Forecasting a 2018 Severe Rainstorm in Shanghai: The Role of Convection across the Gray Zone

Whether to use convection parameterization in the “gray zone” with horizontal grid spacing around 1-10 km has been a complication in high-resolution numerical weather prediction. In this range, convections are not fully resolved and yet many assumptions made in the traditional parameterization are no longer applicable. This study evaluates the effect of convection treatment across the gray zone by parameterization versus explicit solution in forecasting a record-breaking rainstorm in Shanghai on May 25, 2018. The forecasts are made by WRF using a one-way nesting approach with various grid configurations, including 27-9-3-km and 9-1-km, as well as two widely used cumulus parameterization schemes, Kain-Fristch (KF) and Grell-Freitas (GF). These forecasts are compared against observations to examine the utility of parameterized versus explicit convection solution across the gray zone.

In the triple nesting 27-9-3-km configuration, convection parameterizations at the 9-km grid have significant effects on the 3-km forecast. When using explicit convection in the 9-km grid, the 3-km grid predicts a weaker rain belt located much more north of observations. In contrast, using the traditional parameterization KF overestimates rainfall, with fake peaks near the noon. On the other hand, using the scale-aware parameterization GF realistically reproduces the rain belt with the core to the north of Shanghai, well captures the extreme precipitation in the nighttime, and correctly simulates events of light, moderate, large and heavy intensity. Similarly, in the double nesting 9-1-km configuration, explicit convection at the 9-km grid causes the 1-km grid forecast underestimates rainfall amount. When using convection parameterization at the 9-km grid, nesting the 3-km grid effectively improves the forecast for moderate and heavy rain. However, when using explicit convection at the 9-km grid, nesting the 1-km grid improves the forecast only for moderate rain, but excessively inhibits convection development and so underestimates the core rainfall intensity. Therefore, for a realistic forecast of this severe rainstorm, it is necessary to represent the organized convection by parameterization while explicitly resolve smaller scale convection. This can be effectively done by nesting grids with respectively parameterized and explicit convection.