CNN-Based Reanalysis of Optical Turbulence at the Canary Islands Observatories (OCAN)

Abstract

Atmospheric optical turbulence, caused by refractive-index fluctuations driven by wind and temperature inhomogeneities, is the primary source of image degradation in ground-based telescopes. Its vertical distribution, described by the refractive-index structure constant (Cn2(h)), determines essential parameters such as seeing, isoplanatic angle, and coherence time, which characterize astronomical sites and are also crucial for the design and performance of high-resolution instruments. The Observatorios de Canarias (OCAN), comprising the Observatorio del Roque de los Muchachos on La Palma and the Observatorio del Teide on Tenerife, have been extensively characterized through long-term Generalized SCIDAR campaigns. In total, the database comprises nearly 300 observing nights and hundreds of thousands of individual Cn2(h) profiles. We present a reanalysis of the OCAN database using deep-learning techniques. An end-to-end software framework was developed to ingest, preprocess, and transform the original turbulence measurements into standardized numerical representations suitable for machine-learning analysis. Temporal-vertical turbulence heatmaps are generated under multiple temporal configurations and enriched with seasonal information. These representations are processed by pre-trained convolutional neural networks (CNN) used as frozen feature extractors to obtain compact deep embeddings. Unsupervised clustering is subsequently applied to identify latent atmospheric regimes within the dataset. The resulting clusters reveal coherent turbulence structures and offer a complementary perspective on atmospheric conditions at the Canary Islands observatories. This methodology provides new insights into turbulence variability at the Canary Islands sites and represents a novel approach to large-scale turbulence characterization using modern deep-learning techniques.