Optimal Interpolation with CMAQ

Assimilating AIRNOW Ozone
Observations into CMAQ Model to
Improve Ozone Forecasts
Tianfeng Chai1, Rohit Mathur2, David Wong2,
Daiwen Kang1, Hsin-mu Lin1, and Daniel Tong1
1. Science and Technology Corporation, 10 Basil
Sawyer Drive, Hampton, VA 23666, USA
2. U.S. Environmental Protection Agency, Research
Triangle Park, NC 27711, USA
This research is funded by NOAA, under collaboration between NOAA
and US EPA (agreement number DW13921548).
Background
• In meteorology, assimilating real-time
observations is essential in all
weather forecasting systems
• AIRNOW ozone measurements are
available in near real time, and can
be used to improve ozone forecasts
• Optimal Interpolation has potential to
be applied operationally for air quality
forecasting
Optimal Interpolation (OI)
• In a sequential assimilation, at each time
step, we try to solve the following analysis
problem
1
X  X  BH ( HBH  O) (Y  HX )
a
b
T
T
• In OI, we assume only a limited number of
observations are important in determining
the model variable analysis increment.
Domain, Grid, and AIRNOW Stations
Estimate Model Error Statistics w/
Hollingsworth-Lonnberg Method
• At each station, calculate
differences between forecasts
(B) and observations (O)
• Pair up AIRNOW stations, and
calculate the correlation
coefficients between the two
time series at the paired stations
• Plot the correlation as a function
of the distance between the two
stations,
Error Statistics
1
60
EB ~
Pair Density (1/km)
Correlation
0.8
40
0.4
30
0.2
20
0
10
-0.2
0
200
400
600
Distance(km)
800
1000
0
1200
Pair Density (1/km)
0.6
Correlation
14.2 ppbv
50
EO ~
3.3 ppbv
Correlation
length:
60 km
Setup of OI Assimilation Tests
• Model starts at 1200 GMT, 8/5/07
• Hourly AIRNOW observations assimilated in first 24 hours
• Model continues to run another day without observations
1200 UT, 8/5/07
1300
1400
1500
…
Hourly Ozone Observations
…
1200 GMT,
8/7/07
Observation-Prediction (in ppbv)
Day 1
R=0.59
R=0.78
1300 - 2400 Z
R=0.56
R=0.68
Day 2
Surface O3 at 1800Z, 8/5/07
Base Case
OI (Analysis)
Surface O3 at 1800Z, 8/6/07
Base Case
OI (Forecast)
Ozone Bias and RMS error
Bias
RMS error
4D-Var Data Assimilation
1.
2.
3.
4.
CMAQ v4.5 Adjoint was
developed at Virginia
Tech. by A. Sandu et al.
Adjoint available for:
Transport, Chemistry
Assimilation time
window is 15 hours
Only initial O3 are
adjusted to minimize
the cost functional,
OI vs. 4D-Var
Bias
RMS Error
Summary
• CMAQ model error statistics has been
estimated using Hollingsworth-Lonnberg
method
• The model error covariance is used in
optimal interpolation to assimilate AIRNOW
observations
• Assimilating AIRNOW observations into
CMAQ model using Optimal Interpolation
proves to be beneficial for the next-day
ozone forecasting
• The positive effect of assimilation is
throughout the second day, but the effect
on the night-time ozone forecasts is
minimal
• A 4D-Var data assimilation test shows
similar effect as OI
1hr obs?
Bias correction