Earth Mover Research Articles

Assessing Clouds Using Satellite Observations Through Three Generations of Global Atmosphere Models

AbstractClouds are parameterized in climate models using quantities on the model grid‐scale to approximate the cloud cover and impact on radiation. Because of the complexity of processes involved with clouds, these parameterizations are one of the key challenges in climate modeling. Differences in parameterizations of clouds are among the main contributors to the spread in climate sensitivity across models. In this work, the clouds in three generations of an atmosphere model lineage are evaluated against satellite observations. Satellite simulators are used within the model to provide an appropriate comparison with individual satellite products. In some respects, especially the top‐of‐atmosphere cloud radiative effect, the models show generational improvements. The most recent generation, represented by two distinct branches of development, exhibits some regional regressions in the cloud representation; in particular the southern ocean shows a positive bias in cloud cover. The two branches of model development show how choices during model development, both structural and parametric, lead to different cloud climatologies. Several evaluation strategies are used to quantify the spatial errors in terms of the large‐scale circulation and the cloud structure. The Earth mover's distance is proposed as a useful error metric for the passive satellite data products that provide cloud‐top pressure‐optical depth histograms. The cloud errors identified here may contribute to the high climate sensitivity in the Community Earth System Model, version 2 and in the Energy Exascale Earth System Model, version 1.

Steel surface defect detection based on self-supervised contrastive representation learning with matching metric

Defect detection is crucial in the quality control of industrial applications. Existing supervised methods are heavily reliant on the large amounts of labeled data. However, labeled data in some specific fields are still scarce, and it requires professionals to do expensive manual annotations. In this paper, we construct a novel self-supervised steel surface defect detection model by learning better embedding feature representation of the defect on large amounts of unlabeled data, which can achieve excellent results in downstream detection tasks. Commonly used image embeddings strategies in self-supervised contrastive learning methods destroy the spatial structures of the image and are not suitable for pre-training of object detection. To address the aforementioned issue, we preserve convolutional feature maps to mine robust data structures and local features, which can enhance the representation capability of the upstream model and make it applicable for transfer to object detection tasks. Besides, in order to eliminate the effect of random augmentations of contrastive learning, which can introduce noise on multi-target coexistence datasets, the Earth Mover’s Distance (EMD) metric is employed to evaluate the contrastive matching similarity. Finally, a Self-supervised Contrastive Representation Learning framework with EMD (SCRL-EMD) is constructed through learning on large-scale unlabeled data and then transferred to Faster R-CNN and RetinaNet for detection performance validation on two public steel defect datasets. Comparative experimental results show that our method can achieve superior results than the state-of-the-art approaches. Compared to the baseline model, it achieves 4.1% and 6.8% mAP improvement on the two datasets, respectively. More importantly, a further improvement can be achieved on a smaller downstream dataset, revealing the meaningful potential of our method in exploiting more readily available unlabeled data.

Differentially Private Heatmaps

We consider the task of producing heatmaps from users' aggregated data while protecting their privacy. We give a differentially private (DP) algorithm for this task and demonstrate its advantages over previous algorithms on real-world datasets. Our core algorithmic primitive is a DP procedure that takes in a set of distributions and produces an output that is close in Earth Mover's Distance (EMD) to the average of the inputs. We prove theoretical bounds on the error of our algorithm under a certain sparsity assumption and that these are essentially optimal.

Cross-Modality Earth Mover’s Distance for Visible Thermal Person Re-identification

Visible thermal person re-identification (VT-ReID) suffers from inter-modality discrepancy and intra-identity variations. Distribution alignment is a popular solution for VT-ReID, however, it is usually restricted to the influence of the intra-identity variations. In this paper, we propose the Cross-Modality Earth Mover's Distance (CM-EMD) that can alleviate the impact of the intra-identity variations during modality alignment. CM-EMD selects an optimal transport strategy and assigns high weights to pairs that have a smaller intra-identity variation. In this manner, the model will focus on reducing the inter-modality discrepancy while paying less attention to intra-identity variations, leading to a more effective modality alignment. Moreover, we introduce two techniques to improve the advantage of CM-EMD. First, Cross-Modality Discrimination Learning (CM-DL) is designed to overcome the discrimination degradation problem caused by modality alignment. By reducing the ratio between intra-identity and inter-identity variances, CM-DL leads the model to learn more discriminative representations. Second, we construct the Multi-Granularity Structure (MGS), enabling us to align modalities from both coarse- and fine-grained levels with the proposed CM-EMD. Extensive experiments show the benefits of the proposed CM-EMD and its auxiliary techniques (CM-DL and MGS). Our method achieves state-of-the-art performance on two VT-ReID benchmarks.

