The Data Problem. We Need an Evolution in Technology to Get Us Out

VOL. 101 | NO. 8 Europe’s Biodiversity Strategy AUGUST 2020 A Virtual Hackathon Fights Locusts

MH370’s Search Reveals New Science

INNOVATIONS IN TECHNOLOGY GOT US INTO

THE DATA PROBLEM. WE NEED AN EVOLUTION IN TECHNOLOGY TO GET US OUT.

FROM THE EDITOR

Editor in Chief Heather Goss, AGU, Washington, D.C., USA; [email protected] AGU Sta The Rise of Machine Learning Vice President, Communications, Amy Storey Marketing,and Media Relations

e cover the data problem here in Eos quite a bit. But Editorial Manager, News and Features Editor Caryl-Sue Micalizio “the data problem” is a misnomer: With so many Science Editor Timothy Oleson ways to collect so much data, the modern era of News and Features Writer Kimberly M. S. Cartier W Jenessa Duncombe science is faced not with one problem, but several. Where we’ll News and Features Writer store all the data is only the first of them. Then, what do we Production & Design do with it all? With more information than an army of humans Manager, Production and Operations Faith A. Ishii could possibly sift through on any single research project, Senior Production Specialist Melissa A. Tribur scientists are turning to machines to do it for them. Production and Analytics Specialist Anaise Aristide “I first encountered neural networks in the 1980s,” said Kirk Assistant Director, Design & Branding Beth Bagley Senior Graphic Designer Valerie Friedman Martinez, Eos science adviser for AGU’s informatics section Graphic Designer J. Henry Pereira and a professor at the University of Southampton, United Marketing King dom, when he suggested the theme for our August issue. Director, Marketing, Branding & Advertising Jessica Latterman “Neural networks were a ‘biologically inspired’ way to make Assistant Director, Marketing & Advertising Liz Zipse algorithms which could be trained to respond to certain inputs. Since then it has blossomed into a part of machine learning as we know it today.” Advertising Grappling with climate change research is one of the clearest places where machine learn- Display Advertising Dan Nicholas [email protected] ing will be crucial. “Currently, less than 5% of available environmental observational data are Recruitment Advertising Kristin McCarthy used in numerical models of Earth systems,” writes Amy McGovern and colleagues in “Weath- [email protected] ering Environmental Change Through Advances in AI” (p. 15). Recent innovations for artificial Science Advisers intelligence methods will allow researchers to harness more of those data, but realizing the Geomagnetism, Paleomagnetism, Julie Bowles full potential will require interdisciplinary collaboration to build the proper infrastructure. and Electromagnetism What then? Residents in the U.S. Great Plains are already getting AI weather forecasts that Space Physics and Aeronomy Christina M. S. Cohen Cryosphere Ellyn Enderlin give 36 hours of notice before a hailstorm, write McGovern et al. AI is mapping ecological prov- Study of the Earth’s Deep Interior Edward J. Garnero inces in the ocean (see p. 14), and soon it could even decode alien atmospheres (see p. 7), if Geodesy Brian C. Gunter scientists can overcome “the curse of dimensionality.” History of Geophysics Kristine C. Harper Planetary Sciences Sarah M. Hörst Machine learning, said Martinez, is going to be an essential part of data analysis in Earth Natural Hazards Michelle Hummel and space sciences. “It gives us a way to classify images and signals that we would have strug- Volcanology, Geochemistry, and Petrology Emily R. Johnson gled to process before,” but we also need to understand its limitations. Societal Impacts and Policy Sciences Christine Kirchho Seismology Keith D. Koper The discoveries machine learning can help scientists make come at a cost. The processing Tectonophysics Jian Lin power required for these algorithms requires massive amounts of energy—and many in the Near-Surface Geophysics Juan Lorenzo geoscience community are clamoring for even more powerful supercomputers. In “A Greener Earth and Space Science Informatics Kirk Martinez Paleoceanography and Paleoclimatology Figen Mekik Earth Model” (p. 18), Richard Loft writes about the irony of creating a significant carbon foot- Mineral and Rock Physics Sébastien Merkel print to build climate models that tell us what burning all that carbon is doing to climate. Ocean Sciences Jerry L. Miller Instead, he urges scientists to lead by example, offering several ways to think about—and Global Environmental Change Hansi Singh lower—the energy requirements of these systems. Education Eric M. Riggs Hydrology Kerstin Stahl On the theme of analyzing and interpreting data, don’t miss Stephanie Zeller’s and David Tectonophysics Carol A. Stein Rogers’s “Visualizing Science: How Color Determines What We See” (p. 28). They write, “Visu- Atmospheric Sciences Mika Tosca alization’s single most effective encoder—color—remains vastly understudied,” but here they Nonlinear Geophysics Adrian Tuck Biogeosciences Merritt Turetsky offer a masterful introduction of the many considerations involved when conveying data. Hydrology Adam S. Ward Needless to say, the images illustrating the article are page turners. Diversity and Inclusion Lisa D. White We hope you enjoy this issue on data and machine learning and that it makes you think Earth and Planetary Surface Processes Andrew C. Wilcox Atmospheric and Space Electricity Yoav Yair about the ways in which you might use it—or improve its use—in your own work. GeoHealth Ben Zaitchik

©2020. AGU. All Rights Reserved. Material in this issue may be photocopied by individual scientists for research or classroom use. Permission is also granted to use short quotes, figures, and tables for publication in scientific books and journals. For permission for any other uses, contact the AGU Publications Oce. Eos (ISSN 0096-3941) is published monthly by AGU, 2000 Florida Ave., NW, Washington, DC 20009, USA. Periodical Class postage paid at Washington, D.C., and at additional mailing oces. POSTMASTER: Send address changes to Member Heather Goss, Editor in Chief Service Center, 2000 Florida Ave., NW, Washington, DC 20009, USA Member Service Center: 8:00 a.m.–6:00 p.m. Eastern time; Tel: +1-202-462-6900; Fax: +1-202-328-0566; Tel. orders in U.S.: 1-800-966-2481; [email protected]. Submit your article proposal or suggest a news story to Eos at bit.ly/Eos-proposal. Views expressed in this publication do not necessarily reflect ocial positions of AGU unless expressly stated.

SCIENCE NEWS BY AGU // Eos.org 1 CONTENT

18 28

Features

18 A Greener Earth Model 24 Creating Data Tool Kits By Richard Loft That Everyone Can Use Can climate modelers reduce the carbon footprint that By Zhong Liu et al. goes into crunching more and more data? If the data can’t work together, the scientists can’t either. On the Cover Credit: iStock.com/GarryKillian 28 Visualizing Science: How Color Determines What We See By Stephanie Zeller and David Rogers

Color determines what we see. Scientists should explore these new tools to better communicate their data.

2 Eos // AUGUST 2020 CONTENT

15 50

Columns

From the Editor Research Spotlight 1 The Rise of Machine Learning 49 How Much Modification Can Earth’s Water Cycle Handle? | Improving Atmospheric Forecasts with News Machine Learning 50 Machine Learning Improves Weather 4 Europe Launches Biodiversity Strategy for the Coming and Climate Models Decade 51 Space Weather Forecasting Takes Inspiration from 5 Teaching Machines to Detect Climate Extremes Meteorology | Hardwood Forest Soils Are Sinks 6 Hackathon Participants Solve Global Problems— for Plant- Produced Volatiles from Home 52 How Long Was Venus Habitable? | Chinese Swamp 7 Machine Learning Can Help Decode Alien Skies— Core Reveals 47,000 Years of Monsoon History Up to a Point 8 Search for MH370 Revealed Ocean Crust Waves Editors’ Highlights 10 The Future of Big Data May Lie in Tiny Magnets 11 Long Live the Laurentian Great Lakes 53 Health Concerns from Combined Heat and Pollution 13 Are Geysers a Signal of Magma Intrusion Beneath in South Asia | Eruption and Emissions Take Credit Yellowstone? for Ocean Carbon Sink Changes | Using Saturn’s Rings as a Seismometer 14 How Machine Learning Redraws the Map of Ocean Ecosystems Positions Available Opinion 54 Current job openings in the Earth and space sciences 15 Weathering Environmental Change Through Advances in AI Postcards from the Field 57 Scientists are gathering data on nutrient loading AGU News and identifying patterns of impact in the pools of Ofu, in the Manu‘a Islands of American Samoa, in order 36 In Appreciation of AGU’s Outstanding Reviewers to protect coral from algae outbreaks. of 2019

AmericanGeophysicalUnion @AGU_Eos company/american-geophysical-union AGUvideos americangeophysicalunion americangeophysicalunion

SCIENCE NEWS BY AGU // Eos.org 3 NEWS

Europe Launches Biodiversity Strategy for the Coming Decade

s world governments draw up recovery plans in response to the corona- A virus 2019 (COVID-19) pandemic, environmental concerns risk being shunted aside as leaders default to business-as-usual solutions to revive their economies. But in Europe, researchers and policy analysts have recently voiced cautious optimism, as the European Union (EU) appears to be sticking by commitments outlined in its Green Deal. Devised and rolled out last year, the Euro- pean Green Deal is a policy framework to achieve net zero carbon emissions by 2050 within the bloc of 27 EU nations. The EU has pushed ahead with the launch of its Biodiversity Strategy for 2030, which will help shape how the Green Deal is imple - mented over the next decade (see bit.ly/2030 - strategy). Key commitments include increasing the amount of legally protected land, restoring degraded ecosystems, and creating Ancient woodlands, such as the conifer forest in Sweden’s Ycke Nature Reserve, are biodiversity hot spots. new green spaces in urban environments. Credit: iStock/thomasmales Currently, only 26% of the EU’s land and 11% of its marine zones are legally protected areas; both will increase to 30% under the new plans. A third of these protected zones, disproportionate onus on rural communities the new strategy: “I am optimistic it is finally including all remaining primary and old- in those EU nations. a real step with target numbers and signifi- growth forests, will be subject to even stricter Currently, the EU provides roughly 60 bil- cant funding for the protection of our biodi- protections. lion euros per year in farming subsidies versity.” To help rejuvenate freshwater ecosystems, through its Common Agricultural Policy. Much Péter Ódor, one of Báldi’s colleagues, the plans propose that at least 25,000 kilo- of the money is spent on increasing produc- believes that legally protecting Europe’s old- meters of rivers be restored to free-flowing tion. With its Biodiversity Strategy, the EU is est woodlands is a crucial step in preserving status by 2030. This restoration includes the introducing policies and funding that address European biodiversity. Undisturbed forests removal of obsolete dams and the restoration some of the negative impacts of farming on contain “dead wood that is the main micro- of floodplains and wetlands. the environment, such as clamping down on habitat for about 40% of the forest biota,” he chemical pesticides. The strategy states that said. After logging or other disturbances, “the Protected Trees, Political Realities at least 20 billion euros per year should be restoration and creation of dead wood in rel- Some ecologists and policy analysts, although provided to help with this transition, with a atively young secondary forests could not positive about the Biodiversity Strategy, still significant chunk coming from programs in substitute for these old-growth stands.” have concerns about how it will work in prac- the next long-term EU budget (2021–2027). In Europe, old-growth forests account for tice. This budget commitment has been praised roughly 0.7% of the continent’s forests, with “Targets to place more areas of land under by the World Wide Fund for Nature (WWF), the majority found in Finland, the Balkans, protection make a good headline and intu- which urged the EU to make this spending and the Carpathian Mountains. Each nation itively make sense. But our questions are, commitment more concrete and to involve cit- classifies its forests differently, and this Where will the space be found? Who might izens and nonprofit organizations to imple- inconsistency will likely intensify debate with have to be moved or displaced—if anyone—to ment the plans. “The Biodiversity Strategy the introduction of this new legislation, create them? And who will pay for it?” said spells out what WWF deeply believes in: that according to Duffy and colleagues on the Rosaleen Duffy, a political ecologist at the nature protection and restoration will only BIOSEC team. They are concerned that with- University of Sheffield in the United Kingdom lead to lasting positive change if it is planned out adequate compensation mechanisms, and principal investigator at Biodiversity and and implemented with genuine participation nations are unlikely to rush to designate for- Security (BIOSEC), a project funded by the across society,” said Irene Lucius, WWF’s ests as protected areas. European Research Council. regional conservation director for Central and Duffy is concerned that it may be cheaper Eastern Europe. and more politically acceptable to establish András Báldi at the Centre for Ecological By James Dacey (@JamesDacey), Science protected areas in Eastern Europe, placing a Research in Budapest, Hungary, welcomed Writer

4 Eos // AUGUST 2020 NEWS

Teaching Machines to Detect Climate Extremes

xtreme weather events, whether their output. Instead, researchers “train” a work is identifying the right patterns,” said scorching temperatures that ruin crops system with hundreds or thousands of solved Karthik Kashinath, a machine learning scien- Eor killer storms that drown coastal examples that the system then analyzes to tist leading the project at the National Energy towns, are likely to become more frequent create its own relevant rules. Research Scientific Computing Center in and more powerful with climate change. Deep learning was first developed in the Berkeley, Calif. That’s not to say the project Quantifying the increase in these extreme 1970s in the field of computer vision, with the is perfect—the system still yields a number events (and their economic and public health goal of training computers to “see” or recog- of false positives and false negatives—but costs) requires combing through thousands nize certain objects in digital images. the researchers hope it can be improved with of gigabytes of data that climate models gen- “In the ’80s, ’90s, and 2000s, people kept more training data. erate every day. coming up with heuristics for defining what Scientists can’t just look at the results of makes a pedestrian, what makes a car, what Generating Training Data their climate models and count hurricanes makes a face, and so on and so forth,” said Obtaining reliable training data is one of the or droughts. Instead, they are turning to Prab hat, a computer scientist who leads big- main challenges for deep learning applica- machine learning to find such extreme data initiatives at the Lawrence Berkeley tions. For extreme weather patterns, the weather events in their models’ data. National Laboratory in California. “In the last labeling has to be done by experts in the field. For decades, modelers have relied on heu- 10 years, it has been conclusively proved that “We’re going directly to the climate experts, ristics—mathematical definitions of an object AI and, in particular, deep learning tech - the meteorologists, who have been looking at of interest—to pinpoint extreme weather niques are truly well suited for solving [the these patterns for many years, so they have a events. But heuristics cannot capture the computer vision] problem. We felt that we good sense of what they look like,” Kashinath complexity and variability of weather, which could apply the same idea to finding extreme said. But labeling these images is a tedious are very difficult to condense into a set of val- weather patterns.” and time-consuming process. ues and threshold conditions. Although many To test this approach, Prabhat and his col- algorithms based on heuristics can detect leagues attempted to train a deep learning net- atmospheric rivers (air currents that trans- work to recognize and draw labels around two port water from the tropics to the poles), for types of extreme weather in high-resolution “In the last 10 years, it has instance, their results tend to be unreliable, simulation data: tropical cyclones and atmo- with large discrepancies between algorithms. spheric rivers, both of which are associated been conclusively proved Researchers hope to replace them with a new with heavy rainfall. After being trained with that AI and, in particular, generation of artificial intelligence (AI) algo- over 500 labeled examples, their system can rithms. detect most atmospheric rivers and tropical deep learning techniques cyclones in simulation data it has never seen are truly well suited for Computer Visionaries before. Their work is described in a preprint To build better algorithms, climate scientists of an article submitted to the open- access solving [the computer have turned to a particular subset of machine journal Geoscienti c Model Development (see bit learning techniques known as deep learning. . ly/ detect - extreme - weather). vision] problem.” Deep learning systems do not require a set Experienced climate scientists have been of human-defined rules and values to guide “very convinced that the deep learning net-

To speed things up, Kashinath and Prabhat, with many collaborators, created ClimateNet (see bit . ly/ climate - net), a crowdsourced extreme weather training database for deep learning networks. The goal is to share the labeling effort by opening it to other researchers and institutions around the world. The team also created an online labeling tool called ClimateContours, which allows other scientists to upload their own data sets and label them. “If a dozen institutions around the world can provide labels corresponding to maybe a dozen [extreme weather] patterns, then we can just solve these problems once and for all, and we can move on to the next set of problems,” Prabhat said. This image shows a Community Atmospheric Model (CAM) high-resolution simulation used in attributing changes “The unique aspect to ClimateNet is that in the risk of extreme weather and climate. Credit: U.S. Department of Energy instead of relying on one definition of what

SCIENCE NEWS BY AGU // Eos.org 5 NEWS

an atmospheric river or tropical cyclone is, it crowdsources the information using experts, Hackathon Participants Solve Global all with subtly different ideas on how these features should be defined in data sets,” said Problems—from Home Christine Shields, a climate scientist at the National Center for Atmospheric Research in ata journalist Winnie Kamau knew she Boulder, Colo. “All of those viewpoints are wanted to join the desert locust hack- then used as a way to train the deep learning Dathon team after a friend shared video algorithm, thus incorporating, implicitly, of the locust devastation in her childhood these different definitions.” home of Meru, Kenya. Better data sets will lead to better AI mod- “That made me think, ‘Wow, this is coming els and could unlock untapped results from closer [to] home,’” said Kamau, who now climate simulations, said Claire Monteleoni, lives in Nairobi. The practicality of working an associate professor in computer science at on a project that could directly help Kenyan the University of Colorado Boulder who has farmers—and others—made her more curi- applied AI to hurricane track prediction but ous about the opportunity. According to the wasn’t involved in this project. “Much of the Food and Agriculture Organization of the data output by climate models hasn’t been United Nations, East Africa (especially Ethi- analyzed, so AI is really the cheapest way, opia, Kenya, and Somalia) is currently facing without having to run new climate model an unprecedented threat to its food security The nations of East Africa are enduring an invasion simulations, to gain more insights to help because of locust swarms. of desert locusts, which are threatening food secu- with predicting climate and extreme weather Coming up with ways to help farmers rity and economic activity in areas like Samburu events.” during locust swarms was one of the chal- National Reserve in Kenya. Credit: iStock.com/Jenni- lenges for this year’s INSPIRE Hackathon, an fer Watson Enabling High-Precision Analysis annual competition dedicated to solving The most basic thing researchers will be able problems around the globe with open data, to do with a successfully trained deep learn- volunteered geographic information, and cit- ing network will be to count how many izen observations. event, the virtual format “[gave] us more extreme events a model calculates will hap- Kamau’s team, which won the event, used time and space to be able to be more cre - pen in the future and compare the result with imagery from the European Space Agency’s ative,” Kamau said. present-day numbers. Even so, Prabhat said, Sentinel-1 and -2 satellites to map and study Other hackathon attendees ranged from that would barely scratch the surface of what the effects of desert locust swarms. students to big-data experts. Each team was these systems will allow. “We know that there’s a global pandemic required to make use of freely available Earth Researchers can look at underlying data that has taken over the world right now, and observation data to approach challenges like for each event in the model data and extract we are all fighting an unseen enemy. But in maintaining food security, balancing open information such as how much precipita- East Africa and the Horn of Africa, we are also land use and transportation infrastructure, tion is expected to fall in a specific area. This fighting an enemy that we can see,” Kamau anticipating ways climate change might information can be used in many ways, said during a presentation on 6 May. affect agriculture, and helping farmers plan including estimating insurance costs in high- for and mitigate devastating locust swarms. risk areas and managing future drought. In Virtual Hackathon Each team was provided a mentor, an expert places like California, which depend on rain for a Sustainable Africa to consult on whatever issue they were trying delivered by atmospheric rivers, authorities In 2019, the INSPIRE Hackathon was held in to solve. could be alerted that they should make con- Nairobi with more than 200 participants. Although the hackathon is over now, tingency plans if those rivers are going to be This year’s hackathon was supposed to have Kamau and her teammates still meet regu- diverted, Prabhat said. been hosted in Kampala, Uganda, but with larly. They’re even publishing their report “It’s not that you could have never answered recent lockdown orders associated with the with the Global Open Data for Agriculture and these questions” without resorting to AI, Pra- coronavirus-2019 (COVID-19) pandemic, an Nutrition initiative, an INSPIRE Hackathon bhat explained, but the new approach will in-person hackathon just wasn’t possible. partner that supports open-d ata efforts likely make it easier and faster—and possibly However, moving the event online didn’t around the globe (www.godan.info/). better than with heuristics. seem to deter participants, said hackathon Both Kamau and Lilja Bye stressed the “We are able to do precision analytics and organizer Bente Lilja Bye. Frequent webinars importance of a hackathon-style event in get these very high quality results,” he said, helped participants get to know each other bringing people together to solve problems. which heightens the credibility of climate and share ideas. Webinars also helped the “I think the common denominator is, you models’ results and the plans that rely on hackathon recruit even more participants, have a concentrated work on something, just them for preparing for the extreme weather with some participants joining multiple to solve something, identify a problem and events of the future. teams, Lilja Bye said. solve something,” Lilja Bye said. Kamau actually preferred the virtual format. Instead of dealing with the hassle of air- By Javier Barbuzano (@javibarbuzano), Science ports, hotel reservations, and jet lag, not to By JoAnna Wendel (@JoAnnaScience), Science

