Emulators of Earth System Models (ESMs) are complementary to ESMs by providing climate information at lower computational costs. Thus far, the emulation of spatially resolved climate extremes has only received limited attention, even though it is one of the most impactful aspects of climate change. Here, we propose a method for the emulation of local annual maximum temperatures, with a focus on reproducing essential statistical properties such as correlations in space and time. We test different emulator configurations and find that driving the emulations with global mean surface temperature offers an optimal compromise of model complexity and performance. We show that the emulations can mimic the temporal evolution and spatial patterns of the underlying climate model simulations and are able to reproduce their natural variability. The general design and the good performance for annual maximum temperatures suggests that the proposed methodology can be applied to other climate extremes.

Discrimination between tectonic earthquakes and quarry blasts is important for accurate earthquake cataloging and seismic hazard analysis. However, reliable classification is challenging with raw waveforms and no prior knowledge of source parameters. Here we apply deep learning to perform this task in southern California and eastern Kentucky, which differ significantly in available labelled data, class imbalance and waveform characteristics. Accordingly, we adopt different strategies for the two regions. First, we directly train a convolutional neural network (CNN) for southern California due to its data abundancy. To alleviate class imbalance, the blast data are augmented by randomly shifting waveform windows. The model for California yields an accuracy of 91.97% for single-station classification and 97.54% for network-averaged classification. Second, as eastern Kentucky has a much smaller data size, we fine-tune the pretrained California model to fit the Kentucky data. The fine-tuned model yields an accuracy of 97.35% for single-station classification and 99.46% for network-averaged classification. The fine-tuned model outperforms the model trained from scratch. Finally, we use occlusion test and gradient-weighted class activation mapping to illuminate which parts of waveforms are important for model prediction. Our results demonstrate that deep learning can achieve high accuracy in seismic event discrimination with raw waveforms and that transfer learning is effective and efficient to generalize deep learning models across different regions.

For high-resolution regional geodetic applications, the International Terrestrial Reference Frame (ITRF) is complemented by regional densifications. These are realised either as multi-year solutions related to a tectonic plate (e.g., EUREF for Europe) or as epoch reference frames (ERFs) to capture non-linear geophysical effects like earthquakes or loading displacements (e.g., SIRGAS for Latin America). These GNSS-only-based regional frames have in common that their geodetic datum is aligned with the ITRF datum at a specific epoch. Their origin is thus geocentric only in a mean sense and does not always coincide with the instantaneous centre of mass. Here, we present studies on a direct geocentric realisation of regional ERFs. We propose to realise the geodetic datum for each epoch by combining global GNSS, SLR and VLBI networks via measured local ties at co-located sites. An equally-distributed global GNSS network is used to realise the orientation via a no-net-rotation constraint and is densified by the stations of the regional subnetwork. The developed combination and filtering strategy aims to guarantee a stable datum realisation for each epoch-wise solution. The effectiveness of our methods is validated against the current operational realisation of the SIRGAS Latin American reference frame. Comparing with geophysical loading models relating to the Earth’s centres of mass and figure, we show that the realised geocentric displacement time series directly reflect seasonal geophysical processes. Moreover, as the approach does not need to rely on co-location sites in the region of interest, it is conceptually transferrable to other global regions.

Double seismic zones (DSZs), parallel planes of intermediate-depth earthquakes inside oceanic slabs, have been observed in a number of subduction zones and may be a ubiquitous feature of downgoing oceanic plates. Focal mechanism observations from DSZ earthquakes sample the intraslab stress field at two distinct depth levels within the downgoing lithosphere. A pattern of downdip compressive over downdip extensive events was early on interpreted to indicate an unbending-dominated intraslab stress field. In the present study, we show that the intraslab stress field in the depth range of DSZs is much more variable than previously thought. Compiling DSZ locations and mechanisms from literature, we observe that the “classical’ pattern of compressive over extensive events is only observed at about half of the DSZ locations around the globe. The occurrence of extensional mechanisms across both planes accounts for most other regions. To obtain an independent estimate of the bending state of slabs at intermediate depths, we compute (un)bending estimates from slab geometries taken from the slab2 compilation of slab surface depths. We find no clear global prevalence of slab unbending at intermediate depths, and the occurrence of DSZ seismicity does not appear to be limited to regions of slab unbending. Focal mechanism observations are frequently inconsistent with (un)bending estimates from slab geometries, which may imply that bending stresses are not always prevalent, and that other stress types such as in-plane tension due to slab pull or shallow compression due to friction along the plate interface may also play an important role.

