1 of 1

GNN-Surrogate: A Hierarchical and Adaptive Graph Neural Network for Parameter Space Exploration of Unstructured-Mesh Ocean Simulations

Scientific Achievement

Enable scientists to rapidly explore the parameter space of ocean ensemble simulations on unstructured grids

Significance and Impact

We show that a graph network based neural network can learn the mapping from simulation parameters to output fields defined on unstructured grids.

(a) Comparison of the sea level temperature maps generated using GNN-Surrogate, IDW interpolation, and InSituNet with the ground truth maps. Comparison of the vertical cross-sections at (b) the equator (c) 75°E generated using GNN-Surrogate and IDW interpolation with the ground truth cross-sections. (d) Comparison of the isothermal layer (ITL) depth maps with temperature isovalue 25℃ generated using GNN-Surrogate and IDW interpolation with the ground truth maps.

Technical Approach

Perform a graph coarsening algorithm to build a graph hierarchy consisting of graphs at different resolution levels
Cut the graph hierarchy and transform the graphs to adaptive resolutions
Train a GNN-Surrogate based on the simulation parameters as the input and output data of adaptive resolutions
In the Post hoc analysis stage, simulation outputs can be predicted from given input parameters and then visualized

Neng Shi, Jiayi Xu, Skylar W. Wurster, Hanqi Guo, Jonathan Woodring, Luke Van Roekel, and Han-Wei Shen: GNN-Surrogate: A Hierarchical and Adaptive Graph Neural Network for Parameter Space Exploration of Unstructured-Mesh Ocean Simulations, IEEE Transactions on Visualization and Computer Graphics (Proc. IEEE PacificVis 2022), 28(6):2301-2313, 2022.

In this work, We propose a graph neural network-based surrogate model to explore the parameter space of ocean climate simulations. Parameter space exploration is important for domain scientists to understand the influence of input parameters (e.g., wind stress) on the simulation output (e.g., temperature). However, the exploration requires scientists to exhaust the complicated parameter space by running a batch of computationally expensive simulations. Our approach improves the efficiency of parameter space exploration with a surrogate model that predicts the simulation outputs accurately and efficiently. Specifically, our model predicts the output field with given simulation parameters so scientists can explore the simulation parameter space with visualizations from user-specified visual mappings. Moreover, our graph-based techniques are designed for unstructured meshes, making the exploration of simulation outputs on irregular grids efficient. The quantitative and qualitative evaluation results on the MPAS-Ocean simulation demonstrate the effectiveness and efficiency of our apporch.