An official website of the United States government
Abstract The Ice-sheet and Sea-level System Model (ISSM) provides numerical solutions for ice sheet dynamics using finite element and fine mesh adaption. However, considering ISSM is compatible only with central processing units (CPUs), it has limitations in economizing computational time to explore the linkage between climate forcings and ice dynamics. Although several deep learning emulators using graphic processing units (GPUs) have been proposed to accelerate ice sheet modeling, most of them rely on convolutional neural networks (CNNs) designed for regular grids. Since they are not appropriate for the irregular meshes of ISSM, we use a graph convolutional network (GCN) to replicate the adapted mesh structures of the ISSM. When applied to transient simulations of the Pine Island Glacier (PIG), Antarctica, the GCN successfully reproduces ice thickness and velocity with a correlation coefficient of approximately 0.997, outperforming non-graph models, including fully convolutional network (FCN) and multi-layer perceptron (MLP). Compared to the fixed-resolution approach of the FCN, the flexible-resolution structure of the GCN accurately captures detailed ice dynamics in fast-ice regions. By leveraging 60–100 times faster computational time of the GPU-based GCN emulator, we efficiently examine the impacts of basal melting rates on the ice sheet dynamics in the PIG.
more »
« less
A Graph Neural Network Emulator of a Finite Element Ice Flow Model
Emulators of ice flow models have shown promise for speeding up simulations of glaciers and ice sheets. Existing ice flow emulators have relied primarily on convolutional neural networks (CNN’s), which assume that model inputs and outputs are discretized on a uniform computational grid. However, many existing finite element-based ice sheet models such as the Ice-Sheet and Sea-level System model (ISSM) benefit from their ability to use unstructured computational meshes. Unstructured meshes allow for greater flexibility and computational efficiency in many modeling scenarios. In this work, we present an emulator of a higher order, finite element ice flow model based on a graph neural network (GNN) architecture. In this architecture, an unstructured finite element mesh is represented as a graph, with inputs and outputs of the ice flow model represented as variables on graph nodes and edges. An advantage of this approach is that the ice flow emulator can interface directly with a standard finite element –based ice sheet model by mapping between the finite element mesh and a graph suitable for the GNN emulator. We test the ability of the GNN to predict velocity fields on complex mountain glacier geometries and show how the emulated velocity can be used to solve for mass continuity using a standard finite element approach.
more »
« less
- Award ID(s):
- 2107605
- PAR ID:
- 10548835
- Editor(s):
- Tedesco, Marco; Lai, Ching_Yao; Brinkerhoff, Douglas; Stearns, Leigh
- Publisher / Repository:
- AGU23 Online Program
- Date Published:
- Format(s):
- Medium: X
- Institution:
- American Geophysical Union
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Observational estimates of Antarctic ice loss have accelerated in recent decades, and worst-case scenarios of modeling studies have suggested potentially catastrophic sea level rise (~2 meters) by the end of the century. However, modeled contributions to global mean sea level from the Antarctic ice-sheet (AIS) in the 21st century are highly uncertain, in part because ice-sheet model parameters are poorly constrained. Individual ice-sheet model runs are also deterministic and not computationally efficient enough to generate the continuous probability distributions required for incorporation into a holistic framework of probabilistic sea-level projections. To address these shortfalls, we statistically emulate an ice-sheet model using Gaussian Process (GP) regression. GP modeling is a non-parametric machine-learning technique which maps inputs (e.g. forcing or model parameters) to target outputs (e.g. sea-level contributions from the Antarctic ice-sheet) and has the inherent and important advantage that emulator uncertainty is explicitly quantified. We construct emulators for the last interglacial period and an RCP8.5 scenario, and separately for the western, eastern, and total AIS. Separate emulation of western and eastern AIS is important because their evolutions and physical responses to climate forcing are distinct. The emulators are trained on 196 ensemble members for each scenario, composed by varying the parameters of maximum rate of ice-cliff wastage and the coefficient of hydrofracturing. We condition the emulators on last interglacial proxy sea-level records and modern GRACE measurements and exclude poor-fitting ensemble members. The resulting emulators are sampled to produce probability distributions that fill intermediate gaps between discrete ice-sheet model outcomes. We invert emulated high and low probability sea-level contributions in 2100 to explore 21st century evolution pathways; results highlight the deep uncertainty of ice-sheet model physics and the importance of using observations to narrow the range of parameters. Our approach is designed to be flexible such that other ice-sheet models or parameter spaces may be substituted and explored with the emulator.more » « less
