A Multi-Scale Post-Processing Technique for Short-to-Long Range Ensemble Streamflow Prediction

A novel multi-scale post-processor for ensemble streamflow prediction, MS-EnsPost, is described and comparatively evaluated with the existing post-processor in the NWS's Hydrologic Ensemble Forecast Service, EnsPost. MS-EnsPost uses data-driven correction of magnitude-dependent bias in simulated flow, multiscale regression using observed and simulated flows over a range of temporal aggregation scales, and ensemble generation using parsimonious error modeling. For comparative evaluation, 139 basins in 8 River Forecast Centers in the US were used. Streamflow predictability in different hydroclimatological regions is assessed and characterized, and gains by MS-EnsPost over EnsPost are attributed. The ensemble mean prediction results show that MS-EnsPost reduces the root mean square error of Day-1 to -7 predictions with EnsPost of mean daily flow by 5 to 68 percent, and that, for most basins, the improvement is due to both bias correction and multiscale regression. The ensemble prediction results show that MS-EnsPost reduces the mean Continuous Ranked Probability Score of Day-1 to -7 predictions with EnsPost of mean daily flow by 2 to 62 percent, and that the improvement is due mostly to improved resolution than reliability. Examination of the mean Continuous Ranked Probability Skill Scores (CRPSS) indicates that, for most basins, the improvement by MS-EnsPost is due to both magnitude-dependent bias correction and exhaustive utilization of hydrologic memory through multiscale regression. Comparison of the mean CRPSS results with hydroclimatic indices indicates that the skill of ensemble streamflow prediction with post processing is modulated largely by the fraction of precipitation as snow and, for non-snow-driven basins, mean annual precipitation.