Global Calibration of the GEOS-5 L-band Microwave ... · 18/09/2012 · Global Calibration of the...

GMAO-18Sep12 – 1 / 28

Global Calibration of the GEOS-5 L-band

Microwave Radiative Transfer Model over

Land Using SMOS Observations

Gabrielle De Lannoy, Rolf Reichle, Valentijn PauwelsGlobal Modeling and Assimilation Office (Code 610.1), NASA/GSFC, Greenbelt, MD, USA

Laboratory of Hydrology and Water Management, Fac. Bioscience Engineering, Ghent University, Belgium

18 September 2012

Outline

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 2 / 28

Soil Moisture and Microwave Radiances

Calibration

Brightness Temperatures

RTM Parameters

Remaining Biases

Conclusions

Soil Moisture and Microwave

Radiances

Soil Moisture andMicrowaveRadiances

Tb

RTM

SMOS/SMAP

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 3 / 28

Brightness Temperature (Tb)

Soil Moisture andMicrowaveRadiances

Tb

RTM

SMOS/SMAP

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 4 / 28

� measured by passive microwave sensors� dry and warm land surface → high Tb [K]� wet and cold land surface → low Tb [K]

The relationship between brightness tem-perature (Tb) and soil moisture (SM) is verysensitive to local parameters

0.2 0.4 0.6

220

240

260

280

h = 00.3

0.6

0.9

1.2

1.51.8

Ts

Dry <−−−−−−−> Wet SM [−]

Tbh

[K]

0.2 0.4 0.6

220

240

260

280

h = 0 0.3 0.60.9

1.21.5

1.8

Ts

Dry <−−−−−−−> Wet SM [−]

Tbv

[K]

Tau-Omega Radiative Transfer Model (RTM)

Soil Moisture andMicrowaveRadiances

Tb

RTM

SMOS/SMAP

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 5 / 28

Zero-order microwave radiative transfer model:

� soil contributions : moisture (dielectricconstant), roughness (h) → roughreflectivity (rp)

� vegetation contributions : opacity (τp)→ attentuation (Ap); scattering (ω)

� atmospheric contributions

TbTOA,p = Tbau,p + exp(−τatm,p)TbTOV,p

TbTOV,p = Ts(1− rp)Ap

+Tc(1− ωp)(1− Ap)(1 + rpAp)

+Tbad,prpAp2

GEOS-5 Catchment LSM (Fortuna 2.3; 5 cm surface layer):

� soil moisture, surface temperature (Ts = Tc), LAI,...

SMOS and SMAP

Soil Moisture andMicrowaveRadiances

Tb

RTM

SMOS/SMAP

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 6 / 28

SMOS (ESA, Soil Moisture Ocean Salinity)

1q

2q

SMAP (NASA, Soil Moisture Active Passive)

� launched November 2009

� L-band radiometer

� sensing depth = 5 cm

� 40 km resolution

� launch 2014

� L-band radiometer/radar

� sensing depth = 5 cm

� 3-40 km resolution

→ Calibrate radiative transfer model using Tb data from SMOS toprepare for SMAP

SMOS and SMAP

Soil Moisture andMicrowaveRadiances

Tb

RTM

SMOS/SMAP

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 7 / 28

SMOS (ESA, Soil Moisture Ocean Salinity)

1q

2q

SMAP (NASA, Soil Moisture Active Passive)

Each locationseen with multipleincidence anglesper overpassEach snapshot:σ ∼ 4 K

Each locationseen once peroverpass at fixed40o incidenceangle (conicalscan)σ ∼ 1.3 K

Tb[K]

30 6040 50

TBv

TBh

20

Incidence Angle [o]

SMAPSMOS

Calibration

Soil Moisture andMicrowaveRadiances

Calibration

Problem

θ-pol

Parameters

J

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 8 / 28

Before Calibration

GMAO-18Sep12 – 9 / 28

Large time-mean bias

200 220 240 260 280 300

1 Jan 2011 - 1 Jan 2012, 36 km⇐ SMOS observed Tb, H-pol, 42.5o

⇓ Model-minus-Observations , with themodel using prescribed RTM parameters(SMAP, L-MEB literature, ECMWF-SMOS)

