Advanced Volcanic Monitoring: AI Super-Resolution for Thermal Satellite Images

Image Super-Resolution (SR) models are advanced image processing techniques designed to increase the spatial resolution of digital images by reconstructing fine details from low-resolution inputs while preserving essential characteristics of the original data. SR methods are particularly valuable when high spatial detail is needed but not directly available, enhancing the interpretability of degraded or coarse imagery.

In satellite thermal observations, SR is especially relevant. Thermal Infrared (TIR, 8–14 µm) images, used to measure surface thermal radiation, generally exhibit low spatial resolution and higher noise than optical imagery. These limitations hinder the identification and quantification of fine-scale thermal features, including localized hotspots, small eruptive vents, and narrow lava flows.

Here, we propose a super-resolution method for multispectral thermal images based on advanced artificial intelligence, implemented through a deep Residual Neural Network (ResNet) architecture. Trained on paired low- and high-resolution thermal datasets, the model learns the complex non-linear relationships required to recover high-frequency spatial information typically lost in coarse TIR imagery. Residual learning allows the network to focus on reconstructing missing fine-scale structures, improving training stability and enhancing subtle thermal gradients. The architecture mitigates vanishing-gradient issues and enables deeper networks capable of extraxùcting thermally meaningful features without amplifying noise.

The resulting model reconstructs fine thermal structures—such as narrow lava flows and localized hotspots—producing coherent and physically interpretable thermal maps. ResNet-based SR enables the integration of the broad coverage offered by low-resolution sensors with the detail provided by high-resolution platforms.

From a volcanic monitoring perspective, thermal SR improves the detection and tracking of eruptive features, providing more precise and timely information on volcanic activity. Overall, applying advanced SR techniques to satellite thermal imagery enhances active volcano surveillance and contributes to a more accurate understanding of volcanic thermal processes.