Exploring the Ability of Machine Learning to Infer Subgrid-Scale Convective Cloud Fraction from Coarse-Resolution Environmental Conditions

Abstract Global climate models (GCMs), typically run at ∼100-km resolution, capture large-scale environmental conditions but cannot resolve convection and cloud processes at kilometer scales. Convection-permitting models offer higher-resolution simulations that explicitly simulate convection but are computationally expensive and impractical for large ensemble runs. This study explores machine learning (ML) as a bridge between these approaches. We train simple, pixel-based neural networks to predict convective cloud fraction from environmental variables produced by a regional convection-permitting model. The ML models achieve promising results, with structural similarity index measure (SSIM) values exceeding 0.8, capturing the diurnal cycle and orographic convection without explicit temporal or spatial coordinates as input. Model performance declines when fewer input features are used or specific regions are excluded, underscoring the role of diverse physical mechanisms in convective cloud activity. These results demonstrate the potential of ML as both a tool for scientific discovery and a proof of concept that relatively simple model architectures can elucidate relationships between convective cloud characteristics and large-scale environmental controls. The proposed pixel-based approach is computationally inexpensive and is well suited as a diagnostic framework for analyzing GCM and reanalysis datasets rather than as a forecasting system. Unlike convolutional neural networks, which depend on the spatial structure and grid size, the pixel-based model treats each grid point independently, enabling value-to-value mapping independent of spatial context. This design facilitates cross-model and cross-scenario applicability and supports generalization across diverse environmental regimes, providing an efficient pathway for linking coarse-resolution environmental conditions to subgrid-scale convective features in a changing climate. Significance Statement Understanding how clouds and storms form is critical for predicting severe weather in a changing climate. However, current climate models struggle to capture small-scale cloud features due to their coarse resolution. Our research shows that machine learning, trained on detailed weather simulations, can successfully predict where the most storms occur based on larger-scale environmental patterns. This approach bridges the gap between what climate models can resolve and the finer details of clouds that matter most for impacts like severe weather. By combining high-resolution simulations and machine learning, our work offers a powerful new way to better understand storms in both current and future climate.