A 1 km Monthly Multi-Timescale SPEI Dataset for Mainland China Reconstructed from Station Observations and Multi-Source Environmental Predictors

High-resolution multi-timescale SPEI data are important for drought monitoring and climate-impact studies, but long-term high-resolution datasets remain limited over large regions. This study developed and technically validated a station-supervised monthly multi-timescale SPEI dataset for mainland China from 1979 to 2018 on a 1 km grid, including SPEI-1, SPEI-3, SPEI-6, SPEI-12, and SPEI-24. Monthly labels were extracted from daily records at 427 stations using an exact source-calendar month-end rule. Monthly 0.1° CMFD predictors were mapped to stations and 1 km pixel centres using a NaN-aware bilinear sampler and combined with complete-window temporal features and NASADEM terrain covariates. Five timescale-specific LightGBM models generated 2359 nationwide monthly rasters. Random five-fold cross-validation yielded mean R2 values of 0.8807–0.9122; 200 km spatial-block validation with a 100 km exclusion buffer yielded 0.7857–0.8739; and forward blocked-period validation yielded 0.7577–0.8337. Full-period product comparison gave Pearson r = 0.9295 against Zhang-SPEI for SPEI-3, while comparison with HSPEI across five timescales yielded r = 0.7772–0.3267. Spatially buffered drought-condition validation produced F1 scores of 0.7021–0.7838 for moderate-or-worse drought, whereas extreme-drought Recall was 0.2397–0.3869. The dataset provides a long-term, multi-timescale resource for drought monitoring and climate-impact research across mainland China, with validation results characterizing its performance across regions, periods, and drought-severity levels.