Accurate mapping of post-fire surviving trees is important for tracking forest recovery and prioritizing land management decisions. Satellite-based remote sensing is an effective method to assess post-fire forest conditions. Traditionally, differenced satellite-derived burn severity indices are computed by differencing one year pre- and post-fire spectral reflectance values. Differenced burn severity indices are useful for quantifying and mapping the magnitude of ecological change, but their application to detecting and mapping post-fire live trees may not be as appropriate, particularly for delayed tree mortality. Delayed tree mortality (“delayed mortality”) is a phenomenon where trees that initially survive fire then die over an extended period (between one and five years), and it can be challenging to measure and predict. In this study, we demonstrate the potential of mapping delayed mortality using readily available remotely sensed imagery alone. We used random forest models to detect post-fire live trees using 10-m resolution Sentinel-2 data at one-, three-, and five-years post-fire for four fires in the southern Sierra Nevada, California, USA. Using imagery from the National Agriculture Imagery Program (NAIP; 60-cm resolution), we manually classified live tree presence in 6000 Sentinel-2 pixels (500 pixels for each fire-year combination) to calibrate and validate models. Sentinel-2 based model accuracies ranged from 65 % to 86 % with F-scores ranging from 0.52 to 0.86, and their predictions of live pixel area were on average 44 % lower than inferred from more traditional indices such as relative differenced normalized burn ratio (RdNBR). This work represents a promising first step in using freely available post-fire spectral reflectance imagery to detect live trees over an extended period to support post-fire management.