Supporting Data for VISTAR: A Vegetation-Index-Guided Spectral-Temporal Adversarial Reconstruction Network for Gap Filling of Cloud-Contaminated Sentinel-2 Time Series

This dataset contains the supporting data and materials for the VISTAR study, including synthetic cloud configurations, model predictions, evaluation metrics, and source data for the figures and tables. The original Sentinel-1 and Sentinel-2 observations are publicly available through Google Earth Engine. ## Data Scope and Availability A single VISTAR training run uses the complete three-year Sentinel-2 time series, comprising approximately 108 observations at 10-day intervals. During inference and operational application, any window of 30 consecutive observations at 10-day intervals can be selected from within this three-year period. The complete dataset consists of large multiband and multitemporal raster products. Its total volume would exceed the current upload capacity displayed for this Zenodo record, which is limited to 100 files and 50.00 GB. Uploading the complete archive would also require substantial data-transfer time. Therefore, the deposited original Sentinel-2 data cover one complete 30-step input window, while the experimental outputs are provided for one representative time step. These representative outputs correspond directly to the examples presented in the manuscript. All quantitative results, including the method comparisons, sensitivity analyses, ablation studies, window-length analysis, and cross-regional validation, together with the data splits, random seeds, and experimental parameter settings, are fully reported in the manuscript. The complete multitemporal output archive is not included because of the repository capacity limitation but may be requested from the corresponding author, Xiufang Zhu, Beijing Normal University ([e-mail]). Full output archives produced by the external comparison methods are not redistributed. Publicly accessible reference implementations are available at: - HANTS-GEE: https://github.com/jiezhou87/HANTS-GEE- NB_Re3: https://github.com/HeartBread/NB_Re3- U-TILISE: https://github.com/prs-eth/U-TILISE The original Sentinel-1 and Sentinel-2 observations used in this study are publicly available through Google Earth Engine. The deposited files document the data structure and provide representative inputs and outputs for verification of the results presented in the manuscript.