Title: INSPIRATION:
1Retrieving global sources of aerosols from MODIS
by inverting GOCART model
INSPIRATION
INVERSION Oleg Dubovik1
Tatyana Lapyonok2 MODIS Yoram
Kaufman2, GOCART Mian Chin2
Paul Ginoux3
INSPIRATION
1- LOA/CNRS, University of Lille, France 2-
NASA/GSFC, Greenbelt, MD, USA 3 - NOAA/GFDL,
Princeton, NJ, USA
Yoram Kaufman2
Yoram J. Kaufman Symposium On Aerosols, Clouds
and Climate NASA/GSFC, , May 30, 31, and June 1,
2007
2The idea of the approach
- GOCART
- meteorology
- transport
- chemistry
- removal
- emissions?
MODIS observations of ambient aerosol
INVERSION (e.g. via adjoint model)
Improved sources (location and strength)
Improved agreement of modeling with
observations Improved characterization of
climate, air quality, etc.
3Basic principle of inversion
Forward Model
Matrix m -lines, n- columns
Least Square Method (mgtn)
Criterion of solution optimization
4Vector of global aerosol mass
Vector elements - values of aerosol mass
Mpm(ti , xj , y k , zn)
(p 1, ., Np) Time and space coordinates
zn n ? ?z (n 1,,Nz) yk k ? ?y ( k
1,,Ny) xi j ? ?x ( j 1,,Nx ) ti
i ? ?t ( i 1,,Nt )
M
Index convention p (i-1) Nx Ny Nz (j-1) Ny
Nz (k-1) Nz n NpNt Nx Ny Nz
Z
m( t?t, x, y, z )
Z
m( t, x, y, z )
Y
?z
?y
X
?x
Y
Z
t
t
?z
Y
?y
X
X
?x
5GOCART ModelGoddard Chemistry Aerosol Radiation
and Transport Model
- An atmospheric process model using assimilated
meteorological fields from the Goddard Earth
Observing System Data Assimilation System (GOES
DAS) - Including major types of aerosol, sulfate, dust,
BC, OC, and sea-salt from both anthropogenic and
natural sources - Calculating aerosol composition, 4-D
distributions, optical thickness, radiative
forcing
(spatial resolution 20 x 2.50, time resolution 20
min)
6Dimension of the Problem
Longitude Points (144) X Latitude Points
(91) X Vertical Layers (30) X Time
steps (12 days) 5,000,000
Longitude Points (144) X Latitude Points
(91) X Time steps (12 days) 200,000
5,000,000 X 200,000
7Transport Inversion
Transport Operator
Aerosol Mass
Source - Aerosol Emission
Least Square Method
Measured Aerosol Mass
8Steepest Descent Method
for p??, steepest descent method is equivalent
to LSM
???
It converges as
-has predominant diagonal due to
local character of aerosol
transport
9Transposed Adjoint Transport
Transport Model
?
- Adjoint (Transpose) Transport
For Transport -reversed order of time
integration -reversed order of component
processes (e.g., advection ? backward
advection, falling ? uplifting,
updraft ? falling)
10INPUT
Time Integration of Transport Model
initial mass m(t0,x)
emitted mass s(t,x) (t0 t tmax)
m(t1,x)T(t0,x)(m(t0,x)s(t0,x))?t
Ti - operators of isolated transport
processes - advection, -
diffusion, - convection, - wet
deposition, - dry deposition, etc.
m(t2,x)T(t1,x)(m(t1,x)s(t1,x))?t
t
m(t3,x)T(t2,x)(m(t2,x)s(t2,x))?t
m(t5,x)T(t1,x)(m(t1,x)s (t1,x))?t
m(t2,x)T(t1,x)(m(t1,x)s (t1,x))?t
m(t2,x)T(t1,x)(m(t1,x)s (t1,x))?t
ti1 ti?t
Matrix equivalent
OUTPUT
transported mass m(t,x) (t0 t tmax)
M - vector of global aerosol mass, M0 - vector of
global aerosol mass at t0, S - vector of global
aerosol emissions, T - matrix defining mass
transport
11Backward Time Integration of Adjoint Transport
Model
INPUT
Residual obtained using sp(t,x) ?p(t,x)
s-2 (t,x)(mp(t,x)- m(t,x)) (t0 t
tmax) mp(t,x) - mass simulated using
sp(t,x), m(t,x) - mass measurements
tn-1 tn- ?t
?sp(tn,x)T(tn,x)(?p(tn,x))?t
Ti - adjoint operators of isolated transport
processes - advection, -
diffusion, - convection, - wet
deposition, - dry deposition, etc.
