The energy emitted by a flame and its spectra provide critical information about fuels, such as energy type, temperature, and molecular properties. While traditionally used to analyze combustion, we introduce a spectral-time-based index, optical calorific power (OCP), to calculate the total energy released per mass unit (J/Kg), relating to the calorific value of wood-based fuels. Using novel optical variables and machine learning, we identified different wood species during the combustion of the wood samples. Homogeneous samples of four wood species (sapwood and heartwood) were ignited in a temperature-controlled furnace. Spectra were measured using a calibrated spectrophotometer in the 450–900 nm range with a 100ms integration time. Continuous and discontinuous spectral patterns were observed in all samples, used to calculate spectral-time-based optical variables, and separated using the AirPLS algorithm. Discontinuous spectra correlated with Na and K emissions (589.4 nm and 766.5 nm, respectively). Continuous spectra were analyzed to determine optical variables such as flame temperature (K), total continuous radiation (TCR, μW/cm2), and total continuous energy (TCE, μJ). Five supervised classification models, XGBoost, LR, SVM, LDA, and RF, were trained using normalized physical parameters and optical variables with stratified cross-validation. The proposed optical variables yielded an identification accuracy of 93%, precision of 95%, and recall of 93% during combustion using the XGBoost method. These promising results demonstrate the potential of our method for accurately identifying wood species, while offering a cost-effective and novel alternative to NIR-SWIR reflectance spectral identification techniques.