Search for: All records

Abstract Low‐velocity anomalies in the upper mantle beneath eastern North America, including the Northern Appalachian Anomaly (NAA), the Central Appalachian Anomaly (CAA), and the weaker Southern Coastal Anomaly (SCA), have been characterized by many continent‐scale and regional seismic studies. Different models have been proposed to explain their existence beneath the passive margin of eastern North America, variously invoking the past passage of hot spot tracks, modern upwelling due to edge‐driven convection, or other processes. Depending on the nature and origin of these anomalies, they may influence, and/or be influenced by, the mantle transition zone (MTZ) structure beneath them. Previous receiver function studies have identified an overall thinner MTZ beneath the eastern margin of the US than beneath the continental interior. In this study, we resolve the MTZ geometry beneath these low‐velocity anomalies in unprecedented detail using the scattered wavefield migration technique. We find substantially thinned MTZ beneath the NAA and the CAA, and a moderately thinned MTZ beneath the SCA. In all cases, the thinning is achieved via a minor depression of the 410‐km discontinuity and a major uplift of the 660‐km discontinuity, which suggests the presence of a series of MTZ‐penetrating deep upwellings beneath eastern North America. The upwellings beneath eastern North America and a similar style upwelling beneath Bermuda may initiate from ponded thermally buoyant materials below the MTZ fed by hot return flows from the descending Farallon slab in the deep mantle. 

Abstract Artificial neural networks are increasingly used for geophysical modeling to extract complex nonlinear patterns from geospatial data. However, it is difficult to understand how networks make predictions, limiting trust in the model, debugging capacity, and physical insights. EXplainable Artificial Intelligence (XAI) techniques expose how models make predictions, but XAI results may be influenced by correlated features. Geospatial data typically exhibit substantial autocorrelation. With correlated input features, learning methods can produce many networks that achieve very similar performance (e.g., arising from different initializations). Since the networks capture different relationships, their attributions can vary. Correlated features may also cause inaccurate attributions because XAI methods typically evaluate isolated features, whereas networks learn multifeature patterns. Few studies have quantitatively analyzed the influence of correlated features on XAI attributions. We use a benchmark framework of synthetic data with increasingly strong correlation, for which the ground truth attribution is known. For each dataset, we train multiple networks and compare XAI-derived attributions to the ground truth. We show that correlation may dramatically increase the variance of the derived attributions, and investigate the cause of the high variance: is it because different trained networks learn highly different functions or because XAI methods become less faithful in the presence of correlation? Finally, we show XAI applied to superpixels, instead of single grid cells, substantially decreases attribution variance. Our study is the first to quantify the effects of strong correlation on XAI, to investigate the reasons that underlie these effects, and to offer a promising way to address them. 

Abstract Complex machine learning architectures and high-dimensional gridded input data are increasingly used to develop high-performance geoscience models, but model complexity obfuscates their decision-making strategies. Understanding the learned patterns is useful for model improvement or scientific investigation, motivating research in eXplainable artificial intelligence (XAI) methods. XAI methods often struggle to produce meaningful explanations of correlated features. Gridded geospatial data tends to have extensive autocorrelation so it is difficult to obtain meaningful explanations of geoscience models. A recommendation is to group correlated features and explain those groups. This is becoming common when using XAI to explain tabular data. Here, we demonstrate that XAI algorithms are highly sensitive to the choice of how we group raster elements. We demonstrate that reliance on a single partition scheme yields misleading explanations. We propose comparing explanations from multiple grouping schemes to extract more accurate insights from XAI. We argue that each grouping scheme probes the model in a different way so that each asks a different question of the model. By analyzing where the explanations agree and disagree, we can learn information about the scale of the learned features. FogNet, a complex three-dimensional convolutional neural network for coastal fog prediction, is used as a case study for investigating the influence of feature grouping schemes on XAI. Our results demonstrate that careful consideration of how each grouping scheme probes the model is key to extracting insights and avoiding misleading interpretations. 

