Slide 1 Soil schemes: modeling&assimilation - G. Balsamo Slide 1 Land Surface: Modelling & Data...

Soil schemes: modeling&assimilation - G. Balsamo

Slide 1

Slide 1

Land Surface:Modelling & Data Assimilation

Gianpaolo Balsamo

European Centre for Medium-Range Weather Forecasts (ECMWF)

ARPA-SIM, Bologna, 18 February 2008


Slide 2

Slide 2

OutlineIntroduction

- The Earth Integrated Forecast System

- The role of Land Surface (LS)

- The role of data assimilation

- LS observational network

Modelling the land surface - Motivations

- Simplification vs. Realism in LS parameterizations

- TESSEL scheme

Analysing the land surface- Motivations

- Current practice in NWP (OI, EKF)

- New methods (simplified 2D-VAR, EnKF)

Modelling & data assimilation synergy- The example of soil hydrology (HTESSEL)

- The Cal/Val benchmark/strategy (field-site to global simulations + Data Assimilation)

Conclusions and Perspectives


Slide 3

Slide 3

Earth energy cascadeThe sun emits 4 x 1026 W

the Earth intercepts 1.37 kW/m2

This energy is distributed between- Direct reflection (~30%)

- Conversion to heat, mostly by surface absorption (~43%), re-radiated in the infrared

- Evaporation, Precipitation, Runoff (~22%)

- Rest of the processes (~5%, Winds, Waves, Convection, Currents, Photosynthesis, Organic decay, tides, … )

Robinson & Henderson-Sellers, 1999


Slide 4

Slide 4

Earth water cycle

Atmosphere recycling time scales associated with land reservoir

-Precipitation 4.5/107 = 15 days

-Evaporation 4.5/71 = 23 daysEvaporation

71

TerrestrialAtmosphere

Land

4.5

Rain

107

[•] = 1015 kg = teratons

[•] = 1015 kg yr-1

Runoff

36

Chahine, 1992


Slide 5

Slide 5

Role of land surfaceAtmospheric general circulation models need boundary

conditions for the enthalpy, moisture (and momentum) equations: Fluxes of energy, water at the surface.

Water budget

E

P

Y

0.9 mmd-12.21.4

H

LE RT

RS

27 40 65 134 Wm-2

Energy budget

ERA40 land-averaged values 1958-2001

Carbon budget (natural)

NEE


Slide 6

Slide 6

Role of land surface (2)

Numerical Weather Prediction models need to provide near surface weather parameters (temperature, dew point, wind, low level cloudiness) to their customers.

ECMWF model(s) and resolutions

Length Horizontal Vertical Remarks

resolution levels

- Deterministic 10 d T799 (25 km) L91 00+12 UTC

- Ensemble prediction 15 d T399 (50 km) L62 2x(50+1)

- Monthly forecast 1 m T159 (125 km) L62 (Ocean coupled)

- Seasonal forecast 6 m T95 (200 km) L40 (Ocean coupled)

- Assimilation physics 12 h T255(80 km)/ L91 T95(200 km)


Slide 7

Slide 7

How to initialize the Land Surface: Current Practice in NWP centers

Use of 2m observations

with OI Analysis

(ECMWF, Météo-France,

HIRLAM, MSC)

or Simplified

Kalman Filter(DWD)

OFF-LINE Land surface

(NLDAS, GLDAS, UK MetO

Meteo-France)

Who, When and Why ?Who, When and Why ?

•Coiffier et al. 1987 (Use of 2m for land surface)•Mahfouf 1991 (OI / Variational formulation of the land surface analysis)•the operational application comes few years later (94-95 at ECMWF, 99 Météo-France.)


Slide 8

Slide 8

L-band TbL-band Tb C-band TbC-band Tb C-band scat. IR TsC-band scat. IR Ts

EVOLUTION OF LAND SURFACE DATA ASSIMILATION SYSTEMS

T/H 2mT/H 2m

hourly 6-hourly


Slide 9

Slide 9

L-band TbL-band Tb C-band TbC-band Tb C-band scat. IR TsC-band scat. IR Ts

OBSERVATIONS FOR SOIL MOISTURE ANALYSIS

T/H 2mT/H 2m

INFORMATIVITY on SOIL MOISTURE

2008/2012 AVAILABILITY now

+ Large Information content

+ Global Coverage

+ Reduced Atmospheric Contrib.