Writer mention the limited time of an in-person Writer iStock.com/robertiez

6 Eos // AUGUST 2020 NEWS

Machine Learning Can Help Decode Alien Skies— Up to a Point

uture telescopes like the James Webb Space Telescope (JWST) and the Atmo- Fspheric Remote- sensing Infrared Exo- planet Large-survey (ARIEL) are designed to sample the chemistry of exoplanet atmospheres. Ten years from now, spectra of alien skies will be coming in by the hundreds, and the data will be of a higher quality than is currently possible. Astronomers agree that new analysis techniques, including machine learning algorithms, will be needed to keep up with the flow of data and have been testing options in advance. A study that has been accepted by Monthly Notices of the Royal Astronomical Society trialed one such algorithm against the current gold standard method for decoding exoplanet atmospheres to see whether the algorithm could tackle this future big-data problem (bit .ly/machine-learning-exoplanets). “We got really good agreement between [the answers from] our machine learning method and the traditional Bayesian method that most people are using,” said Matthew Nixon. Nixon is the lead researcher on the project and an astronomy doctoral student at the University of Cambridge in the United College London in the United Kingdom who The algorithm compares each artificial spec- Kingdom. was not involved with this study. “Perhaps trum with the real one and chooses the clos- However, he said, “as we increased the unsurprisingly, the more detailed the model, est match. This is a new and modified appli- parameter space, the computational effi- the longer it takes to compute its results. cation of the random forest algorithm for ciency of our method dropped…. As we Today we are rapidly reaching a stage where exoplanetary atmospheres that was originally started to add more parameters, we started to our traditional techniques become too slow to developed by astronomers at the University get hit by the curse of dimensionality.” compute these increasingly complex models.” of Bern in Switzerland. The researchers tested their algorithm on A Random Forest two exoplanets with exceptionally well stud- Breathing in Exotic Air ied atmospheres and found that the random Astronomers measure the spectrum of an forest’s solution matched the one from atmo- exoplanet’s atmosphere when starlight “As we started to add more spheric retrieval. Moreover, “the authors shines through it or when heat from the parameters, we started achieve a much faster interpretation of the planet lights up the atmosphere from within. data than otherwise possible with traditional In either scenario, the atmosphere imprints to get hit by the curse techniques,” Waldmann said. its chemical signature on the light and is of dimensionality.” However, the two exoplanets in question, detected by our telescopes. WASP-12b and HD 209458b, are both very hot The current front-runner for best deci- Jupiter-sized planets. The algorithm could phering a planet’s spectrum is called atmo- easily simplify its decision because each plan- spheric retrieval. It uses statistical inference et’s atmosphere consists mostly of hydrogen to calculate the likelihood that given an Nixon and his adviser and coauthor, Nikku and helium, Nixon said. observed spectrum, an exoplanet’s atmo- Madhusudhan, also at the University of Cam- “Generally speaking,” Madhusudhan sphere has a certain composition, tempera- bridge, tested a type of supervised machine explained, “it is going to be slightly harder to ture, level of cloud cover, and heat flow. The learning algorithm called a random forest, retrieve atmospheric properties of cooler and technique has so far proven very reliable but which is made up of thousands of decision smaller planets,” for example, super- Earths can be computationally expensive. trees. Each decision tree makes its prediction or Earths. “This is because the spectral sig- “The more detailed the data, the more for a likely combination of atmospheric prop- natures are expected to be smaller for such detailed the model needs to be,” said Ingo erties, and then the algorithm generates an planets, which makes it harder to extract the

iStock.com/robertiez Wald mann, an astrophysicist at University artificial spectrum that has those properties. same amount of information as we have for

SCIENCE NEWS BY AGU // Eos.org 7 NEWS

hot Jupiters currently.” For planets with faint signals and those whose base constituents are Search for MH370 Revealed unknown—an ocean world, super-Earth, or temperate- z one planet—the random forest Ocean Crust Waves would lose its computational edge.

A Balanced Approach for the Way Forward This study adds to a growing effort by exoplanet scientists to find an efficient way to handle the upcoming deluge of atmospheric data. “It is great to see a growing group in the community using machine learning methods and cross- checking each other’s results and claims,” said Daniel Angerhausen, an astrophysicist at ETH Zürich in Switzerland who was not involved with this research. Missions like JWST and ARIEL are first at bat, but Angerhausen is also thinking about missions that will come after those. Astron- omers will need to strategize the most efficient ways to observe interesting targets. “This problem is predestined for a [machine learning] approach,” Angerhausen said. A random forest approach is just the “tip of the iceberg” for algorithms to try. Nixon agreed and said that “going forward, looking at different machine learning algorithms is definitely a positive [step], and also looking at how we can combine these machine learning approaches into hybrid The M/V Fugro Equator, seen here, was one of three ships that collected sonar data from June 2014 to June methods to really boost these retrievals to the 2016 while searching for Malaysia Airlines Flight 370 (MH370). Those data led to a new high-resolution bathyme- next level.” try map of an area near the Southeast Indian Ridge. Credit: Justin Baulch, Australian Transport Safety Bureau, As exoplanet atmosphere research moves CC BY 4.0 (bit. ly/ ccby4- 0) into the big-data era, machine learning will become an increasingly important research tool scientists should be trained to use, Mad- husudhan said. Some graduate programs are alaysia Airlines Flight 370 (MH370) processes that control ocean crust formation already integrating more data science learn- disappeared on 8 March 2014 some- over millions of years. ing into students’ training. (Nixon’s doctor- Mwhere in the southern Indian Ocean. ate work is supported by one such program in The tragic loss of all 239 people aboard led Waves on the Seaﬂoor the United Kingdom.) to a multinational, multiyear search to find Most of the ocean floor is only very roughly “On the other hand,” Madhusudhan added, and recover the plane. Despite this effort, it mapped. “Earth is about 70% ocean floor,” “it also needs to be recognized that while remains missing. said Joyce Sim, “but we really don’t know the machine learning is a great research tool in As part of that search effort, three survey bathymetry of the ocean floor, and it’s really various areas, there are also important areas vessels collected multibeam bathymetry data hard to have this sort of high-resolution data of research where other numerical, statisti- using hull-mounted sonars. These data were on all of it.” Sim is a coauthor on the study cal, and analytic approaches are more suitable used to create maps of the seafloor that span and a geodynamics postdoctoral researcher at for some important problems. Therefore, I roughly 88,000 square kilometers near the the Earth and Planets Laboratory at the Car- believe the right balance needs to be met Southeast Indian Ridge west of Australia. negie Institution for Science in Washington, while integrating machine learning into grad- The bathymetry, at a resolution 15 times D.C. “It takes quite a while just to cover a uate programs in the right research areas.” sharper than previous maps of the area, did pretty small patch of the ocean” with map- “Machine learning may never replace an not reveal MH370’s final resting place. How- ping vessels and costs millions of dollars per atmospheric expert,” Waldmann said, “but ever, a recent analysis of those seafloor maps expedition. I’m certain that artificial intelligence will cer- showed that the ocean crust surrounding the Moreover, the seafloor accumulates layers tainly play a role as a helping hand.” ridge grew at a rate that ebbed and flowed on a of sediment that can prevent sonar from see- timescale of hundreds of thousands of years. ing the ocean crust in detail. The area that was The results, published in Geophysical Research searched for MH370, however, has a thinner- By Kimberly M. S. Cartier (@AstroKimCartier), Letters (see bit.ly/oceanic-crustal-formation), than-average sediment buildup that makes Sta Writer will help scientists better understand the the crust underneath visible to sonar.

8 Eos // AUGUST 2020 NEWS

The scans for MH370 revealed that the ocean crust rises and falls in waves that start at the Southeast Indian Ridge and continue outward.

The scans for MH370 revealed that the ocean crust rises and falls in waves that start at the Southeast Indian Ridge and continue outward. The ridge spreads about 35 milli- meters in opposite directions every year, and Two sections of the bathymetry map that covers part of the MH370 search area. Shown are variations in ocean the wave crests, where crust grew faster, are crust depth around 3.5 kilometers below sea level (left), and crust about 4 kilometers below sea level (right). In more than 100 kilometers long and repeat both maps, the seaﬂoor is spreading along the solid black line from bottom left to top right. Small grooves are every 10–14 kilometers. The researchers esti- abyssal hills. The more widely spaced ridges are evidence that porosity waves helped form the ocean crust. mated when each wave formed by comparing Credit: Parnell- Turner et al., 2020, https:// doi . org/ 10 .1029/ 2020GL087349 the bathymetry maps to independent measurements and models of the crust’s magnetic field. The ocean crust is like “a magnetic tape But for the Southeast Indian Ridge, at least, you see are melt- rich pockets forming in recorder,” said lead author Ross Parnell- that appears not to be the case. “We have a waves and moving toward the ridge axis Turner, a marine geophysicist at Scripps very long record of crustal accretion over a where the oceanic crust is created.” Institution of Oceanography in La Jolla, Calif. very large area,” Parnell-Turner said. “We did “The authors of this paper make a convinc- Crustal rocks record the polarity of Earth’s not find any evidence for cycles of crustal ing case for a longer-s cale structure…and magnetic field as they form, allowing scien- accretion on those Milankovitch timescales.” plausibly attribute such features to mantle tists to trace the age of different sections of processes,” said John A. Goff, a senior ocean crust. research scientist in seafloor mapping at the “You might expect that oceanic crust is University of Texas at Austin. “What really formed in this constant process, so you might makes that possible is the exceptionally long have a uniform thickness, for example, of “These kinds of studies record in terms of crustal ages from the oceanic crust being formed,” he said. “What MH370 search.” Goff was not involved with we found, actually, is that the pace at which drive home the point that this research. new crust is formed varies on this…charac- This bathymetry map, although a signifi- teristic times cale of 300,000–400,000 years.” exploration- driven science cant leap forward in resolution, can’t tell the The wavy pattern repeated for 12 million is as valuable as team for sure whether porosity waves, glacial years. cycles, or something else like mantle compo- It’s not so strange that ocean crust topog- hypothesis- driven science.” sition is truly behind ocean crust waves. The raphy has some periodicity, Sim explained. It fact is, it’s just one location where the sea- would be much odder for the crust to form floor is spreading. perfectly flat for millions of years. “We don’t have very good data in a lot of What is odd about these waves, however, is the deep ocean,” Parnell-Turner said. “Col- that the characteristic time scale doesn’t Melted Mantle Through Thick and Thin lecting more bathymetry data like this in match up with glacial cycles that are governed Seeking an alternate explanation, the areas where we have the crust exposed would by changes in Earth’s orbit around the Sun researchers tested whether the periodicity be helpful” in figuring out the extent to and its axis of rotation. So-called Milanko- could be caused by variations in how melted which each process might regulate crust for- vitch cycles regulate ice coverage and sea level the magma is when it comes up through the mation. on timescales of 23,000, 41,000, and 100,000 seafloor. Using simplified computer simula- “These kinds of studies drive home the years. tions, the team found that the phenomenon, point that exploration- driven science is as It’s been fiercely debated in recent years, called porosity waves, created variations in valuable as hypothesis-driven science,” Goff Parnell- Turner explained, whether the wax- ocean crust topography at timescales similar said. ing and waning of sea level modulate how to those seen in the bathymetry. quickly new crust forms at seafloor ridges. “As the mantle upwells, it starts to melt Data from oceanic crust not far from the due to decompression melting” and becomes By Kimberly M. S. Cartier (@AstroKimCartier), MH370 search site support this theory. a fluid-l ike magma, Sim explained. “What Sta Writer

SCIENCE NEWS BY AGU // Eos.org 9 NEWS

The Future of Big Data May Lie in Tiny Magnets

he use of artificial intelligence and machine learning is now widespread, Twith practical applications in many Earth science domains, including climate modeling, weather prediction, and volcanic eruption forecasting. This revolution in computing has been driven largely by rapid improvements in software and algorithms. But now we’re approaching a second computing revolution of redesigning our hardware to meet new challenges, said Jean Anne Incorvia, a professor of electrical and computer engineering at the University of Texas at Austin. Traditional silicon-based computer hardware (like the computer chips found in your laptop or cell phone) has bottlenecks in both speed and energy efficiency, which may impose limits on its use in increasingly intensive computational problems. An artificial neuron, artistically rendered here, that can mimic properties of a biological neuron could accelerate To break these limitations, some engineers developments in big-data computing. Credit: iStock.com/ktsimage are drawing inspiration from biological neural systems using an approach called neuro- morphic computing. “We know that the brain is really energy efficient at doing things like “Now we’re approaching nal contrast in human sensory systems, like recognizing images,” said Incorvia. This is vision and touch, by having more active neu- because the brain, unlike traditional comput- a second computing rons inhibit the activity of the surrounding ers, processes information in parallel, with revolution of redesigning neurons. neurons (the brain’s computational units) Implementing lateral inhibition in artifi- interacting with one another. our hardware to meet cial neural networks could improve certain Incorvia and her research team are now new challenges.” computer algorithms. “It’s preventing errors studying how magnetic—not silicon—com- in processing the data so you need to do puter components can mimic certain useful less training to tell your computer when it aspects of biological neural systems. In a new did things wrong,” said Incorvia. “So that’s where a lot of the energy benefits come study published in the journal Nanotechnology from.” (bit . ly/ neural - systems), her team reports Lateral inhibition can be implemented with that physically tuning the magnetic interac- conventional silicon computer parts, but at a tions between the magnetic nanowires could cost. significantly cut the energy costs of training “There is peripheral circuitry that is computer algorithms used in a variety of needed to implement this functionality,” said applications. Can Cui, an electrical and computer engineer- “The vision is [creating] computers that ing graduate student at the University of can react to their environment and process Texas at Austin and lead author of the study. a lot of data at once in a smart and adaptive “The end result is that as the size of the net- way,” said Incorvia, who was the senior work scales up, there’s much more difficulty author on the study. Reducing the costs of in the design of the circuit and you will have training these systems could help make that extra power consumption.” vision a reality. Researchers in the new study are using magnetic nanowires and their innate physical In this diagram of an artificial neural network imple- Lateral Inhibition at a Lower Cost properties to more efficiently produce lateral menting a “winner-take-all” (WTA) machine learning Artificial neural networks are one of the inhibition. These devices create small, stray algorithm, neurons (blue) are the computation units main computing tools used in machine magnetic fields, which are typically a nuisance receiving data input (red) with synapses that inter- learning. As the name implies, these tools in designing computer parts because they can connect them. The final computational output neu- mimic biological neural networks. Lateral disturb nearby circuits. But here, “this kind of rons (green) are inhibited by a lateral inhibition sig- inhibition is an important feature of biolog- interaction might be another degree of free- nal (purple). Credit: University of Texas at Austin ical neural networks that helps improve sig- dom that we can exploit,” Cui said.

10 Eos // AUGUST 2020 NEWS

By running computer simulations, the researchers found that when these nanowires Long Live the Laurentian Great Lakes are placed in parallel and next to one another, the magnetic fields they produce can inhibit rom some vantage points, the Great one another when activated. That is, these Lakes feel more like vast inland seas LIVING IN GEOLOGIC TIME magnetic parts can produce lateral inhibition Fthan freshwater lakes. But the 6 qua- without additional circuitry. drillion gallons (~23 quadrillion liters) slosh- ing in Superior, Michigan, Huron, Ontario, and Erie represent one fifth of the planet’s fresh water. Evidence of the rift is visible at Lake Supe- Researchers found that How long will the Great Lakes retain their rior, which is ringed by cliffs of billion-year- honor of being the largest collection of fresh- old basalt that erupted into the active rift, when nano wires are placed water lakes on the planet? Thanks to some giving the largest lake a uniquely rugged in parallel and next to one ancient rifting events, probably for a very shoreline. “Lake Superior looks very different long time to come. from the rest of the Midwest,” where volcanic another, they can produce rocks are rare, Stein said. Ancient Rifts and Ice Age Lakes Lake Superior sits within the Midcontinent lateral inhibition without The story of the Great Lakes began over 1 bil- Rift scar, but all five Great Lake basins were additional circuitry. lion years ago, when the ancient superconti- carved out by glaciers during the last glacial nent Laurentia began splitting in half. Over period, when the region was buried under the the course of about 10 million years, the Mid- Laurentide Ice Sheet. As the ice age came to a continent Rift System opened a massive fis- close, the glaciers retreated, and the lake sure on its way to becoming a new ocean basins filled with meltwater to form the five The researchers ran further computer basin. But for reasons geologists don’t entirely Great Lakes as we know them by around modeling experiments to understand the understand, the rift failed and no ocean was 10,000 years ago. Today some scientists refer physics of their devices. They found that formed, leaving a 3,000- kilometer- long scar to the Great Lakes by their more formal name, using smaller nanowires (30 nanometers across what is now North America. the Laurentian Great Lakes. wide) and tuning the spacing between pairs Eons of erosion have hidden this scar, A few summers ago, I drove hundreds of of nanowires were key to enhancing the which runs in two forks from Lake Superior kilometers out of my way to visit Superior’s strength of lateral inhibition. By optimizing down to Alabama and Oklahoma. “A billion billion- year- old rocks on a backpacking trip these physical parameters, “we can achieve years is a very long time, even by geologic through Porcupine Mountains Wilderness really large lateral inhibition,” said Cui. “The standards,” said Seth Stein, a geophysicist at State Park in Michigan, on the south shore of highest we got is around 90%.” Northwestern University in Evanston, Ill. Lake Superior. Unfortunately, the pilgrimage

The Challenge of Real-World Applications Training machine learning algorithms is a large time and energy sink, but building computer hardware with these magnetic parts could mitigate those costs. The researchers have already built prototypes of these devices. However, there are still open questions about the practical applications of this technology to real-world computing challenges. “While the study is really good and combines the hardware-s oftware functionality very well, it remains to be seen whether it is scalable to more large scale problems,” said Pri- yadarshini Panda, a professor of electrical engineering at Yale University who was not involved in the study. But the study is still a promising next step and shows “a very good way of matching what the algorithm requires from the physics of the device, which is something that we need to do as hardware engineers,” Panda added.

By Richard J. Sima (@richardsima), Science Lake Superior’s towering basalt cli¤s stand in contrast to the sandy beaches that ring the other Great Lakes. Writer Credit: Seth and Carol Stein

SCIENCE NEWS BY AGU // Eos.org 11 NEWS

feeding off nutrients delivered by the Mau- mee River create water quality issues each summer. But larger lakes, like larger boats, cannot change course as quickly. It may take decades before we know how the larger lakes will respond to climatic changes, Robertson said. Currently, around 35 million people rely on the Great Lakes for drinking water, and an additional 56 billion gallons are extracted each day for municipal, agricultural, and industrial use. Those staggering numbers are likely to increase, Rood said, as people displaced from other regions of the country by climate- related impacts seek refuge in the Great Lakes Megalopolis. “Compared to some places, like south Florida, the Great Lakes are going to look like a climate winner,” Rood said. “We need to be Holland Beach, on Lake Michigan is one of many sandy swimming beaches around the Great Lakes. Credit: prepared for how those migrations will Mary Caperton Morton change our communities and water usage in the coming decades.”

