Does resolution matter? Satellite imagery comparison for dynamic fuel mapping and wildfire risk assessment

In recent years, the increasing frequency and severity of wildfire have caused widespread devastation to infrastructure, communities, and ecosystems.Effective wildfire risk mitigation strategies require accurate and up-to-date information about the spatiotemporal distribution of vegetation, which serves as the primary fuel for wildfires.However, current approaches to fuel mapping typically rely on static datasets like LANDFIRE, which often lack the spatial precision and temporal relevance needed for accurate wildfire risk assessment.This study, therefore, presents an application-driven evaluation of a dynamic fuel mapping generation model using satellite imagery.Specifically, a K-nearest neighbors (KNN) algorithm is adapted and evaluated for pixel-wise fuel mapping, incorporating enhancements such as active learning and systematic sampling to improve classification robustness.Model performance is assessed across three sources of satellite imagery with different spatial resolutions: Harmonized Landsat and Sentinel-2 (HLS), Sentinel-2, and SPOT-6.Results show that classification success depends not only on resolution but also on the compatibility between input data and ground truth labels.Moreover, findings indicate that different satellite imagery resolutions are better suited for different mapping objectives, with high-resolution imagery offering advantages for fine-scale analysis, while lowerresolution data remain effective for regional assessments, highlighting the need for evaluation metrics and modeling approaches that account for both pixel-level accuracy and broader spatial patterns.This work contributes a scalable, interpretable, and dynamic fuel type classification approach and offers practical insights for sensor selection and data preparation in dynamic fuel mapping.