Lit1 (SMAP) Lit2 (L-MEB) Lit3 (EC-SMOS)

−50 −40 −30 −20 −10 0 10 20 30 40 50 [K]

Before Calibration

GMAO-18Sep12 – 10 / 28

Large differences in temporal variability

5 10 15 20

1 Jan 2011 - 1 Jan 2012, 36 km⇐ SMOS observed Tb, H-pol, 42.5o

⇓ Model-minus-Observations , with themodel using prescribed RTM parameters(SMAP, L-MEB literature, ECMWF-SMOS)

Lit1 (SMAP) Lit2 (L-MEB) Lit3 (EC-SMOS)

−10 −8 −6 −4 −2 0 2 4 6 8 10 [K]

SMOS versus CLSM/RTM

Soil Moisture andMicrowaveRadiances

Calibration

Problem

θ-pol

Parameters

J

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 11 / 28

Ascending multi-angular Tb, annual (2011) global average

32.5 37.5 42.5 47.5 52.5 57.5

200

220

240

260

Inc Angle [o]

Tb H

[K]

a)

32.5 37.5 42.5 47.5 52.5 57.5240

250

260

270

280

Inc Angle [o]

Tb V

[K]

b)

SMOSLit1Lit2Lit3CalD2

� bias is angle- and polarization-dependent� after calibration with 2010 data, CLSM/RTM Tb will be unbiased vs.

SMOS in 2011

RTM-Parameters

Soil Moisture andMicrowaveRadiances

Calibration

Problem

θ-pol

Parameters

J

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 12 / 28

Different calibration scenarios:1) Selected parameters α(Nα), control vector:

Parameter [min, max] A B C D

hmin [0, 2.0] X X X X∆h ≡ hmax − hmin [0, 1.0] X X X Xω [0, 0.3] X XbH [0, 0.7] X X∆b ≡ bV − bH [-0.15, 0.15] X X

� microwave roughness h

� scattering albedo ω

� vegetation opacity τ= bp LEWT LAI

2) Prior information α0(Nα):bounded Gaussian, centralized around Lit1, Lit2, Lit3

⇒ total of 12 calibration scenarios (e.g. CalD2)

Multi-Variate Objective Function J

Soil Moisture andMicrowaveRadiances

Calibration

Problem

θ-pol

Parameters

J

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 13 / 28

For 6 angles, 2 polarizations, 2 orbit directions (∑

θ

∑

p

∑

d), penalize

� differences in long-term mean� differences in long-term standard deviation� deviations from prior (literature) parameter values

Search algorithm: particle swarm optimization; 1 year (2010)

J = Wm

∑

θ

H,V∑

p

A,D∑

d

Nθ,p,d

N

(< Tbo > − < Tb(α) >)2θ,p,dσ2m

}

J<.>,o

+Ws

∑

θ

H,V∑

p

A,D∑

d

Nθ,p,d

N

(s[Tbo]− s[Tb(α)])2θ,p,dσ2s

}

Js[.],o

+Wα

1

Nα

Nα∑

i=1

(α0,i − αi)2

σ2α0,i

}

JαNα parameters simul-taneously optimized ateach gridcell individually

Globally Averaged J

Soil Moisture andMicrowaveRadiances

Calibration

Problem

θ-pol

Parameters

J

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 14 / 28

4Lit1 Lit2 Lit3

Lit CalA CalB CalC CalD10

1

102

103

104

J [−

]

Scenario

a)

� ⇑ J is reduced after calibration� ⇑ CalA (only calibrating h) is

clearly not optimal� ⇒ mean bias (J<.>,o) is the

largest component, and mostreduced through calibration

� ⇒ standard deviation difference(Js[.],o) optimized, but stronglyconstrained by LSM variations

Lit CalA CalB CalC CalD10

1

102

103

104

J <.>

,o [−

]

Scenario

b)

Lit CalA CalB CalC CalD10

1

102

J s[.],

o [−]