Long Live the Great Lakes was rushed. The moment we parked at the started rising because of heavy rainfall and The long-term trajectory of large freshwater trailhead, clouds of mosquitoes swarmed the flooding across the Midwest. lakes often depends on their outlet: Endor- car. It was mid- June, and the bloodsuckers By 2019, lake levels had drastically reversed heic, or terminal, lakes such as the Great Salt were so relentless—biting through bug spray, course, exceeding their highest recorded lev- Lake have no exit to another body of water, layers of clothes, a smoky fire, and a tent— els. High water in the Great Lakes leads to and the majority of their water balance is lost that we cut our 4-day loop in half. Next time shoreline flooding and erosion, which in turn through evaporation, making them saltier I visit, I’ll avoid peak mosquito season. threaten homes, industrial buildings, and over time. The Great Lakes offer a lifetime’s worth of port infrastructure. The Great Lakes are not an endorheic basin; hiking opportunities: I have my eye on the “Now the narrative is that we will need to the outlet that connects them to the Atlantic Superior Hiking Trail, a 499-kilometer-long, be able to cope with both highs and lows in Ocean has a 535-million-year-old history 2- to 4-week backpacking trip along the lake levels,” with those extremes persisting that’s unlikely to peter out anytime soon. The northwestern shore of Lake Superior in Min- for longer time periods, perhaps several sea- Saint Lawrence Seaway, along with the basins nesota. Isle Royale, a large island in Lake sons in a row, Rood said. that hold Lakes Erie and Ontario, sits inside Superior, “offers arguably the best backpack- Higher influxes of water into the Great the Saint Lawrence Rift, a 1,000-kilometer- ing in the Midwest,” Stein said. Lakes also bring more pollutants, including long scar that dates back to the opening of the nitrogen and phosphorus from agricultural Iapetus Ocean between the paleocontinents More Water, More Problems runoff and Escherichia coli bacteria from over- of Laurentia, Baltica, and Avalonia. The Great Lakes have been a nexus of migra- worked water treatment systems. Algal Unlike the long dead and buried Midconti- tion, transportation, fishing, and trade for blooms and toxic water quality issues have nent Rift System, the Saint Lawrence Rift thousands of years, with a series of lakes, riv- increased. System is still seismically active, capable of ers, and waterways connecting the upper The sheer size of the individual Great Lakes generating earthquakes with magnitudes 5 or Midwest to the Atlantic Ocean. Records that means that pollutants can stay in the system greater. As long as water flows into the Great began in 1918 show lake levels fluctuating for a long time: A water droplet or molecule Lakes Basin and out through the Saint Law- seasonally, with the annual water budget of pollutant will reside in Lake Superior for rence Seaway, the Great Lakes are likely to controlled by inputs from rivers and outputs as long as 191 years, in Lake Michigan for retain their status as the world’s largest—and to the Atlantic Ocean and by evaporation. 99 years, and in Lake Huron for 22 years, hopefully freshest—lakes for eons to come. In 2014, after years of drought and higher whereas the smaller Lakes Ontario and Erie temperatures triggered increased evapora- have residence times of 6 and 2.6 years, tion, water levels in the Great Lakes reached respectively. By Mary Caperton Morton (@theblondecoyote), all-time lows. “For a long time, the dominant “Smaller lakes with shorter residence Science Writer climate change narrative was that lake levels times can respond more quickly to changes were going to drop drastically,” said Richard than bigger lakes,” said Dale Robertson, a  Living in Geologic Time is a series of per- Rood, a climate scientist at the University research hydrologist at the U.S. Geological sonal accounts that highlight the past, present, of Michigan in Ann Arbor. “That narrative Survey in Middleton, Wis. That means there’s and future of famous landmarks on geologic ended abruptly in 2015,” when lake levels hope for Lake Erie, where toxic algal blooms timescales.

12 Eos // AUGUST 2020 NEWS

Are Geysers a Signal of Magma Intrusion Beneath Yellowstone?

Wicks suggests a connection between the cause of the deformation (magmatic volatiles) and Steamboat Geyser.

Small Sample Size Mike Poland, scientist-in-charge of the USGS Yellowstone Volcano Observatory who was not involved in the study, thinks that the connection is more complicated: “Why would just a single geyser respond to processes beneath Norris [Geyser Basin]? Wouldn’t that impact all the geysers?” One explanation might be that Steamboat Geyser likely has a more complex and deeper system than many geysers at Yellowstone. Steamboat Geyser erupts on 4 June 2018. Credit: Jamie Farrell, USGS Wicks admitted that so far, scientists have a limited data set for Steamboat Geyser. “With only two episodes of observed deformation near Norris, the eruption frequency teamboat Geyser in Yellowstone Modeling a Deformation appears to respond to the surface deforma- National Park is an enigma. It is the A new study by Wicks and colleagues in the tion. Of course, that’s a very small sample Stallest currently active geyser in the Journal of Geophysical Research: Solid Earth size, so the apparent connection might be world, sometimes blasting superheated examines this restless behavior to try to happenstance,” he said. Any connection water over 90 meters (300 feet) into the air. understand what drives the deformation and between Steamboat Geyser’s activity and the Yet unlike the more famous Old Faithful, how it might be linked to Steamboat Geyser. accumulation of gases near the surface could Steamboat Geyser runs on its own rhythm. (see bit.ly/magma-intrusion). It turns out heighten the risk of hydrothermal explo- Sometimes the geyser is quiet for decades that the rising and falling of the land, the sions, one of the most significant geologic and then suddenly bursts back to life. It is a behavior of some geysers, and the introduc- hazards for tourists visiting the national park. mystery exactly why Steamboat has such tion of new magma underneath Yellowstone Wicks suggested that the ability to have behavior. After a new period of activity that could all be linked. constant geochemical monitoring of the started in 2018, we might have more clues The north rim area and caldera floor at hydrothermal fluids from geysers and hot about what drives these steam-and-water Yellowstone have been deforming for springs would help scientists create more explosions. decades. The ability to monitor this deforma- accurate models: “Continuous monitoring of Yellowstone caldera is a geologic wonder- tion has greatly improved with GPS and inter- volatile flux rates is really needed to put the land. It is the source of three of the largest ferometric synthetic aperture radar tech- geophysical models in context. If surface explosive eruptions in the past 3 million niques. Wicks and others looked at the North inflation is caused by accumulation of pres- years. The caldera itself covers over 1,500 Rim Uplift Anomaly (NRUA) and used defor- surized fluids and deflation occurs when square kilometers (580 square miles) in the mation data from these techniques to model those fluids escape to the surface, we might northwestern corner of Wyoming. As Charles potential sources of these changes. be able to detect changes in volatile flux over Wicks from the U.S. Geological Survey (USGS) Magma, volatiles released from magma, or time.” put it, “Yellowstone’s roots seem to extend hydrothermal fluids could all cause the sur- The ability to link deeper processes like all the way to the core-mantle boundary. In face to rise or fall. Wicks and others used data magmatic intrusion with surface phenomena that dimension, it’s a magmatic system of from 2000 to 2018 to examine how much the is tantalizing for geoscientists monitoring an continental scale.” NRUA either inflated or deflated. What they active caldera like Yellowstone. Wicks and Yellowstone caldera is packed with geo- found is that the deformation in the NRUA Poland both agreed that understanding the thermal features like geysers, hot springs, may represent the accumulation of mag- connection between the magmatic and mud pots, and geothermal pools. This hydro- matic volatiles from a basaltic intrusion that hydrothermal systems at Yellowstone is fun- thermal activity is driven by the vast reser- occurred around 2000. damental for better hazard models. They voirs of heat beneath the Yellowstone area, Steamboat Geyser might be a pressure might end up being more directly linked than most of which comes from the magma found gauge for these intrusions and accumulations geoscientists have previously thought. many kilometers underneath the caldera. All of volatiles. The geyser is in the midst of one of this heat and water cause the land surface of its most active periods on record, having at Yellowstone to rise and fall frequently, erupted 79 times between 2018 and 1 July By Erik Klemetti (@eruptionsblog), Science meaning that Yellowstone is best described 2020. This activity appears to correlate with Writer and Associate Professor of Geosciences, as a “restless caldera.” the deformation cycle in the NRUA, which to Denison University, Granville, Ohio

SCIENCE NEWS BY AGU // Eos.org 13 NEWS

How Machine Learning Redraws the Map of Ocean Ecosystems

n land, it’s easy for us to see divisions between ecosystems: A rain forest’s Ofan palms and vines stand in stark relief to the cacti of a high desert. Without detailed data or scientific measurements, we can tell a distinct difference in the ecosystems’ flora and fauna. But how do scientists draw those divisions in the ocean? A new paper has proposed a tool to redraw the lines that define an ocean’s ecosystems, lines originally penned by the seagoing oceanographer Alan Longhurst in the 1990s. The paper uses unsupervised learning, a machine learning method, to ana- lyze the complex interplay between plankton species and nutrient fluxes. As a result, the tool could give researchers a more flexible definition of ecosystem regions. Using the tool on global modeling output suggests that the ocean’s surface has more than 100 different regions or as few as 12 if aggregated, simplifying the 56 Longhurst regions. The research could complement ongoing efforts to improve fisheries management and satellite detection of shifting plankton under climate change. It could also telling the computer what a square and a cir- of the ocean” is characterized by very small direct researchers to more precise locations cle look like before it began. But in unsuper- cells. “There’s just not a lot of plankton bio- for field sampling. vised learning, the algorithm has no prior mass.” The region that includes Peru’s fertile knowledge of shapes and will sift through coast, however, has “a huge amount of stuff.” Beyond the Human Eye many images to identify patterns of similar If scientists want more distinctions Coccolithophores, diatoms, zooplankton, and shapes itself. between communities, they can adjust the other planktonic life-forms float on much of Using an unsupervised approach gives tool to see the full 115 regions. But having only the ocean’s sunlit surface. Scientists monitor SAGE the freedom to let patterns emerge that 12 regions can be powerful too, said Son - plankton with long-term sampling stations the scientists might not otherwise see. newald, because it “demonstrates the simi- and peer at their colors by satellite from “While my human eyes can’t see these dif- larities between the different [ocean] basins.” above, but they don’t have detailed maps of ferent regions that stand out, the machine The tool was published in the journal Science where plankton live worldwide. can,” first author and physical oceanogra- Advances (see bit.ly/sage-method). Models help fill the gaps in scientists’ pher Maike Sonnewald at Princeton Univer- Oceanographer Francois Ribalet at the Uni- knowledge, and the latest research relies on sity said. “And that’s where the power of this versity of Washington, who was not involved an ocean model to simulate where 51 types of method comes in.” This method could be in the study, hopes to apply the tool to field plankton amass on the ocean surfaces world- used more broadly by geoscientists in other data when he takes measurements on wide. The latest research then applies the fields to make sense of nonlinear data, said research cruises. He said that identifying new classification tool, called the systematic Sonnewald. unique provinces gives scientists a hint of aggregated ecoprovince (SAGE) method, to how ecosystems could react to changing discern where neighborhoods of plankton and Desert of the Ocean ocean conditions. nutrients appear. Applied to model data, SAGE noted 115 dis- “If we identify that an organism is very The SAGE method relies, in part, on a type tinct ecological provinces, which can then be sensitive to temperature, then we can start to of machine learning algorithm called unsu- boiled down into 12 overarching regions. actually make some predictions,” Ribalet pervised learning. The algorithm’s strength One region appears in the center of nutrient- said. Using the tool will help him tease out an is that it searches for patterns unprompted poor ocean gyres, whereas other regions show ecosystem’s key drivers and how it may react by researchers. productive ecosystems along coasts and the to future ocean warming. To compare the tool to a simple example, if equator. scientists told an algorithm to identify shapes “You have regions that are kind of like the in photographs like circles and squares, the regions you’d see on land,” Sonnewald said. By Jenessa Duncombe (@jrdscience), Staff

researchers could “supervise” the process by One area in “the heart of a desert-like region Writer Opposite: Unsplash/Marek Okon

14 Eos // AUGUST 2020 OPINION

Weathering Environmental Change Through Advances in AI

he world’s need for accurate, usable predictions of atmospheric and oceanic Tphenomena—and their impacts—has never been greater. Climate change is affecting weather patterns and sea levels are rising, putting more and more people at risk from environmental disasters. At the same time, the quantity and quality of modeling systems and Earth observations—from satellite platforms, cell phones, connected vehicles, and many other sources—are increasing dramat- ically, offering a deluge of data so rich that only automated intelligent systems can fully exploit them. Improving society’s long-term sustainability with respect to limited environmental resources while ensuring health and safety is a “wicked problem,” a term coined to describe complex real-world challenges that have severe societal consequences, cannot be studied by trial and error, and do not have clear- cut solutions. Addressing this wicked problem effectively will require new artificial intelligence (AI) methods for knowledge discovery and prediction driven by convergent Convergent research could lead to the development of trustworthy artiﬁcial intelligence methods based on envi- collaborations across multiple sectors ronmental data that help individuals, o¥cials, and others make quick decisions during high-impact events like involved in research and enterprise. tornadoes. Credit: sshepard/E+/Getty Images

Artiﬁcial intelligence is uals, businesses, and governments in making Broadly speaking, academia provides rich better decisions to reduce harm to people, expertise and research programs that with poised to help us harness livelihoods, ecosystems, and economies. funding from government or industry can AI has demonstrated value throughout the sustain their focus on a problem for many the power of the big-data observing, predicting, and impact modeling years. Academia also trains new generations frontier by exploiting process, from preprocessing to postprocess- of scientists, engineers, and policy makers. ing and improving forecasts. And with recent Both industry and government benefit when untapped environmental computational hardware innovations like students develop appropriate interdisciplin- observations and graphics processing units (GPUs) and tensor ary knowledge and skills and gain experience processing units, as well as innovations in AI with hands-on problem-solving. enhancing Earth system methods themselves, AI is poised to help us To achieve environmental and public modeling in cost-e¦ective harness the power of the big-data frontier by health goals, governments develop and exploiting untapped environmental observa- enforce laws, policies, and regulations ways. tions and enhancing Earth system modeling informed by environmental sciences. They in cost- effective ways. provide long-term funding for research and Developing AI approaches that revolution- development in science deemed essential to ize our understanding of and ability to predict the public interest, and coordination at a scale Collaborating to Tap AI’s Potential a wide range of environmental phenomena otherwise unachievable. Governments are Currently, less than 5% of available environ- accurately, and that address society’s needs uniquely able to deploy large-scale, long- mental observational data are used in numer- in the process, calls for a convergent effort term operational environmental sensing sys- ical models of Earth systems. Better models across research and operational sectors. Each tems such as radar, satellite, and buoy net- that leverage additional data are essential to sector provides unique expertise, resources, works. Within the U.S. weather enterprise, improving our understanding of and ability to and viewpoints that when deeply collaborat- the federal government also employs and predict atmospheric and oceanic phenomena ing, can result in synergistic and substantial trains forecasters and has the primary

Opposite: Unsplash/Marek Okon and their impacts, as well as to assist individ- progress. responsibility to issue watches and warnings.

SCIENCE NEWS BY AGU // Eos.org 15 OPINION

The private sector has a wide reach and a fundamental tools and techniques. Academ- profit motive to rapidly develop and imple- ics have developed tools such as explainable ment new technology that can help people, AI, which enables researchers to dive into the organizations, and governments make better black box of machine learning methods and decisions. The private sector can often move highlight what the models have learned, as more quickly and with more flexibility than well as interpretation and visualization tech- other sectors, which can lead to faster adap- niques. But these tools have most often been tation. However, this agility is often driven by tested on novices or on high-level users, such quarterly financial reports, which can make as other researchers, rather than on decision- it difficult to devote internal resources to making end users in government and indus- lengthy research projects that may take years try. Critical end user feedback should help to mature. Partnering with academia and clarify design choices and priorities for future government allows industry to identify prob- explainable AI systems. lems of high practical value, to fund and guide Developing AI tools that earn the trust of external research and development, and to all types of end users requires fundamental adopt resulting products that prove valuable research on their different needs and percep- and widely useful. Many private companies Artiﬁcial intelligence can help put the massive and tions. Identifying these varying needs and can also offer collaborators access to unique growing amount of Earth observations to use in cre- learning more about the dynamics of com- data and computational resources. ating accurate, usable predictions of atmospheric municating model results and uncertainty, as Multinational companies like IBM and and oceanic phenomena and in improving society’s well as about users’ risk perceptions and trust Google are deeply ingrained in nearly every long-term sustainability. Credit: NASA Earth Obser- in AI, are essential to the success of AI inte- industry and have global perspectives on vatory (image); J. Henry Pereira (design) gration across disciplines and sectors. what environmental science information is needed and actionable, on what timescales, Cross-Sector Collaboration in Action as well as on how this information can be dis- Integrative approaches have already proven tributed efficiently to the media, the public, What makes environmental science AI successful as force multipliers in a variety of or individual clients. These companies typi- methods trustworthy to different stakehold- efforts. cally focus on simplicity and pragmatism, ers varies depending on their expertise and In the U.S. Great Plains, hailstorms cause cost–benefit ratios, data latency, user expe- experience, however, and it is important to significant property damage each year. The rience, and maintainability more than aca- understand their diverse viewpoints. Gener- University of Oklahoma collaborated with demia and government do. Injecting this per- ally, physical and computer science research- the U.S. National Oceanic and Atmospheric spective into academic and government ers in academia pay more attention to the Administration’s (NOAA) Storm Prediction research and development helps ensure that theoretical performance of AI-based models Center (SPC) to operationalize AI hail fore- high-value knowledge is developed and that and less to human aspects of trust. AI casting, producing hail forecasts 12–36 hours methods, software, and algorithms are suit- researchers develop new models and score in advance. Operational forecasters (the able for quick and effective use, including in them using objective metrics. Meteorologists meteorologists who produce weather reports commercial tools and industry-specific deci- also use performance diagrams and case like those seen on the news) considered the sion platforms. New types of deep collabora- studies, examining an algorithm’s ability to early AI-generated hailstorm probabilities tion have the potential to accelerate such reproduce observations and effects of known produced by this system to be too high com- research- to- operations transitions. past events, often extreme or societally pared with forecasts traditionally issued by impactful events, to assess an AI model’s NOAA’s SPC. As a consequence, the AI fore- Trustworthy AI performance. Although AI researchers gen- casts were rethought and reformulated to for Environmental Sciences erally evaluate new models in a research set- produce probabilistic forecasts that were Realizing the full potential of AI to yield ting, operational meteorologists care about more in line with what human forecasters improved understanding and predictions of what additional guidance value and physical expected. This collaboration highlighted the environmental phenomena requires conver- consistency the algorithm provides in com- need to work closely with end users to develop gent research to develop methods that earn parison with the other sources of information AI that they consider trustworthy. the trust of decision-makers, including indi- available to them. In a collaboration among academic organi- viduals, households, business owners and Definitions and metrics for assessing reli- zations, federal and state agencies, and the private companies, and officials at all levels able and trustworthy methods can be very private sector, Texas A&M U niversity–Corpus of government. This trustworthiness is espe- different for private industry and govern- Christi created an operational AI-based fore- cially critical for extreme and high- impact ment. Government forecasters may prioritize casting system to help save sea turtles that environmental phenomena—ranging from the reliability of AI techniques across weather found themselves in cold water along the tornadoes to heavy snowfall during rush regimes, for example, whereas private indus- Texas coastline. Strong cold fronts can hour—in which life-or-death and other pro- try is incentivized to ensure that products are decrease water temperatures quickly in coas- tective decisions must be made quickly and tailored to specific clients, with client needs tal bays and lagunas. As a result, thousands reliably. Having skillful, user-trusted AI pre- driving the measures of trustworthiness. of threatened and endangered sea turtles can dictions of these hazards can, for example, By working across sectors and disciplines, become lethargic and require rescue to sur- help people decide when to take shelter or the applied needs of end users in government vive. AI-based water temperature predictions when they should (or should not) drive. and industry drive the development of new allow managers to anticipate impacts on sea

