Modeling approach to regional flux inversions at WLE Marek Uliasz Department of Atmospheric Science Colorado State University Who needs data? or
Jan 15, 2016
Modeling approach to regional flux inversions at WLEF
Modeling approach to regional flux inversions at WLEF
Marek UliaszDepartment of Atmospheric Science
Colorado State University
Who needs data?Who needs data?or
CSU RAMSCSU RAMS
LPD modelLPD model
influence functionsinfluence functionsfor concentrationfor concentration for vertical fluxfor vertical flux
Bayesian inversionBayesian inversion
modeling frameworkmodeling framework
*
00 0 0
*00
0 0 0
* * * *
0 00 0 0 0 0 0
( )
ˆ ˆ( ) ( )
yx
yx
y x
x y
LLT
z
LL H
t
L LT H T H
W E S Nx x L y y L
C
C qdxdydt
C C dxdydz
uC C u C C dydzdt vC C vC C dxdzdt
influence function for concentration measurements C*influence function for concentration measurements C*
concentration sample
*
00 0 0
*00
0 0 0
* * * *
0 00 0 0 0 0 0
( )
ˆ ˆ( ) ( )
yx
yx
y x
x y
LLT
z
LL H
t
L LT H T H
W E S Nx x L y y L
C
C qdxdydt
C C dxdydz
uC C u C C dydzdt vC C vC C dxdzdt
surface fluxes
influence function for concentration measurements C*influence function for concentration measurements C*
concentration sample
*
00 0 0
*00
0 0 0
* * * *
0 00 0 0 0 0 0
( )
ˆ ˆ( ) ( )
yx
yx
y x
x y
LLT
z
LL H
t
L LT H T H
W E S Nx x L y y L
C
C qdxdydt
C C dxdydz
uC C u C C dydzdt vC C vC C dxdzdt
surface fluxes
initial concentration
influence function for concentration measurements C*influence function for concentration measurements C*
concentration sample
*
00 0 0
*00
0 0 0
* * * *
0 00 0 0 0 0 0
( )
ˆ ˆ( ) ( )
yx
yx
y x
x y
LLT
z
LL H
t
L LT H T H
W E S Nx x L y y L
C
C qdxdydt
C C dxdydz
uC C u C C dydzdt vC C vC C dxdzdt
surface fluxes
initial concentration
inflow fluxes
influence function for concentration measurements C*influence function for concentration measurements C*
concentration sample
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0 03:00
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0 09:00
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0
y
[km
]
15:00
0 100 200 300 400 500 600 700 800 900 1000 1100
x [km]
-200
-100
0 21:00
R-tracer
1E-011
5E-011
1E-010
5E-010
1E-009
influence functions for surface fluxes: 1D PBLinfluence functions for surface fluxes: 1D PBL
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0 03:00
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0 09:00
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0
y
[km
]
15:00
0 100 200 300 400 500 600 700 800 900 1000 1100
x [km]
-200
-100
0 21:00
R-tracer
1E-011
5E-011
1E-010
5E-010
1E-009
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0 03:00
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0 09:00
0 100 200 300 400 500 600 700 800 900 1000 1100
-200
-100
0
y
[km
]
15:00
0 100 200 300 400 500 600 700 800 900 1000 1100
x [km]
-200
-100
0 21:00
A-tracer
1E-011
5E-011
1E-010
5E-010
1E-009
influence functions for surface fluxes: 1D PBLinfluence functions for surface fluxes: 1D PBL
WLEF tower – July 1997influence function for passive tracer
4 8 5 0 5 2 5 4 5 6 5 8 6 0 6 2 6 4 6 6 6 8 7 0 7 20
0.5
1
1.5
2
hei
gh
t [k
m]
4 8 5 0 5 2 5 4 5 6 5 8 6 0 6 2 6 4 6 6 6 8 7 0 7 2
sam pling tim e [hours]