-Not Available ‘till 2009

+ Large Information content

+ Global Coverage

+ Reduced Atmospheric Contrib.

-Not Available ‘till 2009

+ Global coverage

+ Relatively reduced Atmospheric contrib.

- RFI

- Vegetation masking VCW>1kg/m2

+ Global coverage

+ Relatively reduced Atmospheric contrib.

- RFI

- Vegetation masking VCW>1kg/m2

+ Large coverage

- Cloud Masking

- Model Bias

+ Large coverage

- Cloud Masking

- Model Bias

+ Wide validation

-Coverage

-Variable Information Content

+ Wide validation

-Coverage

-Variable Information Content


Slide 10

Slide 10

OutlineIntroduction







- TESSEL scheme





- The Cal/Val benchmark/strategy (field-site to global simulations + Data Assimilation)



Slide 11

Slide 11

History of ECMWF 2m T errors


Slide 12

Slide 12

The challenges for Land Surface Modeling

Capture natural diversity of land surfaces (heterogeneity) via a simple set of equations

Focus on elements which affects more directly weather and climate (i.e. soil moisture, snow cover).


Slide 13

Slide 13

TESSEL scheme

High and lowvegetationtreated separately

Variable root depth

Revised canopyresistances,including airhumidity stress onforest

No rootextraction ordeep percolationin frozen soils

New treatmentof snow underhigh vegetation

+ 2 tiles (ocean & sea-ice)

Tiled ECMWF Scheme for Surface Exchanges over Land


Slide 14

Slide 14

Vegetation Type (H and L) at T799GLCC(1998)

6 dominant high veg. type (TVH)

9 dominant low veg. type (TVL)

Used to assign:

root-distributionLAI and Rs_minroughness lengths

by a look-up table


Slide 15

Slide 15

Vegetation Cover (H and L) at T799GLCC(1998)

Note: the cover CVH and CVL are fraction of land use by TVH and TVL and their sum is equal the unity

Used to calculate:

bare ground fraction as

Bare_frac=1-ΣCV(TVi)*RCOV(TVi)

with RCOV provided by a look-up table


Slide 16

Slide 16

OutlineIntroduction







- TESSEL scheme





- The Cal/Val benchmark/strategy (field-site to glo



Slide 17

Slide 17

Case study: Europe, May-June1994 (1)

ECMWF

German (DWD)

Day 2 forecasts


Slide 18

Slide 18



Slide 19

Slide 19



Slide 20

Slide 20

Near surface atmospheric errors

In the French forecast model (~10km) local soil moisture patterns anomalies at time t0 are shown to correlate well with large 2m temperature forecast errors (2-days later)

Balsamo, 2003dry soil

wet soil

G nR H LE


Slide 21

Slide 21

Link between soil moisture and atmosphere

The main interaction of soil moisture and atmosphere is due to evaporation and vegetation transpiration processes.

Ws bare ground

Ws bare ground

Wp vegetation

Wp vegetation

Eg Etr

E

0 <SWI< 1

Ws

Wp


Slide 22

Slide 22

Optimum Interpolation land surface analysis(oper. surface analysis at Météo-France/MSC/ECMWF…) Mahfouf 1991, Bouttier 1993, Giard and Bazile 2000, Mahfouf et al. 2003, Belair et al 2003

Sequential analysis (every 6h)Correction of surface parameters (Ts, Tp, Ws, Wp) using 2m increments between

analysed and forecasted values

Optimum Interpolation of T2m and RH2m using SYNOP observations interpolated

at the model grid-point (by a 2m analysis)

T2m

t

Wp

t

RH2m

t

6-h 12-h 18-h 0-h

Tuning of the OI statistics and regressions and accuracy of 2m analyses are key components

T2m = T2ma - T2m

f RH2m = RH2ma - RH2m

f

Tsa - Ts

fT2m

Tpa - Tp

fT2m / 2Ws

a - WsfWsT T2m + WsRH RH2m

Wpa - Wp

fWpT T2m + WpRH RH2m

Wp/sT/RH = f (t, veg, LAI/Rsmin, texture, atm.cds.)