?sp(tn-1,x) T(tn-1,x)(?sp(tn,x)?p(tn-1,x))?t
?sp(tn-2,x) T(tn-2,x)(?sp(tn-1,x)?p(tn-2,x))?t
(-t)
m(t2,x)T(t1,x)(m)s (t1, ?t
Matrix equivalent
OUTPUT
?Sp TTC-1 ? Mp
Solution Correction sp1(t,x) sp(t,x)-
?sp(t,x)
M0 - vector of global aerosol mass at t0, TT -
transpose of transport matrix T, C - measurement
covariance matrix, ? Mp - vector of residuals (?
Mp M(Sp)- M),
?sp(t,x) (t0. t tmax)
12Implementation of Steepest Descent Method via
Adjoint Modeling
Matrix Formulation
Time Integration of Transport Model
Time Integration of Transport Model
Initialization (p0)
m(t0,x)
mp(t,x)
p-th Approximation of Solution
sp(t,x)
sp0(t,x)
Residual calculation ?p(t,x) s-2
(t,x)(mp(t, x)- m(t,x))
yes
no
END
IF (p lt pmax)
Solution correction
Backward Time Integration of Adjoint
Transport Model ?sp(t,x)
Backward Time Integration of Adjoint Transport
Model
sp1(t,x) sp(t,x)- ?sp(t,x)
13Emission Sources
Testing of inversion
Testing of inversion
Assumed emission
Initial guess
Retrieved (40 iterations)
14inverting model output Black Carbon
August 28, 2000 (107 kg of mass/day) (two weeks
inversion)
Assumed Emission
Retrieved Emission
15Optical Thickness August 28, 2000
sfit lt 0.007 (for instantaneous t values)
?BC(0.55) simulated by GOCART using assumed
emission
?BC(0.55) simulated by GOCART using retrieved
emission
16Optical Thickness(real coverage) August 28, 2000
sfit lt 0.009 (for instantaneous t values)
?BC(0.55) simulated by GOCART using assumed
emission
?BC(0.55) simulated by GOCART using retrieved
emission
17inverting model output Desert Dust
August 28, 2000 (108 kg of mass/day) (two weeks
inversion)
Assumed Emission
Retrieved Emission
18Optical Thickness August 28, 2000
sfit lt 0.008 (for instantaneous t values)
?dust(0.55) simulated by GOCART using assumed
emission
?dust(0.55) simulated by GOCART using retrieved
emission
19Model ? Observations
Extra - assumptions in the inversion
- 24 hours constant emissions
- Components
- Mfine(xyzt)
- (BC OC sulfates fine dust fine sea salt)
- Mcoarse(xyzt)
- (coarse dust coarse sea
- salt)
GOCART output
MODISproducts
- MASS M(xyzt)
- 2o?2.5o,20 min
- Components
- BC hydrophilic
- BC hydrophobic
- OC hydrophilic
- OC hydrophobic
- sulfates
- dust (size d.)
- sea salt
- (fine, coarse)
- Aerosol Optical thickness
- ?(xyt)
- 1o?1o,
- global coverage in 2 days
- Components
- ?fine(0.55?m)
- ?????coarse(0.55?m)
-
?
