In the race to understand the Earth's changing climate and the implications for humanity, EAPS researchers are harnessing data in novel ways.
BY LAUREN HINKEL I EAPS NEWS
Every time mercury creeps up in the thermometer, the rain gauge stays dry longer, or a wave encounters an instrument buoy, scientists gain a data point. The Earth system has been providing volumes of this information on the climate and environment for over 4 billion years. Centuries ago, curious individuals began tapping into this data, carefully recording observations into notebooks. Through their nascent record-keeping, early scientists began to make critical inferences about how our planet works. But this stream of information was just a trickle. It wasn't until the inception of the modern instrumental observation record—covering around the last 100 years or so—that data started flooding in, and bigger trends and connections emerged.
But systematizing how data are documented is only a fraction of the story. Geoscience has been further revolutionized by advances in instrumentation and computation. Beyond temperature and precipitation, wind speeds and sea levels, scientists now are able to measure, and model, much more complex and nuanced Earth systems—from the mechanisms of glacier behavior to the biogeochemical cycles which control ocean productivity, to the dynamics of the carbon cycle, and heat transport around the globe.
Together, the collective body of work from contemporary researchers paints a fuller picture of how our Earth functions, its resiliency, and also of climate extremes past and present.
But this same expanding body of work shows that Earth's current climate resides in a delicate equilibrium, and points to a growing issue that threatens to upend it: climate change from anthropogenic carbon emissions. Global change brings with it the potential for cataclysmic effects on humanity—flooding of coasts where roughly 40% of the global population lives, drought and food shortages, stronger and more frequent storms, risks to infrastructure, and implications for human health, to name a few. Risk and impact analyses estimate that climate change could cost the U.S. alone billions of dollars per year, with the global figure far exceeding that.
To understand the capacities and limitations of our global system, and to support research-backed policies and action on how to tackle climate change, researchers like those in MIT's Department of Earth, Atmospheric and Planetary Sciences (EAPS) interrogate enormous datasets— unfathomable to scientists from centuries ago.
A WIRED FUTURE FOR CLIMATE SCIENCE
In a paper written for the Institute of Electrical and Electronics Engineers, EAPS Principal Research Scientist Sai Ravela and co-authors describe how in recent history the geoscience sector has undergone a paradigm shift in research methodology as it transitioned from, "a data-poor field to a data-rich field,"—a transition owing in part to the explosion of innovation in remote sensing tools such as GPS, topographic and bathymetric LIDAR, drone technologies, deep sea drilling vessels, and nano satellites.
In addition, most geoscience datasets are publicly available, making the application of data science methodologies an easier proposition.
"The growing availability of big geoscience data offers immense potential for machine learning...to significantly contribute to geoscience problems of great societal relevance," the authors note.
DRILLING INTO THE DATA
Unlike datasets from other fields, geoscience observations are governed by physical laws, providing the foundation of physics-based models, which generate large volumes of simulation data for researchers to probe for connections.
But when it comes to observation collection, climate science data inherently varies in quality and quantity, creating both challenges and opportunities for computation. For example, datasets may include both direct measurements—like ocean surface temperatures—and proxies, which indicate conditions that cannot be measured directly. They can also be highly dimensional, as phenomena can extend over large areas and through several layers of the oceans, earth, and atmosphere.
When studying a subject as physically large and complex as the Earth, there will also be areas of poor or missing data —there sometimes exist amorphous boundaries in time and space when observing things like eddies and cyclones; infrequent events, by their very nature, may generate datasets which are sparse.
And, of course, as the availability of better data is skewed toward the present day, the difficulties for analysis and modeling are compounded.
These challenges are where the most exciting opportunities for interdisciplinary collaboration between the geo- and data sciences live. Writing for Communications of the ACM, Ravela and colleagues made the compelling argument: "Many aspects of the geosciences pose novel problems for intelligent systems research...Overcoming these challenges requires breakthroughs that would significantly transform intelligent systems, while greatly benefitting the geosciences in turn."
TAKING IT TO THE NEXT LEVEL
A new consortium of researchers from Caltech, MIT, the Naval Postgraduate School and the Jet Propulsion Laboratory are taking on the interdisciplinary challenges for data and geoscience in a big way. Their goal: to build a new breed of climate model, leveraging data assimilation and machine learning to provide more precise and actionable predictions.
And they're starting from scratch.
With support from Schmidt Futures, EAPS Cecil and Ida Green Professors of Oceanography Raffaele Ferrari and John Marshall are leading MIT's contribution to the project. The Climate Modeling Alliance (CLiMA) will combine Earth observations and high-resolution simulations in a model that represents important small-scale features— like cloud patterns, ocean eddies, and shifting sea ice— more reliably than existing models.