Slide 23

Slide 23

Variational surface analysisMahfouf (1991), Callies et al. (1998), Rhodin et al. (1999),

Bouyssel et al. (2000), Hess (2001), Seuffert et al. (2004), Balsamo et al. (2004)

Formalism:

x is the control variables vectory is the observation vectorH is the observation operator

Continuous analysis

T2m

t

Wp

t

RH2m

t

6-h 12-h 18-h 0-h

The analysis is obtained by the minimization of the cost function J(x)

B is the background error covariance matrix

R is the observation error covariance matrix

= ½ (x – xb) T B-1 (x – xb) + ½(y – H(x))T R-1 (y – H(x))

J(x) = J b(x) + J o(x)

Advantages: Easier assim. asynop. obs.Extension on longer assim. Window (24-h)


Slide 24

Slide 24

(Bouyssel et al. 2000)

The shape of the cost function J for full 2D-VAR

J=f(Ws,Wp,Ts,Tp)


Slide 25

Slide 25

How the simplified 2D-VAR method works

Wp W’p=Wp + Wp

p

pp

p

p

p

T

WYWY

WYWY

)(

-1)(

)2(

)1(

...H

)( -1)()1(2)1( ,,...,,)( pPb YYY YxHy

From a perturbation of the initial total soil moisture Wp applied on each model land grid-point.

Y (i) = YG (i) - YG’ (i) Guess G

Guess G’

Y

t

t=0 1 2 … p

Y (i) = YG (i) - YO (i)

Wp

Y=(T2m ,Tb,Ts )


Slide 26

Slide 26

The 2D hypothesis is validated with simulated observations on a real situation

From a prescribed initial error From a prescribed initial error WpWp

The 6-h forecast errors The 6-h forecast errors

on on TT2m2m and and RHRH2m2m

Analysis error Analysis error

2D hypothesis


Slide 27

Slide 27

Convergence of 2D-VAR analysis

Simulated observations (consistent to SWI=0.5) are assimilated over a 10-day period

A 24-h 2D-VAR analysis with optimised settings

Real observations experiments are then considered


Slide 28

Slide 28

Soil Moisture produced by the ELDAS project

Habets et al. (2003)

The same comparison is produced for the ELDAS soil moisture obtained with the ARPEGE model. An improved match of soil moisture patterns and gradients is obtained on the SAFRAN-ISBA-MODCOU validation area.

ELDAS cycle


Slide 29

Slide 29

June July

(CNES, 2003) Images SPOT/VEGETATION

Variation of NDVI 2003 within respect to 2002

+positive Index 2003/2002 +negative

Variation of SWI at 30 June 2003 compared to 30 June 2000 (ELDAS)

30 June 2003 (exp. 2D-Var + Ecoclimap (Masson et al. 2003) after 2-month cycle)

Drought of summer 2003: Comparison of soil moisture and NDVI anomaly


Slide 30

Slide 30

C-band TbC-band Tb

IR TIR Tskinskin

How Microwave and Infra-red Radiances may be informative on soil water content?

Sounding soil depth Frequency Wavelength Atmospheric absorption

~5 cm 1.4 GHz 21 cm Negligible

~1cm 6.9 GHz 5 cm Low (except rainy area)

superficial (27.7 THz) 10.8 μm Important – clear sky only

L-band TbL-band Tb

Ws

Wp

Ws

Wp

C-band TbC-band TbL-band TbL-band Tb

Soil moisture modifies soil dielectric const. emissivity ε

IR TsIR Ts

Tb = ε Ts

Soil moisture affects Skin temperature and heating rate


Slide 31

Slide 31

G G’ Obs.

Tb, H

t

Wp

t

Tb, V

t

0-h 1-h 2-h 3-h … ………… 23-h 0-h 0-h 1-h 2-h 3-h … …… 23-h 0-h

L-bandL-band & C-BandC-Band TB

Every hour (except RFI in C-band)

IR TskinIR Tskin (or HR)

TsIR

Wp

t

t

Morning

(except Clouds)


Slide 32

Slide 32

PHASE I

THE OFF-LINE LSS ISBA is driven by near surface

atmospheric forcing to obtain the LAND SURFACE STATE

PHASE I

THE OFF-LINE LSS ISBA is driven by near surface

atmospheric forcing to obtain the LAND SURFACE STATE

SMOS (&SMAP) L-BAND simulated TBH,V

PHASE II

THE MICROWAVE RT Model LSMEM is used to compute the brightness temperature at 1.4GHz

PHASE II

THE MICROWAVE RT Model LSMEM is used to compute the brightness temperature at 1.4GHz

TB,h from microwave RT model (Drusch et al. 1999, 2001)

PHASE III

SPATIAL and TEMPORAL location of the simulated TB

PHASE III

SPATIAL and TEMPORAL location of the simulated TB t-1/2-h

t+1/2-h

t

Superficial soil moisture from ISBA


Slide 33

Slide 33

OSSE: Assimilation of Simulated Brightness Temperature

The assimilation of HYDROS simulated H and V polarization L-band brightness temperatureis investigated in a 10-day DA experiment using GSWP-II forcing to create a reference landsurface state (from 1-y ISBA model run).