Graph-Based Reconstruction and Analysis of Disease Transmission Networks Using Viral Genomic Data.

Understanding the patterns of viral disease transmissions helps establish public health policies and aids in controlling and ending a disease outbreak. Classical methods for studying disease transmission dynamics that rely on epidemiological data, such as times of sample collection and duration of exposure intervals, struggle to provide desired insight due to limited informativeness of such data. A more precise characterization of disease transmissions may be acquired from sequencing data that reveal genetic distance between viral genomes in patient samples. Indeed, genetic distance between viral strains present in hosts contains valuable information about transmission history, thus motivating the design of methods that rely on genomic data to reconstruct a directed disease transmission network, detect transmission clusters, and identify significant network nodes (e.g., super-spreaders). In this article, we present a novel end-to-end framework for the analysis of viral transmissions utilizing viral genomic (sequencing) data. The proposed framework groups infected hosts into transmission clusters based on the reconstructed viral strains infecting them; the genetic distance between a pair of hosts is calculated using Earth Mover's Distance, and further used to infer transmission direction between the hosts. To quantify the significance of a host in the transmission network, the importance score is calculated by a graph convolutional autoencoder. The viral transmission network is represented by a directed minimum spanning tree utilizing the Edmond's algorithm modified to incorporate constraints on the importance scores of the hosts. The proposed framework outperforms state-of-the-art techniques for the analysis of viral transmission dynamics in several experiments on semiexperimental as well as experimental data.

Earth mover’s distance as a measure of CP violation

We introduce a new unbinned two sample test statistic sensitive to CP violation utilizing the optimal transport plan associated with the Wasserstein (earth mover’s) distance. The efficacy of the test statistic is shown via two examples of CP asymmetric distributions with varying sample sizes: the Dalitz distributions of B0 → K+π−π0 and of D0 → π+π−π0 decays. The windowed version of the Wasserstein distance test statistic is shown to have comparable sensitivity to CP violation as the commonly used energy test statistic, but also retains information about the localized distributions of CP asymmetry over the Dalitz plot. For large statistic datasets we introduce two modified Wasserstein distance based test statistics — the binned and the sliced Wasserstein distance statistics, which show comparable sensitivity to CP violation, but improved computing time and memory scalings. Finally, general extensions and applications of the introduced statistics are discussed.

Refining data–data and data–model vegetation comparisons using the Earth mover's distance (EMD)

Abstract. Comparing temporal and spatial vegetation changes between reconstructions or between reconstructions and model simulations requires carefully selecting an appropriate evaluation metric. A common way of comparing reconstructed and simulated vegetation changes involves measuring the agreement between pollen- or model-derived unary vegetation estimates, such as the biome or plant functional type (PFT) with the highest affinity scores. While this approach based on summarising the vegetation signal into unary vegetation estimates performs well in general, it overlooks the details of the underlying vegetation structure. However, this underlying data structure can influence conclusions since minor variations in pollen percentages modify which biome or PFT has the highest affinity score (i.e. modify the unary vegetation estimate). To overcome this limitation, we propose using the Earth mover's distance (EMD) to quantify the mismatch between vegetation distributions such as biome or PFT affinity scores. The EMD circumvents the issue of summarising the data into unary biome or PFT estimates by considering the entire range of biome or PFT affinity scores to calculate a distance between the compared entities. In addition, each type of mismatch can be given a specific weight to account for case-specific ecological distances or, said differently, to account for the fact that reconstructing a temperate forest instead of a boreal forest is ecologically more coherent than reconstructing a temperate forest instead of a desert. We also introduce two EMD-based statistical tests that determine (1) if the similarity of two samples is significantly better than a random association given a particular context and (2) if the pairing between two datasets is better than might be expected by chance. To illustrate the potential and the advantages of the EMD as well as the tests in vegetation comparison studies, we reproduce different case studies based on previously published simulated and reconstructed biome changes for Europe and capitalise on the advantages of the EMD to refine the interpretations of past vegetation changes by highlighting that flickering unary estimates, which give an impression of high vegetation instability, can correspond to gradual vegetation changes with low EMD values between contiguous samples (case study 1). We also reproduce data–model comparisons for five specific time slices to identify those that are statistically more robust than a random agreement while accounting for the underlying vegetation structure of each pollen sample (case study 2). The EMD and the statistical tests are included in the paleotools R package (https://github.com/mchevalier2/paleotools, last access: 3 May 2023).

