Transcript CASA POster Template

Parameter Range Study of Numerically-Simulated
Isolated Multicellular Convection
Z. DuFran, B. Baranowski, C. Doswell III, and D. Weber
Simplifications to the Problem
Introduction
Convective storms contain areas with marked ascent which is commonly called the
updraft. Although the magnitude of updrafts was once assumed to be constant, more
recent observations have proven otherwise. The goal of this study is to simulate the
non-continuous nature of updrafts in isolated multicell convection similar to
observations, and to relate the timing of the updraft regeneration to our selected
background environmental parameters.
Computing
•Extensive numerical cloud models are necessary for experiments in meteorology.
•This study uses ARPI, the research version of the Advanced Regional Prediction
System (ARPS), developed by Dan Weber of the Center for Analysis and
Prediction of Storms (CAPS).
•The model solution statistically converges for decreasing resolution, but the
calculation and memory requirements increase quickly (see Table 1).
•In order to complete this project in a human lifetime, model runs must be carried out
on the available supercomputing resources, using message passing interface
(MPI) domain decomposition.
•4 processors are being use in each horizontal direction (16 processors) for each job
•5400 second simulations
•48 x 48 x 20 km model domain with 100 m resolution.
•This study will use OSCER’s Topdawg, as well as NCSA and PSC machines.
Table 1. Number of computations and time required for a
single model run with various resolutions (TopDawg)
Resolution
Total Number of
Computations
Time for 1 job
using 1 processor
(hrs)
Time for 1 job
using 1000
processors (hrs)
Time for 1000 Jobs
using 1000
processors (years)
150m
0.2 PetaFLOPS
123
0.12
0.01
100m
1.1 PetaFLOPS
664
0.66
0.08
75m
3.7 PetaFLOPS
2168
2.17
0.25
50m
19.0 PetaFLOPS
11004
11.00
1.26
25m
304.2 PetaFLOPS
176062
176.06
20.10
The Buckingham PI-theorem is used to conglomerate these variables and
reduce the number of model runs required to span the desired parameter
ranges. This cuts the number of runs from potentially tens of thousands to
270. Additionally, Ben Baranowski has developed a script that creates the
necessary model input file and batch script needed to run the model for a
range of non-dimensional parameter values, specified by the user. This script
also submits those jobs to the computing resource being used.
Analysis
Although we are using continuous forcing mechanisms, the buoyant parcels
(called “bubbles”) pinch off from the forcing region and rise through the cloud
at discrete intervals. The last unanswered question is how we define the
bubbles so that we can determine the regeneration rate. Figure 1 (below)
includes contours of cloud water content that act to visually map out the cloud
outline of the simulated storm. This particular image is a vertical slice through
the center of the storm at 2700 seconds (half way through the simulation).
Circle B is a convective “bubble” that has just lifted away from the forcing
region and a new “bubble” is forming within circle A.
The ARPI model includes a plotting routine that allows us to look at many
useful variables throughout the evolution of each simulation. However, the
number of model runs that are being executed in this study will impair our
ability to qualitatively inspect each run. We will devise a method to analyze
the large amount of data generated by these model runs.
Experiment Design
The parameters of interest to this study are those describing the geometry of the
forcing region (horizontal and vertical radii), the magnitudes of each type of forcing
(convergence and thermal buoyancy), the relative buoyancy and the environmental
wind shear. We have chosen to represent the magnitude of thermal buoyancy with
the convective available potential energy (CAPE) of the most unstable parcel. The
CAPE of the environment is used to represent the background instability or buoyancy.
The nature of the CAPE calculation causes some ambiguity in the value computed.
For instance, a typical value of CAPE for a convective environment might be 2500
m2s-2. But this value can be attained by a large amount of buoyancy over a shallow
layer in the atmosphere, or by a deep layer with marginal buoyancy. Therefore, this
study uses a CAPE value that has been normalized with respect to the depth over
which that CAPE was computed. The CAPE computation is part of a pre-processing
step that prepares the input file, adjusts the sounding profile and other job control
parameter for use by the simulation and supercomputing resource.
Table 2. Dimensional parameters and respective ranges
Parameters to be tested
Ranges
Source Horizontal Size
(rx, ry)
500 m – 15 km
Source Vertical Extent
(rz)
700 m – 1.5 km
max normalized parcel CAPE
(NCAPEp)
0.04 - 1 J kg-1 m-1 (ms-2)
max normalized environment CAPE (NCAPEe)
0.008 – 0.33 J kg-1 m-1 (ms-2)
Wind Shear
(u/z)
0 - 0.005 s-1
Convergence
(-)
0.005 – 0.001 s-1
Figure 1. East-West cross section of cloud water content
from a numerical simulation.
Summary
Sensitive numerical experiments for isolated multicellular convection are limited.
There have been several studies concentrating on updraft regeneration in twodimensional storms like squall lines. However, the factors influencing the
regeneration rate for isolated convection is still speculation. It is our hope that there
will be several relationships between the updraft regeneration rate and the chosen
parameters (within the respective ranges). The proposed range of parameter
values will require 270 model simulations. These initial simulations might show a
high sensitivity to one or more of the parameters, motivating additional study and
simulations over a smaller range to properly characterize those relationships.
This work is supported primarily by the National Science Foundation Grant ATM-0350539. Any opinions, findings, conclusions, or recommendations
expressed in this material are those of the authors and do not necessarily reflect those of the National Science Foundation.