The Self-Aggregation of Convection in Idealized Numerical Simulations using Different Cumulus Parameterizations

This study is motivated by the observation that the National Centers for Environmental Prediction Global Forecast System (GFS) systematically loses 10–15% of the total column precipitable water (PW) over the tropical western North Pacific (defined as the region 0–20°N and 120–180°E) in its medium-range (defined here as days 4–7) forecasts compared to the verifying GFS analysis. The loss of PW occurs during quiet regimes in which a tropical cyclone is not present at the initial time. During these quiet regimes, PW decreases with increasing forecast lead. Given that the PW loss worsens during the forecast, we hypothesize that it is driven by the model physics, which includes the cumulus parameterization.

In the present study, we aim to use the WRF-ARW model in a simplified idealized framework to test the behavior and evolution of convection and atmospheric water vapor during long model integrations that employ different cumulus parameterizations. Specifically, we will compare a “GFS-like” simulation that uses the Simplified Arakawa-Schubert (SAS) cumulus parameterization to simulations that use the Tiedtke, Kain-Fritsch, and Grell-Freitas cumulus schemes. The WRF simulations are integrated for 90 days with uniform fixed SST, no coriolis, and no background flow. A uniform horizontal grid spacing of 15 km with 44 vertical levels is used. A companion simulation run at 3 km horizontal grid spacing without cumulus parameterization is used for comparison against the coarser resolution simulations.

Preliminary results show that convection in the 3 km explicit simulation self-aggregates into a single mesoscale convective complex surrounded by extremely dry conditions (referred to as radiative-convective equilibrium; RCE) in about 40 days. The 15 km simulations that use the Kain-Fritsch and Grell-Freitas cumulus schemes behave similarly, while the simulation using the Tiedtke cumulus scheme reaches RCE in about 30 days. Unlike all other simulations, the “GFS-like” simulation reaches RCE in 10–15 days. As early as day 3, the SAS simulation has much lower domain-average PW compared to the other simulations. It appears that convective updrafts early in the model simulation are deeper and more intense in the SAS simulation. The compensating subsidence contributes to rapid drying throughout the rest of the domain, which enhances the rate of aggregation. The similar behavior of PW in the SAS simulation and the operational GFS over the tropical west Pacific during 2012–2014 suggests the possibility that modifying and testing the SAS scheme in the WRF framework may offer avenues to explore in modifying and improving the SAS scheme in the operational GFS.