Soil biogeochemistry across Central and South American tropical dry forests
Not provided.
Environmental Sciences & Ecology↗
SEARCH · Engineering Papers
Results for “spatial pattern”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Muon track reconstruction in a segmented bolometric array using multi-objective optimization
Recent advances in segmented solid-state detector arrays for rare-event searches have allowed the technology to approach the ton-scale in detector mass and the scale of meters in size. Often focused around searches for neutrinoless double-beta decay or direct dark matter detection, such experiments also have the capability to search for exotic particles that leave track-like signatures across their volume. However, the segmented nature of such detector arrays often sets the spatial resolution and makes the problem of reconstructing track-like paths non-trivial. Here, in this paper, we present an algorithm that improves reconstruction of track-like events in segmented detectors using multi-objective optimization — a computational technique that optimizes more than one cost function at a time without specifying a quantitative weighting between them. Such a technique allows the reconstruction of tracks through a detector and the determination of path-lengths through individual elements. When combined with the reconstructed energy depositions in each element this allows for a calculation of the stopping power of track-like particles and opens the door to searches for particles with abnormal stopping power like monopoles or lightly-ionizing particles (LIPs). Results are presented which evaluate the precision of the reconstruction tools as they currently stand against Monte Carlo generated data. The algorithm is presented in the context of the CUORE experiment, but has applications to other segmented calorimeter detectors.
47 OTHER INSTRUMENTATION↗
Future Vision for Autonomous Ocean Observations
Autonomous platforms already make observations over a wide range of temporal and spatial scales, measuring salinity, temperature, nitrate, pressure, oxygen, biomass; and many other parameters. However, the observations are not comprehensive. Future autonomous systems need to be more affordable, more modular, more capable and easier to operate. Creative new types of platforms and new compact, low power, calibrated and stable sensors are under development to expand autonomous observations. Communications and recharging need bandwidth and power which can be supplied by standardized docking stations. In situ power generation will also extend endurance for many types of autonomous platforms, particularly autonomous surface vehicles. Standardized communications will improve ease of use, interoperability, and enable coordinated behaviors. Improved autonomy and communications will enable adaptive networks of autonomous platforms. Improvements in autonomy will have three aspects: hardware, control, and operations. As sensors and platforms have more onboard processing capability and energy capacity, more measurements become possible. Control systems and software will have the capability to address more complex states and sophisticated reactions to sensor inputs, which allows the platform to handle a wider variety of circumstances without direct operator control. Operational autonomy is increased by reducing operating costs. To maximize the potential of autonomous observations new standards and best practices are needed. In some applications, focus on common platforms and volume purchases could lead to significant cost reductions. Cost reductions could enable order-of-magnitude increases in platform operations and increase sampling resolution for a given level of investment. Energy harvesting technologies should be integral to the system design, for sensors, platforms, vehicles, and docking stations. Connections are needed between the marine energy and ocean observing communities to coordinate among funding sources, researchers, and end users. Regional teams should work with global organizations such as IOC/GOOS in governance development. International networks such as EGO for the emerging glider operations should also provide a forum for addressing governance. Networks of multiple vehicles can improve operational efficiencies and transform operational patterns. There is a need to develop operational architectures at regional and global scales to provide a backbone for active networking of autonomous platforms.
47 OTHER INSTRUMENTATION↗
Dynamic land use implications of rapidly expanding and evolving wind power deployment
Abstract The expansion of wind power poses distinct and varied geographic challenges to a sustainable energy transition. However, current knowledge of its land use impacts and synergies is limited by reliance on static characterizations that overlook the role of turbine technology and plant design in mediating interactions with the environment. Here, we investigate how wind technology development and innovation have shaped landscape interactions with social and ecological systems within the United States and contribute to evolving land area requirements. This work assesses trends in key land use facets of wind power using a holistic set of metrics to establish an evidence base that researchers, technology designers, land use managers, and policymakers can use in envisioning how future wind-intensive energy systems may be jointly optimized for clean energy, social, and environmental objectives. Since 2000, we find dynamic land occupancy patterns and regional trends that are driven by advancing technology and geographic factors. Though most historical U.S. wind deployment has been confined to the temperate grassland biome in the nation’s interior, regional expansion has implicated diverse land use and cover types. A large percentage of the typical wind plant footprint (∼96% to > 99%) is not directly impacted by permanent physical infrastructure, allowing for multiple uses in the spaces between turbines. Surprisingly, turbines are commonly close to built structures. Moreover, rangeland and cropland have supported 93.4% of deployment, highlighting potential synergies with agricultural lands. Despite broadly decreasing capacity densities, offsetting technology improvements have stabilized power densities. Land use intensity, defined as the ratio of direct land usage to lifetime power generation of wind facilities, has also trended downwards. Although continued deployment on disturbed lands, and in close proximity to existing wind facilities and other infrastructure, could minimize the extent of impacts, ambitious decarbonization trajectories may predispose particular biomes to cumulative effects and risks from regional wind power saturation. Increased land-use and sustainability feedback in technology and plant design will be critical to sustainable management of wind power.