A Comparison of Multiple Odor Source Localization Algorithms

There are two primary algorithms for autonomous multiple odor source localization (MOSL) in an environment with turbulent fluid flow: Independent Posteriors (IP) and Dempster–Shafer (DS) theory algorithms. Both of these algorithms use a form of occupancy grid mapping to map the probability that a given location is a source. They have potential applications to assist in locating emitting sources using mobile point sensors. However, the performance and limitations of these two algorithms is currently unknown, and a better understanding of their effectiveness under various conditions is required prior to application. To address this knowledge gap, we tested the response of both algorithms to different environmental and odor search parameters. The localization performance of the algorithms was measured using the earth mover’s distance. Results indicate that the IP algorithm outperformed the DS theory algorithm by minimizing source attribution in locations where there were no sources, while correctly identifying source locations. The DS theory algorithm also identified actual sources correctly but incorrectly attributed emissions to many locations where there were no sources. These results suggest that the IP algorithm offers a more appropriate approach for solving the MOSL problem in environments with turbulent fluid flow.

Influence of Data Distribution on Federated Learning Performance in Tumor Segmentation.

To investigate the correlation between differences in data distributions and federated deep learning (Fed-DL) algorithm performance in tumor segmentation on CT and MR images. Two Fed-DL datasets were retrospectively collected (from November 2020 to December 2021): one dataset of liver tumor CT images (Federated Imaging in Liver Tumor Segmentation [or, FILTS]; three sites, 692 scans) and one publicly available dataset of brain tumor MR images (Federated Tumor Segmentation [or, FeTS]; 23 sites, 1251 scans). Scans from both datasets were grouped according to site, tumor type, tumor size, dataset size, and tumor intensity. To quantify differences in data distributions, the following four distance metrics were calculated: earth mover's distance (EMD), Bhattacharyya distance (BD), χ2 distance (CSD), and Kolmogorov-Smirnov distance (KSD). Both federated and centralized nnU-Net models were trained by using the same grouped datasets. Fed-DL model performance was evaluated by using the ratio of Dice coefficients, θ, between federated and centralized models trained and tested on the same 80:20 split datasets. The Dice coefficient ratio (θ) between federated and centralized models was strongly negatively correlated with the distances between data distributions, with correlation coefficients of -0.920 for EMD, -0.893 for BD, and -0.899 for CSD. However, KSD was weakly correlated with θ, with a correlation coefficient of -0.479. Performance of Fed-DL models in tumor segmentation on CT and MRI datasets was strongly negatively correlated with the distances between data distributions.Keywords: CT, Abdomen/GI, Liver, Comparative Studies, MR Imaging, Brain/Brain Stem, Convolutional Neural Network (CNN), Federated Deep Learning, Tumor Segmentation, Data Distribution Supplemental material is available for this article. © RSNA, 2023See also the commentary by Kwak and Bai in this issue.

