Fire radiative power (FRP) over a pixel area has been highlighted as a valuable parameter for quantitatively deriving smoke emissions. However, smoke plume rise forecasts and characterizations of fire intensity require additional information, including the FRP over the fire area (FRPf) and per fire area (or fire radiative flux), both of which can be calculated through a bi-spectral retrieval of sub-pixel fire area and temperature. This study, the second in a two-part series, examines the sources of error and the corresponding uncertainties in a sub-pixel algorithm that calculates FRPf for fire pixels identified at 1 km2 nominal spatial resolution by the MODerate Resolution Imaging Spectroradiometer (MODIS) fire detection algorithm (collection 5). Radiative transfer simulations are incorporated to account for atmospheric effects as a function of Earth-satellite geometry at 3.96 and 11 μm (MODIS fire detection channels), and show that the 11 μm channel is highly sensitive to variations in column water vapor amount. By investigating several fire events in California, considerable variations in retrieved fire area, occasionally by more than 50%, are observed when comparing the mid-latitude summer climatology and observation-based atmospheric profiles that are used in the sub-pixel retrieval algorithm. While regions of dry, brown vegetation may also increase the potential for error via the surface emissivity assumption, the algorithm is much more sensitive to errors in 11 μm background brightness temperature, where an error of only 1.0 K may alter the retrieved fire area by an order of magnitude or more. An independent application, using the Texas wildfire event of September 2011, reveals that the sub-pixel retrieval can even become irrelevant for 17.6% of the available MODIS fire pixels as a result of noise in the background region causing the 11 μm background brightness temperature to become warmer than the fire pixel, especially during daytime scenes. The various sources of uncertainty in the estimates of FRPf and fire area can be reduced through the summation of individual pixel-level retrievals for large clusters of fire pixels, which can be defined based on the resolution of a mesoscale model grid. In comparison to the standard MODIS pixel-based FRP, the flux of FRPf, defined as total FRPf divided by the retrieved fire area, is shown to have a stronger and statistically significant correlation with surface (10-meter) wind speed and air temperature, especially for large fire pixel clusters, where the respective linear correlation coefficients are 0.55 and 0.77. These strong relationships suggest that, while additional studies are warranted, the flux of FRPf may offer the potential for improved characterizations of the meteorological effects on fire intensity compared to the standard pixel-based FRP.