The 2D-VAR analysis is initialized with a background model error of 10% (SWI) and the observations error is set to 3 K.

The analysis plays for about 50% in the convergence towards the ISBA-GSWP-II reference (starting from a medium soil moisture Wp=0.5(Wfc-Wwl)

97.5

0

10

20

30

40

50

60

70

80

90

100

1-Jul-95 2-Jul-95 3-Jul-95 4-Jul-95 5-Jul-95 6-Jul-95 7-Jul-95 8-Jul-95 9-Jul-95 10-Jul-95

date

0.0

5.0

10.0

15.0

20.0

25.0


date

0

2

4

6

8

10

12

14

16

18

20

N points Assimilation Cycle (mean value) RMSE (Ref-Assim) Reference (mean value)

Soil moisture Error (% vol.)Soil moisture Error (% vol.)Analysis – ReferenceAnalysis – Reference


Slide 34

Slide 34

Soil moisture Error (% vol.)Soil moisture Error (% vol.)Analysis – ReferenceAnalysis – Reference

OSSE: Assimilation of HYDROS Simulated Brightness Temperature

The assimilation of HYDROS simulated H and V polarization L-band brightness temperature is investigated in a 10-day DA experiment using GSWP-II forcing to create a reference land surface state (from 1-y ISBA model run).

The 2D-VAR analysis is initialized with a background model error of 10% (SWI) and the observations error is set to 3 K.

The analysis plays for about 50% in the convergence towards the ISBA-GSWP-II reference (starting from a medium soil moisture Wp=0.5(Wfc-Wwl)

97.5

0

10

20

30

40

50

60

70

80

90

100


date

0.0

5.0

10.0

15.0

20.0

25.0


date

0

2

4

6

8

10

12

14

16

18

20

N points Assimilation Cycle (mean value) RMSE (Ref-Assim) Reference (mean value)


Slide 35

Slide 35

Extended / Ensemble Kalman Filter

Extended / Ensemble Kalman Filter Simplified VAR/EKF MethodsSimplified VAR/EKF Methods

VariationalVariational

Optimal Estimation Theory

Data Assimilation Techniques applied for Land Surface

( ) 1111 −−−− += RHHRHBK TT

0)( →∇ xJ

( ) ( ) xHxxx δδ +=+ HH

( ))( bba H xyKxx −+=

Optimum InterpolationOptimum Interpolation

(...)f==

K

QMMAB +=+ T1 tt

H(x)

H(x)


Slide 36

Slide 36

Off-line vs. Atmospheric Coupled LDAS Balsamo et al. 2007

Within CaLDAS an Off-line version of GEM-15km is available, MEC-15km Same dynamical/physical core GEM-15km

In the Off-line version the forcing is applied at ~50 m (28th level of GEM)

The comparison is proper (same innovations, same atmospheric model trajectory).

A SBL (Delage 1997) is implemented and allows to maintain and interactive layer

A multi-observation OSSE using the simplified 2D-VAR scheme is run.

Diagnostics from Jacobians and the information content theory confirm a good approximation over North America (GEM-core domain) with a reduction of noisy signal which seems beneficial (i.e. no convection). Results are still preliminary (1 day considered) and further tests are in progress.


