Near-Real-Time Wildfire Segmentation and Brightness Temperature Estimation from Geostationary Satellite Imagery Using Deep Learning

Monitoring the progression of large wildfires in near-real-time is essential for active-fire situational awareness and emergency response management. Current satellite-based wildfire monitoring systems face a trade-off between temporal and spatial resolution: geostationary satellites such as GOES offer frequent (~5 minutes) but coarse observations (~2 km), while low earth orbit (LEO) instruments such as VIIRS provide fine spatial detail (∼375 m) with limited temporal coverage (twice per day). To bridge this gap, this study introduces a deep learning (DL) approach that enables near real-time, high-resolution wildfire monitoring using GOES data. The proposed approach consists of two main steps: a segmentation step to distinguish active fire regions from background areas and a regression step to estimate the active fire pixels brightness temperature (BT) across a region of interest. The output of these steps is combined to generate a high-resolution fire location and BT maps. To train the DL model, multi-spectral GOES inputs are paired with VIIRS-derived fire observations from several wildfires across the United States. Spatial consistency between GOES and VIIRS data is achieved through parallax correction, reprojection, resampling, and per-image normalization. Ablation studies are performed to demonstrate the impact of different assumptions (e.g., background values in the VIIRS ground truth) and strategies (e.g., loss functions) throughout the development process. The results show that the proposed DL approach effectively enhances GOES imagery, improving both BT estimation and fire boundary localization. Overall, the proposed method offers a practical and scalable solution for wildfire boundary detection and thermal mapping using existing satellite systems.