In order to meet the emerging need for better estimates of smoke plume injection heights in air quality and climate models, based on the probable dependence of injection height on fire intensity, it is suggested that a sub-pixel-based retrieval of fire radiative power (FRP) should be incorporated within the current MODIS and future VIIRS and GOES-R algorithms. In this study, the new FRP calculation is tested on fire pixels detected at 1 km² nominal spatial resolution by the MODerate Resolution Imaging Spectroradiometer (MODIS) fire detection algorithm (collection 5) for several large wildfire events in California. The methodology stems from the heritage of earlier bispectral retrievals of sub-pixel fire area and temperature. However, in the current investigation, a radiative transfer model is incorporated to remove solar effects and account for atmospheric effects as a function of Earth-satellite geometry at 3.96 and 11 µm (MODIS fire detection channels). The retrieved sub-pixel fire (flaming) area is validated via the multispectral, high-resolution data (3-50 meters) obtained from the Autonomous Modular Sensor (AMS), flown aboard the NASA Ikhana unmanned aircraft. Pixel-level fire area comparisons between MODIS and AMS are highly variable regardless of the viewing zenith angle and show a very low bias with a modest correlation (R = 0.59). However, when lower confidence fire pixels and point-spread-function effects are removed, the correlation becomes much stronger (R = 0.84) and the variability between MODIS and AMS is reduced. To account for these random errors via averaging, two clustering techniques are employed and the resulting AMS and MODIS comparisons of fire area, after correcting for overlapping MODIS pixels, are even more encouraging (R = 0.91). Drawing from the retrieved fire area and temperature, the sub-pixel-based FRP is calculated and compared to the current MODIS pixel area-based FRP, resulting in a strong correlation (R = 0.93) with a bias towards higher sub-pixel FRP values. A sensitivity analysis reveals that regions of dry, brown vegetation may increase the potential for error via the background emissivity, but the retrieval is most sensitive to small errors in 11 µm background temperature. In this case, an error of only 1.0 K can result in retrieved fire area errors of an order of magnitude or more.