SemSyn: Semantic-Syntactic Similarity Based Automatic Machine Translation Evaluation Metric

Machine translation evaluation is difficult and challenging for natural languages because different languages behave differently for the same dataset. Lexical-based metrics have been poorly represented semantic relationships and impose strict identity matching. However, translation and assessment become difficult for target morphologically rich languages with relatively free word order. Most of the standard evaluation metrics consider word order but do not effectively consider sentence structure. In this paper, we propose a novel machine translation evaluation metric SemSyn which incorporates both semantic and syntactic similarity. We incorporate the term frequency-inverse document frequency with the earth mover’s distance and word embedding to cover the semantic similarity. The part of speech and dependency parsing tags assist in covering syntactic similarity in the sentence structure. Part of speech and dependency parsing tags are extracted from universal dependencies and trained on the SpaCy library. Experimental results show that SemSyn has a higher correlation with human judgment than other evaluation metrics for morphologically rich language and other languages.

Graph-Regularized, Sparsity-Constrained Non-Negative Matrix Factorization with Earth Mover’s Distance Metric

Non-negative matrix factorization (NMF) is widely used as a powerful matrix factorization tool in data representation. However, the traditional NMF, measured by Euclidean distance or Kullback–Leibler distance, does not take into account the internal implied geometric information of the dataset and cannot measure the distance between samples as well as possible. To remedy the defects, in this paper, we propose the NMF method with Earth mover’s distance as a metric, for short GSNMF-EMD. It combines graph regularization and L1/2 smooth constraints. The GSNMF-EMD method takes into account the intrinsic implied geometric information of the dataset and can produce more sparse and stable local solutions. Experiments on two specific image datasets showed that the proposed method outperforms related state-of-the-art methods.

AI4SeaIce: selecting loss functions for automated SAR sea ice concentration charting

For maritime navigation in the Arctic, sea ice charts are an essential tool, which still to this day is drawn manually by professional ice analysts. The total Sea Ice Concentration (SIC) is the primary descriptor of the charts and indicates the fraction of ice in an ocean surface area. Naturally, automating the SIC chart creation is desired. However, the optimal representation of the corresponding machine-learning task is ambivalent and discussed in the community. In this study, we explore the representation with either regressional or classification objectives, each with two different (weighted) loss functions: Mean Square Error and Binary Cross-Entropy, and Categorical Cross-Entropy and the Earth Mover’s Distance, respectively. While all models achieve good results they differ as the regression-based models obtain the highest numerical similarity to the reference charts, whereas the classification-optimised models generate results more visually pleasing and consistent. Rescaling the loss functions with inverse class weights improves the performance for intermediate classes at the expense of open water and fully-covered sea ice areas.

Comparison of deep learning models for digital H&E staining from unpaired label-free multispectral microscopy images

Background and objective: This paper presents the quantitative comparison of three generative models of digital staining, also known as virtual staining, in H&E modality (i.e., Hematoxylin and Eosin) that are applied to 5 types of breast tissue. Moreover, a qualitative evaluation of the results achieved with the best model was carried out. This process is based on images of samples without staining captured by a multispectral microscope with previous dimensional reduction to three channels in the RGB range.Methods: The models compared are based on conditional GAN (pix2pix) which uses images aligned with/without staining, and two models that do not require image alignment, Cycle GAN (cycleGAN) and contrastive learning-based model (CUT). These models are compared based on the structural similarity and chromatic discrepancy between samples with chemical staining and their corresponding ones with digital staining. The correspondence between images is achieved after the chemical staining images are subjected to digital unstaining by means of a model obtained to guarantee the cyclic consistency of the generative models.Results: The comparison of the three models corroborates the visual evaluation of the results showing the superiority of cycleGAN both for its larger structural similarity with respect to chemical staining (mean value of SSIM ∼ 0.95) and lower chromatic discrepancy (10%). To this end, quantization and calculation of EMD (Earth Mover’s Distance) between clusters is used. In addition, quality evaluation through subjective psychophysical tests with three experts was carried out to evaluate quality of the results with the best model (cycleGAN).Conclusions: The results can be satisfactorily evaluated by metrics that use as reference image a chemically stained sample and the digital staining images of the reference sample with prior digital unstaining. These metrics demonstrate that generative staining models that guarantee cyclic consistency provide the closest results to chemical H&E staining that also is consistent with the result of qualitative evaluation by experts.