0
0.5
1
1.5
2
hei
gh
t [k
m]
0
0.08
0.16
0.24
0.32
0.4
D=100km
D=500km
influence functions for inflow fluxes: 1D PBLinfluence functions for inflow fluxes: 1D PBL
4 8 5 0 5 2 5 4 5 6 5 8 6 0 6 2 6 4 6 6 6 8 7 0 7 20
0.5
1
1.5
2
hei
gh
t [k
m]
4 8 5 0 5 2 5 4 5 6 5 8 6 0 6 2 6 4 6 6 6 8 7 0 7 2
sampling time [hours]
0
0.5
1
1.5
2
hei
gh
t [k
m]
0
0.08
0.16
0.24
0.32
0.4
D=100km
D=500km
30m sample 1100m sample
0 6 12 18 24local time [hours]
-16
-14
-12
-10
-8
-6
-4
-2
0
2
4
6C
O2
flu
x [
mol
m-2
s-1 ]
CO 2 f luxapproxobs
respiration flux
assimilation flux
NEE constrains
used in inversion calculations: NEE=R+AR=R0
A=A0cvegRAD/(RAD+200)R0,A0 – unknown parameters to be estimatedRAD, cveg – from RAMS
CW
1000 km
x
z
D
q
sam
ple
s
D D D
q0
Two-Tower Inversions
• R is very well estimated
• A isn’t bad• NEE very hard to
estimate with unknown inflow
• Best estimates when towers are spaced optimally w.r.t. travel time (daytime)
0 40 80 120 160 200
0
25
50
75
100R fluxA fluxnet CO2 flux
0 40 80 120 160 200
0
25
50
75
100
RM
SE
/flu
x [%
]
0 40 80 120 160 200d - distance between towers [km]
0
25
50
75
100
2x400m towers
400m+30m towers
2x30m towers
200 400 600 800 1000
x [km]
200
400
600
800
1000
y [k
m]
W LEF tower - single level (76m)
200 400 600 800 1000
x [km]
200
400
600
800
1000
W LEF tower - all levels
Climatology of influence functions for August 2000
influence functions derived from RAMS/LPD model simulations passive tracer different configurations of concentration samples - time series from - a single level of WLEF tower - all levels of WLEF tower - WLEF tower + six 76m towers
200 400 600 800 1000
x [km]
200
400
600
800
1000
all towers
0.010.020.050.10.20.51251020
[ppm/umol]
Regional inversionsRegional inversions
reduction of uncertainty in flux estimationreduction of uncertainty in flux estimation
pseudo-data generation and ensemble inversion pseudo-data generation and ensemble inversion
0 300 600 900 1200
0
300
600
900
1200 N
E
S
W
Configuration of source areaswith WLEF tower in the centerof polar coordinates
Example of estimation of NEE averaged for August 2000 Bayesian inversion technique using influence function derived from CSU RAMS and Lagrangian particle model flux estimation for source areas in polar coordinates within 400 km from WLEF tower (better coverage by atmospheric transport) NEE decomposed into respiration and assimilation fluxes: R=R0, A=A0 f(short wave radiation, vegetation class) inversion calculations for increasing number of concentration data (time series from towers) NEE uncertainty presented in terms of standard deviation derived from posteriori covariance matrix inflow CO2 flux is assumed to be known from a large scale transport model in further work, concentration data from additional tower will be used to improve the inflow flux given by a large scale model
0
1
2
3
4
0
1
2
3
4
a-p
rio
ri N
EE
un
cert
ain
ty
0
1
2
3
4
NE
E e
sti
ma
tio
n u
nce
rtai
nty
[m
ol/s
/m2]
d istance [km]0-100100-200200-300300-400
W LEF 76m(single level) W LEF all levels W LEF all levels
+ 6 additional towers
N N NE S W E S W E S W
DIRECTIONAL SECTOR
0
1
2
3
0
1
2
3
a-p
rio
ri N
NE
un
cert
ain
ty
0
1
2
3
NE
E e
sti
ma
tio
n u
nce
rta
inty
[m
ol/s
/m2 ]
