Overly Intense Convective Updrafts Exposed as a Significant Contributor to Model Biases

Varble, A., Pacific Northwest National Laboratory

Cloud Processes

Cloud Life Cycle

Varble A, EJ Zipser, AM Fridlind, P Zhu, AS Ackerman, J Chaboureau, J Fan, A Hill, B Shipway, and C Williams. 2014. "Evaluation of cloud-resolving and limited area model intercomparison simulations using TWP-ICE observations: 2. Precipitation microphysics." Journal of Geophysical Research: Atmospheres, 119(24), 10.1002/2013jd021372.


Median profiles of maximum vertical velocity (a, c) and (b, d) radar reflectivity for 3D defined convective updrafts beginning below 1 km and ending above 15 km for the period of 1310Z to 1750Z 23 January 2006. Solid black lines represent observations and all other lines represent simulations, CRMs in (a-b) and LAMs in (c-d).


Example vertical cross sections through deep updrafts in quarter domain DHARMA-2M simulations showing (a-c) MSE (filled) and (d-f) total condensate (filled) with vertical velocity (thin black contours: 1 and 5 m/s; thick black contours: 10, 15, 20, and 25 m/s) overplotted. The 100 m run is shown in (a) and (d), the 100 m run degraded to 900 m in (b) and (e) and the 900 m run in (c) and (f). Updrafts in the 100 m run have slightly lower mass fluxes and less condensate aloft than in the 900 m run.


Median profiles of maximum vertical velocity (a, c) and (b, d) radar reflectivity for 3D defined convective updrafts beginning below 1 km and ending above 15 km for the period of 1310Z to 1750Z 23 January 2006. Solid black lines represent observations and all other lines represent simulations, CRMs in (a-b) and LAMs in (c-d).

Example vertical cross sections through deep updrafts in quarter domain DHARMA-2M simulations showing (a-c) MSE (filled) and (d-f) total condensate (filled) with vertical velocity (thin black contours: 1 and 5 m/s; thick black contours: 10, 15, 20, and 25 m/s) overplotted. The 100 m run is shown in (a) and (d), the 100 m run degraded to 900 m in (b) and (e) and the 900 m run in (c) and (f). Updrafts in the 100 m run have slightly lower mass fluxes and less condensate aloft than in the 900 m run.

Mesoscale simulations run at cloud-resolving scales often fail to reproduce observed cloud and precipitation structures of convective systems and their dependence on large-scale environmental conditions. This is an important issue because these types of simulations are increasingly used in satellite retrievals and improving representation of convective systems in general circulation models. In this first of a two-part study attempting to find possible causes for cloud and precipitation biases, detailed comparisons of ten 3D cloud-resolving model (CRM) and four 3D limited area model (LAM) mesoscale simulations with radar retrievals of reflectivity and vertical winds were performed for the 23-24 January 2006 active monsoonal mesoscale convective system during the Tropical Warm Pool – International Cloud Experiment in Darwin, Australia. The simulations employed ~1-km horizontal grid spacing and a range of bulk microphysics parameterizations.

In agreement with previous studies, simulated convective radar reflectivity was high biased, and assumptions used in bulk ice microphysics parameterizations modulated the magnitude of reflectivity biases by altering the distribution of condensate mass between different hydrometeor species and sizes. However, high-biased reflectivity in simulations also resulted from deep convective updrafts that were also too strong and lofted too much condensate above the freezing level. Some of the most intense simulated updrafts contained nearly undilute cores and exhibited unexpected supercellular characteristics despite simulated domain-mean thermodynamic and wind profiles that closely resemble those observed. Large rainwater contents in simulated intense updraft cores were lofted above the -4°C level and quickly frozen, which amplified latent heating and significantly increased maximum vertical wind speeds in the upper troposphere. The resultant high ice water contents and large hydrometeors in simulations were efficiently lofted to high altitudes and advected over large regions, producing unrealistically large convective regions and radar reflectivities. Simulated updrafts were weakened in a higher resolution 100-m grid spaced quarter domain CRM simulation, but not enough to match observational retrievals.

In addition to insufficient grid spacing, the overly strong simulated convection appears to be linked to model forcing biases and interactions between model dynamics and parameterized microphysics that promote stronger convection than observed. This conclusion is true for both CRM and LAM simulations, which use different domain sizes, boundary conditions, and large-scale forcings, but produce several similar convective and stratiform precipitation biases. As a part of eliminating these biases, many accurate and coincident convective updraft dynamical and microphysical measurements at mid levels are required for a variety of well-characterized environmental conditions.