1. T. Quaife, P. Lewis, M. Williams, M. Disney and M. De Kauwe.
Assimilating Earth Observation Data into a Vegetation Model using an Ensemble Kalman Filter
T. Quaife, P. Lewis, M. Williams,
M. Disney and M. De Kauwe.

2. DALEC Cf
Csom/cwd
Clit Cr Cw
GPP Af Ar Aw Ra Lf Lr Rh D

3. DALEC NEP

4. Ensemble Kalman Filter
Aa = A + A′A′THT(HA′A′THT + Re)-1(D - HA)
H = observation operator
A = state vector ensemble
A′ = state vector ensemble – mean state vector
D = observation ensemble
Re = observation error covariance matrix

5. Strategies for assimilation
Assimilate EO products
Probably noisey
Linear observation operator
Assimilate reflectance
Errors more easily characterised
Non linear observation operator

6. The “Twin experiment”
Use a more complex model to represent the “truth”
Simulate observations from truth model
Asses ability of DALEC/EnKF to make accurate predictions

7. SDGVM Max Evaporation Soil Moisture Litter Transpiration LAI
Soil C & N
NPP
H2O30
Phenology
Hydrology
Century
Growth

8. DALEC & SDGVM NEP

9. NEP - Assimilating modelled 30 day LAI

10. NEP - Assimilating 30 day FASIR LAI

11. LAI – no assimilation

12. LAI – SDGVM assimilation

13. LAI – FASIR assimilation

14. EnKF – augmented analysis
Aa = A + A′Â′TĤT(ĤÂ′Â′TĤT + Re)-1(D - ĤÂ)
Ĥ = augmented observation operator
 = augmented state vector ensemble
 = h( A )

15. Non-linear observation operator
NDVI = a0 × ( 1 – e( -a1 × LAI ) )
Regressing the FASIR LAI against the FASIR NDVI:
a0 = 0.678
a1 = 0.982

16. NEP - Assimilating FASIR NDVI

17. LAI - Assimilating FASIR NDVI

18. Conclusions Test exercise very promising
Demonstrates ability to use non-linear observation operators
Next step is to couple a full CRM to DALEC to enable assimilation of reflectance data
Accurate characterisation of errors is critical
Models very different
Improve DALEC
Seek other data