16 Eos // AUGUST 2020 OPINION

turtles and save most of them by mobilizing spheric Research (NCAR) and the American come, helping improve the sustainability and resources, gathering and organizing volun- technology company Nvidia to optimize safety of our world. teers, diverting watercraft, and interrupting NCAR’s state-of-the-art Model for Prediction other coastal activities such as shipping. By Across Scales (MPAS) weather model to run Acknowledgments developing a reliable AI method through col- on a GPU-equipped POWER9 supercomputer. The views expressed here are those of the laborations and agreements among the rele- IBM also worked with the University of Wash- authors and do not necessarily represent vant federal, state, and local governments ington to develop AI methods to clean and those of their employers, NOAA, or the U.S. and the private sector, researchers have sig- correct millions of pressure measurements government. nificantly reduced sea turtle deaths. gathered from opted-in cell phone users so On the other side of the world, Google is the measurements could be assimilated into collaborating with the government of India the model. The resulting IBM Global High- By Amy McGovern ([email protected]), Uni- to provide end-to-end flood forecasts, Resolution Forecast System (IBM GRAF) runs versity of Oklahoma, Norman; Ann Bostrom, including AI-based precipitation forecasting hourly and produces forecasts at 3-kilometer University of Washington, Seattle; Imme Ebert- and hydrological and inundation modeling. resolution. This type of high-resolution, rap- Upho¦, Colorado State University, Fort Collins; A partnership between the Indian Central idly updating forecast was previously avail- Ruoying He, North Carolina State University, Water Commission and Google Crisis able only in industrialized countries but is Raleigh; Chris Thorncroft, University at Albany, Response delivers alerts. When AI is inte- now available to consumers and businesses Albany, N.Y.; Philippe Tissot, Texas A&M Univer- grated into the warning system, these alerts around the world, including more than 300 sity–Corpus Christi; Sid Boukabara, National are more personalized to individual users and million users of the Weather Channel mobile Oceanic and Atmospheric Administration, Col- their specific locations and needs. More than app. lege Park, Md.; Julie Demuth and David John 1.5 million alerts were issued in 2019, and Although some of these collaborations are Gagne II, National Center for Atmospheric there are plans to expand coverage tenfold in just beginning, they highlight the ability of Research, Boulder, Colo.; Jason Hickey, Goo- 2020. convergent multisector research to create gle Research, Mountain View, Calif.; and John K. AI approaches can also make hyperlocal trustworthy AI, enabling improved under- Williams, The Weather Company, an IBM Busi- weather forecasts accessible to more regions standing and prediction of a broad range of ness, Andover, Mass. around the world. In a multiyear effort, IBM environmental phenomena. These tools will worked with the National Center for Atmo- be useful to decision makers for years to u Read the article at bit.ly/Eos-AI

SCIENCE NEWS BY AGU // Eos.org 17 A GREENER EARTH MODEL

BY RICHARD LOFT tock.com/4X-image iS

18 Eos // AUGUST 2020 As weather and climate models grow larger and more data intensive, the amount of energy needed to run them continues to increase. Are researchers doing enough to minimize the carbon footprint of their computing? tock.com/4X-image iS

SCIENCE NEWS BY AGU // Eos.org 19 SOME PROJECTIONS INDICATE THAT BY 2030, ELECTRICITY USAGE BY COMMUNICATION TECHNOLOGY COULD ecently, the National Center CONTRIBUTE UP TO 23% build powerful supercomputing machines to for Atmospheric Research tackle long- standing model biases in the (NCAR), where I serve as OF GLOBAL GREENHOUSE water cycle and improve predictions seem to director of technology GAS EMISSIONS. be coming to fruition. All of these long- development in the Com- running, high-resolution models, and the putational and Information big computers needed to run them, will RSystems Laboratory, conducted a carbon carbon dioxide equivalents. That number require investments from governments and footprint analysis. The organization was may give pause to some, while for others it philanthropic organizations. For example, quite pleased with the results, until it real- may represent an acceptable cost of doing in February, the United Kingdom announced ized that the analysis neglected to account business in the modern world. Regardless, £854 million in funding to develop its next- for carbon dioxide emissions related to the unlike a gasoline-powered car with an generation computer, and the U.S. National lab’s modeling activities. When these emis- exhaust pipe, there’s nothing about the act Oceanic and Atmospheric Administration sions were included, the overall picture of sending an email that makes it obvious announced that it was tripling the size of its looked considerably less green. that we’re causing carbon dioxide emis- investment for weather- and climate- How could this oversight have hap- sions, so the environmental cost is easy to related computing. pened? In part, it’s because modern society overlook. With the continuing growth of In this context, the ESS community is very good at obscuring the environmen- cloud computing and the Internet of Things, should consider the question, What is our tal impacts of its activities, whether these more emissions, like those from computing collective responsibility to reduce carbon are impacts from food production, manu- and the communication of data, will be fur- emissions related to these large-scale mod- facturing, or many other activities. The ther virtualized. eling activities? same is true of the information technology Within the Earth system sciences (ESS) (IT) industry. Carbon dioxide emissions community, many initiatives around the Leading by Example from the IT industry already rival those of world are planning the next generation of In discussions with colleagues, three coun- the prepandemic airline industry, and global weather and climate models that will terarguments have been offered: first, that some projections indicate that by 2030, be capable of resolving storms and, ulti- weather and climate modeling activities are electricity usage by communication tech- mately, clouds. Examples of these include only a small contributor to societal emis- nology could contribute up to 23% of global the Climate Modeling Alliance (CliMA) sions overall; second, that ESS research is greenhouse gas emissions [Andrae and model, the Energy- efficient Scalable Algo- too important to let it be slowed by these Edler, 2015]. rithms for Weather and Climate Prediction considerations; and third, that even raising In his book How Bad Are Bananas? The Car- at Exascale (ESCAPE-2) model, and the the subject of emissions from ESS comput- bon Footprint of Everything, Mike Berners-Lee Energy Exascale Earth System Model ing provides ammunition to both ends of estimates that the energy required to trans- (E 3 SM). The repeated calls from the com- the political spectrum to attack the research mit a typical email generates 4 grams of munity [Shukla et al., 2010; Palmer, 2014] to goals.

The central processing unit–based Cheyenne supercomputer, went into service in 2017 at the National Center for Atmospheric Research (NCAR)–Wyoming Supercomputing Center. Credit: University Corporation for Atmospheric Research (UCAR), Carlye Calvin, CC BY-NC 4.0 ( bit.ly/ ccbync4-0)

20 Eos // AUGUST 2020 Fig. 1. A comparison of power usage eciency (PUE) at a variety of energy- ecient data centers. All of these data centers were built within the past few years, with the exception of NCAR’s Mesa Laboratory, which was built in 1967. The average PUE in 2019 among more than 600 data centers surveyed by the Uptime Institute was 1.67.

To be sure, studying the impacts of mate change, and we should laud and try to ture (the computer and data system), the human activity on our planet is both an emulate organizations that do so. It is worth software (the model), and the data motion important and, of late, a risky area of considering, however, that the environ- (the workflow). research: Scientists studying climate, for mental side effects of a future decarbonized Computing facilities: According to a recent example, are frequently caught in political energy portfolio are not well understood report from the Uptime Institute, “Efforts crosshairs and are also subjected to trolling [Luderer et al., 2019], and switching to to improve the energy efficiency of the and other forms of harassment. But the first renewable energy sources often means buy- mechanical and electrical infrastructure of two arguments above sound like rational- ing credits, which can be traded, thereby the data center are now producing only mar- izations that might be made by anyone obfuscating the actual source of the energy ginal improvements. The focus needs to when they are first confronted with their powering a computing facility. move to IT.” The commonly used measure carbon footprint. of facility efficiency is power usage efficiency Certainly, it seems this is an opportunity (PUE), which is basically the total power a for the ESS community to lead by example. THIS IS AN OPPORTUNITY facility uses divided by the power used to So perhaps a more productive question is, run the infrastructure within the facility. Have we done everything we can to mini- FOR THE ESS COMMUNITY TO Whereas older computing facilities still have mize the carbon footprint of our computing LEAD BY EXAMPLE. HAVE WE room for improvement in PUE, recently built activities? DONE EVERYTHING WE CAN data centers are, for the most part, already In answering this question, it’s useful to quite efficient (Figure 1). For example, the separate considerations about decarboniz- TO MINIMIZE THE CARBON NCAR–Wyoming Supercomputing Center ing utilities from those of reducing the FOOTPRINT OF OUR (NWSC), which my organization operates “energy to solution,” a metric for compar- for the ESS research community on behalf of ing high- resolution atmospheric models COMPUTING ACTIVITIES? the National Science Foundation, has a PUE that accounts for the total energy needed to of 1.08, meaning that if the facility were per- run a model simulation from start to finish fect, with no energy usage overhead, it would [Fuhrer et al., 2018]. In this way, questions In the meantime, one way to address the yield only an 8% improvement in PUE com- about the merits of carbon offset schemes problem without creating more problems or pared with its present state. Still, as efficient or the locations of facilities, for example, sacrificing transparency is to improve the as it is, since the facility’s commissioning in can be considered apart from questions efficiency of the modeling enterprise. This 2012, the computers that NWSC houses have about how to reduce the energy consumed idea is perhaps best expressed by the Japa- emitted roughly 100,000 metric tons of car- by our research activities. nese expression “mottainai,” often inter- bon dioxide—roughly the mass of a modern preted to mean “waste not, want not.” aircraft carrier. In such cases, the best path Components of Computing’s We can consider the challenge of improv- to further optimization isn’t the facility. Carbon Footprint ing the energy efficiency of modeling as an Computing infrastructure: Supercomputers Energy sources: Switching to renewable infrastructure stack to be optimized, where on the scale that will be required for high- energy sources like wind, solar, and biogas the components are the computing facilities resolution Earth system models are being must be part of the solution to mitigate cli- (the data center), the computing infrastruc- deployed by the U.S. Department of Energy.

SCIENCE NEWS BY AGU // Eos.org 21 The Summit supercomputer at Oak Ridge onboard storage devices, and drives them to the inherent complexity of Earth system National Laboratory and the planned Aurora Amazon for storage in the cloud. Yes, it is model codes, the lack of workforce trained system at Argonne National Laboratory, actually more efficient and faster to load in programming new technology, the for example, will each consume on the data into a truck and drive them across underresourcing of such programming order of 10 megawatts, annually producing country than to transfer them via the Inter- efforts, and the inherent architectural com- carbon dioxide emissions equivalent to net. Given that a single 1-kilometer 3D plexity of the heterogeneous computing those from more than 30,000 round-trip atmospheric field will produce roughly half systems currently available. flights between Washington, D.C., and a terabyte in single precision, data move- Fourth, chip manufacturers and super- London (calculated using the Avoided Emis- ments of this magnitude surely will be computer vendors should make it easier for sions and Regeneration Tool (AVERT) and required routinely in an exascale computing users to measure the actual amount of power the MyClimate initiative). Using values complex. drawn during execution of models. For some of standard system performance bench- models, the actual power draw can be as marks—specifically the High Performance The Path to Eciency much as 40% less than the “nameplate” Conjugate Gradients (HPCG) and High Per- I propose five steps the ESS modeling com- power of the systems on which they’re run. formance Linpack (HPL) benchmarks— munity should take to move toward a more Currently, the tools to make these measure- from published sources, such as the top500 energy efficient and lower- emissions future. ments are often poorly documented and are .org list of the 500 fastest supercomputers, First, although many modern computing architecture dependent. They are also typi- we find, for example, that the graphics pro- centers are already very efficient, there may cally used only by computer science teams. cessing unit (GPU)-based Summit super- still be ways to improve PUE of older facili- Fifth, the ESS community should increase computer system is 5.7 and 7.2 times more ties that should be examined. Also, when research into techniques that could lead to power efficient on HPCG and HPL, respec- possible, use renewable energy sources to energy savings, including reducing floating tively, than NCAR’s roughly contempora- power facilities. point precision computation; developing neous central processing unit (CPU)-based Second, regarding modeling software, machine learning algorithms to avoid unnec- Cheyenne system at NWSC. GPUs look like you can’t improve what you don’t measure. essary computations; and creating a new an energy-efficient alternative to CPUs for Publishing the energy consumed per simu- generation of scalable numerical algorithms reducing energy to solution. lated day when reporting benchmarking that would enable higher throughput in The software: In its report, the Uptime results, particularly for high-resolution terms of simulated years per wall clock day. Institute notes that optimizing energy- Achieving higher scientific throughput efficient software is the least frequently with reduced energy consumption and used best practice suggested by the Euro- EXPERIENCE WITH reduced emissions is a mottainai approach pean Code of Conduct (CoC) for Data Centre that is not only defensible but is also one of Energy Efficiency. Could Earth system mod- ATMOSPHERIC MODELS which everyone in the ESS community could els, which have been designed to run on THAT HAVE BEEN ADAPTED, be proud. CPUs predominantly, be optimized for OR PORTED, TO RUN GPUs? Experience with atmospheric models References that have been adapted, or ported, to run on ON GPUS HAS SHOWN THAT Andrae, A. S. G., and T. Edler (2015), On global electricity usage of communication technology: Trends to 2030, Challenges, 6(1), GPUs, like the Consortium for Small-scale SIGNIFICANT ENERGY SAVINGS 117–157, https:// doiorg/. 10.3390/ challe6010117. Modeling ( COSMO) model and most recently Fuhrer, O., et al. (2018), Near-global climate simulation at 1 km the Model for Prediction Across Scales CAN BE ACHIEVED. resolution: Establishing a performance baseline on 4888 GPUs with COSMO 5.0, Geosci. Model Dev., 11, 1,665–1,681, https://doi (MPAS), has shown that significant energy . org/ 10.5194/ gmd11-1665- -2018. savings can be achieved. However, refactor- Luderer, G., et al. (2019), Environmental co-beneﬁts and adverse side- eects of alternative power sector decarbonization strat- ing legacy models for GPUs is labor inten- models requiring large-scale computing egies, Nat. Commun., 10, 5229, https:// doiorg/. 10.1038/ s41467 sive, creating an obstacle to this potential resources, would help raise awareness about - 019 -13067-8. energy savings. And because of the substan- energy efficiency in ESS modeling, and Palmer, T. (2014), Climate forecasting: Build high- resolution global climate models, Nature, 515, 338–339, https://doi .org/ 10.1038/ tial complexity of ESS code, Earth system through competition could foster develop- 515338a. modelers have not shown much appetite for ments leading to further energy savings. Shukla, J., et al. (2010), Toward a new generation of world climate research and computing facilities, Bull. Am. adopting these power-efficient devices. This approach should be integrated into Meteorol. Soc., 91(10), 1,407–1,412, https:// doi . org/ 10.1175/ Data motion: Also relevant to the carbon existing model intercomparison projects, 2010BAMS2900.1. footprint of ESS modeling is data motion. such as the Dynamics of the Atmospheric Moving a petabyte of data around a data General Circulation Modeled on Non- Author Information center requires roughly the amount of hydrostatic Domains (DYAMOND) initiative Richard Loft ([email protected]), Computational and energy contained in half a barrel of oil; from the Centre of Excellence in Simulation Information Systems Laboratory, National Cen- burning this much oil produces about 215 of Weather and Climate in Europe. ter for Atmospheric Research, Boulder, Colo. kilograms of carbon dioxide emissions. Third, porting existing models to, and These figures explain the economics of developing models for, new computing The article represents the personal views and Amazon’s Snowmobile, a low-tech but architectures must be made easier. opinions of the author only and does not reect effective “sneakernet” solution to the big- Researchers seeking to program the current the opinions of NCAR/UCAR. data problem in which a data center on generation of energy- efficient GPUs can wheels—in the form of a semitruck—plugs attest to the difficulty of achieving this feat. u Read the article at bit.ly/Eos-greener in to your data center, transfers data to This difficulty comes from multiple sources: - Earth-model

22 Eos // AUGUST 2020 AGU FALL MEETING 2020 Register for #AGU20 online everywhere from 1-17 December!

Highlights include: • Presidential Forum speaker Leland Melvin, former NASA astronaut and NFL wide receiver • Multiple time zones for 1,000+ sessions and hours of networking and poster hall time • Recorded and live content to watch (or binge) at your convenience

Registration opens in August!

agu.org/fallmeeting

Meet the Candidates To be eligible to vote, your membership must be active by 1 September.

agu.org/elections

SCIENCE NEWS BY AGU // Eos.org 23 24

Eos // August 2020

Opposite Page: XXXXXX; This Page: Top Right: XXXX Creating

ToolData Kits That Everyone Can Use

By Zhong Liu, Vasco Mantas, Jennifer Wei, Menglin Jin, and David Meyer

Earth scientists need to make the growing wealth of data more accessible and build data services meant for interdisciplinary use.

s Earth science and the technologies it uses evolve and improve, the data and services that support the science also change and become more complex, often spanning multiple disciplines. The ability to easily find and seamlessly access these data and services in an open and integrated environment is essential to facilitating interdisciplinary research and applications and to broadening data user communities. The sheer amount of available data is growing rapidly as the science community adopts the Findable, Accessible, Interoperable, and Reusable (FAIR) data principles [Wilkinson et al., 2016] op Right: XXXX andA emerging technologies such as cloud computing. Even with recent advances in data archiving and services (e.g., more data sets and related information are available online with his Page: T his Page: customized data services and multiple data access methods), accessing heterogeneous inter-

Multiple special or discipline-oriented tools, often with steep learning curves, are required to handle heterogeneous, complex, and evolving Earth science data sets in interdisciplinary research and applications. Credit:

Opposite Page: XXXXXX; T Opposite Page: Pixabay/ Deborah Breen Whiting

SCIENCE NEWS BY AGU // Eos.org 25 disciplinary data sets (e.g., those with nonuniform data types and formats) still poses challenges to users globally. To address these challenges in Earth science interdisciplinary Access websites for dierent data services, we organized and led sessions titled “Data and Infor- mation Services for Interdisciplinary Research and Applications in data resources are often designed Earth Science” at AGU Fall Meetings 2018 and 2019; at both meetings, this was one of the largest of the Earth and Space Science dierently, and as a result, only Informatics sessions. Groups of international participants pre- those who are already familiar sented data, tools, and services for Earth science interdisciplinary research and applications, as well as work on other topics related to with the repositories can easily big data, artificial intelligence (AI), machine learning (ML), and natural language processing. locate data sets and information.

