forschungszentrum karlsruhe in der helmholtz-gemeinschaft ndacc h2o workshop, bern, july 2006 water...

Forschungszentrum Karlsruhein der Helmholtz-Gemeinschaft

NDACC H2O workshop, Bern, July 2006

Water vapour profiles by ground-based FTIR Spectroscopy:

study for an optimised retrieval and its validationby M. Schneider, F. Hase, and T. Blumenstock

1. Inversion on a logarithmic scale, why ?

2. Error characterization (… towards an optimal strategy for UT H2O retrieval)

3. 7-year record of water vapour above Izaña observatory

4. Summary and outlook

published in ACP: Atmos. Chem. Phys., 6, 811-830, 2006



0 10 20 30 40 500,10,5

2

10

305070

90

9899,5

theoretical 2 cdf

2 cdf for assumption of normal distributed vmrs

2 cdf for assumption of log-normal distributed vmrs

pro

bab

ility

[%

]

2


Examination of a-priori distribution. χ2 test checks for a normal distribution:

)(S)( 1T2aaa xxxx




)(S)()K()K( 1a

TT2aa xxxxxyxy

The cost function

is only correctly posted if state vector is transformed on a logarithmic scale !

On a linear scale a wrong a-priori distribution is assumed !Then the term

)(S)( 1a

Taa xxxx

provides for an overestimation of small x and an underestimation of large x !



2. Error estimations

Error estimation is performed by a „Monte Carlo“ method. Analytic error estimation like proposed by Rodgers (1990) is not appropriate due to non-linearities.

1. Forward calculation of assumed H2O profile2. Introducing of errors (measurement noise, temperature profile, ILS,

spectroscopic parameter, …)3. Inversion of the simulated spectra

This „Monte Carlo“ error estimation works only if we apply a large ensemble of H2O profiles, which obey the real H2O statistics ! We apply 500 ptu-sonde measurements of the years 1999 and 2000.




Errors are commonly presented as a mean (systematic error) and variance (random error).

Here we present them in a more general manner: by least squares fits

→ difference of regression curve from diagonal: offset + slope of regression line (mean + sensitivity as systematic error)

→ scattering around the regression curve (residual variance as random error):

_ 21reg

x



Error estimation by means of linear least squares fit2. Error estimation

Why? — there are two kind of systematic errors: 1. mean error (bias)

2. sensitivity error (slope≠1 or no linear correlation)

0 2 4 6 80

2

4

6

8

0 2 4 6 8-2

-1

0

1

2^

erro

neou

s va

lue

correct value

offset: x-x

xx^

slope: x/x

^___

rem

aini

ng e

rror

bias corrected erroneous value



In the following the errors of lower, middle, and upper tropospheric H2O are

1. estimated by Monte Carlo simulations

and

2. validated by comparing H2O from FTIR measurements with daily ptu-sondes




linear scale logarithmic scale

Estimation of total FTIR error (2.3-3.3km):

0 10 20 30

0

10

20

30

0 10 20 30

0

10

20

30

real partial column [1021/cm-2]

=0.98; m=0.98

retr

ieve

d pa

rt.

col.

[102

1/c

m-2]

lower troposphere (2.3-3.3km)


=0.98; m=0.98





FTIR vs. ptu-sondes at 2.3-3.3km. Problem: detection of different regions (FTIR: boundary layer; sonde: free troposphere)

0 10 20 30

0

10

20

30

0 10 20 30

0

10

20

30

sonde partial column [1021/cm-2]

=0.91; m=1.03

retr

ieve

d pa

rt.

col.

[102

1/c

m-2]

lower troposphere (2.3-3.3km)


=0.88; m=0.96






0 5 10 15 20

0

5

10

15

20

0 5 10 15 20

0

5

10

15

20


=0.97; m=0.83

retr

ieve

d pa

rt.

col.

[102

1/c

m-2]

middle troposphere (4.3-6.4km)


=0.97; m=0.86





FTIR vs. ptu-sonde at 4.3-6.4km:

0 5 10 15 20

0

5

10

15

20

0 5 10 15 20

0

5

10

15

20


=0.95; m=0.96

retr

ieve

d pa

rt.

col.

