Benchmarking Gap-Filling Techniques in Satellite Altimetry-Based Lake Water-Level Time Series

Abstract. Satellite radar altimetry has significantly enhanced inland water monitoring by providing consistent, long-term lake-level observations. However, these datasets often contain substantial gaps caused by sensor malfunctions, orbital limitations, or retrieval errors, complicating hydrological analyses and downstream applications. This study introduces a robust benchmarking framework designed to systematically evaluate gap-filling techniques in satellite-derived lake water-level records through controlled pseudo-gap injection experiments. We investigated three large lakes with distinct hydrological behaviours, the Caspian Sea, Lake Superior, and Lake Tanganyika each representing unique temporal dynamics. Synthetic seven-year gaps (2002–2008) were artificially introduced into complete altimetry datasets, and three sophisticated gap-filling methods were compared: Singular Spectrum Analysis (SSA), Bidirectional Autoregressive (BIAR) models, and Bidirectional Multi-Layer Perceptrons (BiMLP). Method performance was assessed using standard metrics (RMSE, MAE, Bias, and R2), alongside statistical properties including variance, skewness, autocorrelation, and stationarity. BiMLP consistently delivered the highest accuracy across all study lakes, demonstrating exceptional adaptability to both smooth and highly variable signals. SSA performed effectively for lakes exhibiting quasi-periodic behavior, while BIAR showed sensitivity to lag selection and reduced performance under non-stationary conditions. These results emphasize the critical role of bidirectional modeling approaches and rigorous time-series diagnostics in selecting appropriate gap-filling methods for satellite altimetry-based hydrological studies. These findings provide practical guidance for selecting appropriate gap-filling strategies under long data outages in satellite altimetry workflows.