Six Challenges As a result of the presentations, discussions, and feedback from our AGU Fall Meeting sessions, we identified the following questions that address challenges in making Earth science data and data services information is a major challenge in improving interdisciplinary more accessible and useful: research and applications. Another barrier to easy access is that discipline-oriented websites How can we make Earth science interdisciplinary data sets needed • currently exist for their own special disciplinary requirements. To for a specific research project or application easier to find? unify these different websites, an integrated data system must How do we eliminate the need for the many special tools, some • address both general and discipline- specific requirements. On a with steep learning curves, that are currently required to handle larger scale, Earth science data from various U.S. federal agencies, heterogeneous and interdisciplinary data sets? countries other than the United States, private companies, and citi- What data services can we provide in the cloud environment, where • zen scientists must also be easily accessible in an integrated environ- unprecedented access to data sets and data analytics is available? ment. Ideally, all of these data would be accessible without the need What data services can we provide to facilitate AI/ML activities? • to visit different websites, but making this a reality requires collabo- How can we collect metrics to help development and enhancement • ration from domestic and international data scientists, developers, of data services and to benchmark the performance and societal and stakeholders to address such issues as disciplinary vocabulary, impacts of a project or mission? data standards, and usability. How do we ensure the scientific reproducibility of Earth sciences • At present, many websites rely on sorting and filtering of search research? results. In satellite data services, for example, search suggestions, Finding and Accessing the Data research subject, measurement, satellite source, and processing level Data users typically consult various sources (e.g., the Internet, con- are often used to narrow the list of search results. Users’ success in ference proceedings, colleagues, professors) to find where data are finding the data they need can vary significantly in such systems, archived and distributed among many repositories around the depending on the web design, the users’ knowledge, and many other world. However, access websites for different data resources are factors. Websites that focus on a single project or mission and con- often designed differently, and as a result, only those who are tain only a few data sets can eliminate the need for sorting or filter- already familiar with the repositories can easily locate data sets and ing. For more comprehensive resources, inexperienced users often information. Finding the right data and information for a specific need human assistants or a help desk to interact with them, find out research project or application is another challenge, especially for more about what they need, and provide recommendations for data inexperienced data users who may not be familiar with data sets products or services. Therefore, incorporating human expertise with outside their own disciplines and for people searching across disci- AI/ML technologies such as natural language processing in the sys- plines. tem may improve the user experience of finding data for a specific The variety of access website designs can impede data and infor- need. mation searches by users who are unfamiliar with specific repositories. For example, NASA’s Earth Observing System Data and Infor- Simplifying the Tool Kit mation System (EOSDIS) has 12 discipline- oriented Distributed Because interdisciplinary data sets are complex, and their formatting Active Archive Centers (DAACs) that archive and distribute NASA and data structures are not uniform, multiple tools are needed to han- Earth data sets from satellite missions and projects, each with its own dle such data sets for research and applications. For example, more unique web interface. For users doing interdisciplinary research that than 61 data tools are available at the 12 discipline-oriented NASA requires data from multiple DAACs, it can be difficult to become DAACs for search and order; data handling; subsetting and filtering; familiar with all of the interfaces. geolocation, reprojection, and mapping; and data visualization and One solution is to develop an integrated and uniform web interface analysis. It can be a daunting task for a user to learn all of these tools for data access. At present, NASA EOSDIS is developing such an inter- for interdisciplinary activities. face, called Earthdata, which serves as a gateway for all data sets and Heterogeneous data can also present challenges to users and services at the 12 NASA DAACs. When the interface is finished, users stakeholders without access to complex data set processing capabili- will be able to search all NASA Earth science data, along with data ties. For many users, acquiring dedicated software, using multiple services and information, in one place. Building this type of one-stop tools, and performing programming-based data analyses are not via- shop for accessing complex and heterogeneous data, services, and ble. The use of standards like uniform data formatting may address

26 Eos // AUGUST 2020 the problem of heterogeneous data often requiring many tools to One key challenge is to develop metrics that accurately describe a handle. Data tools that integrate more data processing capabilities wide range of data- related activities in research and applications. may also help in reducing the number of tools. Meanwhile, data Another is to develop metric standards for different disciplines so repositories can go beyond their existing services (e.g., providing that metrics are interoperable in interdisciplinary activities. original data as they are) by offering data interoperability services to provide data that meet users’ research or application requirements But Is It Reproducible? with respect to data format, projection, model grid, and spatiotempo- Scientific reproducibility is a cornerstone of Earth science (and of all ral resolution, for example. scientific fields). Providing trustworthy data and results is one of the most challenging and pressing issues in the science community. Putting New Capabilities to Work Reproducing Earth science research requires documentation of all Cloud computing provides new opportunities to address issues elements in the life cycle, including data, algorithms, software, com- related to the unprecedentedly large amount of data and data analyt- puting environment, and other factors. ics available. Governmental and private organizations are putting Several research projects and workshops have focused on ways to significant efforts toward developing cloud-based data services. The improve reproducibility by, for example, collecting best practice NASA Goddard Earth Sciences Data and Information Services Center guidelines—like the community guidelines for open and reproducible (GES DISC) plans to launch its popular online visualization and analy- workflows that the geoscience research modeling community has sis tool Giovanni in the cloud, creating the potential to scale up and worked to develop [Mullendore et al., 2020]. Earth science users are expand its current capabilities by, for example, including Earth sci- also increasingly using open-source software, open standards, and ence data sets from other NASA DAACs, improving performance, and services such as Jupyter Notebook to exchange workflows. For inter- providing new data analytics. disciplinary research and applications, the scope is much larger, Best practices, including user-friendly features and services, requiring participation and collaboration from the entire Earth sci- should be carried over to the cloud environment. These practices can ence community and from stakeholders (e.g., journal publishers). facilitate a smooth or seamless transition from on-premises data Major challenges include the development of standards for the services to the cloud and ensure a satisfactory user experience in the framework and for interoperability. cloud-based environment. One such initiative currently under development is the NASA EOSDIS Cumulus Project, a cloud-based frame- Going Far, Together work for NASA EOSDIS data collections. This project is designed so There are many challenges to improving Earth interdisciplinary data that users will not notice any difference between on-site and cloud- services. Emerging technologies like cloud computing and AI/ML can based data services. potentially enable significant progress in all aspects of interdisciplin- Cloud computing isn’t the only area in which new technologies are ary data services. However, tackling the challenges will also depend introducing significant changes. In recent years, the Earth science on active participation and collaboration from all involved parties in community, like many other sectors and scientific fields, has experi- the community. In the wake of AGU Fall Meeting 2019, we are con- enced a surge in research and applications using AI/ML techniques. tinuing our follow-up activities, which include assembling a special Identifying and adopting the features that data repositories can pro- issue in the journal Computers and Geosciences (with a submission vide to facilitate AI/ML activities are pressing and challenging issues. deadline of 15 September) to report current advances and address the For example, natural language processing– based data systems could challenges as well as planning another session at AGU Fall Meeting simplify access for users who want only visualizations (e.g., images, 2020. maps), facts, or information. These systems handle the interactions between computers and humans using natural language—ordinary Acknowledgments human speech in the form of voice or text rather than arcane computer We thank all session participants of the past two AGU Fall Meetings. commands. On the other hand, more advanced users expect data GES DISC is funded by NASA’s Science Mission Directorate. Vasco repositories to provide analysis- ready or customized data (e.g., train- Mantas received funding from the European Union’s Horizon 2020 ing data) for AI/ML activities. Down the road, standard or customized program under grant agreement GA 776026. AI/ML services—running AI algorithms such as deep learning or random forests, for example—can be integrated into data repositories, References allowing users to conduct AI/ML activities without leaving the system. Mullendore, G. L., M. S. Mayernik, and D. Schuster (2020), Determining best practices for archiving and reproducibility of model data, paper presented at the 36th Conference on Environmental Information Cloud computing may be able to host such services in the future. Processing Technologies, Am. Meteorol. Soc., Boston, Mass., 12–16 Jan. Wilkinson, M. D., et al. (2016), The FAIR guiding principles for scientiﬁc data management and steward- Measuring Performance and Impact ship, Sci. Data, 3, 160018, https:// doiorg/. 10.1038/ sdata2016.18.. Data metrics are frequently used to measure user and system activities—data access and usage in research and applications, for exam- Author Information ple—that are related to the life cycle of a data set and play an import- Zhong Liu ( zhong . liu@ nasa . gov), NASA Goddard Earth Sciences Data and ant role in Earth science. For example, in the satellite community, Information Services Center (GES DISC), Greenbelt, Md.; also at Center data usage metrics are used to benchmark a mission’s or project’s for Spatial Information Science and Systems, George Mason University, success and its societal impacts. Fairfax, Va.; Vasco Mantas, University of Coimbra, Coimbra, Portugal; Among their many applications, data metrics supply key informa- Jennifer Wei, NASA GES DISC, Greenbelt, Md.; Menglin Jin, University tion to satellite data service providers for designing new data services of Maryland, College Park; and David Meyer, NASA GES DISC, Green- and improving existing ones. Satellite product developers rely on data belt, Md. metrics to understand how their products are used. Thus, collecting data metrics is an essential part of a satellite mission or project. u Read the article at bit.ly/ Eos - data

SCIENCE NEWS BY AGU // Eos.org 27 VISUALIZING SCIENCE: HOW COLOR DETERMINES

By Stephanie Zeller WHAT WE SEE and David Rogers

28 Eos // AUGUST 2020 Color plays a major role in the analysis and communication of scientiﬁc information. New tools are helping to improve how color can be applied to data more accurately and eectively.

In a “wave” colormap like this one, selective saturation is used to isolate and “foreground” speciﬁc data ranges so these data are easier to follow over time. In the colormap, desaturated colors surround a saturated section to focus attention while providing context. Credit: Graphic created by Francesca Samsel with data processed and provided by M. Petersen, LANL, using MPAS-Ocean

SCIENCE NEWS BY AGU // Eos.org 29 olor strongly influences the tional rainbow colormap [e.g., Borland and way we perceive information, Taylor, 2007], they pervade visualizations especially when that informa- from basic bar graphs to complex depictions tion is dense, multidimen- of biogeochemical data. sional, and nuanced—as is At AGU’s Fall Meeting 2019, row upon row often the case in scientific of posters in the convention center’s vast Cdata sets. Choosing colors to visually repre- main hall featured the same bright, standard sent data can thus be hugely important colormaps adorning visualizations of tem- in interpreting and presenting scientific perature scales, chlorophyll concentrations, results accurately and effectively. land elevations, and a host of other data. “Language is inherently biased, but “The same colormap applied to a diverse through visualization, we can let the data array of data gets monotonous and confus- speak for [themselves],” said Phillip Wol- ing,” said Rick Saltus, a senior research sci- fram, an Earth system modeler and compu- entist with the Cooperative Institute for tational fluid dynamicist at Los Alamos Research in Environmental Sciences (CIRES) National Laboratory. At Los Alamos, data at the University of Colorado Boulder. visualizations are as ubiquitous as the sage- “You’re trying to communicate both effec- brush that embroiders the nearby desert. tively and efficiently, and that’s impeded if Every day, expert teams wrangle, render, the viewer is presented with a variety of and color encode swaths of data for inter- concepts, all illustrated using identical color pretation with the lab’s Earth and computer mappings.” scientists. Choosing colors to represent var- Color researchers and visualization ious properties of the data, a step that experts around the world are working to ranges from an iterative, responsive process change this status quo. A number of groups to a hasty afterthought, is the final barrier are developing new tools to help scientists between painstaking data collection and image increasingly complex data sets more well-anticipated analysis and discovery. accurately and intuitively, and with higher Most visualization software comes fidelity, using context as a guide to ensure equipped with colormaps—a selection of an appropriate balance of hue, luminance, standard color- encoding gradients that and saturation. researchers can apply, in a matter of sec- onds, to display and evaluate their data. But How We See Color not all data visualizations are created equal, More than a century ago, Albert H. Munsell and despite a proliferation of literature built upon the work of Isaac Newton and denouncing standard maps like the tradi- Johann Wolfgang von Goethe to compose

This image shows a comparison of the same data set displayed using the traditional rainbow colormap (left), a cool–warm colormap (middle), and a “wave” colormap, illustrating how visible detail can change when data are portrayed with dierent color gradients. Credit: Graphic created by Francesca Samsel with data processed and provided by M. Petersen, Los Alamos National Laboratory (LANL), using MP AS- Ocean

30 Eos // AUGUST 2020 Fig. 1. Detail of topography between the Filchner-Ronne and Ross Ice Shelves in West Antarctica. This image illustrates the eects of simultaneous contrast within the traditional rainbow colormap (left), increased detail in a desaturated version of the traditional rainbow (middle), and an analogous color palette for greater aesthetic quality as well as discriminatory power (right). Credit: Graphic created by Francesca Samsel with data processed using E 3 SM

our modern concept of color mapping [Mun- color no longer depended on approximation sell, 2015]. Munsell’s research produced the but could rather be coded numerically, plot- first perceptually ordered color space—a ted along a parabola. This level of accuracy is three-dimensional plot in which the axes necessary because color perception is a represent hue (color), value (lightness or subjective experience dependent on light, darkness), and chroma (intensity of color) simultaneous contrast (the phenomenon [Moreland, 2016]. of juxtaposed colors affecting each color’s Around the same time, in the early 1900s, appearance; see Figure 1), and rod and cone Hermann Grassmann’s theory of linear alge- photoreceptors within the viewer’s eyes bra decrypted abstract math, revealing the [Albers, 2006; Itten, 1970]. origami- like properties of higher dimensions. Grassmann thereby created the con- In the Eye of the Beholder cept of vector space, allowing for the approx- Contemporary color spaces (Figure 2) can be imate calculation of perceived color within a divided into two categories: absolute and defined area [Grassmann, 1844]. The study of nonabsolute. Absolute color spaces define

Fig. 2. Two types of color space established by the International Commission on Illumination (CIE) in 1931: the nonabsolute red–green–blue (RGB) color space illustrated using a cube (left), and the CIELAB color space (right), one of the most widely used absolute color spaces. The full parabola of CIELAB color space illustrates the full visible spectrum of hues typically discernible by the human eye. Within this parabola, the white triangle encloses all hues available in Adobe RGB color space. (CIELAB is a three-dimensional space shown in two dimensions here for simplicity.) Credit: SharkD, CC BY-SA 4.0 (bit.ly/ccbysa4-0) (left); public domain (right)

SCIENCE NEWS BY AGU // Eos.org 31 color in terms of human perception. More available expertise and guidance, prece- familiar nonabsolute color spaces such as dents within the community, and time red–green–blue (RGB) and cyan–magenta– available to researchers to familiarize yellow–black (CMYK) define color based on themselves with new color conventions self-contained models that rely on either [Moreland, 2016]. input devices, like a cam- “When people approach a visualization, era, or output devices, like a they have expectations of how visual fea- monitor or printer. tures will map onto concepts,” said Karen The popular rainbow Schloss, a psychologist at the University of Rainbow mapping can map was defined within a Wisconsin– Madison and the Wisconsin nonabsolute color space— Institute for Discovery. Schloss and her wash out oceans of data in its gradient spans many team are working to tackle these imple- highly saturated hues with- mentation issues and understand trade-offs glaring neon green, creat- out consideration for their between deeply ingrained, communal mathematical separation familiarity and the next generation of color ing false artifacts, color within an absolute (percep- tools. tual) space. This arrange- “We refer to these expectations as interference, and atten- ment creates extremely inferred mappings,” she said. “If someone high contrast between has been working in a particular field their tion bias. hues, visually demarcating whole [life], they have this inferred map- different values in a data ping for colors and the values they repre- set and thereby propelling sent. We are trying to understand these the rainbow map to default status among biases while also being careful about recom- most scientists. mending that these experts just abandon In recent years, however, researchers their conventions completely. We need to have challenged the rainbow’s status and find a balance.” common usage, which proliferated without In search of that balance, visualization full consideration of its influence on data experts are focusing their efforts on under- representation. Rainbow mapping can wash standing what scientists need from their out oceans of data in glaring neon green, data and on how to address those needs creating false artifacts, color interference, without compromising the information and attention bias (see, e.g., the left-hand contained in the data themselves. graph in the image on p. 30) [Borland and Taylor, 2007; Liu and Heer, 2018; Ware et al., What Do Scientists Need? 2019, 2017]. Perceptual scientists have Francesca Samsel, a research scientist at the attempted to remedy this issue through Texas Advanced Computing Center (TACC) “perceptually equalized colormaps”—maps at the University of Texas at Austin, and her made with uniformly spaced values in abso- team describe scientists’ needs with respect lute color space—but these maps, along to data visualization using three catego- with the rainbow colormap and its cousins, ries—feature identification, exploration, were all created independent of the data and communication—as well as subcatego- they are used to represent. This means they ries of pinpointing outliers and determining were made for general use as opposed to relationships. When interpreting a large considering each data set’s unique proper- data set, scientists are usually looking for ties and needs. In an age of rapidly growing specific features (e.g., the direction of flow and increasingly complex data, many visu- of ocean currents or water temperatures in alization experts agree that their utility is certain locations) within known data limited. ranges; exploring the data holistically to make general observations; or looking to Balancing Familiarity with Clarity communicate specific properties of the data Scientists and visualization professionals to colleagues, peers, or the public. Occa- attempting to heed the latest color mapping sionally, a researcher may be interested research struggle, however, to break away only in outliers, or in the way one variable from standard, data- i ndependent maps like affects another—for example, how water the traditional rainbow, viridis, the cool– temperatures change when two currents warm divergent, Jet, and Turbo (Google’s meet. increasingly popular next-generation rain- Visualization is how scientists interact bow). Impeding the adoption of this with the quantitative outcomes of their research are scarcities of convenient and research and how they make data-based user- friendly color mapping resources, arguments, said Wolfram. He regularly uses

32 Eos // AUGUST 2020 Fig. 3. Comparison of the same data portrayed with a desaturated rainbow colormap (top), which is often used to increase visual detail, and with a colormap designed to provide detail in outlying data ranges (bottom). The spe- cialized colormap provides detailed information about the kinetic energy within eddies in the Agulhas Current, a major current in the Indian Ocean, whereas the rainbow colormap simply identiﬁes the eddies. Credit: Graphic created by Francesca Samsel with data processed and provided by M. Petersen, LANL, using MPAS-Ocean

visualizations, including overhead plots, Boulder, have taken concrete steps to create representing the view from above (e.g., such tools, maintaining the goal of intuitive from satellites) to explore Earth systems operation. and climate data in his work at Los Alamos. “What I’m looking for is tightly coupled to A New Set of Color Tools the science question,” he said. “I’m typi- Samsel, who was trained and worked for cally trying to understand geospatial rela- 25 years as an artist before pivoting to visu- tionships, particularly in overhead plot alization, uses her data, for surface features like [ocean] knowledge of color the- eddies.” Advanced color mapping tools help ory in tackling the per- prevent significant losses of feature detail in ceptual challenges of the data Wolfram analyzes, unlike standard crafting colormaps. The importance of visual- maps like Jet (Figure 3). “We’ve discovered in the Samsel and others argue that non- data- course of our research ization throughout the dependent color mapping strategies can that presenting scien- perpetuate bias if the hues are not arranged tists with perceptually research process necessi- in a familiar order (e.g., rainbow order), if equalized colormaps is the luminance is not specifically accounted not always the most tates high- level tools that for, or if multiple gradients within a map are beneficial solution” for not arranged for a specific data set [Borland providing the kind of maximize information and and Taylor, 2007]. The importance of visual- resolution scientists ization throughout the research process need within their data, minimize obstruction. necessitates high-level tools that maximize Samsel said. There are information and minimize obstruction nuance and complexity caused by any of these color-related issues. in color interaction that affect a viewer’s Researchers Samsel, Schloss, and Danielle associative response and how they derive Szafir, an assistant professor and director of the relative importance of different features the VisuaLab at the University of Colorado within the data, she explained.