Scenario

c)

Lit CalA CalB CalC CalD10

0

101

J α [−]

Scenario

d)

Brightness Temperatures

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

Global

Sensitivity

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 15 / 28

CLSM/RTM minus SMOS Tb - Mean

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

Global

Sensitivity

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 16 / 28

H-pol, 42.5 o, ascending, 1/1/2011-1/1/2012 (validation period)Lit1 (SMAP) Lit2 (L-MEB)

Lit3 (EC-SMOS) Calibrated

−50 −40 −30 −20 −10 0 10 20 30 40 50 [K]

� mostly unbiased long-term mean in the period after calibration� bias independent of incidence angle and pol (not shown)

CLSM/RTM minus SMOS Tb - Standard Deviation

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

Global

Sensitivity

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 17 / 28

H-pol, 42.5 o, ascending, 1/1/2011-1/1/2012 (validation period)Lit1 (SMAP) Lit2 (L-MEB)

Lit3 (EC-SMOS) Calibrated

−10 −8 −6 −4 −2 0 2 4 6 8 10 [K]

� preserved temporal variability, while reducing the bias

Sensitivity of Tb H(42.5o) to Soil Moisture

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

Global

Sensitivity

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 18 / 28

Lit CalA CalB CalC CalD

−2

−1.5

−1

−0.5

0

dTb H

/dS

M [K

/ 0.

01 m

3 .m−

3 ]

Scenario

Lit1 Lit2 Lit3

� rule of thumb for bare soil:dSM ∼ 0.01 m3.m−3 corresponds to dTbH (40o) ∼ 2-3 K

� Lit3 clearly lacks sensitivity, because of a too high h-parameter� after calibration, the sensitivity is reasonable and closest to Lit2,

regardless of prior values

→ Important for assimilation: a difference between observed andsimulated Tb will be mapped to a change in soil moisture

RTM Parameters

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Global

N America

Veget. class

τ

Remaining Biases

Conclusions

GMAO-18Sep12 – 19 / 28

Global Averages

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Global

N America

Veget. class

τ

Remaining Biases

Conclusions

GMAO-18Sep12 – 20 / 28

4Lit1 Lit2 Lit3

Lit CalA CalB CalC CalD0

1

<h>

[−]

Scenario

a)

Lit CalA CalB CalC CalD0

0.5

1

<τ>

[−]

Scenario

b)

Lit CalA CalB CalC CalD0

0.2

ω [−

]

Scenario

c)

Parameter estimates are moreconsistent for the mostcomplex calibration scenario.

Param A B C D

hmin X X X X∆h X X X Xω X XbH X X∆b X X

� < h > = time-averaged(soil moisture dependent)

� < τ > = time-averaged(LAI dependent),polarization-averaged

� ω = constant

Spatial Patterns: < τ >, < h >, ω

GMAO-18Sep12 – 21 / 28

BE

FO

RE

(Lit2

)A

FT

ER

(Cal

D2)

0 0.06 0.12 0.18 0.24 0.3

0 0.3 0.6 0.9 1.2 1.5

0 0.04 0.08 0.12 0.16 0.2

SM

OS

(ret

rieva

l)

� from homogeneous to locally optimized parameters

� vegetation/soil/climate patterns

(need RTM recalibration with new CLSM parameters)

� calibrated and SMOS τ have a similar magnitude

Spatial R: < τ >, < h >, ω

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Global

N America

Veget. class

τ

Remaining Biases

Conclusions

GMAO-18Sep12 – 22 / 28

L1 L2 L3 A1 A2 A3 B1 B2 B3 C1C2C3D1D2D3L1L2L3A1A2A3B1B2B3C1C2C3D1D2D3

Scenario

Sce

nario

a)

<h> [−]

L1 L2 L3 A1 A2 A3 B1 B2 B3 C1C2C3D1D2D3L1L2L3A1A2A3B1B2B3C1C2C3D1D2D3

Scenario

Sce

nario

c)

−0.2

0

0.2

0.4

0.6

0.8

1

ω [−]

