Scientists’ Journal

Summer 2023 Issue

Celebrating the arrival of the third-generation machine at the NCAR-Wyoming Supercomputing Center, Derecho, which is a HPE Cray EX supercomputer, this issue focuses primarily on NCAR’s long standing relationship with Cray computers. It includes a brief history of NCAR’s acquisition of the Cray-1 in the 1970’s, remembrances of what scientific computing was like in that era, and summaries of new research that will be conducted on Derecho. This issue took longer than anticipated to produce in part because of the care necessary to record and report that history with the diligence it deserves. As always, this newsletter serves to share stories among NCAR’s labs and generations of scientists. We welcome articles that contribute to this goal.


Signing the contract for the CRAY-1A computer. Photo credit UCAR.
The signing of the contract between UCAR and Cray Research, Inc. to acquire the CRAY-1 on May 12, 1976. Top row: Lionel Dicharry, Paul Rotar, G. Stuart Patterson, Clifford Murino, and John Firor, all of NCAR. Bottom row: Noel Stone, assistant secretary of Cray Research, Inc., Seymour Cray, Francis Bretherton, and Harriet Crowe of UCAR.

Scientific computing has experienced a period of relative stability in the three decades since the adoption of the Message-Passing Interface standard, which saw computations distributed among increasingly powerful central processing units. But with the arrival of Derecho, an HPE Cray supercomputer, NCAR transitions into a new era of computing in which general-purpose graphics processor units make up a significant portion of the machine’s computing power.

The 1970’s were a similar period of computing transition and not only heralded the arrival of NCAR’s first Cray supercomputer, the Cray-1, but also a shift in I/O and data management.

G. Stuart Patterson, then the director of NCAR’s Computing Facility, describes the era, “Where we were going with the Computing Facility was a total change of how it worked, the equipment it had and what you could do with it. Number one at the heart of it was to get people off of punch cards and to get them onto terminals. The second was to have a really massive data management facility. We had 25,000 half-inch tapes sitting beneath the front steps of NCAR where it would leak in a rainstorm.”

The Computing Facility also needed to replace its workhorse supercomputer, the Control Data Corporation (CDC) 7600.

Paul Rotar, a longtime NCAR employee who eventually became head of the Systems Section, knew Seymour Cray, the brilliant engineer who co-founded CDC. At the time, CDC was one of the major computer companies in the United States. Although tech geniuses weren’t as widely known in the 1970’s as they are today, Cray was a revered figure. He graced magazine covers and an article in the Washington Post, reflecting on his life, described him as, “the Thomas Edison of the supercomputing industry.”

Rotar said he would check in with CDC to get some insight into what their next generation computer would look like. He came back from a phone call, and according to Patterson’s retelling, said, “Hey! Guess what? Seymour’s left CDC and he’s working on a new computer up in Chippewa Falls.”

Rotar and Patterson scheduled a trip to visit the newly organized Cray Research, Inc. in Chippewa Falls to get a sense of how near the new machine was to a finished product. Patterson remembers, “It had a lot of aspects of reality about the whole project. It wasn’t close to being finished. The software was way out there but it looked like something that was going to be very doable and would have a whole lot of power to it. Particularly the vector architecture that he was talking about…that was just off the wall!”

The trip convinced Patterson and Rotar that Cray’s machine would be a viable candidate for NCAR’s next supercomputer. Upon returning to Boulder, Patterson and a team consisting of Ed Wolff, then NCAR-director John Firor, and others, worked for a year to properly structure the request for proposal (RFP) for the new machine. “We made sure that we were in total control of putting out an RFP. So when that went out we had full control over what we were benchmarking, how we were benchmarking it and we were divorced from the government regulations on how you did an RFP. That was a big legal step. We had to separate UCAR, make sure UCAR was an independent legal operational entity from NSF and the federal government.”

Structuring the financing for the acquisition also required creativity. Patterson explained, “The price to us for the Cray was $10 million. The way the government works is that you can’t buy anything. So you’d have to lease it from them.” Based on the funding from NSF, it would have taken four years to accumulate enough funds to acquire $10 million but the computer companies expected the funds on a shorter timescale – two years.