distance [km]0-100100-200200-300300-400
1 ppm 2 ppm 3 ppm
N N NE S W E S W E S W
DIRECTIONAL SECTOR
E F F E C T O F D A T A -M O D E L M I S M A T C H E R R O R ( a ll tow er s)
CSU RAMSCSU RAMS
LPD modelLPD model
influence functionsinfluence functionsfor concentrationfor concentration for vertical fluxfor vertical flux
Bayesian inversionBayesian inversion
modeling frameworkmodeling framework
Signature of Lake Superior inWLEF tower CO2 concentration data
Attempt to validate transport modeling
Example of using influence function to analyze observational data
… following data analysis by Noel R. Urban:
2000 WLEF data: CO2 concentration lake and land sectors determined by 396m wind direction wind speed < season median daytime only 10:00-17:00
… following data analysis by Noel R. Urban:
2000 WLEF data: CO2 concentration lake and land sectors determined by 396m wind direction wind speed < season median daytime only 10:00-17:00
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
Repeating analysis for all available CO2 data 1996-2001
0 2 4 6 8 10 121
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6CO2 concentration difference: lake-land
month
0.433
0.964
delta396 m
delta244 m
delta122 m
delta76 m
delta30 m
111 m
Repeating data analysis for August 2000
0 50 100 150 200 250 300 350 4000
0.1
0.2
0.3
0.4
0.5
0.6
0
delta k
39630 levelk
n 743
nlake 66
nland 46
problems: a lot of missing wind data at 396m (only 62% of wind data available during 1996-2001 daytime hours) sectors poorly represent land or water source areas
problems: a lot of missing wind data at 396m (only 62% of wind data available during 1996-2001 daytime hours) sectors poorly represent land or water source areas
Modeling approach to data analysis: RAMS simulation: (August 2000, 2 nested grids)
LPD model influence functions
Modeling approach to data analysis: RAMS simulation: (August 2000, 2 nested grids)
LPD model influence functions
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
0
0.001
0.005
0.01
0.05
0.1
0.5
1
Influence function: August 2000, entire domain
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
0
0.001
0.005
0.01
0.05
0.1
0.5
1
Influence function: August 2000, land
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
0
0.001
0.005
0.01
0.05
0.1
0.5
1
Influence function: August 2000, water
200 300 400 500 600 700 800 900 1000 1100
X [km]
200
300
400
500
600
700
800
900
1000
1100
Y [
km]
0
0.001
0.005
0.01
0.05
0.1
0.5
1
Influence function: August 2000, land
0 124 248 372 496 620 7440
0.2
0.4
0.6
0.8
"land" sector: 120-320 deg1
0
land_landi
land_alli
land_water i
land_alli
7440 i
0 124 248 372 496 620 7440
0.2
0.4
0.6
0.8
"lake" sector: 340-45 deg
lake_landi
lake_alli
lake_water i
lake_alli
i
0 124 248 372 496 620 7440
0.2
0.4
0.6
0.8
"land" sector: 120-320 deg
land_landi
land_alli
land_water i
land_alli
i
0 124 248 372 496 620 7440
0.2
0.4
0.6
0.8
"lake" sector: 340-45 deg1
0
lake_landi
lake_alli
lake_water i
lake_alli
7440 i
0 100 200 300 400 500 600 7000
2
4
6"land" sector: 120-320 deg
land_landi
lake_landi
i
0 100 200 300 400 500 600 7000
2
4
6"lake" sector: 340-45 deg
lake_landi
lake_water i
i
0 100 200 300 400 500 600 7000
2
4
6"land" sector: 120-320 deg
land_landi
