Prototype-enhanced prediction in graph neural networks for climate applications

2025-04-24

AI Summary

To address the need for high-accuracy, low-overhead data-driven surrogate models in atmospheric dispersion modeling for climate simulation, this paper proposes a prototype-augmented graph neural network (GNN) method. Specifically, it introduces learnable and selectable output prototypes as auxiliary inputs into the GNN architecture, enabling effective guidance of high-dimensional predictions without requiring ground-truth labels. Theoretical analysis and empirical evaluation demonstrate that even randomly initialized prototypes yield meaningful improvements, while prototypes derived via k-means clustering further enhance performance. Evaluated on greenhouse gas dispersion simulation, the method significantly improves prediction robustness and accuracy—achieving up to a ~10% gain in key metrics over baseline models—while reducing computational cost. This work establishes a novel paradigm for interpretable and efficient GNN-based modeling in climate science.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks

Abstract

Data-driven emulators are increasingly being used to learn and emulate physics-based simulations, reducing computational expense and run time. Here, we present a structured way to improve the quality of these high-dimensional emulated outputs, through the use of prototypes: an approximation of the emulator's output passed as an input, which informs the model and leads to better predictions. We demonstrate our approach to emulate atmospheric dispersion, key for greenhouse gas emissions monitoring, by comparing a baseline model to models trained using prototypes as an additional input. The prototype models achieve better performance, even with few prototypes and even if they are chosen at random, but we show that choosing the prototypes through data-driven methods (k-means) can lead to almost 10% increased performance in some metrics.

Problem

• Improving high-dimensional emulated outputs using prototypes
• Enhancing atmospheric dispersion emulation for emissions monitoring
• Boosting model performance via data-driven prototype selection

Innovation

• Uses prototypes to enhance emulator outputs
• Improves predictions with few random prototypes
• Data-driven k-means boosts performance by 10%