Fortunately, Ray Chamberlain, then the president of Colorado State University (CSU) was on the UCAR Board of Trustees. In Patterson’s retelling, Chamberlain spoke up at a UCAR board meeting, “You know that’s crazy. You shouldn’t have to pay the lease rate. I tell you what, CSU can float an industrial development bond for $10 million and then we can lease it back to you at a reasonable interest rate and you can pay it off at a reasonable rate.”

With the financing in place and Cray Research, Inc. selected as the winning proposal, Seymour Cray visited NCAR to sign the contract to deliver the newly designed Cray-1. Patterson arranged for Cray to deliver a rare public lecture to NCAR staff after the signing.

“We set up a conference room auditorium and we probably had about fifty people from the Computing Facility come in, some other people.” Patterson continued, “Seymour gave his talk and afterwards I said, ‘Seymour would be happy to answer your questions’ and there was total silence from the audience. Not a word.”

Patterson invented a couple questions to break the tension, the lecture adjourned, and the auditorium emptied. He approached some of the staff afterwards wondering, “Why didn’t you ask any questions?” The staff looked at each other, trying to adequately convey how intimidating it was to be in the presence of such a genius. Finally, one staff member responded, “How do you talk to God?”

After the initial fanfare, the staff got to work making the new computing system available for the scientists. There were some hiccups. Patterson remembers, “We did have big problems with the memory system, the large-scale, the terabit memory system. So I spent a lot of time on that. And we had a lot of problems with the networking. Networking turned out to be extremely difficult to the point where we had to shut down a project that we had going on for well over a year and start it again.”

The Computing Facility had the benefit of having an incredibly capable manager: Margaret Drake. “Margaret at that time was probably managing all of the programming support that the Computing Facility was supporting to scientists. She had about forty programmers, I think, working for her, of which about a third were women,” said Patterson. “In my experience I’ve never seen a better manager in my life.”

For the computer itself, he remembers, “The Cray was not a big deal to get running and running well. I don’t remember any significant issues that I had to deal probably took at least a year to break in things or something like that.”

With the numerical models up and running, producing copious amounts of output, Patterson envisioned NCAR becoming a leader in the field of computer graphics but anticipated some difficulties, “I felt it was going to be a tough sell to the scientists at NCAR who were just really very much focused on computation and not focused on the presentation or analysis of the data. That’s where I wanted to go and I felt that was going to be a hard political battle.”

Patterson remained unsure of how to proceed, until one day Seymour Cray came to him to ask, “How would you like your own R&D lab?”

Seymour Cray developed a pattern over the course of his career of starting up a company, developing a successful computer, getting frustrated with the management and then leaving to start a new company. So, just like he left CDC to start Cray Research, Inc., he planned to start a new company to work on the next iteration of supercomputer design. Patterson joined him, resigning from NCAR to become president of Cray Labs and set out to construct a building to house the new company. After considering a few locations, they settled on a site north of downtown Boulder and began construction of a C-shaped building at 3375 Mitchell Lane.

While the company didn’t last, the building remains. It was purchased by UCAR in 2009 to expand the Foothills Lab campus. Only now it is known by a different name: the Anthes Building.

G. Stuart Patterson. Photo credit UCAR.
G. Stuart Patterson

This article was authored by Ben Johnson (CISL) after an informal conversation with G. Stuart Patterson. Patterson earned his SB in chemical engineering (1957) and SM in nuclear engineering (1959) from MIT. He completed his PhD in mechanics (1966) at Johns Hopkins University under the supervision of Owen Phillips and Stanley Corrsin.

Patterson spent a sabbatical year at NCAR from 1970-1971 while he was a professor of engineering at Swarthmore College. He returned to NCAR as the director of NCAR’s Computing Facility from 1973-1979 before leaving to become president of Cray Laboratories. He became a serial entrepreneur, founding several companies over his career, culminating with a position as Chief Technical Officer for OR Manager, Inc., a company cofounded by his wife, Pat Patterson.