17 WIND ENERGY↗
Climate projections over the Great Lakes Region: using two-way coupling of a regional climate model with a 3-D lake model
Abstract. Warming trends in the Laurentian Great Lakes and surrounding areas have been observed in recent decades, and concerns continue to rise about the pace and pattern of future climate change over the world's largest freshwater system. To date, most regional climate models used for Great Lakes projections either neglected the lake-atmosphere interactions or are only coupled with a 1-D column lake model to represent the lake hydrodynamics. This study presents a Great Lakes climate change projection that has employed the two-way coupling of a regional climate model with a 3-D lake model (GLARM) to resolve 3-D hydrodynamics essential for large lakes. Using the three carefully selected Coupled Model Intercomparison Project Phase 5 (CMIP5) general circulation models (GCMs), we show that the GLARM ensemble average substantially reduces surface air temperature and precipitation biases of the driving GCM ensemble average in present-day climate simulations. The improvements are not only displayed from an atmospheric perspective but are also evident in the accurate simulations of lake temperature and ice coverage. We further present the GLARM projected climate change for the mid-21st century (2030–2049) and the late 21st century (2080–2099) in the Representative Concentration Pathway (RCP) 4.5 and RCP 8.5 scenarios. Under RCP 8.5, the Great Lakes basin is projected to warm by 1.3–2.1 ∘C by the mid-21st century and 4.1–5.0 ∘C by the end of the century relative to the early century (2000–2019). Moderate mitigation (RCP 4.5) reduces the mid-century warming to 0.8–1.8 ∘C and late-century warming to 1.8–2.7 ∘C. Annual precipitation in GLARM is projected to increase for the entire basin, varying from 0 % to 13 % during the mid-century and from 9 % to 32 % during the late century in different scenarios and simulations. The most significant increases are projected in spring and fall when current precipitation is highest and a minimal increase in winter when it is lowest. Lake surface temperatures (LSTs) are also projected to increase across the five lakes in all of the simulations, but with strong seasonal and spatial variability. The most significant LST increases occur in Lakes Superior and Ontario. The strongest warming is projected in spring that persists into the summer, resulting from earlier and more intense stratification in the future. In addition, diminishing winter stratification in the future suggests the transition from dimictic lakes to monomictic lakes by the end of the century. In contrast, a relatively smaller increase in LSTs during fall and winter is projected with heat transfer to the deep water due to the strong mixing and energy required for ice melting. Correspondingly, the highest monthly mean ice cover is projected to reduce to 3 %–15 % and 10 %–40 % across the lakes by the end of the century in RCP 8.5 and RCP 4.5, respectively. In the coastal regions, ice duration is projected to decrease by up to 60 d.
