The southeastern United States experiences frequent wild and prescribed fire activity. Mapped burn severity products in the southeastern U.S. face challenges accurately characterizing fire effects due to rapid post-fire recovery limiting observation windows, limited availability of cloud-free imagery, spectral confusion within wetland areas, and operational constraints. As mapped burn severity datasets are generally focused on large wildfires, the many small and prescribed fires of the Southeastern U.S. are not well-represented in existing burn severity products. Accurate and detailed characterization of burn severity across the region is significant to the estimation of fire-related emissions, measurement of fuel loads and aboveground carbon storage, and guiding land management activities. The U.S. Geological Survey (USGS) developed an algorithm to improve the prediction of post-fire burn severity within the southeastern United States. A burn severity model was developed utilizing over 5000 Composite Burn Inventory (CBI) plots, where post-fire impacts were characterized in the field for 232 unique fire events across the continental US. For each CBI plot location, predictor variables were generated from ARD Landsat scenes capturing first and second-order fire effects, climate norms, and fire seasonality. A gradient-boosted decision tree model was developed to predict post-fire burn severity as a CBI value (0-3), aligning field and satellite observations of fire effects. The model was applied to the extent of burned area identified by the Landsat Burned Area Product to generate annual (2000-2022) burn severity mosaics of predicted CBI burn severity for 78 ARD Landsat tiles encompassing the southeastern United States. These data provide an improved characterization of burn severity in the southeastern United States, with support for small and prescribed fire activity.