[102

1/c

m-2]

middle troposphere (4.3-6.4km)


=0.94; m=1.01






0 1 2

0

1

2

0 1 2

0

1

2


retr

ieve

d pa

rt.

col.

[102

1/c

m-2] upper troposphere (7.6-10.0km)

(LT-slant < 10x1021/cm2)

=0.74; m=0.43


=0.74; m=0.41





FTIR vs. ptu-sonde at 7.6-10.0km:

0 1 2

0

1

2

0 1 2

0

1

2

=0.64; m=0.41

sonde part. col. [1021/cm-2]

retr

ieve

d pa

rt.

col.

[102

1/c


(LT-slant < 10x1021/cm2

and S/N > 200)

=0.56; m=0.51

retrieved part. col. [1021/cm-2]




1110 1111 1112

0

1

2

3

4

1117.5 1121 1122

H2O

measurement retrieval residuum

x 10

abso

lute

rad

ianc

es [W

/(cm

2 ster

ad c

m-1)]

H2O

x 10

wavenumber [cm-1]

H2O

H2O

x 10

Can we reduce the error of the retrieved UT H2O amount ?

Case A: Yes, when absorption lines are unsaturated !





Upper troposphere (Case A): Estimation of total FTIR error for LT slant below 10 x 1021cm-2 (7.6-10.0km):

0 1 2

0

1

2

0 1 2

0

1

2

real partial cololumn [1021/cm-2]

=0.90; m=0.69

retr

ieve

d pa

rt.

col.

[102

1/c




=0.91; m=0.79





Upper troposphere (Case A): FTIR vs. ptu-sonde at 7.6-10.0km for LT slant below 10 x 1021cm-1:

0 1 2

0

1

2

0 1 2

0

1

2


=0.81; m=0.60retr

ieve

d pa

rt.

col.

[102

1/c



and S/N > 200)


=0.81; m=1.02





Upper troposphere (Case A): Estimation of total FTIR error for LT slant below 5 x 1021cm-2 (8.8-11.2km):

0,0 0,2 0,4

0,0

0,2

0,4

0,0 0,2 0,4

0,0

0,2

0,4

=0.89; m=0.61

retr

ieve

d pa

rt.

col.

[102

1/c

m-2]


tropopause (8.8-11.2km)


=0.94; m=0.66






Upper troposphere (Case A): FTIR vs. ptu-sonde at 8.8-11.2km for LT slant below 5 x 1021cm-1:

0,0 0,2 0,4 0,6

0,0

0,2

0,4

0,6

0,0 0,2 0,4 0,6

0,0

0,2

0,4

0,6

=0.83; m=0.53

retr

ieve

d pa

rt.

col.

[102

1/c

m-2]




and S/N > 200)

=0.86; m=0.90




2. Error estimationsCan we reduce the error of the retrieved UT H2O amount ?

Case B: Yes, when strong and moderately strong lines are fitted simultaneously !

1110 1111 1112 1113

0

1

2

3

4

5

1117,9 1121 1122

1221

0

1

2

3

4

5

1252 1253 1325

abso

lute

rad

ianc

es

[W/(

cm2 s

tera

d cm

-1)]

meaurement retrieval residuum x10

H2O

x10

wavenumber [cm-1]

H2O

H2O H

2O

abso

lute

rad

ianc

es

[W/(

cm2 s

tera

d cm

-1)]

HDOCH

4

CH4

wavenumber [cm-1]

HDO

N2O

CH4

HDO




strong lines strong and moderately strong lines

Upper troposphere (Case B): FTIR vs. ptu-sonde at 7.6-10.0km (LT slant below 25 x 1021cm-2):

0 1 2

0

1

2

0 1 2

0

1

2


retr

ieve

d pa

rt.

col.

[102

1/c

m-2]

=0.67; m=0.71


=0.84; m=0.89





Upper troposphere (Case B): FTIR vs. ptu-sonde at 10.0-12.4km (LT slant below 5 x 1021cm-2):

0,0 0,1 0,2

0,0

0,1

0,2

0,0 0,1 0,2

0,0

0,1

0,2


retr

ieve

d pa

rt.

col.