lake_landi
i
0 100 200 300 400 500 600 7000
2
4
6"lake" sector: 340-45 deg
6
0
lake_landi
lake_water i
7440 i
what 400m tower sees in “land” and “lake” sectors in August 2000
Applying modeling approach to data analysis for August 2000
0 100 200 300 400 500 600 700 8000
0.2
0.4
0.6
0.8
hours
0.998
1.622 103
landday396 i lakeday396( )
i
7430 i
Relative contribution from Lake Superior and all land areas
0 100 200 300 400 500 600 700 8000
0.2
0.4
0.6
0.8
hours
landday396 i lakeday396( )
i
i
rows c396( ) 744
nnn2 c396 landday396 0.8( ) 458
nnn2 c396 lakeday396 0.5( ) 27
nnn2 c396 lakeday396 0.6( ) 9
nnn2 c396 lakeday396 0.7( ) 2
Applying modeling approach to data analysis for August 2000
0.1 0.2 0.3 0.4 0.54
2
0
2
4lake-land CO2 concentration [ppm]
fraction of Lake Superior contribution
3.976
2.041
sum2 c396 lakeday396 cc( ) sum2 c396 landday396 0.8( )( )
sum2 c244 lakeday244 cc( ) sum2 c244 landday244 0.8( )( )
0.60.2 cc
Applying modeling approach to data analysis for August 2000
0.1 0.2 0.3 0.4 0.54
2
0
2
4lake-land CO2 concentration [ppm]
fraction of Lake Superior contribution
3.976
2.041
sum2 c396 lakeday396 cc( ) sum2 c396 landday396 0.8( )( )
sum2 c244 lakeday244 cc( ) sum2 c244 landday244 0.8( )( )
0.60.2 cc
Applying modeling approach to data analysis for August 2000
0.1 0.2 0.3 0.4 0.54
2
0
2
4lake-land CO2 concentration [ppm]
fraction of Lake Superior contribution
3.976
2.041
sum2 c396 lakeday396 cc( ) sum2 c396 landday396 0.8( )( )
sum2 c244 lakeday244 cc( ) sum2 c244 landday244 0.8( )( )
0.60.2 cc
0.1 0.2 0.3 0.4 0.50
100
200
300number of obs data analyzed
fraction of Lake Superior contribution
221
4
nnn2 c396 lakeday396 cc( )
nnn2 c244 lakeday244 cc( )
0.60.2 cc
0 100 200 300 400 500 600 7000
0.2
0.4
0.6
0.8
lakeday396 i
i
0 100 200 300 400 500 600 700
340
360
380
c396( )i
i
time series analysis?
lake contributionlake contribution
CO2concentrationCO2concentration
RAMS/LPD simulations for WLEF areaRAMS/LPD simulations for WLEF area
RAMS/LPD simulations for WLEF areaRAMS/LPD simulations for WLEF area
Summer 2000Summer 2000
RAMS/LPD simulations for WLEF areaRAMS/LPD simulations for WLEF area
Summer 2000Summer 2000
Summer 2004Summer 2004
RAMS/LPD simulations for WLEF areaRAMS/LPD simulations for WLEF area
Pseudo-data inversions using the Ring of Towers (Summer 2000)
Pseudo-data inversions using the Ring of Towers (Summer 2000)
Summer 2000Summer 2000
Summer 2004Summer 2004
RAMS/LPD simulations for WLEF areaRAMS/LPD simulations for WLEF area
Pseudo-data inversions using the Ring of Towers (Summer 2000)
Pseudo-data inversions using the Ring of Towers (Summer 2000)
Summer 2000Summer 2000
Summer 2004Summer 2004
Real data inversions using the Ring of Towers (Summer 2004)
Real data inversions using the Ring of Towers (Summer 2004)
RAMS/LPD simulations for WLEF areaRAMS/LPD simulations for WLEF area
Pseudo-data inversions using the Ring of Towers (Summer 2000)
Pseudo-data inversions using the Ring of Towers (Summer 2000)
Summer 2000Summer 2000
Summer 2004Summer 2004
Real data inversions using the Ring of Towers (Summer 2004)
Real data inversions using the Ring of Towers (Summer 2004)
Data analysis using influence functionsData analysis using influence functions