L1 L2 L3 A1 A2 A3 B1 B2 B3 C1C2C3D1D2D3L1L2L3A1A2A3B1B2B3C1C2C3D1D2D3

Scenario

Sce

nario

b)

<τ> [−]

� spatial parameter patterns arereasonable consistent acrossmost calibration scenarios

� big differences with prior patterns

Optimal < τ >, < h >, ω

GMAO-18Sep12 – 23 / 28

Lit1 Lit2 Lit3 CalD2

1 2 3 4 5 6 7 8 9 10 12 14 160

1

<h>

[−]

IGBP Vegetation Class

a)

1 2 3 4 5 6 7 8 9 10 12 14 160

1

2

<τ>

[−]

IGBP Vegetation Class

b)

1 2 3 4 5 6 7 8 9 10 12 14 160

0.1

0.2

ω [−

]

IGBP Vegetation Class

c)

IGBP land cover

1 Evergreen Needleleaf Forest2 Evergreen Broadleaf Forest3 Deciduous Needleleaf Forest4 Deciduous Broadleaf Forest5 Mixed Forest6 Closed Shrublands7 Open Shrublands8 Woody Savannas9 Savannas10 Grasslands12 Croplands14 Crop and Natural Vegetation16 Barren or Sparsely Vegetated

� CalD2 = Lit2 as prior,

calibrate hmin, hmax, bH , bV , ω

� reasonable optimal parameters;

large within-class variability

� when using class-averaged (as

opposed to local) calibrated

parameters, the RTM still performs

better than with Lit1, Lit2 or Lit3→ use aggregate CalD2 pa-

rameters in unobserved regions

Vegetation Opacity τ

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Global

N America

Veget. class

τ

Remaining Biases

Conclusions

GMAO-18Sep12 – 24 / 28

01/01/2010 01/07/2010 01/01/2011 01/07/2011 31/12/2011

0.10.20.30.40.5

τ [−

]

- Little River, Georgia (• SMOS retrievals; – CLSM/RTM)- Walnut Gulch, Arizona (� SMOS retrievals; – CLSM/RTM)

⇒ Vegetation opacity values from the calibrated CLSM/RTM distiguishwell between more and less vegetated areas, and are consistent withSMOS retrievals

Remaining Biases

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 25 / 28

Hovm oller Plots: CLSM/RTM minus SMOS Tb

GMAO-18Sep12 – 26 / 28

Ascending H-pol Ascending V-pol

−10 −5 0 5 10 [K]

� 6-angle average (32.5, . . . , 57.5)o

� using full-pol SMOS data only (early 2010: switch dual-full pol)

Clear seasonal biases remaining,

partly due to CLSM/RTM Tb biases, partly due to SMOS Tb biases

For example: ascending V-pol in North America:

� simulated TbV cannot exceed CLSM Ts, irrespective of the parameters

� ascending SMOS sees military radar, remaining RFI contamination may be present

Hovm oller Plots: CLSM/RTM minus SMOS Tb

GMAO-18Sep12 – 27 / 28

Ascending H-pol Ascending V-pol

Descending H-pol Descending V-pol

−10 −5 0 5 10 [K]→ Warm/cold bias in asc/desc

SMOS instrument calibration

Conclusions

Soil Moisture andMicrowaveRadiances

Calibration

BrightnessTemperatures

RTM Parameters

Remaining Biases

Conclusions

GMAO-18Sep12 – 28 / 28

� Biases between SMOS and CLSM/RTM Tb, when using literature

parameters

� RTM calibration/validation (split sample 2010 / 2011)

� multi-angular; multi-polarization; multi-orbit; local

� objective function: minimize

� differences in long-term mean;

� differences in long-term standard deviation;

� deviations from prior (literature) parameter values.

� long-term unbiased Tb, seasonal to diurnal biases remaining

� realistic effective RTM-parameter patterns

� Future

� new CLSM climatology will need new calibration

� Tb assimilation, SMAP L4 SM product

Questions? In review, Journal of Hydrometeorology; Gabrielle.DeLannoy@nasa.gov