Authors: Dustin Kempton(Georgia State University), Viacheslav Sadykov(Georgia State University), Irina Kitiashvili(NASA Ames Research Center), Rafal Angryk(Georgia State University)

Turbulence in the solar interior and atmosphere plays a crucial role in energy transport, yet modeling its subgrid-scale effects remains a major challenge. This study leverages machine learning (ML) models to predict components of the Reynolds stress tensor using high-resolution StellarBox simulations of the quiet Sun. We compare a Multi-layer Perceptron (MLP) and a 3D Convolutional Neural Network (CNN) against physics-based baselines — Gradient and Smagorinsky models — and observe that both ML models achieve lower Mean Squared Error (MSE) and better generalization across various solar heights and depths. To enhance learning, we incorporate cluster-weighted training using K-Means, Hierarchical Agglomerative Clustering (HAC), and DBSCAN. By weighing the loss function based on cluster-specific prediction errors, we direct the model’s attention to high-error regions. This significantly improves CNN performance, particularly for off-diagonal stress components in the deeper layers of the convection zone. Our results show that integrating deterministic clustering with ML can effectively model subgrid turbulence.