A priori constraints
Using GOCART emissions as a priori estimates (
assimilation)
20Steepest Descent Methodwith a priori
constraints
for p??, steepest descent method is equivalent
to multi-term LSM (Dubovik 2004, Dubovik and
King, 2000)
A priori estimates correction METHODS Kalman
filter, Rogers optimal estimates, Twomey,
Correction by a priori limitation of
derivatives METHODS Phillips-Tikhonov-Twomey
21MODIS Aerosol Product for 24 hours
MODIS ttotal(0.55) (Total AOD (degraded to 20 x
2.50 )
MODIS tfine(0.55) (Fine Mode AOD (degraded to 20
x 2.50 )
22inverting model output sub-sampled according to
actual MODIS coverage Black Carbon
August 28, 2000 (107 kg of mass/day) (two weeks
inversion)
Assumed Emission
Retrieved Emission
23Optical Thickness August 20-28, 2000
sfit lt 0.009 (for instantaneous t values)
?BC(0.55) simulated by GOCART using assumed
emission
?BC(0.55) simulated by GOCART using retrieved
emission
24inverting model output sub-sampled according to
actual MODIS coverage Fine Mode Aerosol
August 20-28, 2000(107 kg of mass/day) (two weeks
inversion)
Assumed Emission BCOCSulfates
Retrieved Emission Single Fine mode aerosol
25Optical Thickness August 28, 2000 (two weeks
inversion)
sfit lt 0.02 (for instantaneous t values)
sMODIS 0.030.05 ?
?BC(0.55) ?OC(0.55) ?sulfates(0.55) simulated
by GOCART using assumed emission
?BC(0.55) simulated by GOCART using retrieved
emission
26inverting model output sub-sampled according to
actual MODIS coverage Desert DUST
August 28, 2000 (108 kg of mass/day) (two weeks
inversion)
Assumed Emission
Retrieved Emission
27Optical Thickness August 28, 2000
sfit lt 0.006 (for instantaneous t values)
?BC(0.55) simulated by GOCART using assumed
emission
?BC(0.55) simulated by GOCART using retrieved
emission
28Fine Mode Aerosol Black Carbon Organic Carbon
Sulfate
August 20 - 28, 2000
MODIS Observations (opt. thickness)
Retrieved Emission
299 day average, Optical Thickness retrieval is NOT
CONSTRAINED to the land
sfit lt 0.04 (for instantaneous t values) sMODIS
0.030.05 ?
Optical Thickness (August 20-28, 2000)
MODISAERONET Observations (Fine Mode opt.
Thickness, (degraded to 20 x 2.50 )
Model output using retrieved emission
30Emission Sources COMPARISON
GOCART Emission Sulfates BC OC
Retrieved Emission (total fine aerosol)
Combination of satellite hotspots (TRMM and
ATSR), satellite burned area (MODIS), and a
biogeochemical model (CASA) Guido van der
Werf Jim Collatz
31Coarse Aerosol Desert Dust Sea Salt etc.
August 20 - 28, 2000
GOCART emissions Desert Dust (mass)
Retrieved emissions Coarse Aerosol (mass)
32Optical Thickness
sfit lt 0.05 (for instantaneous t values)
MODIS AOD (Coarse Mode AOD (degraded to 20 x 2.50
)
Model output using retrieved emission
33 Fine Mode Aerosol Monthly retrievals
February 2001
May 2001
July 2001
34OC BC 0.5 (MG) Inverted emissions
FRP CMG 0.5 (MW) Fire Radiative Power
Fossil Fuel Emissions OC BC adjusted for
anthropogenic sources
Vermote et al. 2007
35Correlation of FRP and emissions
FRP (Fire Radiative Power)
Vermote et al. 2007
36Estimating the Emission Factor
Vermote et al. 2007
37Perspectives
- retrievals over bright surfaces - sensitivity
to aerosol type - sensitivity to vertical
profile
- inverting data from MISR, POLDER,
APS,CALISPO, etc.
- improved coverage- retrievals over bright
surfaces - sensitivity to aerosol type-
sensitivity to vertical profile - sensitivity to
aerosol source variability
- inverting multi-instrument dataMODIS MISR
POLDER APS
- Deriving regional emissions from
geostationary observations
(METEOSAT)
- higher time resolution - higher spatial
resolution
INVERSION REFINEMENTS - using a priori
knowledge about emissions - inverting satellite
radiances