Small uncertainties can have outsized effects when it comes to modeling a system as complex as Earth's climate. "The ocean soaks up much of the heat and carbon accumulating in the climate system. However, just how much it takes up depends on turbulent eddies in the upper ocean, which are too small to be resolved in climate models," says Ferrari, giving one such example.
Next-generation graphics processing units (GPUs) and coding in Julia (the MIT-deveIoped programming language developed for paral let and distributed computing), will allow the team to turbocharge their calculations. The new model will be able to "zoom in" from the current 100 kilometer scale grid to just one kilometer square. But even then, the mountain of computing power it would take to resolve every cell on the globe at this fine resolution, over projections of decades, would be impractical, if not impossible. That's where machine learning comes in.
The team will develop fine-resolution simulations for small-scale phenomena in selected regions of the globe and then nest these simulations in the larger model—informing the modeling of small-scale processes everywhere else. From there, they'll assimilate datasets from real-world observations—like readings from a fleet of thousands of autonomous floats—into the synthetic data and "teach" the model to improve itself in real time. The result, Ferrari says, should enable a leap in accuracy, reducing uncertainty in climate projections by at least half.
"The pace of geoscience investigations today can hardly keep up with the urgency presented by societal needs to manage natural resources, respond to geohazards, and understand the long-term effects of human activities on the planet....recent unprecedented increases in data availability together with a stronger emphasis on societal drivers emphasize the need for research that crosses over traditional knowledge boundaries." 1
"Anything to reduce that margin [of uncertainty] can provide a societal benefit estimated in trillions of dollars," says Ferrari. "If one knows better the likelihood of changes in rainfall patterns, for example, then everyone from civil engineers to farmers can decide what infrastructure and practices they may need to plan for."
MAKING OUR BEST SHOT
At MIT Commencement 2019, former New York mayor Michael Bloomberg charged the MIT community: "The challenge that lies before you—stopping climate change—is unlike any other ever faced by humankind...The stakes could not be higher."
An undertaking of this magnitude echoes that of the Apollo moon landing roughly 50 years ago, which required a culture shift, an "all hands on deck" approach, with intense computational efforts and engagement of keen minds. And MIT scientists and graduates delivered. The Apollo effort established a baseline for future challenges to humanity—the proverbial moonshot.
To confront the challenge of our time, MIT again is stepping up with its "Plan for Action on Climate Change", launched in 2015. Here, EAPS plays a key role—bringing fundamental science to bear on our understanding of the global system. Not only are EAPS researchers investigating crucial phenomena like cloud formation, ocean currents, erosion patterns, and storm behavior, but they are also weighing-in on the plausibility of geoengineering applications and advancements like carbon sequestration and geothermal energy sources.
Computational advancements have catalyzed these efforts, and MIT's latest investment in the new Schwarzman College of Computing will help to further expand data science and machine learning techniques in the geosciences. Investments in research and collaborations like these are closing the knowledge gap on how the Earth system functions and refining what are already good projections of future climate variations—leaving no doubt of the credibility of the science and the risks posed by climate change. And as we look forward, climate findings will be a key factor to help society develop policy and use capital wisely—investing in infrastructure changes and better urban planning, as well as developing mitigation strategies and even reversal technologies.
"All of you are part of an amazing institution that has proven...human knowledge and achievement are limitless," Bloomberg said. "In fact, this is the place that proved moonshots are worth taking."
1. Intelligent Systems for Geosciences: An Essential Research Agenda; Y Gil. S A. Pierce, H. Babaie, A. Banerjee, K. Borne, G. Bust. M. Cheatham. I. Ebert-phon C. Gomes. M Hill,]. Hore!, L. Hsu j Kinter, C. Knoblock, D, Krum. V: Kumar, P. Lermusiaux, Y.: Liu, C North, u Pankratius, S. Peters. B. PIG/e, A. Pope. S Ravela, J. Restrepo. A. Rid!ey, H. Samet_ S. Shekhar; Communications of the ACM, Jonucny 2019; Vol. 62 No.
Story photos courtesy EAPS Assistant Professor Brent Minchew: On on Langjökull Ice Cap, Central Iceland, EAPS alumnus Mark Simons PhD '95, the John W. and Herberta M. Miles Professor of Geophysics at Caltech and Chief Scientist of NASA's Jet Propulsion Laboratory, installs a suite of instruments to measure meteorological conditions and glacier flow, complementing Interferometric synthetic aperture radar data collected from NASA's Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) instrument. From the UAVSAR control station, researchers monitor the flight path of the aircraft over Iceland.
In this issue
For further information on giving opportunities or creating a named fund to benefit the Department of Earth, Atmospheric and Planetary Sciences, please contact:
Senior Development Officer
Earth, Atmospheric and Planetary Sciences at MIT
617 253 5796
Keep up to date with all things EAPS: subscribe to our newsletter - firstname.lastname@example.org