Advancing 3D Cloud Nowcasting: Comparative Analysis of Advection Methods and Machine Learning

Cloud nowcasting, the prediction of short-term (0-3 hour) cloud movements, is crucial for various applications ranging from weather forecasting, renewable energy management, and even Department of Defense (DoD) missions. This study contributes to the Cooperative Institute for Research in the Atmosphere (CIRA) OVERCAST project, focusing on a comparative assessment of traditional cloud advection methods and a machine learning approach that employs a UNET architecture to forecast 3D cloud fields. OVERCAST, sponsored by the US Navy Office of Naval Research (ONR) and Naval Research Laboratory (NRL), aims to create a global, near real-time, atmospheric 3D cloud field analysis. This initiative provides a unique opportunity to enhance cloud nowcasting capabilities through innovative methodologies. In this study, a semi-Lagrangian cloud advection (SLCA) method is employed which utilizes numerical weather prediction wind parameters to simulate cloud movement within a 3D grid. The initial state of 3D clouds is established using cloud top height and cloud geometric thickness from CLAVR-x (Clouds from AVHRR Extended). The SLCA method is compared to a machine learning approach which harnesses a UNET architecture, utilizing the same CLAVR-x parameters as input to predict their values in the future and thereby providing the required data to inform on the future state of the 3D cloud field. By training on historical CLAVR-x data, the UNET model learns to predict cloud movement and potentially cloud evolution. To assess the performance of the two approaches, a validation method is established that compares the resulting 3D cloud fields via the use of traditional forecast metrics such as probability of detection, critical skill index, and false alarm rate as well as a fraction skill score (FSS) to account for spatial variability of predictions. Preliminary results indicate that the machine learning method exhibits promising capabilities in capturing cloud movement and cloud evolution. This study will also endeavor to include novel inputs via feature engineering to better capture cloud evolution.