Tundra fires are projected to increase with anthropogenic climate change, yet our ability to assess key wildfire metrics such as fire severity remains limited. The Normalized Burn Ratio (NBR) is the most commonly applied index for fire severity mapping. However, the computation of NBR depends on short-wave infrared (SWIR) data, which are not commonly available from historical and contemporary high-resolution (≤4 m) optical imagery. The increasing availability of visible near-infrared (VNIR) measurements from proximal to spaceborne sensors/platforms has the potential to advance our understanding of the spatiotemporal patterns of fire severity within tundra fires. Here we systematically assess the feasibility of using VNIR data for fire severity mapping in ten Alaskan tundra fires (cumulatively burned ~1700 km2). We compared the accuracy of 10 published VNIR-based fire indices using both uni-temporal (post-fire image) and bi-temporal (pre-fire and post-fire image difference) assessments against ground-based fire severity data (Composite Burn Index, CBI) at 109 tundra sites. The Global Environmental Monitoring Index (GEMI) had the highest correspondence with CBI (R2 = 0.77 uni-temporal; R2 = 0.85 bi-temporal), with similar performance to NBR (R2 = 0.77 uni-temporal; R2 = 0.83 bi-temporal). Tundra vegetation types affected NBR but not GEMI, as SWIR reflectance was influenced to a greater extent in shrub than graminoid tundra. We applied GEMI to contemporary high-resolution (i.e. Quickbird 2) and historical meso-resolution imagery (i.e. Landsat Multispectral Scanner) to demonstrate the capability of GEMI for resolving fine-scale patterns of fire severity and extending fire severity archives. Results suggest that GEMI accurately captured the heterogeneous patterns of tundra fire severity across fire seasons, ecoregions, and vegetation types.