Division of focal plane (DoFP) polarization sensors can perform linear polarimetric imaging in one shot. However, since they use several neighboring pixels to estimate the polarization state, fast spatial variations of the scene may lead to estimation errors. We investigate the influence of the spatial variations of the three polarimetric parameters of interest (intensity, degree of linear polarization, and angle of polarization) on these errors. Using theoretical derivations and imaging experiments, we demonstrate that the spatial variations of intensity are the main source of estimation errors, much more than variations in the polarization state. Building on this analysis, we show that compensating the intensity variations within a superpixel is sufficient to reach the estimation performance of state-of-the-art demosaicing methods.