Xue, Pengfei (ORCID:000000025702421X)↗
Little vertical and circumferential variations in stem xylem water Δ 2 H and Δ 18 O in three tree species
Vertical and circumferential patterns of xylem water isotope signatures in large trees are rarely characterized but may influence interpretation of water source investigations and soil/xylem water isotope offsets. Furthermore, to examine vertical and circumferential variation in xylem water δ 2 H and δ 18 O, we collected xylem tissue at 5 heights (0, 1, 4, 7 and 10 m) from three replicate trees of three species with contrasting xylem anatomy, two angiosperms (Liquidambar styraciflua and Quercus nigra) and one gymnosperm (Pinus taeda). Concurrently, we also determined δ 2 H and δ 18 O of groundwater, soil water, and recent precipitation. On a different day, we collected circumferential samples at 1 m from the same trees to test for sectorality effects. Water from stem and soil samples were cryogenically extracted and analyzed for δ 2 H and δ 18 O. Mean xylem water δ 2 H and δ 18 O were significantly different between species, both vertically and circumferentially. We did not find significant (p = 0.05) systematic variation in δ 2 H or δ 18 O with height. We found no significant evidence for sectorality effects on δ 2 H, δ 18 O. Variances of vertical synoptic xylem water δ 2 H or δ 18 O were similar between species, 7.2–10.4‰ for δ 2 H and 0.58–0.81‰ for δ 18 O. Circumferential variances were also similar between species, 4.0–6.0 for δ 2 H and 0.37–0.44 for δ 18 O. A mixing model showed that sweetgum, water oak and loblolly pine, were drawing most of their water from deep soil from 45 to 190 cm (84.7, 68.4 and 53.2%, respectively) however, soil water-excess values indicate δ 2 H fractionation effects on these estimates. Dual isotope mixing model evaluation with single and multiple sample configurations showed that source water estimates were not affected by within tree variability of xylem water signatures. Xylem water δ 2 H and δ 18 O variability with height or circumference, was 3.2 and 2.7 times less than between tree variation because of transient temporal and spatial processes and is, therefore, not likely to affect interpretations of water sourcing in these three species.
59 BASIC BIOLOGICAL SCIENCES↗
Drivers and ecological impacts of a wildfire outbreak in the southern Appalachian Mountains after decades of fire exclusion
During one of the warmest and driest droughts of the last century, the southern Appalachian Mountains experienced a regional outbreak of over a dozen large wildfires in late fall of 2016. We provide a synthesis of long-term forest changes leading up to the 2016 wildfires, examine the climatic setting and patterns of burn severity in relationship to topography, and discuss the ecological and management implications of these and future fires. During the pre- and post-European settlement periods, frequent low- and mixed-severity wildfires interacted with complex topographic gradients and maintained heterogenous landscapes dominated by several species of oak (Quercus spp.), pine (Pinus spp.), and the American chestnut (Castanea dentata). Land-use changes associated with European settlement, loss of the American chestnut, and 20th century fire exclusion resulted in large scale shifts towards mesic, fire-intolerant species (i.e. mesophication). Wildfire activity began increasing in the early 1980s, but most fires in the region are small (<1,000 ha). In Fall of 2016, deciduous tree leaf fall occurred concurrently with a period of anomalously dry and warm weather, creating ideal conditions for fire ignition and spread. Thousands of ignitions across the region strained suppression resources and eight fires grew to greater than 5,000 ha. The 2016 fires were larger and burned more area than in the previous three decades combined. In one unique landscape setting, the Chimney Tops 2 Fire, a synoptic wind event drove extreme fire behavior and burned large, high-severity patches resulting in devastating effects in the wildland urban interface. However, immediate post-fire burn severity mosaics for other fires were composed primarily of low- (73 %) and moderate-severity (21 %) fire effects. High-severity fire comprised only 6 % of the area burned, and occurred mostly on steep upper slopes and ridges on south-facing aspects, reflecting the importance of bottom-up topographic drivers in this region. Although the fires will likely enhance biodiversity by restoring fire-dependent species and creating early seral habitat, invasions of non-native plant species, delayed mortality of mature pines and oaks, and rapid re-sprouting of pyrophyllic shrubs pose significant management challenges. Although similar large fire outbreaks may become more common under future climatic conditions, they are unlikely to reverse the effects of mesophication and significantly alter forest dynamics at broad spatial scales. Furthermore, the 2016 fires exposed the vulnerability of the region to wildfire during acute fall drought and demonstrate the potential ecological effects of future wildfires in mixed pine-hardwood landscapes of the southern Appalachians.
54 ENVIRONMENTAL SCIENCES↗
River bank erosion and accretion rates, and planform metrics of the East River, downstream of Gothic Colorado over intervals between 1955 and 2015
This dataset provides the tabular summary of analysis of an alluvial floodplain reach of the East River, downstream of Gothic, CO near Crested Butte. The measurements include bank erosion and accretion rates, channel width, bank curvature, and the aspect/orientation of the river banks. The measurements were derived from binary masks of the location of the river channel from aerial and satellite imagery collected on seven dates over a sixty-year timespan: 1955, 1973, 1983, 1990, 2001, 2011, and 2015. The masks were analyzed using the Spatially Continuous Riverbank Erosion and Accretion Measurements (SCREAM) software detailed in Rowland et al. 2016. The masks used in this analysis can be found an accompanying dataset (DOI: ). Rates of change along the East River were measured over a total of 8-time intervals: 1955-1973; 1955-2015; 1973-1983; 1983-1990; 1990-2001; 2001-2011; 2001-2015; and 2011-2015.In files with “summary” in the name, the data is provided at a pixel level, where each mapped bank pixel has an associated erosion or accretion value, a channel width, a curvature value, and an aspect each river and time period will have an individual file. Files with “Segments” in name provide data that is averaged along segments of the rivers. These segments are approximately 10 channel widths in length. In addition to erosion and accretion rates, the segment-based results include area measurements of erosion and accretion, islands, and channels. The number of islands is also included.