A Brief Recollection of My Early Days of Computing at NCAR

Contributed by Annick Pouquet, Part-Time Research Scientist, LASP; & Emeritus, NCAR

As an Advanced Study Program (ASP) post-doctoral scientist at NCAR starting in the Fall of 1973, I had a discussion with G.S. Patterson (SCD, NCAR) together with U. Frisch (Observatoire de Nice), and I quickly realized the potential of performing accurate pseudo-spectral direct numerical simulations (DNS) of turbulent flows [12], in my case having in mind the problem of the generation of magnetic fields (or dynamo effect) in the Sun, the stars and the Universe at large. I started this project in my second year at NCAR and pursued it for quite a while, using periodic boundaries [47911], while Peter Gilman and his colleagues at the time were tackling at HAO the solar dynamo problem in a rotating spherical shell, a very complex task [6]. A bit earlier, Ulrich Schumann and Jack Herring were comparing properties of closures of turbulence and DNS [3]; Eric Siggia and Stu Patterson were already measuring intermittency property of turbulent flows [5] and of vortex tubes [8], while Greg Holloway was studying, again with models and DNS, the stirring of tracer fields in the ocean [10]. NCAR graphical tools were also being developed at the time and proved very useful to many. One should note that several of these early papers were co-authored by ASP fellows, ASP playing a central role in the dissemination of NCAR savoir-faire for the community at large.

Only much later, after I came back to NCAR in the 2000s, did I move with my team to the task of unraveling some of the properties of rotating and/or stratified turbulence as it occurs in the atmosphere and the ocean, stressing in particular the central role the waves play in dynamically shaping the structures, governing the scaling laws, as well as the transport and dissipation properties of such complex media [121314151617], as Alfven waves are shaping in some way the dynamics of conducting flows.

I did not realize that, in the 70s, not only were the numerical method and impressive computational power entirely new to me, but the pseudo-spectral methodology was also rather new for the open scientific community at large. Indeed, the advance in speed procured by the Fast Fourier Transform algorithm was phenomenal and, following the leadership of G.S. Patterson and S.A. Orszag [12] and under the strong guidance of Stu Patterson, we were able to produce perhaps the first such DNS in the magnetohydrodynamic (MHD) “turbulence” framework, at a grand numerical resolution of 323 grid points on the CDC 7600 [4]. Rather impressive at the time, this computation is feasible today on your smart phone, and of course the turbulence was not quite there yet, due to the lack of a sufficiently large ratio of excited scales.

Soon after, the Cray-1 arrived at NCAR and I pursued this line of work and made use of the Cray with other French colleagues to tackle a few other problems of MHD turbulence, such as showing the growth of magnetic fields even in the absence of helicity [7], or the dynamics of current sheets in two-dimensional ideal (non-dissipative) MHD [9], or the problem of growth of the correlations between the velocity and magnetic field as a signature of the role of Alfven waves in the dynamics of conducting turbulent flows [11]. Progress was still slow at the time, everything had to be learned and the scientific community had to be convinced of the realisability and the reliability of such an approach. Difficult to believe perhaps today, but still this research was not considered as numerical experiments but rather as oddities perhaps. I do recall long evenings going into the night where the room was often filled with the French team on the one end, and the Spanish one (e.g., J.M. Massaguer), working on convective flows with Juri Toomre and Niels Hurlburt, on the other hand. We did achieve and we held for a while a few world records for NCAR, including using the NCAR computers, up until the early 2010s [1215]. Indeed, we have pursued this type of work to this day in the US, in France and elsewhere as a succession of Crays were being made available, then followed by many other computers with numerous architectures and various technical improvements. THAT is another story …

Supporting Cutting-Edge Science in NCAR’s
Computing Facility in the 1970’s

Contributed by Dick Valent

Before the CRAY-1 Serial 3 arrived at NCAR in 1977, I’d worked in the Computing Facility long enough to know that many of our resident and university scientists wanted to solve larger problems than our existing 6600 and 7600 CDC computers could accommodate. I often heard this lament when I was helping the scientists and programmers use the computers. The 7600 offered 65K 60-bit words of small core memory [18, pages 2-4], and the 6600 the same [19].

In fact, one could manage out-of-core computations thanks to the 7600’s Large Core Memory hardware and its accompanying LCM subroutines. These routines allowed you to transfer blocks of data between the 7600’s small core memory and its large core memory, and also to overlap transfers with computation for better runtime performance.  Using this strategy, you could boost the 7600’s usable memory to 250K 60-bit words [18, pages 2-4].