Stochastic slip rupture modeling is a computationally efficient reduced-physics approximation that has the capability to create any number of unique ruptures based only on a few statistical assumptions. Its simplicity and efficiency make it an attractive and viable option for testing early warning systems, hazard assessments, and infrastructure response studies. Yet a fundamental question pertaining to this approach is whether the slip distributions calculated in this way are “realistic”. More specifically, can stochastic modeling reproduce slip distributions that match what is seen in M9+ events recorded in instrumental time? Here, we start with the 2011 M9.1 Tohoku-oki earthquake and tsunami where we test both a stochastic method with a homogeneous background mean model and a method where slip is informed by an additional interseismic coupling constraint. We test two coupling constraints with varying assumptions of either trench-locking or -creeping and assess their influence on the calculated ruptures. We quantify the dissimilarity of slip distribution between the 12,000 modeled ruptures and a slip inversion for the Tohoku earthquake. We also model tsunami inundation for over 300 ruptures and compare the results to an inundation survey along the eastern coastline of Japan. We conclude that stochastic slip modeling can produce ruptures that can be considered “Tohoku-like”, and inclusion of coupling can both positively and negatively influence the ability to create realistic ruptures. We then expand our study and show that for the 1960 M9.4-9.6 Chile and 2004 M9.1-9.3 Sumatra events, stochastic slip modeling has the capability to produce realistic ruptures.

To incentivize the participation and contribution to the growth of an earth-science-based cyberinfrastructure, analytical environments need to be developed that allow automatic analysis and classification of data from connected data repositories. The purpose of this study is to investigate a machine learning technique for automatically detecting shear-sense-indicating clasts (i.e., sigma or delta clasts and mica fish) in photomicrographs, and finding their shear sense (i.e., sinistral (CCW) or dextral (CW) shearing). Previous work employed transfer learning, a technique in which a pre-trained Convolutional Neural Network (CNN) was repurposed, and artificially augmented image datasets to distinguish between CCW and CW shearing. Preprocessing images by denoising, a process in which noise at different scales is removed while preserving edges of an image, improved classification accuracy. However, upon randomizing the denoising parameters, the CNN model didn’t converge due to severe lack of data. While the efforts for acquiring more labeled data is ongoing, this work compensated for it by implementing a pre-processing “detection” system that automatically crops images to regions of image containing the clasts. This is done by utilizing YOLOv3, a CNN based image detection system that outputs a bounding box around an object of interest. YOLOv3 was trained using 93 photomicrographs containing bounding boxes of 344 shear-sense-indicating clasts. The retrained detector was tested on two sets: set A with 10 photomicrographs containing clasts and set B with 100 photomicrographs not containing clasts. All but one of the clasts in set A were correctly detected with an average confidence score of 96.6%. On set B, 72% of images correctly did not indicate presence of clasts. On the remaining images, where clasts were incorrectly identified, an average confidence score of 78.3% was observed. By utilizing a threshold on the confidence scores, the system could be made more accurate. Future work involves utilizing the bounding boxes output by the detection system to refine and improve the CNN model for classifying shear sense of clasts in photomicrographs.

Ocean waves interact with the environment in many ways. They transport energy and mass, and the resultant sea-surface roughness defines the drag coefficients that transmits wind energy to the ocean (Drennan et al., 2003). Through erosion and deposition, waves change the shape and landscape of coastal areas. Storm surge waves can cause flood damage in coastal areas. Recent studies revealed that wetlands are sensitive to the wave condition, which determines the retreat or growth of coastal ecosystems (Green and Coco, 2007; Mariotti and Fagherazzi 2010). Human activities rely on the condition of waves to conduct marine activities such as fishing, shipping, oil extraction, and offshore constructions. Thus, it is important to understand ocean waves to improve earth system modeling, protect the coastline, predict storm surge, preserve coastal ecosystems, and enhance the offshore business. This project will explore the application of synthetic aperture radar (SAR) imagery to predict significant wave height near the coast. High-frequency (HF) radar data of the ocean (aka CODAR) was used as ground truth data set to calibrate and validate the wave height estimator. Off-shore wind data was also included. The developed code will enhance the current capability to process the satellite data and create a new platform to monitor the coastal environment. The collected data will help further our understanding of the wave spectrum in a coastal environment and the data can support other research in the related topics, e.g. the interaction of waves and ice sheets, wetlands, shorelines, wind farm and aquaculture.