SCIENCE NEWS BY AGU // Eos.org 33 Samsel’s research spurred the creation age people to take more care in how they are of ColorMoves [Samsel et al., 2018], an color encoding and presenting their data.” applied tool for interactively fine-tuning As part of an effort to optimize color- colormaps to fit the needs of different data maps for visualizations, Schloss and her sets. The online interface for the tool pro- colleagues created a tool called Colorgorical. vides sets of maps focused on achieving The tool features a series of adjustable slid- increased discriminative power [Ware et al., ers for customizing a palette based on per- 2019] while reflecting the palette of the ceptual distance between hues, the differ- natural world—something called semantic ences between hue names, how close association. For example, many people together the hues are on the color wheel, associate blue with ocean data and green and the specificity of hue names (e.g., pea- with land data. cock or sapphire versus blue) [Heer and Beginning in the 1990s, perceptual Stone, 2012]. Users can also select a hue and researchers established that the human eye luminance range within which the custom is more sensitive to differences in lumi- palette should fall, and they can construct a nance (perceived brightness) than to hue color palette around a “seed color.” when distinguishing patterns within densely packed data points [Ware et al., 2019; Designer in a Box Rogowitz, 2013]. Samsel and her colleagues Schloss, Samsel, and Szafir agree that future have reaffirmed these findings, in part research in this area lies in producing auto- through color theory. Using this informa- matically generated colormaps that are tion in concert with gradations of hue and based on the custom work of designers and saturation, Samsel created linear (one hue that are ready to integrate with the visual- gradient), divergent (two hue gradients that ization software scientists are already using. meet in the middle), and what she calls Szafir and her team at the ATLAS Insti- wave colormaps—maps that cycle through tute’s Department of Computer Science are luminance distribution across many hues already on the path to making this vision a (see image on pp. 28–29). “This creates a reality. “We started thinking about how to greater density of contrast throughout the meet the user where they are,” Szafir said. map, which resolves The attention to detail and creativity that many more features on individual designers can offer are unfortu- continuous data,” said nately not very scalable, so Szafir’s team Samsel. wondered whether they could create a prod- “How can we capture The ColorMoves uct that was effectively a “designer in a interface enables users box.” designer practices with- to drop multiple color- “How can we capture designer practices maps onto their data without asking them to distill their entire out asking them to distill and adjust the result field down to an extremely simplified, and interactively, seeing the therefore incomplete, set of rules?” Szafir their entire field down to changes on their data in asked. real time and allocating Szafir and her colleagues constructed an extremely simplified, hue and contrast wher- machine learning models based on 222 ever appropriate. “It expert-designed color gradients and used and therefore incom- wasn’t necessarily the results to automatically generate maps intended as a tool for that mimic designer practices. The final plete, set of rules?” data exploration, but maps “support accurate and aesthetically scientists have identi- pleasing visualizations at least as well as fied that as a priority” designer [maps] and…outperform conven- and have been using it for exploration, said tional mathematical approaches” to color- Samsel. map automation, Szafir and colleagues Schloss, who runs the Visual Reasoning wrote in a recent study [Smart et al., 2020]. Lab at the University of Wisconsin– M adison, Despite her team’s progress, complete focuses on making visual communication automation has drawbacks, Szafir said. An more effective and efficient through the algorithm needs a set of data to “train” on— study of color cognition, targeting the in this context, a batch of colormaps with trade- offs in color mapping between high similar curves in 3D color space. The algo- contrast and aestheticism. “People see cus- rithm derives a set of rules based on this tomization as a huge asset for creating visu- group of maps, then generates new maps alizations,” she said, “and I think making based on those rules. However, “a lot of what that capability easily accessible may encour- comes out of the artistic community that is

34 Eos // AUGUST 2020 truly expressive, creative, and engaging understanding big data, we just sort of throw comes from breaking these conventions,” the facts out there and wash our hands of she said. “An artist knows where and when the affair,” she said. “Data visualization is a and how to break those rules, and we don’t language, and, foundationally, you’re trying know if we can produce that level of compa- to tell a story. Color is a huge part of this.” rable variation [in a program] successfully.” If researchers do not behave as if they are Both Samsel and Schloss are seeking ave- in conversation with their audience, Bar- nues to merge their diverse expertise with tram said, their work will have little impact. Szafir in the near future, continuing to “The democratization of data visualization structure their respective work around the means that it has become media; it is no data visualization needs of scientists. “We longer just a means to an end for scien- see a future of using new tools to measure tists,” she said. “Visualization has become the spatial density of data, creating an auto- so much more than just a tool. It is now a mation process that interpolates, aligns, part of our conversations and decision- and allocates contrast within a colormap to making as a society at large.” match [those] data,” said Samsel. Such a program would concentrate variance in hue References and luminance where the data are densest, Albers, J. (2006), Interaction of Color, Yale Univ. Press, New Haven, Conn. allowing a scientist to visualize nuance and Borland, D., and R. M. Taylor Ii (2007), Rainbow color map (still) detail where they count. considered harmful, IEEE Comput. Graphics Appl., 27(2), 14–17, https:// doiorg/. 10.1109/ MCG.2007.323435. Grassmann, H. (1844), Die Lineale Ausdehnungslehre ein neuer Conveying an Accurate Message Zweig der Mathematik: Darge stellt und durch Anwendungen, Scientific visualizations play multiple roles Von Otto Wigand, Leipzig, Germany, http:// archive . org/ details/ dielinealeausde00grasgoog. of helping people quantify, interpret, evalu- Heer, J., and M. Stone (2012), Color naming models for color ate, and communicate information. Their selection, image editing and palette design, in Conference importance in the exploration and discovery Proceedings: The 30th ACM Conference on Human Factors in Computing Systems (CHI), p. 1007, Assoc. for Comput. Mach., of data is immense and is growing with New York, http:// visstanford. .edu/ papers/ colornaming- -models. advances in computational power, yet visu- Itten, J. (1970), The Elements of Color: A Treatise on the Color System of Johannes Itten Based on His Book The Art of Color, alization’s single most effective encoder— edited by F. Birren, 96 pp., John Wiley, New York. color—remains vastly understudied. Liu, Y., and J. Heer (2018), Somewhere over the rainbow: An Although the macroscale effects of such empirical assessment of quantitative colormaps, in Proceedings of the 2018 CHI Conference on Human Factors in Computing neglect on public perception of issues like Systems, pp. 1–12, Assoc. for Comput. Mach., New York, https:// climate change are yet unknown, many doi .org/ 10.1145/ 3173574.3174172. groups are beginning to devote greater Moreland, K. (2016), Why we use bad color maps and what you can do about it, Electron. Imaging, 16, 1–6, https:// doiorg/. resources to designing their visualizations 10.2352/ ISSN2470. -1173 .2016.16 .HVEI-133. with the public in mind. Munsell, A. H. (2015), A Color Notation, Scholar’s Choice Edition, 98 pp., Creative Media Partners, https:// booksgoogle. . com/ Ed Hawkins, a lead author of the upcom- books ? id =7dtJrgEACAAJ. ing sixth Intergovernmental Panel on Cli- Rogowitz, B. E. (2013), Perceptual approaches to finding features mate Change (IPCC) report and creator of in data, Proc. SPIE Electron. Imaging, 8651, https:// doiorg/. 10.1117/ 12.2013502. the now- viral “Warming Stripes” graphic, Samsel, F., S. Klaassen, and D. H. Rogers (2018), ColorMoves: said the IPCC pays multiple graphic Real-time interactive colormap construction for scientific visu- designers to transform complex visualiza- alization, IEEE Comput. Graphics Appl., 38(1), 20–29, https:// doi . org/ 10.1109/ MCG2018. .011461525. tions into simplified graphics. “We have to Smart, S., K. Wu, and D. A. Szafir (2020), Color crafting: Automat- be able to reach a very broad audience,” ing the construction of designer quality color ramps, IEEE Trans. Visualization Comput. Graphics, 26(1), 1,215–1,225, https:// doi Hawkins added. “Not just for general com- . org/ 10.1109/ TVCG2019. . 2934284. munication but also to inform policy deci- Ware, C., et al. (2017), Evaluating the perceptual uniformity of sions and to help people respond to risks color sequences for feature discrimination, in EuroVis Workshop on Reproducibility, Verification, and Validation in Visualization that threaten their way of life.” Hawkins (Eur0RV3), Eurographics Assoc., https:// doiorg/. 10.2312/ eurorv3 and his team spend “a lot of time” focusing . 20171107. Ware, C., et al. (2019), Measuring and modeling the feature detec- on color blindness issues in addition to tion threshold functions of colormaps, IEEE Trans. Visualization readability and semantic understanding of Comput. Graphics, 25(9), 2,777–2790, https:// doiorg/ . 10.1109/ color. TVCG .2018 .2855742. Keeping in mind the difference in perception between those publishing reports and Author Information those reading them is imperative, according Stephanie Zeller ([email protected]), to Lyn Bartram, a professor in the School of University of Texas at Austin; and David Interactive Arts and Technology at Simon Rogers, Los Alamos National Laboratory, Los Fraser University and director of the Van- Alamos, N.M. couver Institute for Visual Analytics. “Rather than engaging people in the experience of u Read the article at bit.ly/ Eos - color

SCIENCE NEWS BY AGU // Eos.org 35 AGU NEWS

In Appreciation of AGU’s Outstanding Reviewers of 2019

ued to accept invitations to peer review article submissions. Editorials in each journal (some already published, some upcoming) express our appreciation along with recognition lists. Our thanks are a small recognition of the large responsibility that reviewers bear in improving our science and its role in society.

Additional Thanks In addition, we are working to highlight the valuable role of reviewers through events (though they may be virtual) at Fall Meeting and other meetings. Each reviewer also receives a discount on AGU and Wiley books. We will continue to work with the Open Researcher and Contrib- utor Identification (ORCID) network to provide official recognition of reviewers’ efforts, GU Publications recognizes a number results. This increase in complexity, in turn, so that reviewers receive formal credit there. of outstanding reviewers for their has increased the challenge and role of review- As of 5 May, we have over 59,962 ORCIDs A work in 2019, as selected by the edi- ing. The outstanding reviewers listed here linked to GEMS user accounts, compared with tors of each journal. have all provided in-depth evaluations, often 49,000 at this time last year. Peer review is central to communicating over more than one round of revision, that and advancing science. Although there have greatly improved the final published papers. never been more ways to distribute ideas and Their contributions have helped to raise the research outputs online, a robust peer review quality of submissions received from around Our thanks are a small ensures that we maintain the highest integ- the world, providing valuable feedback that rity in our scientific discourse. The peer elevates the prominence of our journals to the recognition of the large review process is organized by our journal high standards that align with AGU tradition. responsibility that reviewers editors, but every article decision relies on individuals who take time out from their own Many Reviewers: A Key Part bear in improving our research to volunteer their expertise. The of AGU Journals science and its role in work of these reviewers ensures proper eval- Although we note these few outstanding reviewers here, we also acknowledge the society. broad efforts by all AGU reviewers in helping ensure the quality, timeliness, and reputation The outstanding reviewers of AGU journals. In 2019, AGU received over 16,700 submissions, up from 15,600 submis- Getting Your Feedback listed here have all provided sions received in 2018, and published over We are working to improve the peer review in-depth evaluations, often 7,000, up from 6,600 articles in 2018. Many process itself, using new online tools. We of these submissions were reviewed multiple conducted a full survey earlier in 2020, and over more than one round times—in all, 18,173 reviewers completed we continue to provide a short questionnaire 39,368 reviews in 2019 compared with 37,674 for feedback after each review is completed. of revision, that greatly reviews completed in 2018. We value your feedback, including ideas improved the ﬁnal This increase happened in the past year about how we can recognize your efforts even while each AGU journal worked to shorten the more, help improve your experience, and published papers. time from submission to first decision and increase your input on the science. publication or maintained already industry- We look forward to hearing from you. If leading standards. Several AGU journals reg- you’d like to respond directly, feel free to take ularly return first decisions within 1 month of our survey. uation of thousands of articles each year. We submission, and most others do so now Once again: Thanks! are truly thankful for their efforts. within 2 months. Reviewers represent a key As the uses for scientific literature have part of this improvement. We look back at grown, so has the complexity of papers, which 2019 here, but we have already seen that in By Matt Giampoala ([email protected]), now typically include more authors bringing 2020, during the pandemic and unrest, mem- Vice President, Science, AGU; and Carol Frost,

more techniques, data, simulations, and bers of our amazing community have contin- Chair, AGU Publications Committee Opposite: iStock.com/FoxysGraphic

36 Eos // AUGUST 2020 AGU NEWS

Kelly Aho Hoori Ajami Ruzbeh Akbar A. Addison Alford JGR: Biogeosciences editors Martyn Clark Valeriy Ivanov Earth’s Future editors JGR: Biogeosciences Water Resources Research Geophysical Research Letters Earth’s Future

Subramaniam Leif G. Anderson Martin Archer Sylvain Barbot Ananthakrishnan Peter Brewer JGR: Space Physics editors Isabelle Manighetti Sana Salous JGR: Oceans JGR: Space Physics JGR: Solid Earth Radio Science

William Barnhart Flavien Beaud Matthew W. Becker Larry K. Berg Gavin P. Hayes Harihar Rajaram Martyn Clark Minghua Zhang Geophysical Research Letters Geophysical Research Letters Water Resources Research JGR: Atmospheres

Wouter R. Berghuijs Gilles Billen Joachim Birn Russell Blackport Martyn Clark Global Biogeochemical Astrid Maute Gudrun Magnusdottir Water Resources Research Cycles editors Earth and Space Science and Chris Patricola Global Biogeochemical Cycles Geophysical Research Letters Opposite: iStock.com/FoxysGraphic

SCIENCE NEWS BY AGU // Eos.org 37 AGU NEWS

David Bromwich Stephen J. Burges Susannah Burrows Tim Canty Ruby Leung Martyn Clark Joel Thornton James Crawford JGR: Atmospheres Water Resources Research Geophysical Research Letters JGR: Atmospheres

Vachel Carter Valerie Cayol Stephanie Contardo Jessie Creamean Valerie Trouet Rebecca Carey Ryan Mulligan Lynn Russell Geophysical Research Letters Geophysical Research Letters JGR: Oceans JGR: Atmospheres

Jessica N. Cross Kyla M. Dahlin Stephen S. E. Darby Andrew Delman Harihar Rajaram JGR: Biogeosciences editors JGR: Earth Surface editors Janet Sprintall Geophysical Research Letters JGR: Biogeosciences JGR: Earth Surface Geophysical Research Letters

Kristine DeLong Tonya DelSontro Charlotte A. DeMott Simone Di Matteo Valerie Trouet Rose M. Cory Chidong Zhang Astrid Maute Geophysical Research Letters Geophysical Research Letters JGR: Atmospheres Earth and Space Science

38 Eos // AUGUST 2020 AGU NEWS

Andrew P. Dimmock Stepan Dubyagin Susana Enríquez Stefan Eriksson JGR: Space Physics editors JGR: Space Physics editors JGR: Biogeosciences editors Andrew Yau JGR: Space Physics JGR: Space Physics JGR: Biogeosciences Geophysical Research Letters

Laura Ermert Robyn Fiori André Bustor¦ Fortunato Matthew Fox Isabelle Manighetti Space Weather editors Andrea Donnellan Geochemistry, Geophysics, JGR: Solid Earth Space Weather Earth and Space Science Geosystems editors Geochemistry, Geophysics, Geosystems

Amanda Frossard Bea Galladro-Lacourt Juan Pascual García Christoph Geiss Lynn Russell JGR: Space Physics editors Sana Salous Geochemistry, Geophysics, JGR: Atmospheres JGR: Space Physics Radio Science Geosystems editors Geochemistry, Geophysics, Geosystems

Cynthia Gerlein-Safdi Jie Gong Sander Goossens Julia Green Earth’s Future editors Hui Su JGR: Planets editors Alessandra Giannini Earth’s Future Geophysical Research Letters JGR: Planets Geophysical Research Letters

SCIENCE NEWS BY AGU // Eos.org 39 AGU NEWS

Rebecca Greenberger Jennifer Griswold Nicolas Gruber Bin Guan JGR: Planets editors Lynn Russell Global Biogeochemical Ruby Leung JGR: Planets JGR: Atmospheres Cycles editors JGR: Atmospheres Global Biogeochemical Cycles

Hoshin Vijai Gupta Jasper Halekas Heather Handley Yuki Harada Martyn Clark Andrew Dombard Geochemistry, Geophysics, JGR: Space Physics editors Water Resources Research Geophysical Research Letters Geosystems editors JGR: Space Physics Geochemistry, Geophysics, Geosystems

Adrian Adam Harpold Tore Hattermann Liming He Carole Helfter Martyn Clark Laurie Padman JGR: Biogeosciences editors Fabio Florindo Water Resources Research JGR: Oceans JGR: Biogeosciences Reviews of Geophysics

Jennifer Hertzberg Peter Hitchcock Forrest Ho¦man Ricardo Hueso Alonso Paleoceanography and Bill Randel Rose M. Cory JGR: Planets editors Paleoclimatology editors JGR: Atmospheres Geophysical Research Letters JGR: Planets Paleoceanography and Paleoclimatology

40 Eos // AUGUST 2020 AGU NEWS

Kenneth Hughes Adele L. Igel Rachael James Anna Karion Laurie Padman Journal of Advances in Modeling Geochemistry, Geophysics, Fabio Florindo JGR: Oceans Earth Systems editors Geosystems editors Reviews of Geophysics Journal of Advances in Modeling Geochemistry, Geophysics, Earth Systems (JAMES) Geosystems

Joseph R. Kasprzyk Laura Kerber Jessica Keune Lawrence Kleinman Martyn Clark Andrew Dombard Valeriy Ivanov James Crawford Water Resources Research Geophysical Research Letters Geophysical Research Letters JGR: Atmospheres

Maria Klepikova Harald Kucharek James LaBelle Simon Lamb Martyn Clark JGR: Space Physics editors Andrew Yau Isabelle Manighetti Water Resources Research JGR: Space Physics Geophysical Research Letters JGR: Solid Earth

Felix W. Landerer William K. M. Lau Carl J. Legleiter Solène Lejosne Kathleen Donohue Minghua Zhang Martyn Clark JGR: Space Physics editors Geophysical Research Letters JGR: Atmospheres Water Resources Research JGR: Space Physics

SCIENCE NEWS BY AGU // Eos.org 41 AGU NEWS

Ricardo Letelier Michelle L. L’Heureux Laifang Li Lin Li Angelicque White Suzana Camargo Kristopher Karnauskas Tectonics editors Geophysical Research Letters Geophysical Research Letters JGR: Oceans Tectonics

Justin J. Likar Wuyin Lin Gregory A. Lyzenga Douglas MacMartin Space Weather editors Minghua Zhang Andrea Donnellan Earth’s Future editors Space Weather JGR: Atmospheres Earth and Space Science Earth’s Future

Elizabeth H. Madden Penelope Maher Odin Marc Milena Marjanović Isabelle Manighetti Journal of Advances in Modeling JGR: Earth Surface editors Geochemistry, Geophysics, JGR: Solid Earth Earth Systems editors JGR: Earth Surface Geosystems editors Journal of Advances in Modeling Geochemistry, Geophysics, Earth Systems (JAMES) Geosystems

Massimo Mattei Thorsten Mauritsen Stefano Mazzoli Jessica McBeck Tectonics editors Hui Su Tectonics editors Isabelle Manighetti Tectonics Geophysical Research Letters Tectonics JGR: Solid Earth

42 Eos // AUGUST 2020 AGU NEWS

Gavin McNicol Kathryn McWilliams Florian Mekhaldi Annette K. Miltenberger JGR: Biogeosciences editors JGR: Space Physics editors Valerie Trouet Martyn Clark JGR: Biogeosciences JGR: Space Physics Geophysical Research Letters Water Resources Research

Kyungguk Min Kumar Vijay Mishra Toru Miyama Guillaume Morard Gang Lu Sana Salous Leo Oey Steve Jacobsen Geophysical Research Letters Radio Science JGR: Oceans Geophysical Research Letters

Didier Mourenas Juan Muglia Catherine M. Naud Roseanna M. Neupauer Gang Lu and Paleoceanography and Suzana Camargo Martyn Clark JGR: Space Physics editors Paleoclimatology editors Geophysical Research Letters Water Resources Research Geophysical Research Letters Paleoceanography and and JGR: Space Physics Paleoclimatology

Kiwamu Nishida Renee Obringer Yuko M. Okumura Dani Or Isabelle Manighetti Earth’s Future editors Chris Patricola Martyn Clark JGR: Solid Earth Earth’s Future Geophysical Research Letters Water Resources Research

SCIENCE NEWS BY AGU // Eos.org 43 AGU NEWS

Devon A. Orme Scott L. Painter Jaime B. Palter Adriana Paluszny Tectonics editors Martyn Clark Global Biogeochemical Isabelle Manighetti Tectonics Water Resources Research Cycles editors JGR: Solid Earth Global Biogeochemical Cycles

Christopher Paola Jay W. Parker Claudia Pasquero Lenard Pederick JGR: Earth Surface editors Andrea Donnellan Andy Hogg Sana Salous JGR: Earth Surface Earth and Space Science Geophysical Research Letters Radio Science

Tim J. Peterson Carole Petit Robert Porritt Giovanni Porta Martyn Clark Tectonics editors Jeroen Ritsema Martyn Clark Water Resources Research Tectonics Geophysical Research Letters Water Resources Research