-
Measurements of volcano deformation are increasingly routine, but constraining complex magma reservoir geometries via inversions of surface deformation measurements remains challenging. This is partly due to deformation modeling being limited to one of two approaches: computationally efficient semi-analytical elastic solutions for simple magma reservoir geometries (point sources, spheroids, and cracks) and computationally expensive numerical solutions for complex 3D geometries. Here, we introduce a pair of Graph Neural Network (GNN) based elasto-static emulators capable of making fast and reasonably accurate predictions (error upper bound: 15 %) of surface deformation associated with 3D reservoir geometries: a spheroid emulator and a general shape emulator, the latter parameterized with spherical harmonics. The emulators are trained on, and benchmarked against, boundary element (BEM) simulations, providing up to three orders of magnitude speed up compared to BEM methods. Once trained, the emulators can generalize to new reservoir geometries statistically similar to those in the training data set, thus avoiding the need for re-training, a common limitation for existing neural network emulators. We demonstrate the utility of the emulators via Bayesian Markov Chain Monte Carlo inversions of synthetic surface deformation data, showcasing scenarios in which the emulators can, and can not, resolve complex magma reservoir geometries from surface deformation. Our work demonstrates that GNN based emulators have the potential to significantly reduce the computational cost of inverse analyses related to volcano deformation, thereby bringing new insights into the complex geometries of magmatic systems.more » « less
-
Meshing is a critical, but user-intensive process necessary for stable and accurate simulations in computational fluid dynamics (CFD). Mesh generation is often a bottleneck in CFD pipelines. Adaptive meshing techniques allow the mesh to be updated automatically to produce an accurate solution for the problem at hand. Existing classical techniques for adaptive meshing require either additional functionality out of solvers, many training simulations, or both. Current machine learning techniques often require substantial computational cost for training data generation, and are restricted in scope to the training data flow regime. Mesh Deep Q Network (MeshDQN) is developed as a general purpose deep reinforcement learning framework to iteratively coarsen meshes while preserving target property calculation. A graph neural network based deep Q network is used to select mesh vertices for removal and solution interpolation is used to bypass expensive simulations at each step in the improvement process. MeshDQN requires a single simulation prior to mesh coarsening, while making no assumptions about flow regime, mesh type, or solver, only requiring the ability to modify meshes directly in a CFD pipeline. MeshDQN successfully improves meshes for two 2D airfoils.more » « less
-
Elastic continuum mechanical models are widely used to compute deformations due to pressure changes in buried cavities, such as magma reservoirs. In general, analytical models are fast but can be inaccurate as they do not correctly satisfy boundary conditions for many geometries, while numerical models are slow and may require specialized expertise and software. To overcome these limitations, we trained supervised machine learning emulators (model surrogates) based on parallel partial Gaussian processes which predict the output of a finite element numerical model with high fidelity but >1,000× greater computational efficiency. The emulators are based on generalized nondimensional forms of governing equations for finite non‐dipping spheroidal cavities in elastic halfspaces. Either cavity volume change or uniform pressure change boundary conditions can be specified, and the models predict both surface displacements and cavity (pore) compressibility. Because of their computational efficiency, using the emulators as numerical model surrogates can greatly accelerate data inversion algorithms such as those employing Bayesian Markov chain Monte Carlo sampling. The emulators also permit a comprehensive evaluation of how displacements and cavity compressibility vary with geometry and material properties, revealing the limitations of analytical models. Our open‐source emulator code can be utilized without finite element software, is suitable for a wide range of cavity geometries and depths, includes an estimate of uncertainties associated with emulation, and can be used to train new emulators for different source geometries.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Posted Content:
- The DOI is not currently available.