Slide 37

Slide 37

OutlineIntroduction







- TESSEL scheme





- The Cal/Val benchmark/strategy (field-site to Global simulation + Data Assimilation)



Slide 38

Slide 38

TESSEL land surface scheme: + and -

High and lowvegetationtreated separately

Variable root depth

Revised canopyresistances,including airhumidity stress onforest

Inhibited root extraction,or drainagein frozen soils

New treatmentof snow underhigh vegetation

+ 2 tiles (ocean & sea-ice)

Tiled ECMWF Scheme for Surface Exchanges over Land

A single soil textureglobally, excessive drainage

Too little surfacerunoff

Too early snowmelting


Slide 39

Slide 39

Surface Water reservoirs (ERA-40)

DA increments redistribute water and constraint near-surface errors

snow Soilmoisture

Early snowmelting

moisture deficit

anticipate moisture supply


Slide 40

Slide 40

Cold processes I: Snow DA increments

ERA-40 ERA-Interim1992, daily SWE increments


Slide 41

Slide 41

HTESSEL scheme

R1 > R2

D1 < D2

P1 = P2

_1 > _ 2

R2

Fine texture Coarse texture

Hydrology-TESSEL

- Global Soil Map (FAO)

- New formulation of Hydraulic properties

- VIC surface runoff


Slide 42

Slide 42

Soil Type at T799FAO(2003)

6 dominant soil type

Used to assign:hydraulic properties (drainage and surf. runoff)field capacity & wilting point for SM analysis


Slide 43

Slide 43

A revised hydrology scheme (H-TESSEL)

A spatially variable hydrology scheme is being tested following Van den Hurk and Viterbo 2003

Use of a the Digital Soil Map of World (DSMW) 2003

Infiltration based on Van Genuchten 1980 and Surface runoff generation based on Dümenil and Todini 1992

Van den Hurk and Viterbo 2003


Slide 44

Slide 44

Field Capacity and Permanent Wilting Point

Soil DiffusivitySoil ConductivityTESSELTESSEL

TESSELSoil PWP [m³/m³] FC [m³/m³]

1 Loamy 0.171 0.323

HTESSEL Soil PWP [m³/m³] FC [m³/m³]

1 Coarse 0.059 0.242

2 Medium 0.151 0.346

3 Medium-fine 0.133 0.382

4 Fine 0.279 0.448

5 Very fine 0.335 0.541

6 Organic 0.267 0.662


Slide 45

Slide 45

The soil texture classification database

Dominant soil type from FAO2003 (at native resolution of ~ 10 km)

█coarse █medium █med-fine █fine █very-fine █organic

Soil texture percentage occupation as a function of resolution

0%

10%

20%

30%

40%

50%

60%

70%

coarse medium medium-fine fine very-fine organic

input (10 km) T21 T42 T159 T799

The interpolation to model grid is donewithin the IFS by the prepdata (interporoutine) preserving the dominant texture type at various resolution (T21-T799). Important for “upscalability”

FAO 2003 from Freddy.Nachtergaele, after a survey of the available datasets.


Slide 46

Slide 46

The orography runoff generation

Also the standard deviation of orography is scaling with resolution (especially T159-799).

fraction runoff s of the grid-point area S.

b

S

S

w

w

S

s⎟⎟⎠

⎞⎜⎜⎝

⎛−−=

max

11

Runoff as a function of orography (b is based on standard deviation of orography)

10mm/hUp to~30% Surfacerunoff in complex orography

minmax

min

ss

ssb

−−

=;


Slide 47

Slide 47

Verification Strategy for the new Hydrology

Field sites

(Offline)Catchment (Offline)

Global (Offline)

Coupled GCM

Coupled GCM + DA


Slide 48

Slide 48

Field site verification of HTESSEL

observed atmospheric forcing

observed SM/LE/H

observed Tb

Ancillary data as in operational (no local readjustment)


Slide 49

Slide 49

SEBEX (sandy soil)

HTESSEL show a consistent improvement of soil moisture and evaporation with respect to TESSEL

Savannah, desert climate


Slide 50

Slide 50

BERMS (Boreal Forest)

HTESSEL show a consistent improvement of Top 1m soil moisture with respect to TESSEL and a better represented interannual variability

Forest, snow dominated site


Slide 51

Slide 51

Insitu Network

Agrometeorologic Network

-Russia: 63 stations

-Ukraine: 96 stations

Soil Property Measurements

-volumetric density, total water holding capacity, field capacity, wilting level

-sampled at several cross-sections in 3 fields for each station

-evaluated periodically

By Klaus Scipal


Slide 53

Slide 53

Soil Properties - Ukraine

TESSEL

HTESSEL

By Klaus Scipal


Slide 54

Slide 54

Basin-scale verification of HTESSEL

observed atmospheric forcing (GSWP2)

observed total Runoff (GRDC)

ERA-40 atmospheric TCWV variations combined with GRDC runoff to obtain Terrestrial Water Storage change

dS/dt = (P - E) - R

Ancillary assigned by closest IFS grid resolution (T255 for 1x1 regular lat lon grid).


