At this time, AGU is planning the 2021 Fall Meeting as a "hybrid" meeting, aimed to optimize both in-person and worldwide virtual participation and to present a best-in class experience for all attendees. Join us for Fall Meeting 2021 in New Orleans, Louisiana.
The latest generation of Earth system models (ESMs) participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) has improved owing to a combination of increased model resolution and more advanced parameterization of unresolved physical processes, such as those for convection, cloud microphysics, or surface processes. More advanced models may not be synonymous with reduced uncertainty; increasing complexity may be associated with increased uncertainty in model parameters or structural choices. It is crucial that methodologies are developed for systematically improving models to ensure the reliability of climate projections and refinement of climate sensitivity estimates. In this session, we welcome presentations on advances in developing and calibrating models using observations (e.g. satellite, airborne, or ground-based) and/or high-resolution model simulations. Presentations that focus on ESM parameter estimation, autotuning, and emulator/surrogate based tuning, as well as methods that directly address structural errors in models (e.g., using machine learning) or observational uncertainties are welcomed.
High resolution Earth system modeling on large supercomputers
High resolution Earth system modeling offers significant opportunities for improving the quality of simulations, particularly for water cycle processes and extreme events. Increase in model resolution and complexity presents significant challenges on many fronts. These include evaluation, diagnosis, and analysis of model simulations, as well as uncertainty quantification, computational performance, and model initialization and data assimilation. This session aims to bring together scientists who develop, run, evaluate, and analyze coupled global and regional Earth system models and their components at high resolution on high performance computers. With submissions particularly encouraged on ~10 km and ultra-high (order km) scales, this session welcomes contributions on development of parameterizations to improve model components and the coupled system at these resolutions. The session also encourages presentations on high-resolution simulations from model intercomparison projects such as HighResMIP, DYAMOND, and OMIP, as well as development of observational datasets and metrics for evaluation of such high-resolution simulations.
A086 - Numerical coupling of atmospheric processes in current and future models: challenges and paths forward
The atmosphere is a complex system involving physical, chemical, and biogeochemical processes spanning a wide range of spatial and temporal scales. Accurate representations of the interactions between processes and/or scales are important for high-quality predictions made by numerical models. Ongoing increases in model resolution and complexity present both new opportunities for more accurate results and new challenges in efficient use of computing resources. This session discusses process coupling in weather and climate models, with an emphasis on addressing computational and numerical concerns. Relevant topics include, e.g., (1) identification of crucial interactions in numerical simulations, (2) computational methods for coupling different processes and/or scales, (3) problems to be solved in current and emerging models, (4) idealized models and test cases to facilitate the development of complex models, and (5) model reduction studies that produce alternative model formulations or novel computational methods to improve solution accuracy and efficiency.
IN015 - Climate Model Computational Performance: State of the Practice (combined session including AI below)
Coupled climate models present a singular challenge in terms of computational complexity and efficient utilization of state of the art supercomputing resources. There has been a Cambrian explosion of hardware architectures and associated software toolchains that are deployed or planned for the leadership-class supercomputers around the world. This has led to a multitude of programming and modeling approaches to target accelerators like Graphical Processing Units (GPUs) while incrementally integrating such sub-models for coupled climate model simulations. This session is intended to share current status of computational performance and best practices from diverse modeling groups. We solicit papers addressing computational performance of Earth system models preferably in real-world settings (production simulation campaigns in lieu of kernel benchmarks) with emphasis on metrics including model throughput, energy efficiency, I/O volume and read/write bandwidths etc. informed by initiatives like the CPMIP effort. Furthermore, we encourage submissions describing early performance from pathfinding work targeting future architectures.
Accelerating Earth System Predictability: Advances in High Performance Computing, Numerical modeling, Artificial Intelligence and Machine Learning
As highlighted in the recent White House Earth System Predictability R&D Framework and Roadmap, enhanced Earth system predictions are critical to inform societal resilience and mitigate impacts from extreme events. Thus, advances in Earth System Predictability (ESP) can bring clarity and inform necessary investments to address the monumental challenge of global climate change. Concurrently, there have been dramatic changes in the computing landscape with the advent of new hardware architectures including Graphical Processing Units (GPUs) that enabled significant advances in Artificial Intelligence (AI)/Machine Learning (ML) and numerical modeling.
This session focuses on novel approaches that address the ESP challenge on multiple fronts including computational readiness work on leadership-class supercomputers, scalable AI/ML methodologies addressing large scale training and data challenges and process-based methods for accurate modeling of interconnected human-natural systems. High-resolution earth system models are in active development to incorporate the requisite complexity with the goal of enabling global cloud resolving capabilities through efficient utilization of new supercomputer architectures. The availability of large earth science datasets augurs well for application of scalable AI and data analytics techniques despite the challenges of training, pre/post-processing and curating well-represented, unbiased, and comprehensive training datasets. Additionally, process-based models (e.g., radiative transfer, hydrology, fire behavior) facilitate understanding causal effects of interconnected human-natural systems through capture of feedbacks and processes connecting land, water, atmosphere, and biosphere.
The goal of this interdisciplinary session is to bring together practitioners to share best practices on building performant methodologies, scalable end-to-end workflows and build partnerships to address open problems.
This session focuses on process-based modeling of glaciers and ice sheets that improves our understanding of ice dynamics or enhances the capabilities of current models. We invite contributions from a broad range of theoretical, numerical, or experimental studies that explore new or improved representations of physical processes relevant to ice flow. The range of topics that is encouraged includes, but is not limited to: basal processes (such as glacial hydrology, erosion, surging, the onset of ice streams), mechanical and thermodynamic processes (constitutive relationships, material behavior, ice fracturing), and ice-ocean interactions (for example calving, grounding line dynamics).