54 ENVIRONMENTAL SCIENCES↗
Using graph neural networks to reconstruct charged pion showers in the CMS High Granularity Calorimeter
A novel method to reconstruct the energy of hadronic showersin the CMS High Granularity Calorimeter (HGCAL) is presented. TheHGCAL is a sampling calorimeter with very fine transverse andlongitudinal granularity. The active media are silicon sensors andscintillator tiles readout by SiPMs and the absorbers are acombination of lead and Cu/CuW in the electromagnetic section, andsteel in the hadronic section. The shower reconstruction method isbased on graph neural networks and it makes use of a dynamicreduction network architecture. It is shown that the algorithm isable to capture and mitigate the main effects that normally hinderthe reconstruction of hadronic showers using classicalreconstruction methods, by compensating for fluctuations in themultiplicity, energy, and spatial distributions of the shower'sconstituents. The performance of the algorithm is evaluated usingtest beam data collected in 2018 prototype of the CMS HGCALaccompanied by a section of the CALICE AHCAL prototype. Thecapability of the method to mitigate the impact of energy leakagefrom the calorimeter is also demonstrated.
46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
Particle hit clustering and identification using point set transformers in liquid argon time projection chambers
Liquid argon time projection chambers are often used in neutrino physics and dark-matter searches because of their high spatial resolution. The images generated by these detectors are extremely sparse, as the energy values detected by most of the detector are equal to 0, meaning that despite their high resolution, most of the detector is unused in a particular interaction. Instead of representing all of the empty detections, the interaction is usually stored as a sparse matrix, a list of detection locations paired with their energy values. Traditional machine learning methods that have been applied to particle reconstruction such as convolutional neural networks (CNNs), however, cannot operate over data stored in this way and therefore must have the matrix fully instantiated as a dense matrix. Operating on dense matrices requires a lot of memory and computation time, in contrast to directly operating on the sparse matrix. We propose a machine learning model using a point set neural network that operates over a sparse matrix, greatly improving both processing speed and accuracy over methods that instantiate the dense matrix, as well as over other methods that operate over sparse matrices. Compared to competing state-of-the-art methods, our method improves classification performance by 14%, segmentation performance by more than 22%, while taking 80% less time and using 66% less memory. Compared to state-of-the-art CNN methods, our method improves classification performance by more than 86%, segmentation performance by more than 71%, while reducing runtime by 91% and reducing memory usage by 61%.
