Today?s advanced science applications often process huge volumes of data in real-time under extreme ambiances such as in spacecrafts and surveillance equipment. Faults manifesting in this raw data may corrupt the input given to these applications, causing flawed results. Hence we show that highly significant improvements in application reliability and precision can be obtained if the data is proactively preprocessed, using statistical analysis of input for possible recovery from the bit errors at input. In this paper we implement application-specific algorithms for preprocessing the input datasets for two major space applications. The Next-Generation Space Telescope (NGST) and the Orbital Thermal Imaging Spectrometer (OTIS) are both from NASA?s REE suite, and yet have su.cient contrasts in their datasets to demonstrate the versatile range of applications for which our approach is suitable. Our preprocessing algorithms utilize application-semantics, inherent data redundancy, absolute natural bounds, and temporal and/or spatial locality coherence in the natural data for setting dynamic thresholds to identify data-faults. We have compared the performance gain in relation to an ideal system for these algorithms with two standard algorithms used for image smoothing in graphics applications.