As you may guess, using this larger memory was costly.  The reads and writes were slow compared to those that could fit in the 7600’s small core memory and setting up the calls to the routines was exacting and time-consuming. And you would need to facilitate the program’s save-restarts in LCM, as well. As anyone who has worked with this sort of out-of-core computation knows, it makes one wish, I mean really wish, for a computer with larger memory.  Happily, this desire was realized in the CRAY-1 with its million 64-bit words of “core”.  Also very important, the CRAY-1 offered greater computational speed via vectorization. Many NCAR scientists made the additional effort of learning how to write vectorized code for their applications. 

You can imagine the community’s elation when scientists and programmers were at last able to run larger and more efficient applications. Our overall excitement and sense of unity was enhanced by seeing Seymour Cray walking the halls here at the NCAR Mesa Lab, checking on the new computer. Also, for several months after the CRAY-1’s arrival, the Computing Facility hosted a “War Room” at the Mesa Lab staffed by our engineers and programmers, where CRAY-1 users could bring their codes and discuss their current problems.

Readers interested in the era of the CRAY-1 Serial 3 at NCAR will find substantially more information in NCAR’s 1978 Scientific Report [20, pages 180-191]. I’d like to give a nod here to the UCAR Opensky archives: there’s huge amount of historical information on these pages, both for NCAR computing and the science done here. And please consider archiving some of your own materials, for the projects you are working on.

In closing, I thank the Scientists’ Assembly for inviting me to contribute these memories. So many of the old-timers who helped bring the CRAY-1 to NCAR are gone now.  Also gone are many of the early users, the scientists and programmers who used the Serial 3 within its first few years at NCAR.  If they were here to help, this note would be more informative. If you have questions or corrections about it, you may contact me, Dick Valent at


The Accelerated Scientific Discovery (ASD) program provides early access to users to complete computationally ambitious projects on NCAR high-performance computing systems during the first few months after acceptance testing has been completed. Six university-led projects and ten NCAR-led projects (nine primary and one alternate) were selected for ASD allocations.

Data-inspired MURaM Simulations of Flares Resulting from Sunspot Collisions 
Matthias Rempel, Yuhong Fan, Anna Malanushenko (HAO), Georgios Chintzoglou, Mark Cheung (Lockheed/LMSAL), MURaM GPU team (CISL, University of Delaware, & Max Planck Institute for Solar System Research)
Major solar eruptions often originate from complex active regions, specifically in active regions that are composed of several bipolar spot groups that interact with each other. AR 11158 (Feb 2011) is a well-studied example in which two opposite polarities collide and form a flare productive collisional polarity inversion line (cPIL). We propose a data inspired simulation of AR 11158 with the MURaM radiative MHD code in which the observed spot motions will be imposed at the lower sub-photospheric boundary condition. Synthetic observables covering visible to EUV observations will be computed and compared to the available observations from NASA/SDO. The investigation will focus on connecting changes in the magnetic topology prior to flares to available observables, specifically constraining the build-up and release of magnetic energy. Unlike earlier MURaM simulations, this simulation aims for the first time at reproducing processes in a specific observed active region through data-constrained boundary driving.

Urban Air Quality Across the Globe with MUSICA
Louisa Emmons (ACOM), Simone Tilmes, Gabriele Pfister, Rebecca Buchholz, Duesong Jo, Wenfu Tang, & David Edwards (ACOM); Behrooz Roozitalab (University of Iowa)
Air quality is primarily driven by local anthropogenic emissions sources, but it can also be strongly influenced by long-range transport of pollutants and regional influences (natural emissions, chemistry, meteorology, climate). In turn, local air quality can have impacts that extend all the way to the global scale. Global models including CESM2(CAM-chem) with comprehensive chemistry in the troposphere and stratosphere usually perform well in reproducing distributions of important air pollutants, including those of ozone and particulate matter (PM2.5). However, over highly polluted urban regions the model’s coarse horizontal resolution is often unable to capture local peak emissions of specific precursors for ozone. So far however, global models are not able to increase the horizontal resolution sufficiently because of needed large computer resources for transporting 200-300 chemical tracers. MUSICAv0, a configuration of CESM2.2(CAM-chem) with variable resolution, now has the unique capability to simultaneously simulate urban-scale air quality with high horizontal resolution and regional-to-hemispheric-to global influences and impacts of pollutants, while still using an expensive chemistry and aerosol scheme. 
This project will perform simulations of MUSICAv0 with a custom variable resolution mesh with 3 refined regions of special interest targeting the United States, Europe and southern and eastern Asia. Our plan is to use a base resolution of ne60 (0.5 degree), zooming into ne240 (~1/8 degree) resolution over the 3 regions.  Using this unique setup will allow us to better quantify the impact of urban pollution simultaneously on local, regional and hemispheric scales.

