankton phenology and benthos communities. This work highlights the importance of comparing and integrating long-term ecological data collected in different ecosystems, for a more comprehensive overview of the ecological effects of climate change. Nevertheless, there is a need for i) adopting co-located monitoring site networks to improve our ability to obtain sound results from cross-site analysis, ii) carrying out further studies, with fine spatial and temporal resolutions to improve understanding of responses to extreme events, and iii) increasing comparability and standardizing protocols across networks to clarify local from global patterns.

Improvements in experimental and computational technologies have led to significant increases in data available for analysis. Topological data analysis (TDA) is an emerging area of mathematical research that can identify structures in these data sets. Here we develop a TDA method to detect physical structures in a cell that persist over time. In most cells, protein filaments (actin) interact with motor proteins (myosins) and organize into polymer networks and higher-order structures. An example of these structures are ring channels that maintain constant diameters over time and play key roles in processes such as cell division, development, and wound healing. The interactions of actin with myosin can be challenging to investigate experimentally in living systems, given limitations in filament visualization \textit{in vivo}. We therefore use complex agent-based models that simulate mechanical and chemical interactions of polymer proteins in cells. To understand how filaments organize into structures, we propose a TDA method that assesses effective ring generation in data consisting of simulated actin filament positions through time. We analyze the topological structure of point clouds sampled along these actin filaments and propose an algorithm for connecting significant topological features in time. We introduce visualization tools that allow the detection of dynamic ring structure formation. This method provides a rigorous way to investigate how specific interactions and parameters may impact the timing of filamentous network organization.

The question of whether deep neural networks are good at generalising beyond their immediate training experience is of critical importance for learning-based approaches to AI. Here, we consider tests of out-of-sample generalisation that require an agent to respond to never-seen-before instructions by manipulating and positioning objects in a 3D Unity simulated room. We first describe a comparatively generic agent architecture that exhibits strong performance on these tests. We then identify three aspects of the training regime and environment that make a significant difference to its performance: (a) the number of object/word experiences in the training set; (b) the visual invariances afforded by the agent's perspective, or frame of reference; and (c) the variety of visual input inherent in the perceptual aspect of the agent's perception. Our findings indicate that the degree of generalisation that networks exhibit can depend critically on particulars of the environment in which a given task is instantiated. They further suggest that the propensity for neural networks to generalise in systematic ways may increase if, like human children, those networks have access to many frames of richly varying, multi-modal observations as they learn.

Applications based on synergistic integration of optical imagery and LiDAR data are receiving a growing interest from the remote sensing community. However, a misaligned integration between these datasets may fail to fully profit the potential of both sensors. In this regard, an optimum fusion of optical imagery and LiDAR data requires an accurate registration. This is a complex problem since a versatile solution is still missing, especially when considering the context where data are collected at different times, from different platforms, under different acquisition configurations. This paper presents a coarse-to-fine registration method of aerial/satellite optical imagery with airborne LiDAR data acquired in such context. Firstly, a coarse registration involves extracting and matching of buildings from LiDAR data and optical imagery. Then, a Mutual Information-based fine registration is carried out. It involves a super-resolution approach applied to LiDAR data, and a local approach of transformation model estimation. The proposed method succeeds at overcoming the challenges associated with the aforementioned difficult context. Considering the experimented airborne LiDAR (2011) and orthorectified aerial imagery (2016) datasets, their spatial shift is reduced by 48.15% after the proposed coarse registration. Moreover, the incompatibility of size and spatial resolution is addressed by the mentioned super-resolution. Finally, a high accuracy of dataset alignment is also achieved, highlighted by a 40-cm error based on a check-point assessment and a 64-cm error based on a check-pair-line assessment. These promising results enable further research for a complete versatile fusion methodology between airborne LiDAR and optical imagery data in this challenging context.

Fully convolutional networks (FCNs) have been recently used for feature extraction and classification in image and speech recognition, where their inputs have been raw signal or other complicated features. Persian signature verification is done using conventional convolutional neural networks (CNNs). In this paper, we propose to use FCN for learning a robust feature extraction from the raw signature images. FCN can be considered as a variant of CNN where its fully connected layers are replaced with a global pooling layer. In the proposed manner, FCN inputs are raw signature images and convolution filter size is fixed. Recognition accuracy on UTSig database, shows that FCN with a global average pooling outperforms CNN.

In this paper, we propose a broad comparison between Fully Convolutional Networks (FCNs) and Mask Region-based Convolutional Neural Networks (Mask-RCNNs) applied in the Salient Object Detection (SOD) context. Studies in the SOD literature usually explore architectures based in FCNs to detect salient regions and objects in visual scenes. However, besides the promising results achieved, FCNs showed issues in some challenging scenarios. Fairly recently studies in the SOD literature proposed the use of a Mask-RCNN approach to overcome such issues. However, there is no extensive comparison between the two networks in the SOD literature endorsing the effectiveness of Mask-RCNNs over FCN when segmenting salient objects. Aiming to effectively show the superiority of Mask-RCNNs over FCNs in the SOD context, we compare two variations of Mask-RCNNs with two variations of FCNs in eight datasets widely used in the literature and in four metrics. Our findings show that in this context Mask-RCNNs achieved an improvement on the F-measure up to 47% over FCNs.