Isoelectric point barcode and similarity analysis with the earth mover’s distance for identification of species origin of raw meat

In this work, by combining the microcolumn isoelectric focusing (mIEF) and similarity analysis with the earth mover’s distance (EMD) metric, we proposed the concept of isoelectric point (pI) barcode for the identification of species origin of raw meat. At first, we used the mIEF to analyze 14 meat species, including 8 species of livestock and 6 species of poultry, to generate 140 electropherograms of myoglobin/hemoglobin (Mb/Hb) markers. Secondly, we binarized the electropherograms and converted them into the pI barcodes that only showed the major Mb/Hb bands for the EMD analysis. Thirdly, we efficiently developed the barcode database of 14 meat species and successfully used the EMD method to identify 9 meat products thanks to the high throughput of mIEF and the simplified format of the barcode for similarity analysis. The developed method had the merits of facility, rapidity and low cost. The developed concept and method had evident potential to the facile identification of meat species.

An Uncertainty-driven Peak-integration (UP) Strategy for 3D Borehole Layout Planning

Equal-spaced borehole drilling is commonly adopted in practice for site investigation, which ignores spatial uncertainty and might result in insufficient site characterization. Model-driven precise drilling is regarded as a potential solution. In this study, an uncertainty-driven peak-integration (UP) strategy is proposed to improve decision-making in 3D borehole planning. The UP strategy consists of stratigraphic uncertainty modelling and peak-integration method. The random field-based uncertainty modelling approach has been improved by employing stratum-informed Scales of Fluctuation and a novel site division method to capture the site-specific uncertainty distribution with sufficient interpretation. The peak-integration method is proposed to consider the uncertainty around peaks, as improvement of existing peak uncertainty-based borehole planning strategy. To provide an objective evaluation for borehole layout planning, a new index “Contribution of Borehole” (CoB) is proposed for the first time, using the Earth Mover’s Distance, to quantify the net changes of stratigraphic distribution revealed by additional boreholes and offer validation for every decision-making. The feasibility and practicability of the proposed strategy are demonstrated by applying to an artificial geological model created based on real borehole data. The superiority of the proposed strategy is elucidated by comparing with equal-spaced strategy, via both CoB and conventional entropy reduction with detailed discussion.

Understanding the Convergence of the Preconditioned PDHG Method: A View of Indefinite Proximal ADMM

The primal-dual hybrid gradient (PDHG) algorithm is popular in solving min-max problems which are being widely used in a variety of areas. To improve the applicability and efficiency of PDHG for different application scenarios, we focus on the preconditioned PDHG (PrePDHG) algorithm, which is a framework covering PDHG, alternating direction method of multipliers (ADMM), and other methods. We give the optimal convergence condition of PrePDHG in the sense that the key parameters in the condition can not be further improved, which fills the theoretical gap in the-state-of-art convergence results of PrePDHG, and obtain the ergodic and non-ergodic sublinear convergence rates of PrePDHG. The theoretical analysis is achieved by establishing the equivalence between PrePDHG and indefinite proximal ADMM. Besides, we discuss various choices of the proximal matrices in PrePDHG and derive some interesting results. For example, the convergence condition of diagonal PrePDHG is improved to be tight, the dual stepsize of the balanced augmented Lagrangian method can be enlarged to 4/3 from 1, and a balanced augmented Lagrangian method with symmetric Gauss-Seidel iterations is also explored. Numerical results on the matrix game, projection onto the Birkhoff polytope, earth mover’s distance, and CT reconstruction verify the effectiveness and superiority of PrePDHG.

In‐Situ Observation of Magnetic Null on 19 September 2015 Event Using Magnetospheric Multiscale Mission

AbstractIn this paper, we detect a magnetic null inside the MMS space‐craft (SC) tetrahedron using a linear interpolation that might be nearly impossible for very small inter‐SC separation like MMS (previously implemented on Cluster mission which is larger separation SC). We also report, successive or identical magnetic nulls that enter/reenter and cross the MMS spacecraft tetrahedron, and the “bifurcations” (change of the null type) of magnetic nulls are captured when it crosses one of four MMS tetrahedral surfaces during the 19 September 2015 event. During this event, we observe the eigenvalues of the radial null that enters and crosses the MMS tetrahedron changes from all real to one conjugate complex pair and one real eigenvalues, when the null exits through one of the tetrahedral boundary cells. This bifurcation is also evidenced by measuring the topological distance between two nulls before and after “bifurcation” using the Earth Mover's Distance.