Adina Pusok Annie L. Putman Yun Qian Qubin Qin Geochemistry, Geophysics, Harihar Rajaram Zhanqing Li Marjy Friedrichs Geosystems editors Geophysical Research Letters JGR: Atmospheres JGR: Oceans Geochemistry, Geophysics, Geosystems

44 Eos // AUGUST 2020 AGU NEWS

Julianne D. Quinn Licia Ray Ann Marie Reinhold Carl Renshaw Martyn Clark JGR: Space Physics editors Rose M. Cory Isabelle Manighetti Water Resources Research JGR: Space Physics Geophysical Research Letters JGR: Solid Earth

Camilo Rey-Sanchez Cesar B. Rocha Chris Rollins Brian Rose JGR: Biogeosciences editors Andy Hogg Isabelle Manighetti Alessandra Giannini JGR: Biogeosciences Geophysical Research Letters JGR: Solid Earth Geophysical Research Letters

Zachary Ross Xiaozhou Ruan Mojtaba Sadegh Takuya Sagawa Isabelle Manighetti Janet Sprintall Earth’s Future editors Paleoceanography and JGR: Solid Earth Geophysical Research Letters Earth’s Future Paleoclimatology editors Paleoceanography and Paleoclimatology

Adam Scaife Daniel N. Scott Matthew Siegfried Eric Small Bill Randel Martyn Clark Harihar Rajaram Valeriy Ivanov JGR: Atmospheres Water Resources Research Geophysical Research Letters Geophysical Research Letters

SCIENCE NEWS BY AGU // Eos.org 45 AGU NEWS

Aleksey Smirnov Karen Smith Robert Sohn Michael Spall Monika Korte Alessandra Giannini and Hui Su Rebecca Carey Leo Oey and Laurie Padman Geophysical Research Letters Geophysical Research Letters Geophysical Research Letters JGR: Oceans

Paul W. Staten Philip Stau¦er Scott Steinschneider Claudia Stolle Suzana Camargo Harihar Rajaram Martyn Clark Andrew Yau Geophysical Research Letters Geophysical Research Letters Water Resources Research Geophysical Research Letters

Zhan Su Zhenpeng Su Tianhaozhe Sun Nicoletta Tambroni Leo Oey and Lei Lou JGR: Space Physics editors Isabelle Manighetti JGR: Earth Surface editors JGR: Oceans JGR: Space Physics JGR: Solid Earth JGR: Earth Surface

Silvia Trini Castelli Daniel Trugman Luke D. Trusel Natalie Umling James Crawford Gavin P. Hayes Mathieu Morlighem Paleoceanography and JGR: Atmospheres Geophysical Research Letters Geophysical Research Letters Paleoclimatology editors Paleoceanography and Paleoclimatology

46 Eos // AUGUST 2020 AGU NEWS

Ype van der Velde Erik van Sebille John T. van Stan Fátima Viveiros Martyn Clark Marjy Friedrichs Martyn Clark Gabriel Filippelli Water Resources Research JGR: Oceans Water Resources Research GeoHealth

Marissa Vogt Fabian Wadsworth Laura Wallace Baoshan Wang Andrew Yau Rebecca Carey Isabelle Manighetti Isabelle Manighetti Geophysical Research Letters Geophysical Research Letters JGR: Solid Earth JGR: Solid Earth

Nicholas Ward Naomi Wells Earle Williams Mathew Williams JGR: Biogeosciences editors JGR: Biogeosciences editors Minghua Zhang Journal of Advances in Modeling JGR: Biogeosciences JGR: Biogeosciences JGR: Atmospheres Earth Systems editors Journal of Advances in Modeling Earth Systems (JAMES)

Phillip J. Wolfram Teng-fong Wong Pei Xin Shiqing Xu Journal of Advances in Modeling Isabelle Manighetti Martyn Clark Isabelle Manighetti Earth Systems editors JGR: Solid Earth Water Resources Research JGR: Solid Earth Journal of Advances in Modeling Earth Systems (JAMES)

SCIENCE NEWS BY AGU // Eos.org 47 AGU NEWS

Wenchang Yang Andrew Yool Chunlüe Zhou Hejun Zhu Gudrun Magnusdottir Global Biogeochemical Kaicun Wang Jeroen Ritsema Geophysical Research Letters Cycles editors Geophysical Research Letters Geophysical Research Letters Global Biogeochemical Cycles

2019 Cited Referees Not Pictured

Rebecca Harrington Foteini Vervelidou Isabelle Manighetti JGR: Planets editors JGR: Solid Earth JGR: Planets

Robert Zimmerman Ildiko Horvath Shuguang Wang Isabelle Manighetti JGR: Space Physics editors Chidong Zhang JGR: Solid Earth JGR: Space Physics JGR: Atmospheres

Read it first on Articles are published on Eos.org before Geoscience Commits to Racial Justice. they appear in the magazine. Now We’ve Got Work to Do bit.ly/Eos-racial-justice Visit Eos.org daily for the latest news and perspectives. Internal Compass Guides Salmon’s Incredible Journey bit.ly/Eos-salmon

Evolution of the Asian Monsoon bit.ly/Eos-Asian-monsoon

The Ticking Time Bomb of Arctic Permafrost bit.ly/Eos-permafrost

Clearing the Haze Around Aerosol Assessments bit.ly/Eos-aerosols

How to Improve Space Weather Forecasting https://bit.ly/Eos-space-weather-forecasting

48 Eos // AUGUST 2020 RESEARCH SPOTLIGHT

How Much Modiﬁcation Can Earth’s Water Cycle Handle?

arth’s fresh water is essential: It helps regulate climate, support ecosystems, Eand sustain human activities. Despite its importance to the planet, humans are modifying fresh water at an unprecedented scale, and because of their damming rivers, pumping groundwater, and removing forests, humans represent the primary source of disturbances in the world’s freshwater cycle. Extensive modifications to the water cycle and perturbations to other Earth processes prompted the development of the planetary boundaries framework roughly a decade ago. Portions of Europe (foreground) and North Africa are seen in this view from the International Space Station. The framework defines the “safe operating Human activities at local and regional scales all over the planet inﬂuence the global water cycle. Credit: NASA space” for essential Earth system processes, including fresh water, and sets limits beyond which we risk rapidly and maybe irreversibly disrupting critical global systems. cycle particularly vulnerable? And how do systems around the world—we can shape In a new study, Gleeson et al. argue that the local changes in water stores and fluxes policy to avoid reaching tipping points that current definition of the planetary boundary affect regional and global processes, and vice would undermine the planet’s resilience. related to fresh water—and the methodology versa? The study invites scientists to meet the sci- for determining it—is insufficient, neither The authors delve into each of these ques- entific and ethical grand challenge of exam- capturing water’s role in Earth system resil- tions. In addressing the second question, for ining how water cycle modifications affect ience adequately nor offering a way to mea- example, they examine where ecological Earth system resilience. In outlining the sure perturbations on a global scale. The regime shifts would alter the water cycle. As grand challenge, the authors suggest initia- authors provide an overview of how water a case study, they assess the Amazon rain for- tives that can be addressed immediately by supports Earth’s resilience and propose an est. Research has suggested that if the Ama- collaborative working groups to illuminate approach for analyzing and better under- zon loses between 10% and 40% of its current better the current extent of water modifica- standing global water cycle modifications forested extent, it may transition perma- tions and the state of the water cycle. (Water focused on three central questions: What nently into savanna and disrupt water and Resources Research, https://doi.org/10.1029/ water- related changes could lead to global carbon cycles. By understanding the limits of 2019WR024957, 2020) — Aaron Sidder, Science tipping points? How and where is the water the Amazonian ecosystem—and other eco- Writer

Improving Atmospheric Forecasts with Machine Learning

eather forecasting has improved significantly in recent compared the machine learning forecasts to three benchmark fore- decades. Thanks to advances in monitoring and computing casts: daily climatology; a persistence model, which assumes that the W technology, today’s 5-day forecasts are as accurate as 1-day atmospheric state will remain constant throughout the forecast; and forecasts were in 1980. Artificial intelligence could revolutionize the Simplified Parameterizations, Privitive- Equation Dynamics weather forecasts again. In a new study, Arcomano et al. present a (SPEEDY) model, a low-resolution version of numerical weather pre- machine learning model that forecasts weather in the same format as diction models. classic numerical weather prediction models. They found that the machine learning model typically forecast the Previously, the team developed an efficient machine learning algo- global atmospheric state with skill 3 days out. It outperformed both rithm for the prediction of large, chaotic systems and demonstrated daily climatology and the persistence model in the extratropics, how to incorporate the algorithm into a hybrid numerical machine though not in the tropics, and bested the SPEEDY model in predicting learning model for dynamical systems like atmospheric conditions. temperature in the tropics and specific humidity at the surface in both In the new proof- of -concept study, the researchers build on their pre- the tropics and the extratropics. However, the SPEEDY model still out- vious work by using a reservoir computing–based model, rather than performed the machine learning model for wind forecasts more than a deep learning model, to reduce the training time requirements for 24 hours out. The authors note that overall, the reservoir computing– their machine learning technique. based machine learning model is highly efficient and may be useful The researchers trained their model using data from the European in rapid and short- term weather forecasts. Geophysical( Research Let- Centre for Medium-Range Weather Forecasts and prepared 171 sepa- ters, https://doi.org/10.1029/2020GL087776, 2020) —Kate Wheeling, rate 20-day forecasts, each of which took just 1 minute to prepare. They Science Writer

SCIENCE NEWS BY AGU // Eos.org 49 RESEARCH SPOTLIGHT

Cumulonimbus clouds advance over the Baltic Sea. Incorporating machine learning techniques into weather and climate models could improve both. Credit: Arnold Paul, CC BY-SA 2.5 (bit.ly/ ccbysa2-5)

Machine Learning Improves Weather and Climate Models

oth weather and climate models have uncertainty in subgrid-scale processes, and tion in L96. The authors found that the suc- improved drastically in recent years, as they are capable of producing fairly accurate cess of the GANs in providing accurate Badvances in one field have tended to weather forecasts and climate projections. weather forecasts was predictive of their per- benefit the other. But there is still significant But it’s still a mathematically challenging formance in climate simulations: The GANs uncertainty in model outputs that are not method. Now researchers are turning to that provided the most accurate weather quantified accurately. That’s because the pro- machine learning to provide more efficiency forecasts also performed best for climate cesses that drive climate and weather are to mathematical models. simulations, but they did not perform as well chaotic, complex, and interconnected in ways Here Gagne et al. evaluate the use of a class in off-line evaluations. that researchers have yet to describe in the of machine learning networks known as gen- The study provides one of the first practi- complex equations that power numerical erative adversarial networks (GANs) with a cally relevant evaluations for machine learn- models. toy model of the extratropical atmosphere—a ing for uncertain parameterizations. The Historically, researchers have used approx- model first presented by Edward Lorenz in authors conclude that GANs are a promising imations called parameterizations to model 1996 and thus known as the L96 system, approach for the parameterization of small- the relationships underlying small- scale which has been frequently used as a test bed scale but uncertain processes in weather and atmospheric processes and their interactions for stochastic parameterization schemes. The climate models. ( Journal of Advances in Mod- with large-s cale atmospheric processes. Sto- researchers trained 20 GANs, with varied eling Earth Systems (JAMES), https:// doi .org/ chastic parameterizations have become noise magnitudes, and identified a set that 10.1029/2019MS001896, 2020) —Kate Wheeling, increasingly common for representing the outperformed a h and- tuned parameteriza- Science Writer

50 Eos // AUGUST 2020 RESEARCH SPOTLIGHT

Space Weather Forecasting Takes Inspiration from Meteorology

very so often, the Sun unleashes powerful bursts of plasma particles and Emagnetic field structures toward Earth. These solar storms can wreak havoc on power grids, satellites, and other infrastructure, but they are difficult to predict more than a few days in advance. In a new review, Dikpati and McIntosh show- case mounting evidence that solar storms arise from solar Rossby waves, a type of wave associated with rotating fluids. Just as the 1939 discovery of Rossby waves in Earth’s atmosphere paved the way to accurate weather prediction, Rossby waves in the Sun could be key to predicting disruptive space weather in time Large- scale Rossby waves propagating around the Sun may play a role in triggering solar storms (yellow and to prepare for it. deep red regions indicate where sunspot activity is unlikely and more likely, respectively). These waves are akin On Earth, atmospheric Rossby waves arise to Rossby waves in Earth’s atmosphere, which inﬂuence weather, including potential heat waves (red arrows) from the planet’s rotation, and these large- and cold snaps (blue arrows). Credit: Simmi Sinha, University Corporation for Atmospheric Research scale meandering features help transport warm air toward the poles and cold air toward the tropics. Earth’s Rossby waves sometimes have extreme effects, such as those from 2019’s polar vortex. The researchers suggest that computational techniques developed Rossby waves in the solar plasma arise from the star’s rotation and for meteorology could inform strategies to improve solar storm predic- originate within a transitional layer known as the tachocline. Unlike tions. In the future, scientists could use observations of the Sun’s sur- Earth’s Rossby waves, solar Rossby waves are strongly influenced by face as indicators of Rossby wave dynamics deep below, potentially powerful magnetic fields. Recent observations and theoretical model- revealing harbingers of solar storms weeks, months, or even a few ing suggest that these magnetically modified Rossby waves interact with years ahead of their eruption. (Space Weather, https:// doi . org/ 10.1029/ the differing rates of rotation of the Sun’s plasma to trigger solar storms. 2018SW002109, 2020) —Sarah Stanley, Science Writer

Hardwood Forest Soils Are Sinks for Plant- Produced Volatiles

iogenic volatile organic compounds symbiotic fungus that penetrates the cells of ECM soils absorbed more BVOCs than AM (BVOCs) are carbon- containing mole- tree roots to form a network of nodes where soils, especially as temperatures warmed. Bcules released into the air by plants. sugar, gas, and nutrients are exchanged. The ECM soils were also much more active when They act as signaling molecules between trees, second type of soil was beneath trees that moisture levels were higher. Finally, ECM in a kind of airborne chemical communication, associate with ectomycorrhizal fungi (ECM), soils showed stronger seasonality, acting as and play important roles in larger climate pro- which form tiny exchange nets around plant sources of BVOCs before the growing season cesses by facilitating aerosol formation and roots but do not breach the root cells them- but then becoming strong sinks after leaf out. ozone production. Forest soils can store BVOCs selves. Scientists often find substantial differ- AM soils, on the other hand, were weak BVOC and influence the compounds’ exchange with ences in soil biogeochemistry between AM- sinks year-round. the atmosphere, functions that are affected by and ECM- d ominated forest stands. For The scientists conclude that characteriz- variables that scientists are keen to under- example, AM-associated tree species drop lit- ing forest soil by tree-associated mycorrhizal stand, such as the types of trees present and ter that decays more rapidly and cycle nutri- associations may be a good first step in cap- associated fungi growing in the soil. ents faster than ECM-associated tree species. turing landscape-scale variation in BVOC In a new study, Trowbridge et al. analyzed Overall, the new analysis shows that both transport between the land and the atmo- BVOCs in situ from two types of soil within a soil types studied work as net BVOC sinks sphere. ( Journal of Geophysical Research: hardwood forest in south central Indiana. The during the growing season; however, other Biogeosciences, https:// doi . org/ 10.1029/ first type was beneath trees that associate with factors, like temperature and moisture, are 2019JG005479, 2020) —David Shultz, Science arbuscular mycorrhizal fungi (AM), a type of critical to understanding the larger picture. Writer

SCIENCE NEWS BY AGU // Eos.org 51 RESEARCH SPOTLIGHT

How Long Was Venus Habitable?

arth and Venus are “sister worlds,” sharing a similar size, mass, examine how variations in the planet’s rotation rate and surface water and bulk composition. You wouldn’t want to visit modern-day levels might have influenced its early climate. Assuming that Venus’s EVenus, though, with its atmosphere of carbon dioxide and nitro- early atmosphere, like early Earth’s, was carbon rich and cool and that gen and surface temperatures hovering around 450°C. But our neighbor its rotation rate was slow, the team found that Venus’s climate could probably wasn’t always so inhospitable. have been stable for most of the planet’s more than 4-billion- y ear his- Deciphering what early Venus looked like isn’t easy—in part because tory—a strike against the gradually warming Sun theory. the planet’s surface is relatively young, just 300–7 00 million years old— The authors believe that simultaneous eruptions of LIPs over the past but indications from the Pioneer Venus mission suggest that its atmo- few hundred million years could have led to a runaway greenhouse sphere once contained more water than it does today. Venus also might effect by releasing large amounts of carbon dioxide into the atmosphere. have hosted liquid water at its surface, as well as plate tectonics and a The resultant drying of the planet’s surface could have driven it into a stable, temperate climate; some studies even indicate that Venus’s cli- new interior- surface dynamics regime, with newly exposed basalts— mate may have been more stable than early Earth’s, avoiding Earth’s evident on Venus today—acting as an efficient oxygen sink. icy snowball periods. In Earth’s past, LIPs have emerged sequentially in a random stochas- Theories abound about what led to Venus’s drastic transformation: tic process rather than simultaneously, which the authors note is “for- A gradually warming Sun may have left the planet hot and desiccated tuitous for life as we know it today.” But not enough is known about after a short period of habitability, or a very early magma ocean and Venus’s interior to speculate whether an uninhabitable end state is the an atmosphere of carbon dioxide and steam could have given way to the inevitable product of internal processes on Venus- like planets or even planet’s current state nearly 4 billion years ago. on Earth, for that matter. Researchers need more observations from In a new study, though, Way and Del Genio provide evidence that a Venus’s surface to better constrain its early history and further chal- shallow water ocean and habitable conditions may have persisted on lenge the magma ocean theory. Venus for as long as 3 billion years, until volcanic large igneous provinces Ultimately, a better understanding of Venus’s history will provide (LIPs) emerged simultaneously and ended the planet’s temperate period. insights into both terrestrial processes and those of exoplanets, includ- The team ran several simulations of Venus’s history using NASA’s ing whether the window of habitability is wider than currently thought. ROCKE-3D (Resolving Orbital and Climate Keys of Earth and Extrater- ( Journal of Geophysical Research: Planets, https://doi.org/10.1029/ restrial Environments with Dynamics) general circulation model to 2019JE006276, 2020) —Kate Wheeling, Science Writer

Chinese Swamp Core Reveals 47,000 Years of Monsoon History

very summer, one third of the world’s to the hydrologic and climatic processes in terns of fluctuation between relatively wet population receives rainfall from the play when the materials were deposited. and dry periods in the region over the past EEast Asian monsoon. Variations in Findings from the mineral-magnetic anal- 47,000 years. These monsoonal rainfall pat- monsoon behavior can pose flood or drought ysis enabled the scientists to reconstruct patterns are consistent with data from north- risk, so understanding how it has changed ern China and are in line with past climate over time could help clarify future risks. Wei changes resulting from gradual shifts in et al. now provide new insights into 47,000 Earth’s orbit and orientation. years of East Asian monsoon history. The results also add to mounting evidence The new research addresses a lack oflong- contrary to the traditional view that climate term, high-resolution data on past monsoon processes at high latitudes were the primary variability from southern China. To help fill driver of paleoclimate monsoon trends. The this gap, the researchers collected an 8.6- new findings suggest that tropical climate meter- long sediment core from Dahu Swamp patterns associated with the El Niño– in the Nanling Mountains of southern China; Southern Oscillation have played an import- the region’s topography makes it particularly ant role in driving long- term monsoon rain- sensitive to shifts in monsoon rainfall pat- fall patterns. terns. This research could help refine climate The research team took samples of mate- models and improve predictions of future rial every 2 centimeters along the length of shifts in monsoon rainfall patterns as climate the core and analyzed each sample’s mag- New insights into ancient trends in East Asian mon- change progresses. (Paleoceanography and netic properties to produce snapshots of min- soon rainfall could help reﬁne predictions of ﬂood Paleoclimatology, https:// doi .org/ 10.1029/ eral composition in the swamp at different and drought risk as global warming progresses. 2019PA003796, 2020) —Sarah Stanley, Science time periods. These snapshots provided clues Credit: Kamiswlswoo, CC BY-SA 3.0 (bit.ly/ ccbysa3-0) Writer