[102

1/c

m-2]

=0.75; m=0.57


=0.78; m=0.45



Recipe for retrieval of upper tropospheric H2O:

• Only retrieval on a logarithmic scale provides for a correctly posted cost function → this reduces the systematic errors if compared to a retrieval on linear scale. It leads to consistent time series and a good sensitivity for UT H2O amounts

• Simultaneous fit of strong and moderately strong lines provides for best exploitation of spectra → makes continuous observation of UT H2O feasible

• Most important errors are: smoothing error, ILS, line parameter, and temperature profile




3. 7-year record of water vapour above Izaña observatory

Above 9km retrieval requires unsaturated H2O lines → LT slant below 5 x 1021 cm-2 (in Izaña only for 10% of all measurements)

Above 7.5km retrieval requires moderate LT H2O amounts and simultaneous fit of strong and moderately strong lines

Up to 7km retrieval possible under all measurement conditions

0

10

20

0

5

10

15

200

1

2

0,0

0,1

0,2

0,00

0,02

0,04

0,06

LT (2.3-3.3km)

1999 2000 2001 2002 2003 2004 2005

MT (4.3-6.4km)

UT (7.6-10.0km)

for LT slant < 25x1021cm-2

pa

rtia

l co

lum

n a

mo

un

ts [1

021

/cm

2 ]

0

5

0

5

prec

iptib

le H

2O

[mm

]

0,0

0,5


for LT slant < 5x1021cm-2




Our study shows the following:

1. FTIR well-suited for continuous monitoring of lower and middle tropospheric H2O amounts.

2. For upper tropospheric H2O amounts a continuous monitoring is tricky, but feasible by FTIR, if: A: the inversion is perform on a log-scale B: strong and moderately strong lines are fitted simultaneously



Future work:

1. Produce continuous time series of UT water vapour from many historic FTIR measurements (e.g. measurements at Jungfraujoch or Kitt Peak could produce unique continuous 20-year records !)

2. Further improvements are still possible for dryer or higher situated sites: we could apply stronger H2O lines → improved sensitivity at higher altitudes. Detection of lower stratospheric water vapour variability from Jungfraujoch ?

3. Retrieval of HDO/H2O




Thank You !!!!!!!!!




Summary of differences between FTIR and ptu sonde data (sum of errors of sonde and FTIR). Theoretical estimations of FTIR errors are in brackets:

total 2.3-3.3km 4.3-6.4km 7.6-10.0km 8.8-11.2km

random 25 (4) 40 (21) 32 (24) 54 (50) 56 (47)

systematic(A) +6 (0) +3 (-3) -4 (-6) -40 (-31) -47 (-38)

total 2.3-3.3km 4.3-6.4km 7.6-10.0km 8.8-11.2km

random 25 (4) 47 (22) 33 (24) 58 (49) 51 (42)

sytematic(A) +6 (-1) -4 (-4) +1 (-1) +2 (-23) -10 (-33)

linear scale:

logarithmic scale:

(A): systematic spectroscopic line parameter errors are not considered





Upper troposphere (Case B): Estimation of total FTIR error at 7.6-10.0km (LT slant below 25 x 1021cm-2):





Upper troposphere (Case B): Estimation of total FTIR error at 7.6-10.0km (LT slant below 25 x 1021cm-1):



3. Vertical resolution

In the case of water vapour the averaging kernels only contain limited information about the sensitivity of the measurements.

Reason:

1. Non-linearities (kernels depend strongly on actual H2O profile)

2. variability of H2O decreases with height by 4 orders of magnitude → comparison of kernels for different heights is not straight forward !

Alternative representation of sensitivity:

Correlation matrices (correlation between original profiles and inverted profiles). The matrices give a realistic overview of the detectable atmospheric regions. Furthermore, they allow us to perform this sensitivity analysis for a realistic error scenario.

What about the averaging kernels ?