An Improved SVM with Earth Mover’s Distance Regularization and Its Application in Pattern Recognition

A support vector machine (SVM) aims to achieve an optimal hyperplane with a maximum interclass margin and has been widely utilized in pattern recognition. Traditionally, a SVM mainly considers the separability of boundary points (i.e., support vectors), while the underlying data structure information is commonly ignored. In this paper, an improved support vector machine with earth mover’s distance (EMD-SVM) is proposed. It can be regarded as an improved generalization of the standard SVM, and can automatically learn the distribution between the classes. To validate its performance, we discuss the necessity of the structural information of EMD-SVM in the linear and nonlinear cases, respectively. Experimental validation was designed and conducted in different application fields, which have shown its superior and robust performance.

Spatial and Texture Analysis of Root System distribution with Earth mover’s Distance (STARSEED)

PurposeRoot system architectures are complex and challenging to characterize effectively for agronomic and ecological discovery.MethodsWe propose a new method, Spatial and Texture Analysis of Root SystEm distribution with Earth mover’s Distance (STARSEED), for comparing root system distributions that incorporates spatial information through a novel application of the Earth Mover’s Distance (EMD).ResultsWe illustrate that the approach captures the response of sesame root systems for different genotypes and soil moisture levels. STARSEED provides quantitative and visual insights into changes that occur in root architectures across experimental treatments.ConclusionSTARSEED can be generalized to other plants and provides insight into root system architecture development and response to varying growth conditions not captured by existing root architecture metrics and models. The code and data for our experiments are publicly available: https://github.com/GatorSense/STARSEED.

Offshore wind energy forecasting sensitivity to sea surface temperature input in the Mid-Atlantic

Abstract. As offshore wind farm development expands, accurate wind resource forecasting over the ocean is needed. One important yet relatively unexplored aspect of offshore wind resource assessment is the role of sea surface temperature (SST). Models are generally forced with reanalysis data sets, which employ daily SST products. Compared with observations, significant variations in SSTs that occur on finer timescales are often not captured. Consequently, shorter-lived events such as sea breezes and low-level jets (among others), which are influenced by SSTs, may not be correctly represented in model results. The use of hourly SST products may improve the forecasting of these events. In this study, we examine the sensitivity of model output from the Weather Research and Forecasting model (WRF) 4.2.1 to different SST products. We first evaluate three different data sets: the Multiscale Ultrahigh Resolution (MUR25) SST analysis, a daily, 0.25∘ × 0.25∘ resolution product; the Operational Sea Surface Temperature and Ice Analysis (OSTIA), a daily, 0.054∘ × 0.054∘ resolution product; and SSTs from the Geostationary Operational Environmental Satellite 16 (GOES-16), an hourly, 0.02∘ × 0.02∘ resolution product. GOES-16 is not processed at the same level as OSTIA and MUR25; therefore, the product requires gap-filling using an interpolation method to create a complete map with no missing data points. OSTIA and GOES-16 SSTs validate markedly better against buoy observations than MUR25, so these two products are selected for use with model simulations, while MUR25 is at this point removed from consideration. We run the model for June and July of 2020 and find that for this time period, in the Mid-Atlantic, although OSTIA SSTs overall validate better against in situ observations taken via a buoy array in the area, the two products result in comparable hub-height (140 m) wind characterization performance on monthly timescales. Additionally, during hours-long flagged events (< 30 h each) that show statistically significant wind speed deviations between the two simulations, both simulations once again demonstrate similar validation performance (differences in bias, earth mover's distance, correlation, and root mean square error on the order of 10−1 or less), with GOES-16 winds validating nominally better than OSTIA winds. With a more refined GOES-16 product, which has been not only gap-filled but also assimilated with in situ SST measurements in the region, it is likely that hub-height winds characterized by GOES-16-informed simulations would definitively validate better than those informed by OSTIA SSTs.