calibration and fitting methods↗
LLNL Kimberlina 1.2 NUFT Simulations June 2018 (v2)
This dataset contains the output 6,000, 3-dimensional reactive multi-phase flow and transport aquifer simulations of brine and CO2 leakage into a protective aquiver in California’s San Joaquin Valley and input data files detailing the geologic mesh, aquifer physical properties and CO2 and brine injection rates. This data set was generated as an ongoing effort with the US DOE National Risk Assessment Partnership (NRAP) to evaluate the effectiveness of monitoring techniques to detect brine and CO2 leakage from legacy wells into underground sources of drinking water overlaying a CO2 storage reservoir. Each simulation contains a unique set of input parameters, generated stochastically. The outputs consist of these upper three geologic layers (from top): the Etchegoin, Macoma-Chanac, Santa Margarita-McLure formations. These simulations span the several distances (1, 3 and 6 km or wells W31-0.2, W31-0.5 and W31-1.0, respectively) from the CO2 injector, initiated from bottom hole pressure and saturation to calculate wellbore leakage from the storage reservoir, with low and high regional groundwater gradients and wellbore leakage into 5 leaky nodes. The dataset includes 1,000 unique simulations for each distance, which each contain a unique aquifer heterogeneity, aquifer and caprock permeability, and two model generations are included with a high permeability (prod07) and hybrid permeability (prod09). The range of permeability distributions is listed in Table 1. Each model generation consists of 3,000 simulations. Included in the dataset are the leakage rates determined from 2D wellbore models which utilize the pressure and CO2 saturation from LBL's reservoir simulations, NUFT mesh files with distributed lithology, NUFT rocktab files which describe the material properties for the geologic layers and the NUFT input files and post-processed output 'ntab' files. Each ntab file contains spatial (rows) and temporal (columns) model output tables for each model cell, the locations (x,y,z) and dimensions for each cells (dx, dy, dz). Table 1. Permeability distribution ranges for prod07 and prod09 model generations Geologic Layer: Permeability Range (log10 m^2) prod07 prod09 Etchegoin -12.92 to -10.92 -13.70 to -11.44 Macoma-Chanac -12.72 to -10.72 -13.50 to -11.24 Santa Margarita-McLure -12.70 to -10.70 -13.48 to -11.22 The input files used to generate the model include which are included in the dataset are: Time series of CO2 leakage input into the model (ex: Q_brn.W31-0.2.sim1000.layers123.tab) Time series of CO2 leakage input into the model (ex: Q_CO2.W31-0.2.sim1000.layers123.tab) Physical properties of the aquifer materials detailing the aquifer porosity, solid density, partitioning coefficients, permeabilities and van-Genuchten parameters detailed in a NUFT rocktab file: (ex: sim1000.usnt.rocktab) Numerical mesh and geologic data assigned to each model cell detailed in a NUFT genmsh format (ex: sim1000.mesh_k16.prod07.trans.genmsh) The primary output parameters are: pH (use absolute value) Change in TDS (mg/kg) Change in Pressure (Pa) Change CO2 gas saturation (fraction range 0.0-1.0) for example, the directory /p/lscratchh/mansoor1/nrap/kimberlina/prod09/mainfiles/sim1000/W31- 0.2 contains: sim1000.W31-0.2.trans.pH.red.ntab sim1000.W31-0.2.no_bg.trans.TDS.red.ntab sim1000.W31-0.2.usnt.P.deltabg.red.ntab sim1000.W31-0.2.usnt.CO2_sat.deltabg.red.ntab Each row in the NTAB files consist of model output per numerical grid cell. Each output file contains 33 columns (variables), including the information of numerical records, geologic location and sizes and the simulated parameter values over time. The first 13 variables are about numerical records and relative geologic information for a simulation grid: 1. index: simulation index 2. i: the ith grid of x-axis 3. j: the ith grid of y-axis 4. k: the ith grid of z-axis 5. element_ref: element reference 6. nuft_ind: nuft index 7. x: grid location in the x axis direction 8. y: grid location in the y axis direction 9. z: grid location in the z axis direction 10. dx: grid length in the x axis direction 11. dy: grid length in the y axis direction 12. dz: grid length in the z axis direction 13. volume: volume of the simulation grid The remainder (14, 15, 16...) variables are the simulated parameter values over time, take Pressure as an example, are: 14. 0.0y: initial pressure per cell. 15. 10.0y: simulated pressure at the end of the 10th year. 16. 20.0y: simulated pressure at the end of the 20th year. ... (repeated for every 10 years until 200 years)... The model extends 10,000 m, 5,000 m and 1,411 m in the x,y and z dimensions, respectively. The mesh consists of 164,832 cells with mesh dimensions of 101 x 51 x 32 (nx, ny, nz), with cell dimensions ranging from 100 m laterally (along x and y-axis) and model layers are as designated in the z-axis: Layer 1: atmosphere (1e-30 m thick) Layer 2: upper caprock (10 m thick) Layers 3-13: Etchegoin (536.23 m thck) Layers 14-27: Macoma-Chanac (679.04 m thick) Layers 28-32: Santa Margarita-McLure (185.94 m thick) The wellbore is placed along node i=51, j=26, and extends vertically along 5 nodes from the top to the bottom of the model. Special instructions when extracting files: Each Gzip archive (ex: prod07.sim1000-sim00099.tar.gz) contains 100 simulations. Gzip archives should be transferred into base directories (ie. In Linux: mkdir prod07; mv prod07.