Slide 55

Slide 55

Quantitative estimate of Global Water budget:Dataset for Mid-latitude River Basins

Seneviratne et al. 2004, J. Climate, 17 (11), 2039-2057Hirschi et al. 2006, J. Hydrometeorology, 7(1), 39-60

“BSWB”

http://iacweb.ethz.ch/data/water_balance/

Courtesy of Sonia Seneviratne


Slide 56

Slide 56

Case Study: Mississippi & Illinois

Water-balance Estimates

Observations(soil moisture+groundwater+snow)

Seneviratne et al. 2004, J. Climate, 17 (11), 2039-2057

corr=0.8, r2=0.71

Courtesy of Sonia Seneviratne


Slide 57

Slide 57

European catchments: Validation using ERA-40 derived BSWB (Basin Scale Water Budgets)

HTESSEL increases the storage w.r.t. TESSEL, closer to Annual variations estimated by the BSWB dataset

TESSEL is better offline than in ERA-40 due to P6h bias over Europe

DA works efficiently to correct soil moisture by adding water and preserving evaporation

SM

ET

dS

P


Slide 58

Slide 58

Total Runoff per basin

20. tura21. ob22. volga23. don24. dnepr25. neva26. baltic27. elbe28. odra29. wisla30. danube31. northeast_europe32. po33. rhine34. weser35. ebro36. garonne37. rhone38. loire39. seine40. france41. central_europe

1. yukon2. mackenzie3. columbia4. arkansas5. mississippi6. missouri7. ohio8. murray_darling9. changjiang10. syrdarya11. selenga12. irtish13. amudarya14. amur15. lena16. yenisei17. podkamennaya_tunguska18. vitim19. Tom

BIAS

RMSE

Soil Moisture dominant

Snow dominant

TESSELHTESSEL


Slide 59

Slide 59

Interaction of early snow

melt and runoff is

particularly evident

In Siberian basins

Snow melt and runoff

HTESSEL better reproducethe runoff peak

Early snow melt displacethe peak by 1 month

RMSE in runoff increases due to double penalty

R

P

E


Slide 60

Slide 60

Rhone Agg experiment: Daily runoff8 8 km and 1 runs for a 4-year period (1 August 1985 – 1 August 1989)

By Bart van den Hurk (KNMI)

TESSEL showlow-pass-filterbehavior

HTESSEL morealike obs


Slide 61

Slide 61

Global-scale verification of HTESSEL

At global scale land surface can be compared to climatological datasets

Result obtained offline should be also verified in climate run experiments where there is a feedback from the atmosphere.

a better dS/dt and R --> improved (P - E), dW/dt, ..

Data assimilation experiments although expensive offer an excellent verification looking at land surface data assimilation increments.

dS/dt = (P - E) - R + dA

a better model --> reduced dA increments


Slide 62

Slide 62

Total runoff (Qualitative) GSWP2 model output vs.GRDC-composite estimate 1986-1995)

H-TESSEL

TESSEL

GRDC


Slide 63

Slide 63

Climate runs (1-year run coupled): surface T2m compared with analysis

H-TESSELTESSEL


Slide 64

Slide 64

Reduction [%] of RMSE against all the climatological datasets used in CLIMPLOT for the 4ens 2000/09-2001/08 period

Climate runs (extensive validation): Scores of HTESSELHow to quantify climate improvements?

Relying on datasets

- need to be aware we are not scoring against observations

Assuming RMS can be reduced “asymptotically”

- (RMS_new < RMS_old)

If we normalized RMS reduction (with RMS_old)

- HTESSEL vs. TESSELin 32R3


Slide 65

Slide 65

Long DA cycle with HTESSEL

A long DA experiment at T159L91 is done with TESSEL and HTESSEL (01/04-01/11/2906)

Differences in soil moisture analysis increments can be interpret as improvements of the slow model component

- |ΔSM(HTESSEL)| > |ΔSM(TESSEL)|

- |ΔSM(HTESSEL)| < |ΔSM(TESSEL)|


Slide 66

Slide 66

OutlineIntroduction







- TESSEL scheme





- The Cal/Val benchmark/strategy (field-site to glo



Slide 67

Slide 67

Summary and conclusionLand Surface Modelling and Data Assimilation are closely related

A simplified variational method (Balsamo et al. 2004) which make use the “model” as observation operator to establish the link between soil moisture and the observation

The good news is that LDAS can work offline (Balsamo et al. 2007)

Land surface analysis should not compensate for model bias

An improved hydrology (HTESSEL) shows a better match to field site observations of soil moisture while preserving good performance of TESSEL in LE/H.