Linking large-eddy simulations to local observations of clouds enables us to investigate clouds and microphysical processes and to improve our understanding as well as the respective representation in coarser models. Insights gained with large-eddy simulations can be applied within the development and evaluation of parameterizations for larger scale models, to bridge the gap between those models and the detailed local observations. In this study, various approaches for large-eddy simulations around local observations are explored and investigated with respect to remote sensing observations of clouds and their representation. By adding more ‘realism’ to the simulations and defining a well constrained setup, the representation of the daily variability at a mid-latitude site could be improved and shows promising results for continued research in the future. Especially the shown potential to investigate representativeness of column measurements will provide new insights into the analysis and construction of observational experiments.

High sea surface chlorophyll concentration on the Argentine Continental Shelf (ACS) frequently extends to the deep ocean in the vicinity of the Brazil/Malvinas Confluence (BMC). The offshore transport of shelf waters likely plays a key role in the biogeochemical balance of the western South Atlantic and promotes the offshore transport of planktonic species. We analyze data from an oceanographic survey carried out in the western South Atlantic shelf between 31° and 38°S in October 2013. We describe the distribution and circulation of the water masses and focus on the exchanges with the open ocean. On-shelf subsurface intrusions of oceanic waters and river discharge supply nutrients to the shelf. A low-salinity tongue of Río de la Plata (RDP) waters extends northward to 32°S. Below these waters Subantarctic and Subtropical Shelf waters (SASW, STSW) meet to form the Subtropical Shelf Front. The main SASW branch, oversaturated in oxygen and with high-fluorescence mixes with a detachment of Brazil Current waters at 38°S and is exported offshore along the BMC. A second branch of SASW reaches 33°S mixing along its way with RDP and STSW, and returns southward after splitting into an onshore and an offshore branch. The offshore branch is exported to the open ocean through the BMC. These export routes are in overall qualitative agreement with those indicated by a high-resolution reanalysis. We estimate a net off-shelf transport (geostrophic plus Ekman) of 3.33Sv to the open ocean between 32.1° and 37.7°S. The majority of the offshore flow occurs between 34.7° and 37.7°S.

One of the requirements for inferring local mantle redox states via zircon Ce/Ce* oxybarometry and Ti-in-zircon thermometry is to select mantle-equilibrated zircon (i.e., unaltered magmatic zircon without interaction to hydrosphere). Traditional protocols for identifying mantle-equilibrated zircon require a comprehensive examination of whole rock geochemistry in addition to zircon morphology, texture, and trace element and isotopic measurements to obtain information for zircon saturation context, source melt petrogenesis, and post-magmatic history. This study proposes a simple filter scheme for selecting mantle-equilibrated zircon using trace element systematics alone (P, Ti, Y, Nb, REE, Hf, Th, U). A total of 13 filtering criteria are synthesized from previous studies based on simulations, experiments, and compiled global datasets, all of which help to provide geological context and ultimately constrain the inferred melt redox state from three perspectives: (1) mineral inclusions; (2) source melt petrogenesis; and, (3) zircon saturation context. The filter scheme presented here, which is based on detailed classifications of zircon morphology/texture and host rock compositions of 2317 zircon analyses from 30 independent references, is shown to distinguish non-magmatic zircon (Group III), magmatic zircon with significant inclusions and/or sourced from highly enriched source melt (Group II), and mantle-equilibrated zircon (Group I). A case study of Group I mantle-equilibrated zircon from Greenland suggests that their respective mantle source has reached FMQ -0.5 since ca. 2950 Ma. Future applications of this filter scheme include studies of out-of-context detrital and/or xenocrystic zircon.