*.tar.gz prod07/.) before extracting, or files will be overwritten. Each sub-simulation tree should have the following file structure pattern (using the linux 'tree' command): |-- prod07 | |-- sim0001 | |-- W31-0.2 | | |-- Q_brn.W31-0.2.sim0001.layers123.tab | | |-- Q_co2.W31-0.2.sim0001.layers123.tab | | |-- sim0001.W31-0.2.no_bg.trans.TDS.red.ntab | | |-- sim0001.W31-0.2.trans.pH.red.ntab | | |-- sim0001.W31-0.2.usnt.CO2_sat.deltabg.red.ntab | | |-- sim0001.W31-0.2.usnt.P.deltabg.red.ntab | |-- W31-0.5 | | |-- Q_brn.W31-0.5.sim0001.layers123.tab | | |-- Q_co2.W31-0.5.sim0001.layers123.tab | | |-- sim0001.W31-0.5.no_bg.trans.TDS.red.ntab | | |-- sim0001.W31-0.5.trans.pH.red.ntab | | |-- sim0001.W31-0.5.usnt.CO2_sat.deltabg.red.ntab | | |-- sim0001.W31-0.5.usnt.P.deltabg.red.ntab | |-- W31-1.0 | | |-- Q_brn.W31-1.0.sim0001.layers123.tab | | |-- Q_co2.W31-1.0.sim0001.layers123.tab | | |-- sim0001.W31-1.0.no_bg.trans.TDS.red.ntab | | |-- sim0001.W31-1.0.trans.pH.red.ntab | | |-- sim0001.W31-1.0.usnt.CO2_sat.deltabg.red.ntab | | |-- sim0001.W31-1.0.usnt.P.deltabg.red.ntab | |-- sim0001.mesh_k16.prod07.trans.genmsh Disclaimer This document was prepared as an account of work sponsored by an agency of the United States government. Neither the United States government nor Lawrence Livermore National Security, LLC, nor any of their employees makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States government or Lawrence Livermore National Security, LLC. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States government or Lawrence Livermore National Security, LLC, and shall not be used for advertising or product endorsement purposes. Lawrence Livermore National Laboratory is operated by Lawrence Livermore National Security, LLC, for the U.S. Department of Energy, National Nuclear Security Administration under Contract DE-AC52-07NA27344. This report was reviewed and released as LLNL-MI-753464.
aquifer↗
Self-Supervised Cloud Classification
Abstract Low-level marine clouds play a pivotal role in Earth’s weather and climate through their interactions with radiation, heat and moisture transport, and the hydrological cycle. These interactions depend on a range of dynamical and microphysical processes that result in a broad diversity of cloud types and spatial structures, and a comprehensive understanding of cloud morphology is critical for continued improvement of our atmospheric modeling and prediction capabilities moving forward. Deep learning has recently accelerated our ability to study clouds using satellite remote sensing, and machine learning classifiers have enabled detailed studies of cloud morphology. A major limitation of deep learning approaches to this problem, however, is the large number of hand-labeled samples that are required for training. This work applies a recently developed self-supervised learning scheme to train a deep convolutional neural network (CNN) to map marine cloud imagery to vector embeddings that capture information about mesoscale cloud morphology and can be used for satellite image classification. The model is evaluated against existing cloud classification datasets and several use cases are demonstrated, including training cloud classifiers with very few labeled samples, interrogation of the CNN’s learned internal feature representations, cross-instrument application, and resilience against sensor calibration drift and changing scene brightness. The self-supervised approach learns meaningful internal representations of cloud structures and achieves comparable classification accuracy to supervised deep learning methods without the expense of creating large hand-annotated training datasets. Significance Statement Marine clouds heavily influence Earth’s weather and climate, and improved understanding of marine clouds is required to improve our atmospheric modeling capabilities and physical understanding of the atmosphere. Recently, deep learning has emerged as a powerful research tool that can be used to identify and study specific marine cloud types in the vast number of images collected by Earth-observing satellites. While powerful, these approaches require hand-labeling of training data, which is prohibitively time intensive. This study evaluates a recently developed self-supervised deep learning method that does not require human-labeled training data for processing images of clouds. We show that the trained algorithm performs competitively with algorithms trained on hand-labeled data for image classification tasks. We also discuss potential downstream uses and demonstrate some exciting features of the approach including application to multiple satellite instruments, resilience against changing image brightness, and its learned internal representations of cloud types. The self-supervised technique removes one of the major hurdles for applying deep learning to very large atmospheric datasets.
54 ENVIRONMENTAL SCIENCES↗