Deep Learning-based Large Ensemble for Subseasonal Prediction of Global Precipitation
Lead: Maria J. Molina (CGD), Co-Lead: Katie Dagon (CGD); Collaborators: Jadwiga Richter (CGD), David John Gagne (CISL), Gerald Meehl (CGD), Kirsten Mayer (CGD), Judith Berner (MMM/CGD), John Schreck (CISL), William Chapman (ASP), Aixue Hu (CGD), Anne Glanville (CGD), and Abby Jaye (MMM)
Every year, extreme precipitation and drought disrupt life, destroy infrastructure, and result in fatalities across the United States and the world. Skillful precipitation forecasts with a lead time of several weeks (i.e., subseasonal) can help stakeholders of societally-relevant public sectors (e.g., water management, agriculture, and health) understand imminent threats and take protective actions to mitigate harm. Our proposal aims to improve subseasonal prediction of precipitation using a data-driven, deep learning approach. With capabilities provided by Derecho, and as part of the Accelerated Scientific Discovery opportunity, we will use a data-driven, deep learning approach to create a 100-member ensemble of subseasonal forecasts of global precipitation. We will leverage deep learning approaches with observational and reanalysis products to improve already existing subseasonal reforecasts created using the Community Earth System Model version 2 (CESM2). Motivating our large ensemble approach is that an ensemble mean of subseasonal precipitation prediction can yield more skill than individual forecasts, but the large computational cost of simulating global numerical model subseasonal hindcasts precludes its creation. Moreover, recent studies have shown that deep learning models can produce subseasonal-to-multiyear forecasts with skill that exceeds current dynamical forecasting systems, making this an ideal time to take advantage of the unique opportunity that the Accelerated Scientific Discovery program presents.

High-resolution Simulations of Wildland Fires and Long-Range Smoke Transport During the 2020 August Complex Fires in California
Timothy Juliano, Masih Eghdami, Rajesh Kumar, & Branko Kosović (RAL); Gabriele Pfister & Rebecca Buchholz (ACOM); Hamed Ebrahimian & Kasra Shamsaei (University of Nevada)
Wildfires are some of the most destructive natural disasters on Earth, many times leading to devastating effects. There is no doubt that wildland fire activity in the United States (U.S.) and across the world has increased significantly over recent decades, and is projected to increase in forthcoming years. Currently, for air quality and solar energy forecasts, the fire emissions are estimated based on satellite observations that lag behind the actual emissions and therefore may not accurately represent the wildfire evolution. In light of this scientific gap in knowledge, we will use a multiphysics modeling approach and focus on improving air quality forecasts during the 2020 U.S. wildfire season. A very intense period of wildfire activity plagued a large portion of the U.S. in August and September 2020.
We propose to conduct coupled simulations using a multiscale (i.e., spanning the mesoscale and microscale) approach. The WRF-Fire wildfire-atmosphere modeling system will simulate fire behavior at fine resolution. While the proposal team has extensive experience using the WRF-Fire model, we have been limited computationally to conducting relatively small domain (order 10s of km) simulations at large-eddy simulation (LES) resolutions. However, during large conflagrations, such as the 2020 August Complex, much larger LES domains are required to accurately capture the wildfire spread and smoke production and transport. We will then use the biomass burning results from WRF-Fire to inform WRF-Chem and the MUlti-Scale Infrastructure for Chemistry and Aerosols (MUSICA) configuration of the Community Earth System Model (CESM), both of which contain complex chemical processes. Such hierarchical multiscale and multiphysics simulations will advance predictive science in support of air quality management and enhance warning systems protecting human health.