Meteoroids smaller than a microgram constantly bombard the Earth, depositing in the mesosphere and lower thermosphere. Meteoroid ablation, the explosive evaporation of meteoroids due to erosive impacts of atmospheric particles, consists of sputtering and sublimation. This paper presents the first atomic scale modeling of sputtering, the initial stage of ablation where hypersonic collisions between the meteoroid and atmospheric particles cause the direct ejection of atoms from the meteoroid surface. Because meteoroids gain thermal energy from these particle impacts, these interactions are important for sublimation as well. In this study, a molecular dynamics simulator calculates the energy distribution of the sputtered particles as a function of the species, velocity, and angle of the incoming atmospheric particles. The sputtering yield generally agrees with semi-empirical equations at normal incidence but disagrees with the generally accepted angular dependence. Λ, the fraction of energy from a single atmospheric particle impact incorporated into the meteoroid, was found to be less than 1 and dependent on the velocity, angle, atmospheric species, and meteoroid material. Applying this new Λ to an ablation model results in a slower meteoroid temperature increase and mass loss rate as a function of altitude. This alteration results in changes in the expected electron line densities and visual magnitudes of meteoroids. Notably, this analysis leads to the prediction that meteoroids will generally ablate 1 - 4 km lower than previously predicted. This affects analysis of radar and visual measurements, as well as determination of meteoroid mass.

Nature-based solutions (NBS) can act as risk redistribution elements within risk management strategies in which insurances play an important role. The (re)insurance industry has been developing its own catastrophe models since the late 1980s for monitoring the ongoing challenge of climate change for decades. The integration of preventive measures, like concrete dikes, into those models is partially effective. But the complex effect of NBS, as preventive measures, on hazard reduction is a scientific challenge and in its infancy. Nevertheless, the (re)insurance industry has started assessing nature-based solutions’ effectiveness (as preventive measures) in reducing risks and their co-benefits. NBS could secure long-term affordability of climate risks insurance schemes in a changing climate.

Using bibliometric analysis techniques, we trace the evolution of climate and climate-change related articles in major oceanographic journals, 1987-2017. We use these bibliometric tools (network mapping, cluster analysis, alluvial analysis, corpus keyword detection) to document trends in growth, integration and centralization of climate-related research within ocean sciences over the past three decades. Such analysis methods offer an objective and complementary methodology, in contrast to the traditional “expert panel” approach, for guiding long-term strategic science planning. But how does the macro trend compare to scientific outputs supported by large ocean observatory facilities? Have scientists making use of these facilities followed, led or diverged from the general trend? We compare the macro trend to corpora of published science from two such facilities, Australia’s Integrated Marine Observing System (IMOS) and Ocean Networks Canada (ONC). The goal is to discern the extent to which these “big science” ocean observatories have been able to support or lead research that helps inform policy, management and the public about critical societal issues such as long term ocean change.

Because remote sensing (RS) data are spatially and temporally explicit and available across the globe, they have the potential to be used for predicting runoff in ungauged catchments and poorly gauged regions, a challenging area of research in hydrology. There is potential to use remotely sensed data for calibrating hydrological models in regions with limited streamflow gauges. This study conducts a comprehensive investigation on how to incorporate gridded remotely sensed evapotranspiration (AET) and water storage data for constraining hydrological model calibration in order to predict daily and monthly runoff in 30 catchments in the Yalong River basin in China. To this end, seven RS data calibration schemes are explored, and compared to direct calibration against observed runoff and traditional regionalization using spatial proximity to predict runoff in ungauged catchments. The results show that using bias-corrected remotely sensed AET (bias-corrected PML-AET data) for constraining model calibration performs much better than using the raw remotely sensed AET data (non-bias-corrected AET obtained from PML model estimate). Using the bias-corrected PML-AET data in a gridded way is much better than using lumped data, and outperforms the traditional regionalization approach especially in headwater and large catchments. Combining the bias-corrected PML-AET and GRACE water storage data performs similarly to using the bias-corrected PML-AET data only. This study demonstrates that there is great potential in using bias-corrected RS-AET data to calibrating hydrological models (without the need for gauged streamflow data) to estimate daily and monthly runoff time series in ungauged catchments and sparsely gauged regions.

