9th Incremental Development of National Air Quality Forecasting Capability at 4 km horizontal resolution for the CONUS

1
9th Incremental Development of National Air
Quality Forecasting Capability at 4 km horizontal
resolution for the CONUS
Pius Lee1, Hsin-Mu Lin2, Fantine Ngan3,
Hyuncheol Kim4, David Wong5, Daniel Tong4,
Tianfeng Chai4, Yunsoo Choi4, Daewon Byun1, Rick
Saylor1, Ariel Stein4, Youhua Tang2, Jeff
McQueen6, Marina Tsudlko2, Jeff Young5, Ho-Chun
Huang2, Sarah Lu2, Catarina Tassone2, Ken Carey7,
and Ivanka Stajner7
(a)
(b)

• INTRODUCTION
• National Air Quality Forecasting Capability
  (NAQFC) is providing 48 hour forecasts of
  surface ozone concentration and its corresponding
  daily one-hour and 8-hour maxima for the
  CONtiguous United States (CONUS). The NAQFC
  numerical modeling system consists of coupling
  the National Centers for Environmental Prediction
  (NCEP) Weather Research and Forecasting
  Non-hydrostatic Mesoscale Model (WRF-NMM) with
  CMAQ. Both WRF-NMM and CMAQ are running at 12 km
  horizontal resolution.
• The Earth Systems Modeling Framework compliant
  National Environmental Modeling System NMMB still
  at 12 km horizontal grid spacing will replace the
  current WRF-NMM. However, multiple finer inner
  nested grids will be provided, including a 4km
  horizontal resolution grid covering CONUS. In
  anticipation of this planned upgrade of the
  meteorological model, hardware requirements of
  CMAQ running at 4km horizontal resolution over
  CONUS (Fig. 1) is investigated in this study.
• A prototype version of NAQFC on a finer
  horizontal grid is used to investigate challenges
  and advantages of refined spatial resolution for
  the CMAQ model. Comparison of such a 4km and 12
  km CMAQ forecasts for a late summer case 2009
  were performed. Fig. 2 shows initialization Fig.
  3 shows emission Figs. 4 and 5 show CMAQ output
  hourly averaged surface O3 concentration at
  forecast hours 1 and 5, respectively.
• This study focuses on hardware resource required
  to through-put CMAQ 4.6 with cb05 gas mechanism
  and aero4 configuration. The model runs in NCEPs
  Power6 IBM super-computer (Table 1).
• Deficiency of the emission handling procedure in
  this makeshift 4km run were manifested Namely,
  in this study the 4km emission is simply a
  smearing thin of the 12 km merged emission file
  using a 1/9 factor, as in proportion to the
  reduced cell-areas. Therefore although the
  NOx/VOC ratios are retained, the chemical regimes
  are very different as shown in Figs. 3 a and b.
• This study reinforces the importance of accurate
  emission modeling as shown in the very different
  forecast results shown in Figs. 4 and 5. A sample
  4 km grid spacing NO emission derived properly
  utilizing GIS technology is shown in Fig.6b.
  Preparation of such emission inputs for the 4km
  CONUS grid entails a considerable data processing
  effort as numerous surrogate files need to be
  reprocessed.

Table 1 Hardware requirement for 24 h 4km CONUS
(22 Layer 132679523191740 grid
cells)
RAM Disk No. Nodes PE configuration Projected Wall-clock to finish 24 h
RAM 55G Disk/24h 295G 10 13131431 8 h 10 min
RAM 55G Disk/24h 295G 12 15121801 8 h 28 min
(a)
Fig.2 Initialization for O3 (ppb) at surface for
12UTC September 2010 run (a) 12 (b) 4 km
(b)
(a)
(b)
Fig.3 NOx emission (mole s-1) at surface at
12UTCSeptember 2010 (a) 12 (b) 4 km
(c)
(a)
(b)
Fig. 6 GIS derived NO emission (mole s-1) for
(a) 12km and (b) 4 km grids, and (c) with b
minus a for 12UTC 21 September 2009 over
region Houston and vicinity
(b) 4 km horizontal grid spacing

• Summary
• An engineering test to investigate the
  feasibility of running 4 km horizontal grid
  spacing for CONtiguous U.S. (CONUS) with CMAQ
  version 4.6 configured with cb05 gas phase
  mechanism and aero4 totaling 103 species has been
  tried out in NCEPs Power6 IBM super-computer.
• At 22 vertical levels, the required run time
  memory is about 55 Gbytes. And for a 24 hour
  forecast, the needed disk space for I/O is about
  0.3 Tbytes.
• The parallelization speed-up showed negative
  return beyond 144 Processing Elements (PE). It is
  demonstrated that parallelization efficiency
  requires improvement to take advantage of
  massively parallel computers.
• The 4 km run with 144 PE completed about 11
  forecast hours worth of simulation when it
  exhausted the 4 hours CPU limit allowed on the
  NCEP Power6 computer. Further effort of
  aggressively optimize CMAQ is desired to fit the
  48 hour forecast with the 4 hours limit.
• Deficiency of the emission handling procedure in
  this makeshift 4km engineering test run were
  manifested. Emission derived properly utilizing
  GIS technology is required to prepare emission
  inputs for the 4km CONUS grid.

Fig.4 Model predicted hourly averaged surface O3
(ppmV) at surface valid at 13UTC September 2010
(a) 12 (b) 4 km
795 GC
(b)
(a)
1326 GC
Fig.1 Decomposition _at_ 12x12 PEs for CONUS _at_ (a)
12 (b) 4 km

• ACKNOWLEDGEMENTS
• The authors are grateful to Drs. Rohit Mathur,
  Jon Pleim, Tanya Otte, Shaocai Yu, George
  Pouliot, and Ken Schere of the Atmospheric
  Modeling Division, USEPA , for their technical
  input and discussion. Advice from Dr. Daiwen Kang
  of CSC is deeply appreciated.
• 1 NOAA/OAR/ARL, 1315 East West Hwy, Room 3316,
  Silver Spring, MD 20910 pius.lee _at_noaa.gov
• 2 I. M. System Group, Inc. , Rockville, MD.
• 3 University Corporation for Atmospheric
  Research, Boulder, CO.
• 4 Earth Resources Technology, Annapolis
  Junction, MD.
• 5 EPA, Research Triangle Park, NC.
• 6National Centers for Environmental Prediction,
  Camp Springs, MD.
• 7Noblis, Inc., Falls Church, VA.

Fig.5 Same as Fig. 4 but for 17UTC September
2010