Ocean-bottom-node (OBN) data are being used successfully in the deepwater environment to acquire time-lapse surveys due to their low 4D noise level. To help maximize the benefits of using such data, we built an acoustic model based on a deepwater subsalt field from the Gulf of Mexico. We generated synthetic base and monitor data using a realistic acquisition geometry and a laterally variable, random water-velocity layer. This study focuses on investigating a number of sources of time-lapse noise, including water-velocity variations, source and receiver repeatability, and position accuracy, and on evaluating ways to correct for these errors during processing. Results indicate that variable water velocities, if not corrected for during processing, produce the majority of residual 4D noise observed on the difference images. With currently achievable node placement accuracy, repeatability is fairly good and produces the smallest amount of noise. Sources, which are the least repeatable, produce almost three times the amount of noise compared to nodes in our experiment. We applied a similar processing workflow to the up- and downgoing wavefields and imaged both types of data using one-way and two-way wave-equation-migration techniques. In this particular synthetic example, image quality and 4D noise level were similar in the extra-salt area on both processed up- and downgoing images, but both measures in the subsalt were much better on the upgoing data.