Extracting Thermal Patterns in Volcanic Areas from Thermal Infrared Satellite Data: A Case Study at the Campi Flegrei Caldera

In volcanic regions, Thermal InfraRed (TIR) remote sensing is a well-established technique for detecting ground thermal anomalies. The analysis of thermal properties, particularly of Land Surface Temperature (LST) time series, represents a valid tool to achieve a rapid characterization of the shallow thermal field, supporting ground-based surveillance networks in the monitoring of volcanic activity, especially in areas that are inaccessible due to high volcanic hazard.

However, in complex active volcanic and hydrothermal settings, the coexistence of processes of different natures that interact and mutually interfere can significantly affect the distribution of the LST parameter, making it challenging to interpret its spatio-temporal variations. In this context, the extraction of the main thermal patterns of volcanic areas from satellite-derived LST time series represents a further step for a more detailed characterization of the shallow thermal field.

In this study, the extraction of the main thermal patterns from satellite-derived LST time series is addressed through decomposition techniques such as the Independent Component Analysis (ICA) and the Dynamic Mode Decomposition (DMD). ICA is a statistical method aimed at identifying a linear transformation of the data that maximizes the statistical independence between its components, defining the signal’s independent components (ICs). DMD is a data-driven technique aimed at decomposing spatio-temporal data for the extraction of coherent features, defining a set of dominant dynamic modes (DMs). 

The investigated area is the Campi Flegrei caldera (southern Italy), a well-known complex volcanic system. The LST time series is retrieved from cloud-free nighttime TIR images acquired by Landsat-8 and Landsat-9 missions (L8 and L9) during the 2018–2025 time interval. Specifically, the LST parameter is estimated through the Radiative Transfer Equation (RTE) applied to a single thermal band (Band 10 for both L8 and L9) and with known information on the surface emissivity and atmospheric conditions of the investigated area. Subsequently, the application of ICA and DMD methods allowed the identification of the main components, revealing the dominant thermal patterns influencing the LST distribution and providing insights into the endogenous and exogenous processes characterizing the volcanic site.