Evaluation of Precipitation Forecasts Based on a Convective-Scale Ensemble Prediction System over the Yangtze-Huaihe River Basin

In this study, a convective-scale ensemble prediction system (EPS) has been established over the Yangtze-Huaihe river basin, China. The EPS was implemented at a horizontal grid spacing of 3 km based on the WRF model (version 3.9), which was initialized daily at 00 UTC and ran out to 36 h during June 2015. First, using the NCEP global forecast system (GFS) analysis as the background, radar and in-situ observations were assimilated to ARPS data assimilation system (ADAS) with cloud analysis, which was applied to produce the diabatic initial condition for the convective-scale EPS. Satellite-gauge precipitation analysis at 0.1 degree resolution was used to verify precipitation forecasts. Compared with the direct downscaling from the GFS initialization, the forecasts with cloud analysis greatly reduced the spin-up time in the WRF model, through the adjustment of hydrometeors and vertical velocity. Multiple combinations of physical parameterization schemes, including microphysics, planetary boundary layer and land surface parameterizations, were used in ensemble members to account for model uncertainties. However, the experiment of a heavy rainfall event indicated that the spread of multiple physics ensembles alone was insufficient and ensemble mean forecasts by the convective-scale EPS did not improve the forecast errors in the GFS forecasts. Therefore, the perturbations derived from the initial conditions (ICs) of the global ensemble forecast system (GEFS) were added to the cloud analysis (with two microphysical schemes) in the convective-scale EPS. One control run and 14 perturbed members were generated by the EPS, forced by randomly selected 14 GEFS members as lateral boundary conditions (BCs). Short-range (3 h or 24 h) ensemble precipitation forecasts were evaluated for one month, including several heavy precipitation events.

In general, the convective-scale EPS provides skillful precipitation forecasts over the Yangtze-Huaihe river basin, in terms of rainfall location and intensity. Cloud analysis is a critical factor to improve 0-12 h forecasts, especially for strong convective events. Combination of IC perturbations from the large-scale EPS and multiple physical schemes improves the spread-skill relationship of ensemble precipitation forecasts, by mitigating the underdispersion and forecast errors in the convective-scale EPS. The improvements of cloud analysis (such as assimilation of satellite information) and convective-scale IC/BC perturbations need further investigations.