Estimation for LT slant column below 10 x 1021cm-2:

5 10 15

5

10

15

real atmosphere [km]

retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00





FTIR vs. sonde for LT slant below 10 x 1021cm-2:

5 10 15

5

10

15

sonde atmosphere [km]

retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00





Estimation for LT slant column below 5 x 1021cm-2:

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00





FTIR vs. sonde profiles for LT slant < 5 x 1021cm-2:

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00



5 10 15

5

10

15

H2O profiles: sonde vs. FTIR


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

5 10 15

5

10

15


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

5 10 15

5

10

15


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

5 10 15

5

10

15

common approach PROFFIT v9.4

LT

sla

nt

< 5

x1021

cm-2

L

T s

lan

t <

2.5

x1022

cm-2


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

Empirical validation of H2O profiles: ptu-sonde vs. FTIR:

Inter-species constraint not only improves quality of δD profiles but also quality of H2O profile:

H2O retrieval benefits from additional information present in HDO lines

Detection of UT/LS H2O with the proposed lines ?

For a definitive conclusion we need a larger ensemble of compared profiles (the correlation shown here bases on only 7 profiles) !

Slide from talk “Ground-based remote sensing of tropospheric HDO/H2O ratio profiles”



3. Validation of FTIR profiles with ptu-sondes


FTIR versus sonde profiles for all measurement days:

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00





Corresponding theoretical sensitivity assessment (whole ensemble):

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00





Corresponding theoretical sensitivity assessment (LT slant < 5 x 1021cm-2):

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00



5 10 15

5

10

15

H2O profiles: sonde vs. FTIR


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

5 10 15

5

10

15


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

5 10 15

5

10

15


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

5 10 15

5

10

15

common approach PROFFIT v9.4

LT

sla

nt

< 5

x1021

cm-2

L

T s

lan

t <

2.5

x1022

cm-2


FT

IR a

tmos

pher

e [k

m]

-0,64-0,52-0,400,400,520,640,760,881,00

Empirical validation of H2O profiles: ptu-sonde vs. FTIR:

Inter-species constraint not only improves quality of δD profiles but also quality of H2O profile:

H2O retrieval benefits from additional information present in HDO lines

Detection of UT/LS H2O with the proposed lines ?

For a definitive conclusion we need a larger ensemble of compared profiles (the correlation shown here bases on only 7 profiles) !

Slide from talk “Ground-based remote sensing of tropospheric HDO/H2O ratio profiles”




5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00


Whole ensemble:



0 1 2 3 4 50

1

2

3

4

5

0 1 2 3 4 5

-1

0

1

^

slope:x/x^

xx

erro

neou

s va

lue

correct value

erro

r (e

rron

eous

-cor

rect

)

eroneous value (retrieved value)

Error estimation by means of linear least squares fit2. Error estimation

Why? — there are two kind of systematic errors: 1. mean error (bias)

2. sensitivity error (slope≠1)





Slant column of lower troposphere (2.3-4.3km) below 10 x 1021cm-2:

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00

5 10 15

5

10

15


retr

ieve

d at

mos

pher

e [k

m]

-0,64

-0,52

-0,40

0,40

0,52

0,64

0,76

0,88

1,00





Summary of random error component of smoothing + measurement noise error:

5

10

0 50 100

5

10

0 50 100

random error [%]

altit

ude

[km

]

whole ensemble LT slant

<10x1022cm-2

LT slant

<5x1022cm-2

random error [%]





Summary of random error component parameter error (whole ensemble):

5

10

0 50 100

5

10

0 50 100

random error [%]

altit

ude

[km

] pressure broadening

phase error T profile

(no retr. of T) T profile

(simultaneaous retrieval of T)

random error [%]





Summary of random error component parameter error (LT slant below 10 x 1021cm-1):

5

10

0 50 100

5

10

0 50 100

pressure broadening




random error [%]

altit

ude

[km

]

random error [%]





Summary of random error component parameter error (LT slant below 5 x 1021cm-1

5

10

0 50 100

5

10

0 50 100

pressure broadening




random error [%]random error [%]

altit

ude

[km

]

forschungszentrum karlsruhe in der helmholtz-gemeinschaft ndacc h2o workshop, bern, july 2006 water...

Documents

forschungszentrum karlsruhe

error characterization

error estimations errors

analytic error estimation

mean systematic error

monte carlo error estimation

variance random error

systematic error scattering