HiPACE-S2S: Historical quantile conditioning for probabilistic correction and scale-consistent downscaling of subseasonal weekly precipitation totals

Systematic biases and coarse spatial resolution limit the direct use of subseasonal ensemble precipitation forecasts. We present HiPACE-S2S, a framework for probabilistic correction and statistical downscaling of ECMWF (EC) forecasts from 1.5◦ to a 0.1◦ observation grid. Historical reforecasts and precipitation observations are used to learn conditional distributions and generate 256 full-field scenario trajectories of weekly precipitation totals for six consecutive, non-overlapping forecast weeks. EC member percentiles within historical reforecast distributions characterize dynamical anomalies, while local observed climatological quantiles define the precipitation scale. Area-weighted projection ensures that the area-weighted mean of the fine-grid values within each parent cell equals the corresponding corrected parent-cell sample in every scenario. Training directly optimizes local probabilistic losses computed after projection and the continuous ranked probability score (CRPS) of regional mean precipitation. Complete historical six-week trajectories, selected using forecast conditions, then organize scenario pairing across parent cells. For each parent cell, a single permutation of scenario indices is applied to the parent and all its fine-grid cells, identically across all six forecast weeks. On 345 initializations during 2021–2023, HiPACE-S2S achieves the best overall point estimates for CRPS, the ten-category ranked probability skill score (RPSS), and the Brier skill score (BSS) at Q10, Q50, Q90 and Q95 among the compared methods when scores are pooled across the study domain and the six non-overlapping weekly forecast targets. Relative to raw EC, CRPS decreases by 20.86%, and Brier scores for Q90 and Q95 events decrease by 44.45% and 46.70%, respectively. Further evaluations cover consecutive multiweek precipitation accumulations starting at W1 over 7–42 days and basin-averaged precipitation in the Pearl and Yellow River basins. In both basins, CRPS pooled across the weekly forecast targets is lower than that of the training-period climatological reference.