52 Eos // AUGUST 2020 AGU ADVANCES EDITORS PRESENT THE LATEST RESEARCH // EDITORS’ HIGHLIGHTS

Health Eruption and Emissions Take Credit Concerns from for Ocean Carbon Sink Changes

Combined Heat he ocean’s capacity to take up carbon McKinley et al. now provide a model- based is a balance between the amount of analysis that evaluates the ocean’s response and Pollution Tcarbon dioxide in the atmosphere and when these two effects coincided, to explain the state of the ocean. Reduced emissions in the slowing of the ocean carbon sink during in South Asia the 1990s resulted in lower atmospheric car- the 1990s. The confluence of these events bon dioxide, which slowed the ocean’s yielded a net carbon uptake by the ocean that uptake of carbon. At the same time, the was less than expected. These results high- ising global temperatures and, corre- eruption of Mount Pinatubo in 1991 produced light the importance of external processes in spondingly, increasing incidences of cooler sea surface temperatures followed by controlling decadal variability in the ocean Rextreme heat events are occurring warming in the latter part of the decade that carbon sink and point to the uncertainty across most of the world. This is even more changed the state of the ocean, which, in introduced by extreme events. (AGU Advances, concerning in South Asia, with many other turn, shifted the timing of the carbon sink’s https://doi.org/10.1029/2019AV000149, 2020) geographic factors leading to prolonged peri- slowing. —Eileen Hofmann ods of hazardous weather. The examination of potential health exposure by Xu et al. is unique. The researchers studied projections for heat events in combination with one aspect of air quality—particulate matter—for Using Saturn’s Rings the current and projected climate of India and a few other countries in South Asia. as a Seismometer The authors show that such jointed events will increase by 175% in frequency by midcen- eismology tury. The fraction of land exposed to pro- is a power- longed high particulate pollution increases by Sful tool for a factor of more than 10 in 2050. The alarming probing a planet’s increases in health exposures over just a few or star’s interior decades in South Asia pose great challenges to structure. Almost adaptation. Action addressing the combined 40 years ago, it impacts of climate change is needed across was recognized the world. (AGU Advances, https://doi.org/10 that internal oscil- .1029/ 2019AV000103, 2020) — Donald Wuebbles lations in Saturn could produce wave patterns in its rings: The rings could act as a seismometer. Now, as Manko vich de- scribes in a Com- Schematic of how oscillations (colors) inside Saturn generate spiral density waves in mentary, 20 such the rings. In reality, the waves are much more tightly wound. Observations of these patterns are known waves can be used to probe Saturn’s internal structure. Credit: Mankovich, 2020 and lead to two surprising conclu- sions. The first is that the interior half of Saturn and more is likely to be learned as modeling is not convecting, as was expected, but is techniques become more precise. Further- Decadal variability in the ocean carbon sink is stably stratified. The explanation is most more, this work opens up the possibility of attributed to (a) slowing of the partial pressure of car- likely a deep but diffuse rock-rich region. using similar approaches elsewhere, notably bon dioxide (pCO2) in the atmosphere in the 1990s The second conclusion is not only that Sat- in the ring system of Uranus. (AGU Advances, (black line) and the rapid cooling and slower warming urn is rotating somewhat faster than was https:// doi . org/ 10.1029/ 2019AV000142, recovery in response to the eruption of Mount Pina- previously thought but also that the rota- 2020) —Francis Nimmo tubo (red arrows) and (b) increasing atmospheric tion rate varies pCO2 in the 2000s (black line). The strength of the with depth. The ocean carbon sink (black arrows) was reduced by use of rings as u Sign up for the AGU Advances Digest: conditions in the 1990s and strengthened by condi- seismometers is agu.org/advances-digest tions in the 2000s. Credit: McKinley et al., 2020 still in its infancy,

SCIENCE NEWS BY AGU // Eos.org 53 POSITIONS AVAILABLE The Career Center (ﬁndajob.agu.org) is AGU’s main resource for recruitment

advertising. Atmospheric Sciences as PM2.5, ozone, deposition, and visibility. We are looking for an expert in emissions modeling, dis- AGU oers online and printed recruitment The Faculty Mathematics and Nat- persion modeling, chemical transport ural Sciences invites in cooperation modeling, data visualization, and advertising in Eos to reinforce your online with the Leibniz Institute of Atmo- advanced data analysis techniques. spheric Physics Kühlungsborn (IAP) job visibility and your brand. Visit This position is located in Reno, NV. applications for a W3- Professorship The successful candidate will have in Atmospheric Physics at the IAP employers.agu.org for more information. demonstrated expertise in simulation and the Institute of Physics, start- of atmospheric systems (hypothesis ing at the earliest possible date setting and testing using simula- (within budget considerations). tions; analysis and exposition of We are looking for an outstanding simulation results; application of Packages are available scientist in the field of atmospheric Eulerian air quality models), manip- physics. Emphasis should be on the ulating large datasets (e.g., observa- for positions that are experimental or theoretical explora- tions collected using aerometric tion of the stratosphere, mesosphere instruments, emissions data, and and lower thermosphere. The main model output data), and conducting SIMPLE TO RECRUIT task is leading the Leibniz Institute of experiments with numerical models Atmospheric Physics (IAP). In addi- ◆◆ online packages to access our Career Center that are guided by empirical data. The tion, the candidate is expected to col- research activities associated with audience laborate with the Institute of Physics this position will also build the of the University, to participate in ◆◆ knowledge base related to air quality 30-day and 60-day options available joint projects and to contribute to the management strategies. activities of the Interdisciplinary Fac- ◆◆ prices range $475–$1,215 Required Qualifications ulty. The teaching duties comprise • Ph.D. in chemical engineering, courses and seminars in atmospheric mechanical engineering, civil engi- physics for master and PhD pro - neering, atmospheric science, chem- CHALLENGING TO RECRUIT grams. istry, or a related discipline from an Additional Information: ◆◆ online and print packages to access the accredited institution. Prof. Dr. Stefan Lochbrunner • 2- year minimum experience in wider AGU community Phone: +49 381/ 498- 6960 atmospheric or closely related science email: stefan . lochbrunner@ uni research (candidates with less or no ◆◆ 30-day and 60-day options available - rostock .de experience will be considered for a **** ◆◆ postdoctoral fellowship). prices range $795–$2,691 Qualifications are as per §58 of the • Demonstrated expertise in using Higher Education Act of the State of regional air quality models such as Mecklenburg- Vorpommern (LHG CMAQ and CAMx including ability to M-V): completed university studies, DIFFICULT TO RECRUIT use emissions processing software as doctoral degree, post- d octoral thesis ◆◆ well as post- p rocessing of model our most powerful packages for maximum (“Habilitation”), teaching experience results. or equivalent qualifications, usually multimedia exposure to the AGU community • Demonstrated ability to conduct earned within the context of junior sensitivity simulations using brute- ◆◆ 30-day and 60-day options available professorship. force methods or using various sen- A special focus is placed on aca- sitivity tools available with the ◆◆ prices range $2,245–$5,841 demic achievement and teaching regional air quality model. qualifications as well scientific orga- For more detailed information nization and academic administra- about DRI, please visit the careers tion. For this reason, candidates FREE TO RECRUIT section of our website at www . d ri should describe previous teaching . edu. ◆◆ these packages apply only to student and results, ideas regarding future teach- graduate student roles, and all bookings are ing (including didactic lesson planning) and their prior experience in SUSTech (http:// www .sustech . edu subject to AGU approval academic and scientific management. . cn/en) was founded in 2011 with Applications with the usual docu- public funding from the Municipal ◆◆ eligible roles include student fellowships, ments (full CV, a complete list of aca- Government of Shenzhen. A thriving internships, assistantships, and scholarships demic and professional background, metropolis of over 20 million people publications, teaching experience, bordering Hong Kong, Shenzhen any additional qualifications, a sum- has often been referred to as the “Sil- mary of grants and sponsored icon Valley of China” with strong research activities and a description telecommunication, biotechnology of future research plans) should be and pharmaceutical sectors. Widely Eos is published monthly. sent no later than 31st July 2020 to the regarded as a pioneer of higher- University of Rostock, Dean of the education reform in China, SUSTech Deadlines for ads in each issue are published at sites . agu.org/ Faculty of Mathematics and Natural aims to become a top- tier interna- media-kits/eos-advertising-deadlines/. Sciences, Wismarsche Straße 45, tional university that excels in inter- 18057 Rostock or by e- mail to dekan disciplinary research, talent devel- Eos accepts employment and open position advertisements from . mnf@ uni - rostock . de. opment and knowledge discovery. In governments, individuals, organizations, and academic institutions. the latest Times Higher Education We reserve the right to accept or reject ads at our discretion. The Desert Research Institute (THE) World Universities Rankings (DRI) Division of Atmospheric Sci- 2020, SUSTech was ranked as the 9th ences is seeking a Postdoctoral among the mainland China univer- Eos is not responsible for typographical errors. Fellow or an Assistant Research sities and the No. 1 young university Professor to advance a research under 50-y ear old. Internationaliza- program on air quality topics such tion is a hallmark of SUSTech where

54 Eos // AUGUST 2020 POSITIONS AVAILABLE

1. OVERVIEW OF SUSTECH Shenzhen, also known as The Great Eagle City, is China’s ﬁrst Special Economic Zone and recognized as the driving force for the reforming and opening up of the Chinese economy over the past thirty-forty years. In 2017, Shenzhen was selected by the Chinese Government to develop into one of the two International Ocean Cities in China.

Southern University of Science and Technology (SUSTech is a public research university established in 2011, funded by Shenzhen Municipality. Widely regarded as a pioneer and innovator in collectively moving China’s higher education forward to match China’s growing role in the international arena, SUSTech aspires to become a globally-renowned university that contributes signiﬁcantly to the advancement of science and technology by excelling in interdisciplinary research, nurturing creative future leaders and creating knowledge for the world. More information can be found in website: http://www.sustech.edu.cn.

2. OVERVIEW OF DOSE The Department of Ocean Science and Engineering (DOSE) was founded in July 2015, aiming to build a team of high quality and international faculty, an “Into the deep ocean” scientiﬁc research platform, an international research program in Oceanography, and an internationally leading center in Ocean Engineering. Our long-term vision is to station in Shenzhen, devote our research to the three oceans of the globe!

Research areas in Ocean Science include Marine Geophysics/Geology (Solid), Physical Oceanography (Liquid) and Microbial Oceanography/Biogeochemistry etc. Those in Ocean Engineering include Offshore and Coastal Engineering, Offshore Energy & Resource and Ocean Acoustic & Fiber Technology etc.

As in January 2020, DOSE has 17 faculty staff, including ﬁve Chair Professors, two Professors, one Associate Professor and nine Assistant Professors. Our plan is to increase the number to 40 within a short period of time.

3. OCEAN SCIENCE & FACULTY OPENINGS We plan to appoint a Chair Professor in Physical Oceanography (Liquid), and about 10 Assistant/Associate/Full Professors in all ﬁelds in Ocean Science.

4. OCEAN ENGINEERING & FACULTY OPENINGS Ocean Engineering is a strategic discipline for development in DOSE and SU Tech. We plan to appoint at least 12 faculty staff at all levels including Assistant/ Associate/Full/Chair Professors in the following areas:

· Structural Engineering · Offshore Engineering & Resource · Fluid mechanics · Pipelines · Wind engineering · Ocean Acoustic & Fiber Technology · Wave-structure interaction · Ocean Engineering equipment · Geotechnics · Ocean Engineering surveying · Ocean Engineering materials

5. REMUNERATION & START-UP PACKAGES I. Income and Beneﬁts 1. Globally competitive (including US & HK) salary 2. Apartment inside campus (depending on remaining apartment quantity) or housing allowance of leasing outside; 3. Social Insurance: Retirement insurance, medical insurance, unemployment insurance, industrial injury insurance, maternity insurance and housing accumulation funds. Special health insurance negotiable. 4. Shenzhen living subsides of CNY 1.6~3 million and other living subsidies according to your talent level. II. Lab space no less than 150 square meters. III. SUSTech start-up fund of CNY 1 million and other research funds according to your talent level.

6. HOW TO APPLY Candidates must have a proven and consistent track record of high-quality scientiﬁc publications, good communication skills and relevant teaching experience corresponding to their stages of career. English are required languages for teaching.

To apply, please submit the following materials electronically to [email protected]: 1) Cover letter; 2) Curriculum vitae (with a complete list of publications); 3) Statement of research, including justiﬁcation of your suitability for being a faculty member in DOSE and your future research plan; 4) Statement of teaching, including your teaching philosophy and teaching interests; 5) Three representative full papers; and 6) Names and contact information of ﬁve referees.

All applications will be considered shortly after received and offers will be made to those who are qualiﬁed.

SCIENCE NEWS BY AGU // Eos.org 55 POSITIONS AVAILABLE

English is a primary instructional planning to fill additional two dozen earliest possible date. The success- • Weather and climate extremes language. tenure- track/tenured positions over ful applicant will be assigned to the • High- performance scientific The SUSTech | School of Environ- the next 3- 4 years to enhance and University of Rostock as a university computing and data science in cli- mental Science and Engineering expand existing faculty and research professor with a reduced teaching mate research (ESE) (http:// ese . sustc . edu . cn/ en/) strengths. Globally competitive obligation (2 SWS) and in parallel will Appointment: Each position will was established in May 2015. The (including US and Hong Kong) sala- undertake the position of the Direc- be supported for two years, contin- mission of ESE is to become: an inno- ries and benefit packages will be tor of the DLR Institute for Solar- gent upon funding availability and vative training ground for cultivating offered. New hires may also be eligi- Terrestrial Physics in Neustrelitz. satisfactory performance. top talent in environmental fields; an ble for additional government sup- Women with essentially equiva- Application Instructions: international center of excellence for port such as the Shenzhen City’s Pea- lent qualifications will be given pri- Applications must include the follow- environmental research; a leading cock Program and many others ority provided that the personal qual- ing: platform for innovation and industri- (http:// www . sustech .edu .cn/ en/ ities of a male candidate are not 1. A cover letter containing a state- alization of advanced environmen- faculty _ en). better suited for the position. ment of interest. tal protection technologies; and an Applicants are required to have a Further information about the 2. A curriculum vitae. influential think- tank for environ- Ph.D. in environmental science and position, and application process can 3. Three references with complete mental sustainability. Currently, ESE engineering, earth and atmospheric be found under https:// www . uni contact information. has over 65 f ull- time faculty and sciences, or related disciplines. Post- - rostock . de/ en/ stellen/ professuren/ Review of applications will begin research staff, including the recipi- doctoral experience is preferred but immediately and the positions will ents of numerous national and inter- not required. Candidates must have a remain open until filled. Application national awards and honors. ESE is proven and consistent track record of Ocean Sciences submission information can be organized into three broadly-defined high- quality scientific publications found: groups (programs): Environmental, and good communication skills. To Postdoctoral Positions at iHESP, https:// tamus . wd1 Water, and Global Change (including apply, submit the following materials Texas A&M University . myworkdayjobs .com/ en -US/ atmospheric science). The school is electronically to iese@ sustech .edu Overview: The International Lab- TAMU _ External/ job/ College- Station home to the State Environmental . cn: 1) Cover Letter; 2) Curriculum oratory for High- Resolution Earth - TAMU/Postdoctoral -Research Protection Key Laboratory of Inte- Vitae (with a complete list of publica- System Prediction (iHESP, https:// - Associate - 3_R - 017416 - 1 grated Surface Water- Groundwater tions); 3) Statement of research and ihesp .tamu .edu) – a QNLM- TAMU- https:// tamus . wd1 Pollution Control as well as the Shen- teaching interest; 4) PDFs of three NCAR partnership seeks to hire two . myworkdayjobs .com/ en - US/ zhen Institute of Sustainable Devel- recent publications; and 5) Names postdoctoral researchers in high- TAMU _External/ job/ Bryan opment. and contact information for 3-5 ref- resolution climate modeling, analysis - TAMHSC/ Postdoctoral - Research Applications are invited for faculty erences. All positions remain open and prediction research. The postdocs - Associate - 1_R - 029797 - 3 positions at all ranks. Areas of inter- until filled. For additional informa- will be supported and managed iHESP is within the Department est include, but are not limited to, tion, please contact Yuanyuan Su through the iHESP main office at of Oceanography which is a part of environmental toxicity, soil and (email: suyy@sustec h. edu. cn, phone: Texas A&M University, College Sta- the College of Geosciences at Texas groundwater contamination and + 86- 755- 8801- 0822). tion, Texas, but are expected to work A&M University along with Depart- remediation, ecohydrology and eco- closely with scientists at NCAR and ments of Atmospheric Sciences, logical restoration, environmental The Faculty of Mathematics and QNLM. Geography, and Geology & Geophys- health, environmental microbiology Natural Sciences at the University Area of research: We seek candi- ics, as well as the Berg Hughes Cen- and biotechnology, atmospheric of Rostock is involved in a joint dates with strong background in ter, the Texas Center for Climate chemistry and air pollution control, appointment procedure with the ocean, atmosphere and climate Studies. The Department web site wastewater and solid waste treatment German Aerospace Center dynamics, and/or high- performance http://ocean . tamu. edu contains a full and recycling, remote sensing and (Deutsches Zentrum für Luft- und scientific computing. Specific areas of description of our program. environmental monitoring, earth Raumfahrt e.V.) (DLR) and invites expertise include, but are not limited The Texas A&M System is an Equal system modeling, macroecology applications for a University Pro- to: Opportunity/Affirmative Action/ and biodiversity, global change and fessorship (W3) for Solar Terrestrial • Subseasonal- to- decadal climate Veterans/Disability Employer environmental sustainability. ESE is Physics to commence work at the prediction and predictability committed to diversity.

CLIMATE SCIENTIST Use science to help transform the climate conversation. PLACE Climate Central seeks a highly motivated, creative, mid- to senior-level natural or social scientist to spearhead a strategic effort to build an unprecedented assemblage of ways to link diverse local observations and events to climate science, impacts, YOUR AD and solutions for dissemination to social media, journalists, and other key messengers, in near real-time. HERE The successful candidate will have a Ph.D. in a relevant natural Visit agu.org/advertise to learn more or social science plus more than 10 years of research and media about employment advertising with AGU. experience.

For more information, visit: https://www.climatecentral.org/what-we-do/jobs#climatescientist.

56 Eos // AUGUST 2020 POSTCARDS FROM THE FIELD

Talofa from Ofu!

The U.S. Geological Survey (USGS) and the National Park of to Ofu’s pools and is thus a likely pathway for transporting American Samoa (NPSA) are collecting data on Ofu, in the nutrients that may be driving the detrimental algal outbreak. Manu‘a Islands of American Samoa. NPSA’s Ofu unit contains more than 80 species of corals that are uniquely tolerant to The photo shows Cordell Johnson tending to a USGS- developed thermal stress: Daily temperatures in the pools frequently radon-measuring buoy that enables scientists to quantify SGD exceed the local coral bleaching threshold of 30°C. Corals rates to the coral reefs. The goal is to constrain the SGD- growing here provide insights into potential coral acclimati- associated nutrient loading and identify the patterns of impact zation and adaptation to climate change. in Ofu’s pools, thus leading to future management decisions to protect these important corals. Ofu’s pools are experiencing an outbreak of Valonia fastigiata algae that is overgrowing and killing healthy coral colonies, —Curt Storlazzi, Pacific Coastal and Marine Science Center, implying a shift in the local nutrient regime toward eutrophi- U.S. Geological Survey, Santa Cruz, Calif. cation. Submarine groundwater discharge (SGD) is increasingly recognized as an important vector for transporting nutrients into the coastal ocean, often leading to eutrophication View more postcards at bit.ly/Eos-postcard and algal blooms. SGD is the dominant source of fresh water