Estimating SMOS error structure using triple collocation.ppt

Estimating SMOS error structure using triple

collocation

Delphine Leroux, CESBIO, FranceYann Kerr, CESBIO, FrancePhilippe Richaume, CESBIO, France

1

Soil moisture products at global scale

2

AMSR-E (NSIDC)

ERS-ASCAT

(TU Wien)

Model output

(ECMWF)

AMSR-E

(VUA)

TMI (VUA)

SSM/I (VUA)

Aquarius

SMAP

How to evaluate SMOS ???

SMOS?

Inter comparison between SMOS soil moisture and …

o Ground measurements (point scale)

o Other global products (point scale)

3

Statistics -> triple collocationo Global scale ?

Structure

1. Triple Collocation method-> Theory and requirements

2. Chosen datasets-> Characteristics and differences

3. Global maps of relative errors-> Maps of errors-> Maps of bias and scale factors

4

Triple Collocation – theory (Caires et al., 2003)

Starting equation

Taking the anomalies

5

Final equation

Maps of the std of the errors

Maps of the bias Maps of the scale factors

1) Triple Collocation

Theory Requirements

r: bias s: scale factor ε: error

Triple Collocation - requirements

oStrong assumptions : Mutually independent errors

(ε) No systematic bias between

the datasets

o Requirements : 100 common dates

(Scipal et al., IGARSS 2010)

o Results : Relative errors

6

-> choose properly the 3 datasets-> TC applied to the anomalies and not to the variables directly

-> including the 6 closest grid nodes

1) Triple Collocation

Theory Requirements

Datasets

Frequency (GHz)

Incidence angle (°)

Instrument resolution (km)

Crossing time (A/D)

Grid resolution (km)

SMOS 1.4 0-55 40 6am / 6pm

15

AMSR-E 6.9 – 10.7 - …

55 57-6.25 1:30pm/ 1:30am

25

7

AMSR-E soil moisture derived with the VUA algorithm (Vrije University of Amsterdam)

ECMWF product from SMOS Level 2 product (at SMOS resolution and crossing time)

2) Datasets Chosen datasets Number of triplets

Number of triplets for 2010

82) Datasets Chosen datasets Number of triplets

Difficulties for regions with mountains, forests, wetlands, …

Std of SMOS errors

93) Global maps of …

relative errors bias scaling factors

Good results in North America, North Africa, Middle East, Australia.Land contamination in Asia (Richaume et al., RAQRS, 2010).

Std of AMSR-E(VUA) errors

103) Global maps of …

relative errors bias scaling factors

Good results in the same areas as SMOS.

Std of ECMWF errors

113) Global maps of …

relative errors bias scaling factors

Comparison over continents

123) Global maps of …

relative errors bias scaling factors

RELATIVE ERRORS!

SMOS is often between or close to the two values except in Asia

133) Global maps of …

relative errors bias scaling factors

Bias : AMSR-E(VUA) - SMOS

Very high bias for high latitudes (mainly due to the vegetation)Mean bias around 0.1

143) Global maps of …

relative errors bias scaling factors

Bias : ECMWF - SMOS

High bias for high latitudes but more homogeneousMean bias around 0.2-0.3

Scale factor AMSR-E(VUA)

153) Global maps of …

relative errors bias scaling factors

Scale >1 higher dynamic than SMOSScale <1 lower dynamic than SMOS

Scale factor ECMWF

163) Global maps of …

relative errors bias scaling factors

Unlike the bias maps, there is no obvious structure for the scale factor

Conclusions

o As part of the validation process, triple collocation compares 3 different datasets at a global scale : SMOS, AMSR-E/VUA and ECMWF

o SMOS and AMSR-E/VUA have the same performance areas, but ECMWF and VUA give the best results

o SMOS algorithm is still improving and it can be considered as a good start

o Further work : apply triple collocation to other triplets (SMOS-NSIDC-ASCAT, etc…) and apply it with 2011 data

17

Thank you for your attention

18

Any questions ?