GSWP2 runs 1986-1995 show that HTESSEL addresses the issue of storage.

Runoff bias is reduced and Runoff timing is improved (when not combined with snow early melting)

Long DA runs: reduction of sm increments at mid-latitudes.


Slide 68

Slide 68

Near Future model&DA developmentsImprove treatment of land surface water

- Continuous assessment of LSM hydrology (e.g. GLACE-2)

- 2008 SMOS/ASCAT year (lakes)

Vegetation seasonal cycle- Introduction of monthly LAI

- Prepare for CTESSEL (GEOLAND-2)

Snow “patches”- SnowMIP2 results now available allow to tune the snow scheme

(adding simple parameterizations as water refreeze.

Offline suite- GSWP2 experimental framework can be extended to E4 and OD

- Link with land surface data assimilation

Data Assimilation via simplified 2D-VAR/EKF- ASCAT soil moisture assimilation

- L-band brightness temperature (SMOS 2009)


Slide 69

Slide 69

Acknowledgements

Thanks to:

- Anton Beljaars , Pedro Viterbo, Martin Hirschi (ETH), Bart van den Hurk (KNMI), Alan Betts for HTESSEL

- Matthias Drusch and Patricia Derosnay for the L-band feedback and DA discussions

- Klaus Scipal for Soil moisture verification (from Global Soil Moisture Data Bank)

- Seasonal/monthly forecast teams (Tim Stockdale, Paco Doblas-Reyes, Frederic Vitart, Laura Ferranti, …)


Slide 70

Slide 70

Compatibility and information to users

Web-page inECMWF site (Paul, Umberto)

Soil moisturerescale scriptin IFS(Jan, Joerg, Nils)


Slide 71

Slide 71

Ecoclimap ECOCLIMAP merge climatic zone and Land Use datasets making

use of AVHRRNDVI to identify relevant biomes.

The NDVI seasonality is used to obtain derived properties (eg. LAI)

January

June


Slide 72

Slide 72

Influence MATRIX (HK)Cardinali et al. 2003 ; Chapnik et al. 2004

This diagnostic tool allows to separate the information content of each observation in the analysis

xa=xb+K(y-Hxb) analysis equation

with A=(I-KH)B analysis error covariance Matrix

Hxa=(I-HK)Hxb+HKy Analysis in observation space

if we calculate the derivative ∂ … / ∂y0 we have :

Tr ( ∂ Hxa / ∂y0) = Tr (HK) Amount of information extracted from observation y0

if H is linear = Tr ((B-A) B-1) Analysis Error Reduction

H


Slide 73

Slide 73

L-band TbL-band TbC-band TbC-band Tb

IR TsIR Ts

T/H 2mT/H 2m

Contribution of each observation source in the Soil Moisture Analysis (N. America): 2D-VAR on 24-h time window using hourly distributed observations (date 01/07/1995)

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

Observation Source

N. of observations

0.0

5.0

10.0

15.0

20.0

25.0

30.0

35.0

40.0

45.0

50.0

% of contribution [ 100 (B-A) B-1 ]

N. Analysis Influence (%)

N. 7396 7392 7904 7874 3301 8757 8757

Analysis Inf luence (%) 20.8 15.1 11.0 7.4 9.3 6.1 7.5

sigma Obs. 3.0 3.0 3.0 3.0 3.0 2.0 0.002

Tb (L-Band)

H

Tb (L-Band)

V

Tb (C-Band)

H

Tb (C-Band)

V

T skin (IR)

AM+PMT2m (6-h) Q2m (6-h)


Slide 74

Slide 74

Improvements in the 3 latest IFS cycles

32R1 vs. 31R2 32R2 vs. 32R1 32R3 vs. 32R2

McRad: RRTM SW, No Physics change New convection Modis Alb, McICA V.Diff, HTESSEL

Date post:	20-Dec-2015
Category:	Documents
View:	220 times
Download:	0 times

Slide 1 Soil schemes: modeling&assimilation - G. Balsamo Slide 1 Land Surface: Modelling & Data...

Documents