Atmospheric correction over turbid waters can be problematic if atmospheric haze is spatially variable. In this case the retrieval of water quality is hampered by the fact that haze variations could be partly mistaken for variations in suspended sediment concentration (SSC). In this study we propose the suppression of local haze variations while leaving sediment variations intact. This is accomplished by a multispectral data projection (MDP) method based on a linear spectral mixing model, and applied prior to the actual standard atmospheric correction. In this linear model, the hazesediment spectral mixing was simulated by a coupled water-atmosphere radiative transfer (RT) model. As a result, local haze variations were largely suppressed and transformed into an approximately homogenous atmosphere over the MERIS top-of-atmosphere (TOA) radiance scene. The suppression of local haze variations increases the number of satellite images that are still suitable for standard atmospheric correction processing and subsequent water quality analysis.