Spatial Resolution Enhancement of Geostationary Thermal Observations for Wildfire Monitoring

Wildfires represent one of the most critical threats to Mediterranean forests, making timely detection and continuous monitoring a priority for risk mitigation and environmental management. Despite significant advances in satellite-based fire monitoring, current approaches remain constrained by a fundamental trade-off between spatial and temporal resolution in available remote sensing data. Geostationary satellite systems offer high-frequency observations that are well suited for near-real-time monitoring, yet their coarse spatial resolution limits their effectiveness for applications requiring fine-scale spatial detail. Addressing this limitation is particularly relevant for wildfire monitoring, where early-stage events often occur at small spatial scales. In this presentation, we introduce a learning-based framework for spatial resolution enhancement of high-temporal infrared satellite observations. The approach explores multiple model families, including autoencoder-based architectures, residual channel attention networks, and generative models such as neural operator diffusion, to reconstruct fine-scale thermal structure from coarse measurements while preserving temporal consistency. The best model configurations are tested in the context of wildfire monitoring, using higher-resolution thermal products from NASA VIIRS as reference data. Results indicate improved representation of fire-related signals, with implications for better early detection and monitoring applications. Detailed methodological developments and quantitative evaluations will be presented in a forthcoming publication.