Abstract Satellite altimetry has provided continuous measurements of reservoir water levels since 1992. However, pulse-limited radar altimetry missions have large footprints, which can lead to inaccuracies, particularly in reservoirs with changing water extents, narrow widths, low water levels, and complex topography. In these scenarios, waveforms often contain signals from both water and the surrounding land, complicating the accurate retrieval of water levels. Sentinel-3 (S3) satellites with smaller footprints and Synthetic Aperture Radar (SAR) mode enable more precise measurements by mitigating land contamination effects. This study aims to enhance water level measurements from pulse-limited radar altimetry missions by leveraging data from S3. We choose locations where S3 and pulse-limited altimetry Jason-3 (J3) are co-located. This study employs a deep learning approach, specifically a 1D U-Net model, to transform J3 LRM waveforms into S3 SAR-like waveforms, utilizing S3 data as a reference. A dataset of paired J3/S3 waveforms from 22 crossover locations over reservoirs was created. S3 waveforms were filtered using K-means clustering to select water returns as the training target. The 1D U-Net was trained on 80% of the data and validated on the remaining 20%. The model achieved a Mean Absolute Error (MAE) of 0.061 and Mean Squared Error (MSE) of 0.016 on the validation set. The model replicates the target S3 SAR waveform morphology, sharpening the leading edge and reducing noise at the trailing edge. This study demonstrates that deep learning can be used for waveform-to-waveform translation. This approach provides a method to enhance pulse-limited waveforms, thereby improving multi-mission altimetry harmonization at the waveform level for inland water bodies.