Abstract. Due to the increasing availability of high-performance computing over the past few decades, numerical models have become an important tool for research in geodynamics.Several generations of mantle convection software have been developed, but due to their differing methods and increasing complexity it is important to evaluate the accuracy of each new model generation to ensure published geodynamic research is reliable and reproducible.Here we explore the accuracy of the open-source, finite-element codes ASPECT and CitcomS as a function of mesh spacing using low to moderate-Rayleigh-number models in steady-state thermal convection.ASPECT (Advanced Solver for Problems in Earth's ConvecTion) is a new-generation mantle convection code that enables modeling global mantle convection with realistic parameters and complicated physical processes using adaptive mesh refinement (Kronbichler et al., 2012; Heister et al., 2017).We compare the ASPECT results with calculations from the finite-element code CitcomS (Zhong et al., 2000; Tan et al., 2006; Zhong et al., 2008), which has a long history of use in the geodynamics community.We find that the globally averaged quantities, i.e., root-mean-square (rms) velocity, mean temperature, and Nusselt number at the top and bottom of the shell, agree to within 1 % (and often much better) for calculations with sufficient mesh resolution.We also show that there is excellent agreement of the time evolution of both the rms velocity and the Nusselt numbers between the two codes for otherwise identical parameters.Based on our results, we are optimistic that similar agreement would be achieved for calculations performed at the convective vigor expected for Earth, Venus, and Mars. 

A national focus on Diversity, Equity, and Inclusion (DEI) has increased public awareness and acknowledgement of challenges faced by members of Under-represented Minority (URM) groups pursuing academic careers. This case study of a multi-institutional partner-ship explores the development, implementation, and evolution of a replicable model to transition a cohort of STEM URM dissertators into the professoriate. The model structured cohort engagement around an Individual Development Plan (IDP), cohort participant engagement with multiple mentors, monthly scholarly learning community meetings, research and teaching immersion experiences at Historically Black Colleges and Universities (HBCUs) and international institutions, and support and training around transferable skills necessary for job preparation and success, such as grant development, job search, portfolio preparation, interview skills, and online course development. Program evaluation results emphasized the evolution of collaborative practice among stakeholders in promoting the success of the model and among cohort participants as these participants transitioned into academic careers. Discussion of best practices to design and fine-tune the model included engagement of cohort participants in refining the implementation of the model activities, offering personalized services to the cohort members, and engaging research and practitioner communities using multiple dissemination strategies. The results of this work include publicly available virtual resources curated as part of the dissemination plan that can be explored for implementation at other institutions and use by individuals. 

Teachers’ integration of the Next Generation Science Standards and corresponding Science and Engineering Practices (SEPs) illustrate current science education reform in the United States. Effective teacher professional development (PD) on SEPs is essential for reform success. In this study, we evaluated the Nebraska STEM Education Conference, a PD program for middle school, high school, and first- and second-year post-secondary STEM teachers. This SEP-oriented PD program focused predominantly on the SEPs ‘developing and using models’ and ‘using mathematics and computational thinking.’ An electronic survey was used to measure participants’ (n = 45) prior integration of SEPs, influential factors and barriers to using SEPs, and changes to interest and confidence in using SEPs as a result of attending the PD program. Our results showed that teachers had limited prior use of SEPs in their teaching. Student interest and learning outcomes were the factors found to be most influential to teachers’ use of SEPs, while limited knowledge, confidence, and resources were the most commonly identified barriers. As a result of attending the PD program, participants significantly improved their confidence and interest to incorporate SEPs. We recommend continued SEP-oriented PD to foster successful NGSS integration and to advance reforms in science education. 

Constraining the thermal and compositional state of the mantle is crucial for deciphering the formation and evolution of Mars. Mineral physics predicts that Mars’ deep mantle is demarcated by a seismic discontinuity arising from the pressure-induced phase transformation of the mineral olivine to its higher-pressure polymorphs, making the depth of this boundary sensitive to both mantle temperature and composition. Here, we report on the seismic detection of a midmantle discontinuity using the data collected by NASA’s InSight Mission to Mars that matches the expected depth and sharpness of the postolivine transition. In five teleseismic events, we observed triplicated P and S waves and constrained the depth of this discontinuity to be 1,006 ± 40 km by modeling the triplicated waveforms. From this depth range, we infer a mantle potential temperature of 1,605 ± 100 K, a result consistent with a crust that is 10 to 15 times more enriched in heat-producing elements than the underlying mantle. Our waveform fits to the data indicate a broad gradient across the boundary, implying that the Martian mantle is more enriched in iron compared to Earth. Through modeling of thermochemical evolution of Mars, we observe that only two out of the five proposed composition models are compatible with the observed boundary depth. Our geodynamic simulations suggest that the Martian mantle was relatively cold 4.5 Gyr ago (1,720 to 1,860 K) and are consistent with a present-day surface heat flow of 21 to 24 mW/m 2 .