It is difficult to capture the early signs of a forest fire at night using current visible-spectrum sensor technology. Infrared (IR) light sensors, on the other hand, can detect heat plumes expelled at the initial stages of a forest fire around the clock. Long-wave IR (LWIR) is commonly referred to as the “thermal infrared” region where thermal emissions are captured without the need of, or reflections from, external radiation sources. Mid‑wave IR (MWIR) bands lie between the “thermal infrared” and “reflected infrared” (i.e. short-wave IR) regions. Both LWIR and MWIR spectral regions are able to detect thermal radiation; however, they differ significantly in regards to their detection sensitivity of forest-fire heat plumes. Fires fueled by organic material (i.e. wood, leaves, etc.) primarily emit hot carbon dioxide (CO2) gas at combustion. Consequently, because CO2 is also present in the atmosphere, re-emission restricts the spectral transmittance and hence spectral radiance over a wide range of frequencies in the MWIR region. Moreover, as the distance between the detector and fire’s heat plume becomes greater, the additional CO2 introduced into the detection path leads to further attenuation of photon emittance. Since these absorption frequencies also lie within the response bandwidth of the MWIR sensor material, captured heat plume radiation manifests itself as a group of “flooded” or saturated pixels that exhibit very little dynamic behavior. Meanwhile, since the LWIR spectral region is not significantly affected by atmospheric gas absorption, its sensor is able to capture the forest fire’s heat plume thermal signature at far range without such complications. By exploiting the underlying spectral differences between LWIR and MWIR regions, this study aims to achieve early forest fire heat plume detection via direct identification of its dynamic characteristics whist concurrent attenuation of detected non-fire-related radiation. A land‑based, co‑located, cooled-LWIR/cooled-MWIR (CLWIR/CMWIR) detector camera is used to capture and normalize synchronized video from which sequential spatial-domain difference frames are generated. Processed frames allow for effective extraction of the heat plume’s “flickering” features, which are characteristic to the early stages of a forest fire. A multilayer perceptron (MLP) neural network classifier is trained with feature points generated from known target samples (i.e. supervised learning). Resulting detection performance is assessed via detection time, error metrics, computation time, and parameter variation. Results indicate that heat plumes expelled at the early stages of a forest fire can be identified with high sensitivity, low false alarm, and at a farther range than commercial detectors.