Nonlinear Multiscale Coupled Data Assimilation: Designing the Future of Air Quality Forecasting
B. Gaubert, W. Tang, F. Lacey, L. K. Emmons, S. Tilmes, M. Dawson, M. Barth, & G. Pfister (ACOM); K. Raeder, M. Gharamti, & J. L. Anderson (CISL); A. Arellano (University of Arizona)
This project is designated as an alternate and will be elevated in the event that other projects failed to make progress.
Applying concurrent data assimilation of chemical and physical observations in coupled chemistry meteorology models is often overlooked because of computational limitations. Unstructured grids with regional refinements have never been explored in chemical Data Assimilation (DA). This project aims to apply ensemble DA to an global online coupled chemistry-meteorology variable-resolution model. We will assess how improving the dynamics and physics via higher resolution impacts the chemical surface fluxes and state of the atmosphere. We will explore ensemble representations of physical and chemical uncertainties to disentangle errors stemming from emissions, transport and chemistry. The system is built on the coupling between the Multi-Scale Infrastructure for Chemistry and Aerosols (MUSICA) and the Data Assimilation Research Testbed (DART).
It uses the spectral element (SE) dynamical core of Community Atmosphere Model with full chemistry (CAM-chem) with horizontal mesh refinement defining a Regionally Refined (RR) domain, denoted as CAM-chem-SE-RR. The global grid has a resolution of ne30 (∼111 km) and the refinements reach ne240 (30×8, or ∼14 km) over the conterminous United States. The first objective is to assess the performance of the meteorological data assimilation and its comparison to current specified dynamics approaches. The second objective is to evaluate the impact of spatial resolution on initial state optimization and fluxes inversion of carbon monoxide (CO). The set of chemical data assimilation experiments will focus on the assimilation of CO from the TROPOMI instrument.

Extreme Weather Events Under a Wide Range of Climates in High-Resolution Coupled CESM
Bette Otto-Bliesner (CGD), Jiang Zhu (CGD), Esther Brady (CGD), Jesse Nusbaumer (CGD), Chijun Sun (ASP), Jessica Tierney (University of Arizona), Ran Feng (University of Connecticut), Clay Tabor (University of Connecticut), Andrew Walters (University of Arizona)
We propose an unprecedented, landmark set of fully coupled high-resolution (HR) climate simulations for past greenhouse and icehouse climates to study the dynamics that govern the characteristics of extreme weather events in both atmosphere and ocean under altered climate states. We target well-studied paleoclimate intervals with higher and lower atmospheric CO2, including the preindustrial, the Last Glacial Maximum, the Pliocene, and the Early Eocene. We employ scientifically validated and extensively tested CESM code and configuration, the iHESP (International Laboratory for High-Resolution Earth System Prediction) HR CESM1.3 (~0.25° atmosphere/land and ~0.1° ocean/sea ice) with water isotopes. The unique water isotope capability enables unprecedented integration of information from model and paleoclimate observational data. The paleo-HR simulations will complement the preindustrial, historical and RCP8.5 future simulations available from the iHESP project, resulting in HR simulations to investigate the dynamics that connect past and future climate changes. The proposed work will greatly expand our fundamental understanding of how elevated CO2 levels affect the pattern and intensity of extreme weather events, thus contributing to future projections of climate change and the physical science basis for actionable policies.

Benchmark Simulations Using a Lagrangian Microphysics Scheme to Study Cloud-Turbulence interactions: from Direct Numerical Simulations of a Laboratory Cloud Chamber to High-Resolution Large-Eddy Simulations of Clouds
Hugh Morrison (MMM), Kamal Kant Chandrakar (MMM), Wojciech W. Grabowski (MMM), George H. Bryan (MMM), Lulin Xue (RAL), Sisi Chen (RAL), Raymond A. Shaw (Michigan Technological University), and Greg McFarquhar (University of Oklahoma)
Clouds involve an enormous range of scales from the ~1 mm dissipation microscale to synoptic scales. Accurate representation of clouds in atmospheric models across these scales poses a significant challenge and is a critical source of uncertainty in weather and climate models. Our high-resolution simulations on Derecho will use a novel approach to simulating cloud/rain droplets and lead to better understanding of multi-scale processes in clouds. They can serve as benchmarks for developing and testing “traditional” cloud parameterizations in weather and climate models. These datasets can also be used to train artificial intelligence and machine learning algorithms for parameterization development.
The proposed simulations will utilize a Lagrangian particle-based microphysics scheme called the “super-droplet method” (SDM). SDM provides a major advancement for representing cloud microphysics in models compared to traditional bin and bulk microphysics schemes. For example, it is free from numerical diffusion, unlike bin schemes. SDM is available in NCAR’s CM1 model and runs efficiently on other supercomputing systems. The CM1-SDM framework was successfully applied to study the effects of turbulence and entrainment on cloud droplet size distributions. We will use three related model configurations in a hierarchical approach from direct numerical simulation (DNS) of small-scale turbulence to cloud-scale and mesoscale dynamics using large eddy simulation (LES).

