A Dimension-Reduction Approach to Space-Time Kalman Filtering