A rare disturbance of the stratospheric Antarctic polar vortex in September 2019 led to a significantly higher than usual polar total ozone column. We use assimilation of ozone, HCl, and NO data from the Microwave Limb Sounder with the Global Earth Observing System Constituent Data Assimilation System driven by reanalysis meteorology to study the evolution of the 2019 Antarctic polar ozone. We find that the maximum 2019 ozone hole area was near 10 million km, and as little as 20% of that in 2018 in mid-September. However, the magnitude of vortex-averaged chemical ozone depletion was not significantly different between the two years despite earlier chlorine deactivation in 2019. The assimilation results show that most of the differences between 2018 and 2019 Antarctic ozone resulted from two factors: (1) the geometry of the 2019 vortex, with ozone-rich middle-stratospheric air masses overlying the lower portion of the vortex and leading to a significant reduction of the total column; (2) significantly reduced vortex volume. The anomalously small ozone hole of 2019 was comparable in size to the record breaking 2002 case and the mechanisms responsible were similar in the two cases. While the 2019 sudden stratospheric warming is classified as minor, its impact on ozone was very significant.

Air-sea exchange of mercury (Hg) is influenced by meteorological factors that have substantial interannual variability. Here we investigate its interannual variability and influcing factors by using the MITgcm ocean model. We elucidate a latitudinal pattern with a relatively small variability in the mid-latitudes (6.4%-8.5%) and a large one in the Arctic and Equator (15%). Wind speed, salinity, and sea ice dominate the patterns in equatorial, mid-, and high-latitudinal oceans, respectively. A in equatorial Pacific is found between El Niño and La Niña events, owing to wind speed anomaly caused by the variation of Walker circulation. A higher atmospheric Hg concentration (1-2%) in northern hemisphere is found by the GEOS-Chem simulation due to higher ocean evasion fluxes, consistent with the CAMNet and EMEP (1-11%) observations. Besides, a slight fluctuation in the upper tropospheric (±0.5%) reveals a potential contribution from the ocean evasion for interannual variability of tropospheric Hg.

The Qiangtang Basin is a large endorheic basin as the inner part of the Tibetan Plateau and has been thought to be a dry region in contrast with its wet surrounding outer region that feeds all the major Asian rivers. Combing surface hydrological data with modelling and satellite data between 2002 and 2017, our study reveals that an enormous amount of water of 54.52±15.36 km is unaccounted for annually in the Qiangtang Basin. The amount of this missing water is comparable to the total annual discharge of the Yellow River. Data from the Gravity Recovery and Climate Experiment (GRACE) show no increase of the local terrestrial water storage. Thus the missing water must have flowed out of the basin through underground passages. Interpreting this result with recent seismic and geological studies of Tibet, we suggest that a significant amount of meteoric water in the Qiangtang Basin have leaked out by way of groundwater flow through deep normal faults and tensional fractures along the nearly N-S rift valleys that are oriented sub-normal to and cross the surficial hydrological divide on the southern margin of the basin. Cross-basin groundwater outflow of such a magnitude defies the traditional view of basin-scale water cycle and leads to a very different picture from the previous hydrological view of the Qiangtang Basin. The finding calls for major rethinking of the water balance in Tibet and the nearby regions.

The natural sound generated by an unmanned aerial vehicle is used in conjunction with tomography to remotely sense atmospheric temperature and wind profiles simultaneously. Sound fields recorded onboard the aircraft and by an array of microphones on the ground are compared and converted to sound speed estimates for the ray paths intersecting the intervening medium. Tomographic inversion is then used to transform these sound speed values into vertical cross-sections and 3D volumes of virtual temperature and wind vectors, which enables the atmosphere to be visualised and monitored over time up to altitudes of 1,200m and over baselines of up to 600m. This paper reports on results from two short campaigns during which 2D and 3D profiles of wind and temperature obtained in this way were compared to: measurements taken by co-located mid-range Doppler SODAR and LIDAR; and temperature measurements made by instruments carried by unmanned aircraft flying through the intervening atmosphere. Large eddy simulation of daytime atmospheric boundary layers were also used to examine the anticipated performance of the instruments and the nature of any errors. The observations obtained using all systems are shown to correspond closely.