Global Convection-Permitting Simulations with GPU-MPAS
Falko Judt (MMM), Andreas Prein (MMM), Bill Skamarock (MMM), Supreeth Suresh (CISL), Roy Rasmussen (RAL), Tim Schneider (RAL)  
We will produce a series of global convection-permitting simulations using GPU-MPAS. Our main goal is to assess the “added value” of convection-permitting resolution in (1) simulating structure and life cycle of mesoscale convective systems across different climate zones, (2) capturing the diurnal cycle and the duration, frequency, & intermittency of precipitation, (3) predicting extreme weather from local to global scales, and (4) representing orographic precipitation. Our secondary goal is to better understand the dynamics of tropical convection, and the predictability of the atmosphere in different climate zones.
We will simulate 4 pairs of 40-day long simulations on a globally quasi-uniform 3.75-km mesh, where one pair consists of a control run and a stochastically perturbed run. The number of simulation days on the 3.75-km mesh will be 4*2*40 = 320 (i.e., almost one year of global 3.75-km resolution data). In addition, there will be 15 km, 30 km, and 120 km mesh counterparts (again 4 pairs à 40 days) for added value assessments. These simulations will be identical to the 3.75-km runs except with reduced horizontal resolution.
The four 40-day long simulations will include the following time periods/events: April 2011, simulating the super Outbreak, the largest, costliest, and one of the deadliest tornado outbreaks ever recorded; August/September 2017, simulating Hurricanes Harvey, Irma, and Maria; December 2018–January 2019, simulating a Madden-Julian oscillation event initiated in the Indian Ocean and propagated across the Maritime Continent into the western Pacific; June/July 2021, simulating a series of record shattering extreme events that happened within a 4-week period in early summer 2021.

Response of Tropical Cyclone Rainfall to Thermal Forcing in Long-Term Convection-Permitting Simulations
George H. Bryan (MMM), Andreas Prein (MMM), Brian Medeiros (CGD), Jonathan Martinez (ASP), Kelly Nunez Ocasio (ASP), and Kerry Emanuel (Massachusetts Institute of Technology)
Recent Hurricanes Maria, Harvey, Lane, and Florence had something in common: they brought record-breaking, catastrophic rainfall to their landfall locations. Their associated rainfall amounts were considered “extreme” in our current climate, but those amounts may become more common if our planet continues to warm at the projected rates. At the same time, an increasing body of literature suggests that interactions between clouds and radiation—known as cloud-radiative feedbacks (CRFs)—impact several processes in the tropical atmosphere, including convective organization, precipitation extremes, and tropical cyclone formation. Given the impactful nature of tropical cyclone rainfall, this proposed study will use convection-permitting idealized simulations to investigate if tropical cyclone rainfall will increase under a projected 4-K warming while also investigating if CRFs affect extreme rainfall in tropical cyclones.

Enhancing Earth System Predictability by Diagnosing Model Error in Mode Water Regions
Ben Johnson (CISL), Moha Gharamti (CISL), Anna-Lena Deppenmeier (CGD), Ian Grooms (University of Colorado)
Major modes of climate variability such as the Pacific Decadal Oscillation (PDO) have off-equatorial dipole cores that coincide with regions of mode water formation. Mode waters are ocean mixing pathways that connect near-surface waters to the deeper central and intermediate waters beneath them. Simulations of these regions produced by eddy-parameterizing ocean models diverge from observations to such an extent that data assimilation schemes fail to assimilate many observations. This project conducts twin data assimilation experiments using the Data Assimilation Research Testbed (DART) with eddy-parameterizing (~1.0° horizontal resolution) and eddy-resolving (~0.1° horizontal resolution) eighty-member ensembles of POP2. The ensembles, which are forced by the CAM6 Reanalysis, are designed to diagnose model error in these regions and improve earth system predictability. These experiments use DART’s capability to identify observations that exceed an ensemble’s outlier threshold in an attempt to associate specific model processes with model errors.